STP1290 Eb.1415051 1 PDF

Copyright by ASTM Int'l (all rights reserved); Mon Jan 16 18:33:32 EST 2012
Downloaded/printed by
(PDVSA Los Teques) pursuant to License Agreement. No further reproductions authorized.
STP 1290
Fiber, Matrix, and

Interface Properties
Christopher J. Spragg and Lawrence T. Drzal, Editors
ASTM STP Publication Code Number (PCN):

04-012900-33
ASTM
100 Barr Harbor Drive
West Conshohocken, PA 19428-2959
Printed in the U.S.A.
Library of Congress Cataloging-in-Publication Data
Fiber, matrix, and interface properties / Christopher J. Spragg and

Lawrence T. Drzal, editors.
p. cm.--(STP ; 1290)
"ASTM Publication Code Number (PCN) : 04-012900-33."
"The Symposium on Fiber, matrix, and interface properties was held
14 November, 1994 in Phonenix, Arizona. ASTM Committee D-30 on High
Modulus Fibers and Their Composites sponsored the symposium"--
Foreword.
Includes bibliographical references and index.
ISBN 0-8031-2046-X
1. Composite materials--Congresses. 2. Polymeric composites--
Congresses. 3. Fibrous composites--Congresses. I. Spragg,
Christopher J., 1959- . II. Drzal, Lawrence T. Ill. ASTM Committee
D-30 on High Modulus Fibers and Their Composites. IV. Symposium on
Fiber, matrix, and interface properties (1994 : Phoenix, Ariz.)
V. Series: ASTM special technical publication ; 1290.
TA418.9.C6F45 1996
620.1' 18---dc20 96-22167
CIP
Copyright 9 1996 AMERICAN SOCIETY FOR TESTING AND MATERIALS, West Consho-
hocken, PA. All rights reserved. This material may not be reproduced or copied, in
whole or in part, in any printed, mechanical, electronic, film, or other distribution and storage
media, without the written consent of the publisher.
Photocopy Rights
Authorization to photocopy items for internal, personal or educational classroom use, or

the internal, personal, or educational classroom use of specific clients, is granted by
the American Society for Testing and Materials (ASTM) provided that the appropriate fee
is paid to the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923;
Tel: 508-750-8400; online: http://www.copyright.com/.
Peer Review Policy
Each paper published in this volume was evaluated by three peer reviewers. The authors
addressed all of the reviewers' comments to the satisfaction of both the technical
editor(s) and the ASTM Committee on Publications.
The quality of the papers in this publication reflects not only the obvious efforts of the
authors and the technical editor(s), but also the work of these peer reviewers. The
ASTM Committee on Publications acknowledges with appreciation their dedication and
contribution of time and effort on behalf of ASTM.
Printed in Scranton, PA
July 1996
Foreword
The Symposium on Fiber, Matrix, and Interface Properties was held 14 November 1994 in
Phoenix, Arizona. ASTM Committee D-30 on High Modulus Fibers and Their Composites
sponsored the symposium. Christopher J. Spragg, Amoco Performance Products, Inc., and
Lawrence T. Drzal, Michigan State University, presided as cochairmen and are editors of
this publication.
Contents
Overview 1
Nonlinear Elastic Characterization of Carbon Fibers---P. ARSENOVIC 2
Methods of Determining the Temperature Dependence of Primary Creep--
C. A. LEWINSOHN, C. E. BAKIS, AND R. E. TRESSLER 9
Interracial Measurements and Fracture Characteristics of Single and Multi-
Fiber Composites by Remote Laser Raman Microscopy--c. GALIOTIS,
V. C H O H A N , A. PAIPETIS, AND C. VLATTAS 19
The Use of lnterfaeial Test Methods in Composite Materials Development--
M. J. PITKETHLY 34
Analysis of Stress Transfer from the Matrix to the Fiber Through an Imperfect
Interface: Application to Raman Data and the Single-Fiber Fragmentation
Test--J. A. NAIRN, Y. C. LIU, AND C. GALIOTIS 47
Modeling the Dynamic Response of the Fiber/Matrix Interphase in Continuous
Fiber Composite Materials--J. J. YUAN,J. M. KENNEDY,AND D. D. EDIE 67
Direct Observation of Debonding in Fiber Pull-Out Specimens--M. R. PIGGOan"
AND YU X I O N G 84
Interlaboratory Study of Adhesion Using Voltage Contrast XPS--J. D. MILLER,

G. W. ZAJAC, A N D T. N G U Y E N 92
Effect of Temperature and Fiber Coating on the Strength of E-Glass Fibers
and the E-Glass/Epoxy Interface for Single-Fiber Fragmentation Samples
Immersed in Water--c. R. SCHULTHEISZ, C. L. SCHUTTE, W. G. M c D O N O U G H ,
K. S. MACTURK, M. McAULIFFE, S. KONDAGUNTA, AND D. L. HUNTSON 103
A New Technique to Study the Interfacial Strength and Transverse Cracking

Scenario in Composite Materials---c. WOODAND W. BRADLEY 132
Characteristics of Fatigue Life and Damage Accumulation of Short-Fiber
Reinforced Polymer Composites--A. T. Y O K O B O R I , JR., n . TAKEDA, T. ADACHI,
J. C, HA, AND T. Y O K O B O R I 152
Fiber/Matrix Interface Studies Using Fragmentation Test--J. P. ARMISTEAD
AND A. W. SNOW 168
Fiber/Matrix Interface Effect on Monotonic and Fatigue Behavior of
Unidirectional Carbon/Epoxy Composites--B. LARGE-TOUMI,M. SALVIA,
AND L. VINCENT 182
STP1290-EB/Jul. 1996
Overview
The D-30 Committee on High Modulus Fibers and Their Composites of the American
Society for Testing and Materials (ASTM) sponsored and organized the Symposium on Fiber,
Matrix, and Interface Properties, held 14 November 1994 in Phoenix, AZ. This symposium
focused on the subjects of fibers, matrices, and their interface/interphase in composite materials
with an emphasis on fiber-matrix adhesion and its characterization.
Applications of composite materials, whether they be polymeric, metallic, or ceramic, grew
dramatically over the last decade. Rapid developments in glass and carbon fibers, matrices,
and processing methods occurred during the late 80s and early 90s. These new materials and
processes opened up a myriad of new applications. As the industry matured the pace of
improvement slowed. Throughout this time, the constituent properties were improved or modi-
fied based on testing of the finished composite mechanical performance. This method yielded
new composite materials capable of providing an advantage in density, strength, and stiffness,
but only with considerable effort. Now, researchers are studying the interaction between fiber
and matrix and the role of the fiber-matrix interface/interphase in composite performance.
With an understanding of this relationship we can extract the maximum performance and value
from current materials and reduce the cost and risk of new fiber or matrix development.
The research community has been actively pursuing a clearer understanding of the structure
and role of the interface/interphase in composite performance. In particular, test methods have
been the subject of a great deal of attention so that interfacial properties could be measured
in a meaningful way and related to composite performance. This has forced a reexamination
of the assumptions under which many of the "standard" fiber-matrix adhesion test methods
have been developed. This symposium brought together leading researchers in the field to
present and discuss the latest developments for the benefit of the composites community.
This symposium contains research papers presented by various authors who have directed
their attention towards this new class of interfacial problems. This volume contains results from
the application of the most commonly used test methods (fragmentation, pull-out, indentation) to
high-performance composites and their constituents, as well as their analysis. It also contains
several applications papers in which fiber-matrix adhesion was evaluated as part of the overall
assessment of composite performance. Two notable additions to the field are the development
and use of voltage contrast XPS for assessing adhesion of fiber to matrix in real composite
materials and the use of dynamic mechanical methods to evaluate interfacial properties in
high-volume fraction composites.
This symposium volume will be a significant addition to the personal or corporate library
of scientists and engineers concerned with either the fundamentals of fiber-matrix adhesion
or practitioners concerned with the evaluation of surface treatments and methods for optimizing
adhesion in composite materials.
The papers are timely and contain substantial reviews of the current literature so that they
can serve as a benchmark for future work. Some authors have identified future directions and
promising approaches to advancing the understanding of the interface/interphase.
1
Downloaded/printed
Copyright* by International
1996 by ASTM www.astm.org
Petar Arsenovic 1
Nonlinear Elastic Characterization of

Carbon Fibers
REFERENCE: Arsenovic, E, "Nonlinear Elastic Characterization of Carbon Fibers," Fiber,
Matrix, and Interface Properties, ASTM STP 1290, C. J. Spragg and L. T. Drzal, Eds., American
Society for Testing and Materials, 1996, pp. 2-8.
ABSTRACT: The non-linear elastic behavior of a series of pitch-based graphite fibers was
studied using the method of laser generated ultrasound. A Q-switched Nd-YAG laser was used
to produce pulses which generated stress waves in the fibers. The velocity of the acoustic wave
was determined from the time of flight of the wave at the point of impact to a piezoelectric
transducer positioned near one end of each specimen. The effective Young's modulus of the
fibers was determined for a wide range of static tensile stresses and varying temperatures. X-
ray diffraction measurements were also made on these fibers, to determine such factors as
crystallite size, orientation, and modulus. From these data, theoretical calculations of fiber
modulus were performed, and the results compared to the ultrasonic modulus measurements.
These results are useful for characterizing the elastic behavior of composites in which the fibers
are used.
KEYWORDS: graphite, crystal, modulus, X-ray diffraction, uniform stress, crystallite
High-performance carbon fibers are important in many applications, particularly in aerospace

composites. In such materials, the properties of the fibers compared to the matrix material are
generally quite different. For example, graphite fibers exhibit non-linear elasticity [1] as well
as negative coefficients of thermal expansion parallel to their axis [2]. The nature of a composite
material subjected to variations in stress and temperature can therefore be significantly affected
[3]. An understanding of the non-linear elastic properties of the fibers is therefore clearly
important in the design and materials selection of these composites.
From consideration of bonding types in solids, it is expected that the strongest materials
are likely to be those with a high density of covalent bonds. The carbon-carbon bond is
particularly strong, this being known from spectroscopic studies of bond energies in compounds
containing C-C linkages. The structure of the graphite fiber can be envisioned as a carbon
"chain" extending in two dimensions instead of one, comprised of a series of "crystallites" at
varying orientations with respect to the fiber axis. The arrangement within each crystallite is
one of planar sheets of covalently-bonded carbon atoms arranged in hexagonal arrays. These
sheets are located atop each other in sequences ranging from totally random to fully ordered
ABABAB type stacking, with bonding between layers being of the weaker Van Der Waals type.
This paper will examine some of the non-linear elastic properties of carbon fibers by using
laser generated ultrasound to determine their Young's modulus as a function of temperature
and applied static tensile stress. X-ray diffraction is used to examine crystallite orientation,
modulus, and size. The results are combined into a model to explain the non-linear elastic
behavior of these fibers in terms of their internal structure.
Materials Engineer, NASA Goddard Space Flight Center, Greenbelt, MD 20771.
2
Downloaded/printed
Copyright by International
s 1996 by ASTM www.astm.org
ARSENOVIC ON NONUNEAR ELASTIC BEHAVIOR 3
Experimental
The material under investigation consisted of a series of pitch-based graphite fibers designated
P25, P55, P75, P100, and P120. The samples were in the form of bundles of extremely fine,
black filaments. P25 was the least brittle of the group and had the lowest Young's modulus.
PI20, while having the highest modulus, was so brittle that even a minimal shear caused
breakage of the fibers.
The Young's modulus of the fibers was determined by laser generated ultrasound. From a
measurement of the ultrasonic velocity, the effective Young's modulus may be calculated by
the equation
E--- pv2 (1)
where p is the density and v is the ultrasonic velocity [4]. With this method, a set of moduli
for each specimen can be obtained under conditions of varying temperature and initial applied
load on the fibers. A Q-switched Nd-Yag laser was used to produce pulses of approximately
15 ns in duration and 20 mJ of energy. The pulses generated stress waves in the fibers by the
deposition of energy caused by their impact. The fibers were mounted in a temperature cell
and positioned vertically and perpendicular to the direction of the beam. A piezoelectric
transducer was clamped near the end on the fiber bundle to detect the acoustic wave generated
by the impact of the laser pulse upon the fibers. A sampling oscilloscope obtained a time of
flight of this wave as it traveled up the fiber bundle to the transducer. The velocity of the
acoustic wave was determined by dividing the difference between two distances from laser
impact to transducer by the difference in corresponding times (A~/At).
X-ray diffraction measurements were made using an automated Picker diffractometer system
controlled by a VAX computer. The computer controlled each of the four goniostat angles
(20, oJ, X, and 4)) by means of a motor used to drive the angle and an encoder which read the
angle positions to 0.01 degrees. The control program, written in FORTRAN, accepted com-
mands which were converted by the computer to appropriate electrical signals and sent to the
hardware. Data output was logged to both a video terminal and a printer. These experiments
were performed using an X-ray tube with a copper target and a nickel filter, providing CuKet
radiation at a wavelength of 1.54 A. This system was used for orientation studies, crystallite
size determination, and crystal modulus measurement, which in the case of the graphite fibers
was 1/stl. The sij represent the elastic compliances for graphite.
Results and Discussion

Figure 1 shows the effective Young's modulus obtained by laser-generated ultrasound as a
function of static tensile stress at room temperature. Each data point represents an average of
five measurements. The modulus changes in all of the fibers tested have proven to be reversible
in stress. In addition, testing of several samples of P55 and P75 with a surface layer of
amorphous carbon has shown this coating to have no apparent effect on the non-linear elastic
behavior of these fibers. Clearly, the non-linear elastic portion of plots in Fig. 1 decreases with
samples having higher initial modulus. The effect is not present at all for the P120 specimens.
Figure 2 shows the effective Young's modulus of the carbon fibers as a function of temperature
with no external applied load. The modulus of all the fiber types initially, decreases with
increasing temperature before leveling off at ~ 200~ What is therefore generally observed
is an increase in Young's modulus with increasing stress and a decrease in the modulus with
increasing temperature.
A number of parameters must be considered in developing a model that explains the behavior
exhibited by the carbon fibers. X-ray diffraction has been a valuable tool for investigating
4 FIBER, MATRIX, AND INTERFACE PROPERTIES
*-INCREASING STRESS
o-DECREASING STRESS
8 5 0 i-
~ PI20
800 I
750~ _.
- - I ~ - ~ - x ~ ~ PIO0
7oo ' - " -
o9
:D 5OO _ P75 o*
z:) r- ' < ~ * ~ '~ - " | P75S "
o
o
450 -
4OO
co
-o 55O OX
z
z) P555
0 300 -
>-
25O | P25
20O
150 I I I I 1 I I I I I I I I I i f I t I 1
0 20 40 60 80 ~00 i20 ~40 ~60 ~80 200
STRESS (MPo)
FIG. 1--Young's modulus versus stress at room temperature.
some of these variables, such as the size, orientation, and modulus of the crystallites of which
the fibers are composed. The first step was to take standard equatorial and meridional 20 scans
samples from the fiber series P25-P120. As the fiber modulus increases, the X-ray diffraction
peaks get sharper and narrower, indicating higher degrees of crystallite order and perfection.
CrystaUite orientation was determined by measuring the intensity of the (002) equatorial
reflection as a function of diffraction angle X. The orientation is generally expressed in terms
of half-width at half maximum, and can also be stated conveniently in terms of Hermann's
orientation factor [5] or average orientation angle. The orientation factor F is given by the
expression:
F = (f I(3 sin2• - l)cos • d• f l cos x d• (2)
where I is the intensity and X is the diffractometer angle chi measured from the equator. An
orientation factor of - 0 . 5 indicates perfect alignment. The average orientation angle r of the
crystallites relative to the fiber axis is determined by the expression:
F= [3<cos2r) - l]/2 (3)
where F is again the Hermann's factor described above. A value of r of 90 ~ indicates perfect
alignment, since in this case r indicates the angle perpendicular to the fiber axis.
ARSENOVIC ON NONLINEAR ELASTIC BEHAVIOR 5
900
800 PI20
700
P tOO
600
500
0
400 P75
o
>-.
500
P55
200
P25
IO0
I
O 50 IOO 150 200 250 3OO 350 400 450
T (~
FIG. 2--Young's modulus versus temperature at zero stress.
To obtain the data, equatorial 20 scans were run to locate the relevant peak, i.e., the (002)
reflection, precisely. Variation of the angle X was made between 0 and 90 ~ at the position of
maximum intensity in 20. Table 1 summarizes the orientation results.
As can be seen from Table 1, crystallite orientation improved both with processing (different
types of fiber) and with increasing stress. The higher modulus fibers had a superior crystallite
orientation to begin with, and all fibers showed further alignment with the application of stress.
Changes in crystallite alignment are therefore shown to be at least partially responsible for
the variation of modulus with stress and processing observed by the ultrasonic method described
earlier. The change of alignment does not completely encompass the change of modulus with
stress and from one fiber to the next. This can be seen in Fig. 3 in which the effective Young's
modulus is plotted versus the half-width at half-maximum of the crystal orientation distribution
for various tensile stresses. The line through the data is a "best fit" through the zero stress
modulus measurements. The variation of modulus and half-width with stress for each fiber
does not completely coincide with the change of modulus and half-width from one fiber to
the next at fixed stress. Clearly, there are other parameters involved, and one must begin to
look at the individual elastic constants of graphite. The hexagonal structure has five independent
elastic constants.
One candidate is the compliance, $44 for shear on the basal plane [6]. It increases with an
increasing degree of graphitization. X-ray measurements of the spacing of the change of the
(110) planes with tensile stress indicate that the compliance, s~ t, for tension on the basal planes
also changes. The measurements were performed by using a scan range consisting of seven
points located symmetrically on the (110) peak. 20 values for each point were determined
by a least squares fit of a Gaussian curve to the scanned points. With changing stress, peak
shift and hence change in lattice parameter spacing are observed. The fibers were mounted
on the X-ray diffractometer using a fiber deformation device so that tension could be applied
to the specimens while held in the diffracting position. Changes in load were determined by
FIBER, MATRIX, AND INTERFACE PROPERTIES
TABLE 1--Comparison of stress with average orientation angle.

Stress, HWHM, Avg. Orientation,
Fiber (MPa) (Deg) (Deg)
P25 12.0 13.86 72.82

26.0 13.71 72.99
41.7 13.60 73.12
333.4 12.43 74.62
500.1 12.00 75.27
P55 53.9 7.49 82.24
150.0 7.31 82.24
269.5 7.09 82.25
538.9 6.96 82.39
718.6 6.76 82.44
P75 14.0 5.00 83.68
64.2 4.88 83.75
321.2 4.83 83.76
642.4 4.73 83.93
PI00 1.0 2.00 85.22
55.0 1.95 85.24
110.0 1.90 85.26
162.0 1.85 85.27
P120 27.0 1.56 87.12
100.2 1.56 87.12
414.1 1.51 87.16
828.2 1.51 87.16
I100
IO00
9OO
STATIC TENSILE STRESS
120 9 O MPa
n 800
(.9
700
600
O
O ^ P75
500
o') #00 " ~ . . ~ ,,_ P55

-[D
7
300
9
>.-
2OO
I00
t L f t L I I l L I I L I t I I
0 2 5 ~ 5 6 7 8 9 IO I I 12 I3 14 I5 I6 17
HALF WIOTH AT HALF MAXIMUM (OEGREES)
FIG. 3--Young's modulus versus half-width at half-maximum of X scan.
the calibration of the relationship between the output voltage of a strain gage bridge on the
fiber deformation device tension arm and the tension applied to the ann. This was determined
to be 2665 gm/mV. Plots for each specimen of d spacing vs. tension were made, and a "crystal
modulus," defined as llsl 1 was calculated by
ARSENOVIC ON NONLINEAR ELASTIC BEHAVIOR
TABLE 2--1/sH (Gpa).

Fiber 1/sl J
P25 290
P55 582
P75 719
P100 832
P120 1012
Y = 0.0882 doSpl(mDN) (4)
where Y is in Gigapascals, m is the slope of the d spacing versus tension curve in Angstroms/
millivolt, do is the average lattice spacing in Angstroms, p is the density of the fibers in gm/
cc, S = 2665 gm/mv, D = denier per fiber, and N is the number of fibers in the sample. Table
2 shows llsll for each type of fiber.
From these data, in particular the crystal moduli and orientation distributions, the macroscopic
Young's modulus was calculated by the uniform stress model [8] and compared to the ultrasonic
data previously acquired. The results based on the assumption of uniform stress are shown in
Fig. 4. This model assumes that a uniform stress is applied to each crystallite in the direction
of the fiber axis. The strain developed in each crystal will depend upon orientation. The
crystallites are assumed to be strongly linked along the fiber axis, and the total fiber strain is
the sum of the individual crystal strains. The effective Young's modulus is then determined by
lIE = $33 q- (2s13 - 2s33 + s44)I3[11 + ( s l l - 2s13 + $33 - s44)/5]/1 (5)
where
I. = I l(~)cos"(,~) a~ (6)
1(6) is the orientation distribution as determined by the • scans described earlier. The s~j are
the elastic constants of graphite [9]. As can be seen by a comparison of Figs. 4 and 1, the
Calculated Modulus (GPa)

800
700 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .) ~ .......................................................
600 ..................................... ~ ............................................. o~7 .................................. :O. . . . .
500 ......................................................................................................................................................
400 ........... ~ ................................... . . * ................................................................... '~ ........................
300 ................................. ~......................................................................................... ~ ......
2 0 0 .................................................................................................................................................
100 ..........................................................................................................................................
1 l i i t 1 i
20 40 60 BO 100 120 140 160
Stress (MPa)
--- P25 --~-- P55 -,K- P75 -e-- P100 --x-- P120
FIG. 4---Calculated modulus by Uniform Stress model versus applied stress.
8 FIBER,MATRIX,AND INTERFACE PROPERTIES
TABLE 3--Crystal size in meridional direction (110) peak and equatorial direction (002) peak.
Fiber I,~(1I0), Angstroms Lc(O02), Angstroms
P25 103 17
P55 219 Ill
P75 233 131
P100 433 194
PI20 821 227
calculated results by the uniform stress model and the ultrasonically determined values of the
modulus are markedly similar. The calculated modulus is consistently below the ultrasonically
measured modulus by about 15% in all cases with the exception of P25, where the variation
is more marked, especially at higher stresses. The model is fairly effective in predicting the
actual elastic behavior of the fibers with respect to Young's modulus.
Another variable that can influence fiber modulus is crystallite size. This can be calculated
by the Scherrer equation
L = size = ICMB cos 0 (7)
where K = shape factor ~ 0.89, h = X-ray wavelength, B = breadth of diffraction peak, and
0 = Bragg angle. The crystallite size increases with higher modulus and superior orientation.
Table 3 summarizes these data.
Conclusions
The effective Young's modulus of a series of pitch-based graphite fibers was determined
by laser-generated ultrasound as a function of stress and temperature. X-ray diffraction measure-
ments of the crystallites making up the fibers were made in an attempt to explain some of
the mechanisms involved. Changes in crystallite orientation were shown to account for some
of the changes observed in the moduli of the fibers when various external loads were applied.
Crystallite size was also shown to increase as one goes to the higher modulus fibers. These
data were applied to the model of uniform stress with the calculated results comparing favorably
to the experimental values.
References
[1] Curtis, G. J., Milne, J. M., and Reynolds, W. N., Nature, Vol. 220, 1968, pp. 1024-1025,
[2] Jenkins, G. M. and Kawamura, K., Polymeric Carbons, Cambridge University Press, Cambridge,
1976.
[3] Liu, J. M., Applied Physical Letter, Vol. 48, No. 7, 1986, pp. 469-472.
[4] Jiang, H., Arsenovic, P., Eby, R. K., Liu, J. M., and Adams, W. W., Polymer Preprints, Japan, Vol.
36, Nos. 5-10, 1987.
[5] Lenhert,P. G., Obrien, J. E, and Adams, W. W., "A Users Guide to the Picker Diffractometer for
Polymer Morphology Studies," AFWAL-TR-86-4024,December 1986.
[6] Reynolds,W. N. and Moreton, R., Philosophical Transactions of the Royal Society, London, Ser.
A 294, 1980, pp. 451-461.
[7] Lenhert,P. G., "X-ray Modulus Determination of Ordered Polymers," AFWAL-TR, I985.
[8] Brydges,W. T., Badami, D. V., Joiner, J. C., and Jones, G. A., Applied Polymer Symposia, No. 9, 1969.
[9] Goggin, P. R. and Reynolds W. N., "The Elastic Constants of Reactor Graphites," 1967.
C. A. Lewinsohn, 1 C. E. Bakis, t a n d R. E. Tressler 1
Methods of Determining the

Temperature Dependence of Primary
Creep
REFERENCE: Lewinsohn, C. A., Bakis, C. E., and Tressler, R. E., "Methods of Determining
the Temperature Dependence of Primary Creep," Fiber, Matrix, and Interface Properties,
ASTM STP 1290, C. J. Spragg and L. T. Drzal, Eds., American Society for Testing and Materials,
1996, pp. 9-18.
ABSTRACT: The service life of high-temperature composites and reinforcement fibers can be
limited by creep, therefore it is desirable to understand the relationships between the creep
strain and external variables. Due to load sharing between constituents or internal microstructural
changes, a steady-state creep rate is never observed in some materials. Usually in these materials
the creep rate decreases with increasing time. This is known as primary creep. Determination
of the temperature dependence during primary creep is complicated by its dependence on time
and/or strain. The suitability of various methods of analyzing primary creep data has not been
addressed in detail before.
In this paper, three methods of determining the temperature dependence during primary creep
are compared: the Arrhenius, the cross-cut, and the Sherby-Dorn methods. The methods are
applied to creep data for silicon carbide fibers intended as reinforcements in high-temperature
ceramic or metal matrix composites. The sensitivity of each method to measurement resolution
and the capability of the creep laws associated with each method for predicting measured creep
behavior are evaluated. The reliability of the computer software fitting routine is independently
confirmed. The Sherby-Dorn and cross-cut methods are mathematically similar but the Sherby-
Dorn method does not involve an arbitrary selection of experimental conditions, is less cumber-
some to apply to large amounts of data, and uses a broader range of the available data.
KEYWORDS: ceramic fibers, primary creep, temperature dependence, silicon carbide fibers
Many materials exhibit time-dependent deformation, known as creep, when subjected to a

combination of heat and stress. In some materials a steady-state creep rate is reached when
the rate of stress-induced deformation is balanced by a resistance or relaxation mechanism.
In other materials, however, a steady state is never reached and the rate of deformation is
always finite yet decreasing. These materials are said to exhibit only primary creep. The
dependence of the primary creep rate on time and/or strain complicates calculation of the
temperature dependence of the creep rate and, hence, inference of the mechanism responsible
for creep. In this paper, three techniques that can be used to determine the temperature
dependence o f primary creep will be discussed. Important differences resulting from steady
state and time-dependent analyses of materials exhibiting primary creep will be illustrated.
The three methods will be used to determine the temperature dependence of a material that
exhibits only primary creep. The creep data that will be used are from SCS-6 silicon carbide
Graduate assistant, Department of Materials Science and Engineering, associate professor of Engi-
neering Science and Mechanics, professor and department head, Department of Materials Science and
Engineering, respectively, The Pennsylvania State University, University Park, PA 16802.
Downloaded/printed
10 FIBER,MATRIX, AND INTERFACE PROPERTIES
fibers 2 that have been tested in tension. The fibers were fabricated via chemical vapor deposition
of silicon carbide onto a carbon core and they have been used as reinforcements in ceramic
matrix composites for high-temperature engines. The chemical composition and microstructure
of the silicon carbide fibers have been described in detail elsewhere [1,2]. Ceramic matrix
composites offer greater reliability than monolithic ceramics, but under certain conditions their
high-temperature lifetime may be limited by creep of the reinforcing fibers. The mechanism
responsible for the creep behavior of the silicon carbide fibers used in this study has been
discussed elsewhere [3] and is not relevant to this analysis, other than it leads to primary
creep behavior [4].
The details of the experimental method appear elsewhere [5], but will be summarized here.
Single fibers were tested, under constant load, in rigid mechanical test frames. The fibers were
held in 32-mm-long stainless steel shafts by glass beads. The steel shafts were gripped by the
jaws of pin-vises. The pin-vise grips were outside the furnace hot-zone and the fiber hung
vertically through an electrically heated resistance furnace. The total displacement was mea-
sured by a linear capacitative displacement transducer 3 and the output was monitored by either
a personal computer or chart recorder. The resolution of the transducers was 13 Ixm and the
response exhibited no more than 0.05% deviation from linearity over the full range of deflection
(5 p~m). More than one test frame and transducer was used and the total displacement after
many tests was confirmed to within 20 I~m by measurements using a traveling-stage optical
microscope. 4 The tests were performed under constant load, 437 _ 0.5 g (278 MPa), in air.
Due to minor variations in the fiber diameter along its length, approximately -+ 1 I~m, the
stress is actually only accurate to within 4 MPa. The creep strain was measured continuously
at temperatures of 1200, 1300, and 1400~ for up to 100 h. Although experiments were
performed at other stress levels, to determine the stress dependence of the creep behavior,
they will not be discussed in this paper which focuses on the temperature dependence only.
A commercially available software package, SigmaPlot,5 was used to perform all the data
analysis, except where otherwise noted.
Methods of Determining Temperature Dependencies

The physical mechanisms responsible for creep deformation require mass transport and are
therefore thermally activated. In the case of steady-state creep, the strain rate is often described
by the empirical relationship [6],
where G is the creep rate, Ao is a constant, Q is the activation energy (J/mol), R is the universal
gas constant (8.3144 J/mol-K), and T is the absolute temperature (K). The conventional method
of determining the activation energy for steady-state creep is to calculate the constant of
proportionality between the natural logarithm of the creep rate and the inverse of the tempera-
ture. This quantity is equal to -Q/R as shown by taking the natural logarithm of Eq I.
2 Manufactured by Textron Specialty Materials, Inc., Lowell, MA.

3 ASE Model 1082 SLVC, Milton Keynes, UK.
4 Questar QM-1, New Hope, PA.
5 SigmaPlot version 5.0, Jandel Scientific Inc., Corte Madera, CA.
LEWINSOHN ET AL. ON TEMPERATURE AND PRIMARY CREEP 11
In ~c = In Ao - ~ (2)
This technique will be referred to as the Arrhenius method. The method can be represented
graphically on a plot of the natural logarithm of the strain rate versus the inverse of the
temperature (Fig. 1). The slope of the line representing the creep rate as a function of temperature
is - Q / R , and would be independent of creep strain if the specimen was undergoing steady
state creep.
During primary creep, however, the creep rate is dependent on time. In terms of the parameters
given in Eq 1, chemical or physical changes in the material undergoing creep are causing
changes in the value of the pre-exponential term, Ao. Therefore, the Arrhenius method is not
valid although creep itself is still thermally activated. It has been suggested that if the creep
strain is representative of the structure of the material then the creep rate at a fixed strain and
varying temperatures should be used to determine the activation energy for that structure [6].
It has even been suggested that the method be applied to the strain rates after a fixed time at
varying temperatures [6], but this seems even more arbitrary than using the strain rates at a
constant strain. In either case the activation energy calculated will only be accurate for the
specific strain or time chosen. A demonstration of this statement and its implications will be
presented in the following section.
Another method of determining activation energy from creep data is referred to as the cross-
cut method [7]. As the temperature increases, the time to obtain a constant strain decreases.
Using Eq 1, the relationship between the time, fi, to reach a constant strain at temperature Tl
and the time, t2, to reach the same strain at temperature T2 is,
\ q ] = -R ~22 - ~
1] (3)
This method can be represented graphically on a plot of the natural logarithm of the creep
strain versus the inverse of the temperature (Fig. 2). Isochronic lines are constructed by plotting
the corresponding creep strain, at varied temperatures, for specific values of time (i.e., 10,
100, 1000 min). The collection of additional data that could be used with confidence was not
possible due to the length of time required to collect experimental data. The distance between
the intersection of a horizontal line (constant strain) with the isochronic lines is equivalent to
the term in square brackets in Eq 3. Equation 3 applies to thermally-activated steady-state
-10
r
-12
-14
-16
Q=
-18
-20 , r f h
5.75 6.00 6.25 6.50 6.75 7.00
1/T (K -1 x 10 '~)
FIG. I - - A graph of the natural logarithm of the creep strain rate versus the reciprocal absolute
temperature used to apply the Arrhenius technique at ~c = 0.0027.
-4
-5
0 mln"
In ~c --8
m'~
-7 ~ m m
_~ i i i i ,
5.75 6.00 6.25 6.50 6.75 7.00

1/T (K -1 x 10 4)
FIG. 2 A graph of the natural logarithm of the creep strain versus the reciprocal absolute
temperature, used to apply the cross-cut method.
creep (Eq 1) and certain forms of primary creep including that measured in this study. For
other forms of primary creep, the terms in Eq 3 may differ slightly but the approach will be
similar. The cross-cut method is also only accurate for the particular strain chosen for the
calculation, as will be demonstrated in the following section and as is clearly apparent by the
difference in the slopes of the isochronic lines shown in Fig. 2.
The last technique that will be discussed in this paper has been described by Sherby and
Dom [8] for cases where the stress-strain relationship of a material is dependent on both the
strain rate and temperature through a single, dimensionless parameter. In many cases, the
primary creep strain is found to be a single-valued function of the temperature compensated
time, |
ec = F(O) (4)
The temperature-compensated time is defined by the equation,
O=flexp(-#T)dt=t[exp(-#T)] (5)
where t represents time and the other symbols have the same meaning as described previously.
The function, F(O), can be found by treating the exponential term as a constant and finding
a function of time that describes each individual creep curve. Several types of equations have
been proposed to describe primary creep [6,9,10] and these were investigated first because
they have been shown to be valid on either an empirical or semi-empirical basis. Therefore,
the activation energy for creep can be determined by finding the value of Q for which the
creep strain is described by Eq 4 at all temperatures. This technique can also be represented
graphically on a plot of the creep strain versus the temperature compensated time, | (Fig. 3).
The best value of Q is that which causes all the creep curves to coincide, provided Q is
actually a true activation energy and independent of temperature. The method used to find
the best value of Q will be described later.
Application and Comparison of the Techniques

The three methods described in the previous section will be used to determine the temperature
dependence of SCS-6 silicon carbide fibers that exhibit only primary creep. The accuracy of
0.02
I Q = 552 kJ/real I
9 .
1400"C " "
...-"
Creep 300"C.* ~
(m/re)Strain0.01 I, ~
O0~
0.00 i ,
0 1 2 3 4-
Temperature compensated Trine (minutes x 10 TM}
FIG. 3--A graph of the creep strain versus the temperature compensated time used to apply the
Sherby-Dorn technique.
the temperature dependence will be evaluated in the discussion. A plot of typical creep data
for two SCS-6 fibers tested at 1300~ in air, is shown in Fig. 4. The symbols represent actual
data and the solid lines represent computer generated best-fit time power laws,
~r = At p (6)
The data and the computer-generated curves illustrate the time-dependent creep rate of the
fibers (i.e., p is not equal to one). For the purpose of comparing the three techniques, the
effect of material variability was removed by averaging the experimental data and fitting a
curve describing the strain as a function of time, with the form of Eq 6, through the average
data at each temperature. The same curves were used in each of the analyses to determine
the temperature dependence. The effect of measurement resolution on the analyses will be
considered later.
To apply the Arrhenius method it was necessary to determine the instantaneous strain rates
for a fixed value of strain at each different temperature. The instantaneous strain rate was
calculated by dividing the difference between the chosen strain value and a strain value
immediately before or after it by the corresponding difference in time. Figures 1 and 5
demonstrate the Arrhenius method at two different strain values, e = 0.0027 and e = 0.004,
respectively. The lower strain value was chosen because it was the lowest value where the
creep rate could be measured with confidence in the highest-temperature data set and the higher
strain value was chosen because it was the highest common strain value at all temperatures. The
0.020
0.015
Creep
Strain 0.010
(m/m]
0.005
0.000 ~ , ~ i f
1000 2000 5000 4-000 5000 6000
Time (m~nutes)
FIG. 4 Comparison of experimental data at 1300~ (symbols), from two separate tests, with
power law curve fits (solid lines) of the form ~c = Atp through each set of data.
-10
t==O,O04
-12
in ir -14_16
-16
Q=566kJ/mo~~
-20 ~ ' ' '
.5.75 6.00 6.25 6.50 6.75 7.00
1/r (K -1 x 10 4 )
FIG. 5--A graph of the natural logarithm of the creep strain rate versus the reciprocal absolute
temperature used to apply the Arrhenius technique at Ec = 0.004.
slope of the least-squares and best-fit first order monomial using the Marquadt-Levenberg
technique (both determined by SigmaPlot) were identical up to four significant digits. Based
on these slopes, the activation energy was found to be 624 kJ/mol at e = 0.0027 and 568 kJ/
mol at e = 0.004.
In Fig. 2, representing the cross-cut method, the lines shown represent the best-fit given by
the least squares method. The temperatures at which it would take 10, 100, or 1000 min to
reach e = 0.0027 and e = 0.004 were calculated from the slope and intercept of the best-fit
line. The strain values were chosen to correspond to those used in determining the activation
energy via the Arrhenius method. The activation energy was determined by rearranging Eq 3
and solving for Q. Values of the activation energy, Q, were determined at both strain levels
for the intervals between the 10 and 100 min, 100 and 1000 min, and 10 and 1000 min
isochronic lines (Table 1). The average value of Q at creep strains of 0.0027 and 0.0040 was
found to be 610 and 660 kJ/mol, respectively.
The third method considered, described by Sherby and Dorn, is demonstrated in Figs. 3,
6, and 7. In these figures the creep strain is plotted versus the temperature-compensated time
for three different activation energies. An activation energy of 550 kJ/mol was determined
from the parameters that gave the best-fit (using the Marquadt-Levenberg algorithm) of the
data of the time, temperature, and average strain at 1200, 1300, and 1400~ to the equation,
(7)
TABLE 1--Temperature dependence determined by the cross-cut method.

Creep Time Temperature
Strain, m/m Interval, min Dependence, kJ/mol
0.0027 10-100 680

0.0027 100-1000 554
0,0027 10-1000 611
0.0040 10-1000 727
0.0040 100-1000 593
0.0040 10-1000 653
0.02 r
= 4o0 kJ/mol I
1400"C
Creep . 9 * .....
Strain 0.01 ..- .............
(m/m)
1200"C
0.00 ' ' ' '

1 2 3 4- 5
Temperature Compensated T i m e ( m i n u t e s x 10 ~~
FIG. 6---A graph of the creep strain versus the temperature compensated time demonstrating the
Sherby-Dorn technique for an incorrectly low (400 kJ/mol) temperature dependence.
where A and p are constants and Q is the activation energy. Equation 7 corresponds to the
function of temperature compensated time that describes the data best (Eq 4). This empirical
equation was proposed by DiCarlo and Morscher [11] to explain the creep behavior of SCS
fibers and was initially suggested by Sherby and Dorn to describe primary creep [9]. To
demonstrate the failure of an incorrect activation energy in describing the data by Eq 7,
activation energies of 400 kJ/mol and 700 kJ/mol were used to generate the plots of strain
versus temperature compensated shown in Figs. 6 and 7, respectively. These plots graphically
illustrate that these values do not describe the data with a single-valued function.
To determine whether the software used primarily in this study, SigmaPlot, provided reliable
results the temperature dependence of the creep strain of SCS-6 fibers tested at a stress of
278 MPa was evaluated by the Sherby-Dorn technique using a second software package,
CONSAM [12,13]. This program was developed for kinetic modeling and numerical analyses
in biological fields and the data analysis was performed by Eric Ramberg as part of a separate
investigation into determining the magnitude and temperature dependence of rate constants
[14]. The value of the temperature dependence for SCS-6 fibers determined by CONSAM
agreed with that determined by SigmaPlot (Table 2) which gave further confidence in the results.
Discussion
The broad range of values determined for the activation energy indicates the importance of
critically evaluating the technique used to calculate the temperature dependence of primary
creep. The significance of the different values determined by each technique was compared
0.02 . , , , . . . . . ,
[o = 700 kj/mo, I
Creep
Strain 0.01 ~ . 1300"C. 9 4
(m/m) ~ . . . . 99 " I oo'(
0.00 ! I i I r I I I i I
0 I 2 3 4 5 6 7 8 9 I0
Temperature Compensatea Time (minutes x 1019)
FIG. 7--A graph of the creep strain versus the temperature compensated time demonstrating the
Sherby-Dorn technique for an incorrectly high (700 kJ/mol) temperature dependence.
TABLE 2--Calculated values of the temperature dependencefor creep of SCS-6.

Method Strain, m/m Average Value, kJ/mol
Arrhenius 0.0027 624

Arrhenius 0.0040 568
Cross-cut 0.0027 610 -4- 30
Cross-cut 0.0040 660 -+ 30
Sherby-Dorn ~ all 550 + 30
Sherby-Dorn b all 550 _+ 30
~Determined by SigmaPlot.
bDetermined by CONSAM.
with the range in values caused by measurement resolution errors. In addition, the capability
of the temperature dependence determined by the cross-cut and Sherby-Dorn methods to
describe the experimental data was investigated.
The resolution of the strain measurements of the data used was 130 I~e which corresponds
to approximately 3-4% of the strain values used. Adding this value to the value of the strain
at the highest temperature (1400~ and subtracting it from the strain at the lowest temperature
(1200~ generates the greatest variation. This procedure was applied to the experimental data
and the calculations were repeated to determine the effect the measurement resolution had on
the value of the temperature dependence. The Arrhenius technique was not evaluated since
all evidence indicated that the fibers exhibited only primary (time-dependent) creep.
The sensitivity of the cross-cut method was examined in the manner described above. For
each time interval evaluated the strain was increased by 4% at 1400~ and was decreased by
4% at 1200~ At e = 0.0027 the Q values for three time intervals ranged from 586 to 710
kJ/mol and averaged 646 kJ/mol. At e = 0.004 the values ranged from 625 to 758 kJ/mol
and averaged 689 kJ/mol. These are equivalent to adding 30 kJ/mol, or approximately 4-5%,
to the activation energies determined from the original data.
The activation energy determined by the Sherby-Dorn method changed from 550 to 590
kJ/mol, which is roughly 7%. Hence, the Sherby-Dorn method appears to be equally sensitive
to variability in the data as the cross-cut method. A summary of the values of the activation
energy determined by each method is shown in Table 2 along with the range due to measurement
resolution. The activation energies predicted by the cross-cut method were noted to vary 8%
between the different strain values that were used. Therefore, an advantage of the Sherby-
Dorn method is that it does not involve variability in the results caused by arbitrary choices
of the time intervals used to determine the temperature dependence.
Mathematically, the cross-cut and Sherby-Dorn methods are similar. In fact, in this paper,
the equations used to fit the data for the two methods (Eqs 3 and 4) were derived from the
same creep law (Eq 7). Both techniques involved fitting parameters to values described by
curves fit to experimental data. The Sherby-Dorn method, however, appears to have three
advantages over the cross-cut method. Firstly, it does not involve the arbitrary choice of time
values and time intervals, which can lead to variability in the results, associated with the cross-
cut method. Secondly, for the case of analyzing continuous creep strain data as a function of
time, improving the accuracy of the cross-cut method would involve using more and more
strain values and time intervals leading to extremely cumbersome and time-consuming calcula-
tions. Lastly, inspection of Figs. 2 and 3 illustrate how the cross-cut method utilizes only a
narrow range of data (that between the strain values that occur at the shortest time at high
temperature and the longest time at low temperature) whereas the Sherby-Dorn method finds
a function that applies over the whole range of times and temperatures. There is a skew to
smaller temperature-compensated times, however, in the Sherby-Dorn technique since the
LEWlNSOHN ET AL. ON TEMPERATURE AND PRIMARY CREEP 17
TABLE 3--Parameters used to predict creep strains from Eq 7for comparison to the experimental
data (tr = MPa, t = min, T = K).
Method A, min -p p Q, kJ/mol
Sherby-Dorn 31.6 0.24 550

Cross-cuta 213.0 0.26 615
aValues for e = 0.0027.
lowest temperature data only exists there. Increasing the range of values for analysis via the
cross-cut method involves increasing the range of experimental conditions (time and tempera-
ture) which may be time-consuming and expensive. Nevertheless, for the case of data limited
to a few discrete values of strain or time the cross-cut method is as valid as the Sherby-
Dorn method.
As a final comparison of the techniques, the activation energies determined by the cross-
cut and the Sherby-Dorn techniques (Table 2) were used to predict creep curves which were
compared to the original creep data. Since the Arrhenius method is only applicable to constant
strain levels and does not describe the observed time dependence, it was not used to make
predictions of creep strain. The creep law proposed by DiCarlo and Morscher (Eq 7) was
assumed to describe the creep behavior of the Silicon carbide fibers. Furthermore, the expression
used to determine the activation energy via the cross-cut method, Eq 3, can also be derived
from this creep law.
The cross-cut method does not provide unique values of the adjustable parameters in the
creep law. To determine the parameters A and p in the creep law, the activation energies
calculated by the cross-cut method at e = 0.0027 and e = 0.004 were used to fit the fiber
creep data at 1300~ The creep strains at 1200 and 1400~ were calculated by substituting
these values of A and p and the corresponding activation energies that fit the data at 1300~
into Eq 7 (Table 3). The parameters associated with the activation energy determined by the
cross-cut method, at both levels of strain, predicted the creep strain at 1200~ but overestimated
it at 1400~
The Sherby-Dorn technique provides unique values for all the adjustable parameters of the
chosen creep law (Table 3). The parameters provided by the Sherby-Dorn method described
the data reasonably well at all temperatures. These results are represented graphically in Fig.
8 where the strains predicted by the cross-cut (ec = 0.0027) and Sherby-Dorn parameters are
compared to the creep data. The creep strains predicted by the parameters obtained by the
0.020
. .+ : i_.ii.....'....:+
Creep~
Strain
(ml=)
o
0.0! 0
0.005
~176176176 ,o;o o'oo +o'oo .o'..'oo +ooo

Time (mlnukes)
FIG. 8--A comparison of the average creep strain of SCS-6 fibers at 278 MPa (solid line,*) and
predicted creep strains using Eq 7 and the parameters from the Sherby-Dorn technique (dashed
lines) and the cross-cut method (symbols).
18 FIBER,MATRIX, AND INTERFACE PROPERTIES'
cross-cut method at ec = 0.004 were slightly lower at 1200~ and slightly higher at 1400~ than
those predicted by the parameters determined at ec = 0.0027. The inaccuracies in predictions by
the cross-cut method are attributed to the temperature dependence being ascertained for the
relatively low values of strain needed to include the 1200~ data.
Conclusions
This analysis has demonstrated that although primary creep may be thermally activated, the
value of the thermal activation energy is dependent on the technique used to calculate it.
The Arrhenius method is inappropriate for describing time-dependent creep phenomena. The
Sherby-Dorn and cross-cut methods are equally sensitive to measurement resolutions but the
Sherby-Dorn method has distinct advantages over the cross-cut method for analysis of continu-
ous data sets. The Sherby-Dorn method does not involve arbitrary choices of experimental
conditions, is less cumbersome to apply to large amounts of data, and maximizes the use of
the data collected. The parameters determined from the Sherby-Dorn analysis also described
the experimental data better than those from the cross-cut method as used here.
Acknowledgments
The authors would like to thank Dr. J. A. DiCarlo and Mr. G. N. Morscher of the NASA
Lewis Research Center for many useful discussions. The authors would also like to thank Mr.
Morscher for providing materials. This work was supported by the National Aeronautical and
Space Administration under grant NAGW-1381.
References
[1] Ning, X, J. and Pirouz, P., "The Microstmcture of SCS-6 SiC Fiber," Journal of Materials Research,
Vol. 6, No. 10, 1991, pp. 2234-2248,
[2] Ning, X. J., Pirouz, P., and Farmer, S. C., "Microchemical Analysis of the SCS-6 Silicon Carbide
Fiber," Journal of the American Ceramic Society, Vol. 76, No. 8, 1993, pp. 2033-2041.
[3] Lewinsohn, C. A., "Primary Creep of CVD Silicon Carbide Fibers," Ph.D. Thesis, The Pennsylvania
State University, 1994.
[4] Lewinsohn, C. A., Bakis, C. E., and Tressler, R. E., "Microstructural Influences on the Creep
Behavior of CVD Silicon Carbide Monofilaments," Advances in Ceramic-Matrix Composites,
Ceramic Transactions,The American Ceramic Society,Westerville,OH, Vol. 83, 1994, pp. 667-678.
[5] Lewinsohn, C. A., Bakis, C. E., Hahn, H. T., and Tressler, R. E., "Creep Testing of CVD Silicon
Carbide Fibers," Proceedings of the American Society for Composites, 7th Technical Conference,
Technomic Publishing, Lancaster, PA, 1992, pp. 779-791.
[6] Cadek, J., Creep in Metallic Materials, Elsevier, NY, 1988.
[7] Damask,A. C. and Dienes, G. J., Point Defects in Metals, Gordon and Breach, Science Publishers
Inc., New York, 1963.
[8] Sherby,O. D., Orr, R. L., and Dorn, J. E., "Creep Correlations of Metals at Elevated Temperatures,"
Transactions of the American Institute of Mining and Metallurgical Engineers, Vol. 1, 1954,
pp. 71-80.
[9] Sherby, O. D. and Dorn, J. E., "Creep Correlations in Alpha Solid Solutions of Aluminum,"
Transactions American Institute of Mining and Metallurgical Engineers, Vol. 9, 1952, pp. 959-964.
[10] Garofalo,F., Fundamentals of Creep Rupture of Metals, MacMillan, New York, 1965.
[11] DiCarlo, J. A. and Morscher, G. N., "Creep and Stress Relaxation Modeling of Polycrystalline
Ceramic Fibers," Failure Mechanisms in High Temperature Composite Materials, American Society
of Mechanical Engineers, New York, 1991, pp. 15-22.
[12] Boston,R. C., Greif, P. C., and Berman, M., "Conversational SAAM--An Interactive Program for
the Kinetic Analysis of Biological Systems," Computer Methods and Programs in Biomedicine,
Vol. 13, 1981, pp. 111-119.
[13] Berman, M., Shahn, E., and Weiss, M. R., "The Routine Fitting of Kinetic Data to Models--A
Mathematical Formalism for Digital Computers," Biophysics Journal, Vol. 2, 1962, pp. 275-287.
[14] Ramberg, C. E., Cruciani, G., Spear, K. E., Tressler, R. E., and Ramberg, Jr. C. F., "Passive
Oxidation of High Purity Silicon Carbide from 800 to 1100~ '' 1994, in preparation.
Costas Galiotis, l Varinder Chohan, t Alkis Paipetis, l and Cosmas
Vlattas
Interfacial Measurements and Fracture

Characteristics of Single and Multi-
Fiber Composites by Remote Laser
Raman Microscopy
REFERENCE: Galiotis, C., Chohan, V., Paipetis, A., and Vlattas, C., "Interracial Measure-
ments and Fracture Characteristics of Single and Multi-Fiber Composites by Remote
Laser Raman Microscopy," Fiber, Matrix, and Interface Properties, ASTM STP 1290, C. J.
Spragg and L. T. Drzal, Eds., American Society for Testing and Materials, 1996, pp. 19-33.
ABSTRACT: The micromechanics of stress transfer in single-fiber as well as multi-fiber

composites were investigated. The material system under investigation consisted of high modulus
carbon fibers embedded in an epoxy resin. The point-by-point stress in the fiber was measured
using the newly developed technique of remote laser Raman microscopy (ReRaM). The compos-
ite specimens were loaded incrementally in tension and the stress transfer profiles emanating
from fiber discontinuities, such as fiber breaks, were closely monitored. At each applied stress
level, the interfacial shear stress (ISS) distribution was derived by means of a balance of shear-
to-axial forces argument. In the single-carbon fiber/epoxy system, a maximum interfacial shear
stress of 30 MPa was reached at the point of first fiber fracture. In the multi-fiber carbon fiber/
epoxy system, the maximum interfacial shear stress developed at the point of first fiber fracture
was of approximately the same magnitude. Finally, the local stress concentration in the intact
fibers, as a result of an adjacent fiber fracture, was determined as a function of distance from
the fiber fracture for three distinct levels of applied stress. A maximum stress concentration of
approximately 1.2 was measured at the maximum applied composite strain level of 0.5%. This
value compared well with existing analytical models.
KEYWORDS: stress-transfer, interfacial shear stress, carbon fiber, epoxy resin, laser Raman
spectroscopy, stress concentration
Fundamental to the discussion of composite micromechanics is the representative volume

element (RVE) [1]. This is the smallest region or piece of material over which the stresses
and strains are macroscopically uniform. For perfect fiber/matrix bond, the RVE concept can
be employed to predict composite stiffnesses and hence Poisson's ratios. Microscopically,
however, the stresses and strains in the RVE are non-uniform owing to the heterogeneity of the
material. For example, in polymer-based composites under tension, fiber fracture(s) normally
precede composite failure and, therefore, the exact stress distribution within the RVE should
determine the mode of failure. This obviously leads to an enlargement of the RVE with applied
stress to accommodate one or more broken fibers. The axial stress in the fiber is zero at the
broken fiber site and reaches the applied stress value at transfer length, l, [2]. The interfacial
Reader, research student, and research fellows, respectively, Materials Department, Queen Mary &
Westfield College, University of London, El 4NS, United Kingdom.
19
Downloaded/printed
shear stress is maximum at the fiber fracture and decays to zero at lt. The ratio of the axial
stress in a fiber adjacent to ' r ' broken fibers to the fiber stress at infinity is called the "stress
concentration factor," Kr [3-6], and is directly related to the interfiber distance, the fiber and
matrix properties, the number of broken fibers, as well as the transfer length, l t . Since the
latter evidently depends upon the strength of the fiber/matrix bond, it follows that the fracture
characteristics in composites can be modified by careful control of the interface. Other important
parameters, which are also affected by the strength of the interface, are the off-axis shear and
the compressive strengths of continuous fiber composites.
A number of single-fiber test methods have been employed over the past 30 years to assess
the interfacial shear strength of certain composite systems and a large volume of sometimes
conflicting data has been produced. The most important of these tests are: (a) the pull-out
tests [7-9], (b) the fragmentation test [10,11], and (c) the microindentation test [12,13]. In
spite of the apparent simplicity of these tests, large discrepancies exist between results of
different tests on the same fiber/matrix system and between results of different laboratories
using the same type of micromechanical tests for the same material system [14]. This lack of
reliability of the micromechanical test methods can be attributed to problems related to the
experimental procedures and to the data reduction schemes employed to derive values of
nominal interfacial shear strength in each case [14].
More recently, experimental stress/strain transfer profiles along the fiber have been obtained
for a number of commercial fiber/matrix systems by employing the technique of Laser Raman
Spectroscopy (LRS) [2,15-17]. This technique is based on the fact that the Raman frequencies
of the atomic vibrations of commercial reinforcing fibers, such as aramid or carbon, are stress-
dependent [2]. Thus, unique Raman frequency versus stress calibration curves can be produced
for each fiber and are used to convert the Raman frequencies obtained along an embedded
fiber to values of axial stress (or strain). Furthermore, the stress-transfer profiles obtained
using the Raman technique can be converted to interfacial shear stress (ISS) profiles along
the length of the fiber by means of a straightforward balance of shear-to-axial forces argument
[2,15-17]. This leads to a simple analytical expression between the ISS, T(rx), and the gradient
d~rf/dx of the stress transfer profiles:
r dcrf (1)
'r(m = 2 dx
where r is the radius and x distance along the length of the fiber. The ISS profiles, T(rx), are
derived by (a) fitting a cubic spline to the raw data, (b) calculating the derivatives dtry/dx
from the spline equations and, finally, (c) employing equation (1) [18]. The procedure for
applying a cubic spline fit involved the addition of a triple knot at the location of each fracture
point to ensure continuity of the fitted curves and extra knots at the middle of each fragment
and at the position of half-maximum stress.
To date the stress transfer profiles and the corresponding ISS distributions have been obtained
for discontinuous as well as continuous fiber/epoxy systems [17]. For discontinuous aramid
fiber/epoxy geometries, the ISS in the elastic region is maximum at the ends and decays to
zero at the middle of the fiber [17]. When interfacial failure occurs, the ISS drops to zero at
the fiber ends, then builds to a maximum and finally decays to zero at the middle of the fiber
[17]. The distance over which the ISS builds to its maximum value provides a good estimate
of the zone of interfacial damage in each case [15]. In certain aramid/epoxy systems [18], the
strain-transfer profiles after interfacial failure appear trapezoidal in shape and the maximum
ISS fluctuates around a value of 45 MPa, then decays to zero. This behavior was attributed
to resin or interphase plasticity due to high von Mises stresses/strains developed near the fiber
GALIOTIS ET AL. ON INTERFACIAL MEASUREMENTS 21
ends and has been modeled satisfactorily by Guild et al. [19] by means of finite element
analysis (FEA).
The micromechanics of reinforcement during fiber fragmentation of a high-modulus carbon
fiber/epoxy system under tensile loading has been studied by Melanitis et al. [16]. The fiber
strain was monitored point-by-point at each level of load until full fragmentation. After fracture,
the strain in the fiber was found to build from the tips of the fiber breaks reaching a maximum
value in the middle of each fragment. The shape of the derived ISS profiles indicated that
interfacial failure in these systems initiates at the locus of fiber failure and propagates in the
direction of the fiber axis. The average maximum value of ISS value increases to approximately
40 MPa at 1.8% (full saturation) and then gradually decreases with further increments of
applied strain. In an attempt to ascertain whether interfacial failure occurs as a result of fiber
fracture, Melanitis et al. [20] employed LRS to monitor the strain transfer and the ISS
distribution along a short high-modulus carbon fiber/epoxy system. An examination of the
shape of the fiber strain profiles indicated that some sort of interfacial failure initiates at the
fiber ends just prior to fiber fracture. This clearly confirms that, in certain fragmentation
experiments, the energy required for interfaciat failure can potentially be reached prior to first
fiber fracture and, therefore, unstable propagation of interfacial damage is likely to take place
in tandem with fiber fracture [20].
In this work, the stress transfer profiles are examined in two identical material systems
consisting of (a) a single-fiber coupon geometry and (b) a standard tensile full-composite
coupon. In both cases, the stresses in the fibers are measured with remote laser Raman
microscopy (ReRaM), developed in our laboratory. In addition, the interfacial shear stress
distributions in the single-fiber as well as multi-fiber geometries are derived. Finally, the values
of the stress concentration in fibers adjacent to a fractured fiber site in a full unidirectional
coupon are measured for the first time and compared to those derived analytically. Thus, the
link between single-fiber model composites and full composites is established and the effect
of the interface on the fracture characteristics of the full composites is fully assessed.
Material Systems
Sized M40B-3K-40B | carbon fibers supplied by Soficar j were used in this study. The fibers
were 6.6 I~m in effective diameter and had a standard level of oxidative treatment. For both
the single-fiber and the multi-fiber composites, a two-part LY-HY 5052 | Ciba-Geigy2 epoxy
resin was used. The ratio of the resin to hardener was 4:1 by weight. The mixture was cured
at 80~ for 8 hours. The specimen preparation for the single-continuous-fiber model composite
(Fig. la) involved (a) applying a thin resin layer, (b) laying/aligning the fiber, and (c) topping
up the mold with resin. In the case of multi-fiber composites, a 0.3 by 0.3 m ~ prepreg sheet
was first made by filament-winding, then 4-ply unidirectional composites were made by
vacuum bagging. The volume fraction of the composite was measured to be approximately
57%. A sketch of the "window" of observation in the case of full composites is given in Fig. lb.
Remote Laser Raman Microscopy (ReRaM)

A remote laser Raman microprobe was built and tested in our lab. The new system, which
is shown schematically in Fig. 2, utilizes flexible fiber-optic cables for laser delivery and
collection. The use of the flexible fiber optics permits the operation of the microprobe in
horizontal, vertical, and multi-angle positions (Fig. 2). Thus, specimens of any size and shape,
1Sol]car (Toray) Industries, France.

2 Ciba-Geigy, Cambridge, England, UK.
FIG. l--Schematic illustrations of the (a) single-continuous-fiber coupon geometry and (b)
surface "window" from which Raman data were collected in the case of the multi-fiber 4-ply
unidirectional coupons.
and under a variety of different environments, can be easily interrogated. The input laser light
of an argon-ion laser excited at 514.5 nm is directed to the objective of a conventional
microscope via a single-mode fiber optic and then focused onto the specimen, which undergoes
mechanical deformation on a 20 kN screw-driven mechanical tester. The 180 ~ scattered light
is collected by another fiber optic and then is directed to a single monochromator for analysis
and Raman detection (Fig. 2).
The Raman Frequency Stress-Gage
The relationship between fiber Raman frequency and tensile stress was obtained by stressing
individual M40B-3K-40B | filaments on the mechanical tester according to the ASTM D3379-
75 procedure. The results for 11 different specimens are shown in Fig. 3. The slope of the
least-squares-fitted straight line was found to be - 3 . 0 cm-]/GPa. This value represents the
FIG. 2--The experimental set-up of the Remote Raman (ReRam) probe employed in these
experiments.
sensitivity of the stress sensor and was used to convert the Raman frequency into stress in all
subsequent composites measurements.
Stress Mapping and lnterfacial Shear Stress Distribution in the Single-Fiber Coupons
In the single-fiber geometry (Fig. la), the stress transfer mechanism is activated at a fiber
break. Prior to this, stress mapping along the whole length of the fiber is carried out to ascertain
FIG. 3--Raman frequency shift as a function of applied tensile stress for 11 different M4OB-3K-
40B | filaments.
uniform distribution of stress in the single fiber. When a first-fiber fracture is detected optically,
a fiber length on either side of the fracture of approximately 1.5 mm is interrogated with the
laser Raman microprobe.
In Fig. 4a the fiber stress as a function of distance along the fiber where the first fracture
occurred is presented. It is interesting to note that in the unloaded specimen the fiber is
subjected to a residual compressive stress of - 0 . 7 GPa as a result of the curing procedure.
The overall distribution of stress along a section near the middle of the fiber specimen is
satisfactory (Fig. 4a). The average fiber stress, at each level of tensile loading prior to fracture,
relates linearly to the applied composite strain (Fig. 4b). The observed local fluctuations of
fiber stress are partly due to experimental error and partly due to the non-uniformity of the
fiber cross-sectional area.
At an applied composite strain of 1.0%, a number of fiber fractures could be clearly detected
along the length of the fiber. In Fig. 5a the stress distribution along the selected window of
Raman measurements is presented. As can be seen, fiber stress builds up from the tip of the
fiber break and reaches a maximum value at the middle of each fragment. It is worth noting
Fiber Stress/GPa
~1~ AppliedStrain e: y~
e=0.00% 4~,e=0.2% !e:=0.4% "~e=-0.6%"]~.e=0.8
. . . . I . . . . I . . . . | . . . . I . . . . I , . . |
-2
22,5 23.5 24.5 25.5 26.5 27.5 28.5
(a) Distance along the FGL / mm
Average Fiber Stress/GPa

MEBSILYS052 1
Z High~Low
Average
.T
-' E " "
0 .--
"r
-2 9 9 9 ~ I , 9 9 9 I i i , 9 i . . . . ~ , . .
-0.25 0 0.25 0.5 0.75
(b) Applied Composite Strain / %

FIG. 4---Stress mapping prior to fiber fracture. (a) fiber stress along the selected .fiber gage
length, (b) average fiber stress as a function of composite applied strain. The error bars represent
2 standard deviations.
ISS / MPa
-40
9 . . J l . . . . I . . i I I . . . . I . . . . I , 9 , i I . . . . . . . .
24.5 25 25.5 26 26.5 27 27.5 28 28.5
(b) Distance along the FGL / m m

FIG. 5--Stress mapping at an applied composite strain of 1% (single-fiber coupon). (a) f b e r
stress along the selectedfiber gage length, (b) interfacial shear stress distribution along the selected
fiber gage length.
that the fiber stress at the location of fiber fracture is compressive and of magnitude equal to
the initial compressive residual stress (Fig. 4). Hence, the fiber at the moment of fracture must
be recoiling back to its initial compressed state. The fact that the initial stress state is retained
indicates that the interface can still support moderately high levels of shear stresses. This
behavior contrasts with other high-modulus (unsized) carbon fiber/epoxy systems where the
stress at the moment of fracture is fully released and, therefore 'Mode II' or mixed mode type
of fiber/matrix debonding is thought to be initiated [16]. The corresponding interfacial shear
stress distribution derived via equation (1), (seen in Fig. 5b), exhibits a small "knee" on either
side of the fracture point, then reaches a maximum of 26 __ 8 MPa before it decays to zero
at the middle of each fragment. At an applied composite strain of 1.4% (Fig. 6), the broad
fragments between 25.7-26.5 mm and 27.4-28 mm (Fig. 5a) have fractured further as a result
of the increase in applied stress. It is interesting to note that the "knee" of the ISS distribution
near all fiber fractures (Fig. 6b) is still present. This is indicative of the presence of some
form of interfaciat damage initiated from each fiber break. Finally, at the last strain level of
1.7% just prior to resin fracture, two additional fragments are observed (Fig. 7a). The maximum
value of ISS at that point is 29 +-- 3 MPa (Fig. 7b). The interfacial damage zone on each side
26 FIBER,MATRIX,AND INTERFACEPROPERTIES
ISS / MPa
Applied Slrain . 1.40%
40 9 )nterfacial Shear Stre~
20
-20
-40
24.5 25 25.5 26 26.5 27 27.5 28 28 5
(b) Distance along the FGL / mm

FIG. 6---Stress mapping at an applied composite strain of 1.4% (single-fiber coupon). (a) fiber
stress along the selected fiber gage length, (b) interface shear stress distribution along the selected
fiber gage length.
of the fracture point extends, on average, 100 Ixm along the fiber. No significant variations
of the damage zone with applied composite strain are observed, within experimental error, up
to an applied composite strain of 1.7%. Finally, as shown in Fig. 8, the average maximum
interfacial shear stress measured on the single M40B-3K-40B| 5052 | coupon, does
not vary appreciably with applied composite strain, within experimental error.
Stress Mapping, lnterfacial Shear Stress Distribution and Stress Concentrations in Full
Composites
The 4-ply unidirectional tensile coupons (according to ASTM D 3039-75) of M40B-3K-
40B| 5052 | composite were loaded incrementally up to fracture. The fiber stress
within a window of seven fibers (Fig. lb) was mapped point-by-point with the laser Raman
microprobe. Two such windows were employed; window A was employed at applied composite
strains of 0.4, 0.45, and 0.5%, whereas window B was mapped only at 0.5%. Stress/strain
mapping prior to fiber fracture along the fibers ensured that the fiber stress closely followed
the applied composite strain. After fiber fracture, the stress along the fractured fiber as well
FIG. 7--Stress mapping at an applied composite strain of 1. 7% (single-fiber coupon). (a) Fiber
stress along the selectedfiber gage length, (b) interracial shear stress distribution along the selected
fiber gage length.
Average Max. ISS / MPa

50
4O
30
20
10
I-l--II . . . . . . . . . . . . . . . . .
MEBS/LY5052 ]
9I. High~Low 0 AverageJ
I I I I
0
0.9 l.l 1.3 1.5 1.7 1.9
Applied Composite Strain/%
FIG. 8--Average maximum interfacial shear stress as a function of the composite applied strain.
as in three fibers situated on the same (surface) plane just above and just below the fiber
fracture (Fig. lb) was closely monitored. A representative stress mapping of one of the fractured
fibers (fiber 0-location A) at the applied composite strain level of 0.5% is shown in Fig. 9a.
As expected, the fiber stress drops to zero at the fiber fracture and then reaches a maximum
on either side of the fiber break value (Fig. 9a). The distance over which the stress is built
to the far field is defined as the transfer length, l,. As mentioned earlier, the interfacial shear
stress distribution is obtained by differentiating the raw stress data with respect to fiber distance
(Eq 1). However, it must be stated that in the case of full composites, this procedure may not
be as accurate as in the case of single-fiber composites due to the effect that possible fiber
breaks located on planes beneath the surface of the composite may have upon the examined
stress profiles. In future experiments an attempt will be made to examine the layer of fibers
beneath the selected window of stress mapping by means of microtoming or optical methods.
The present results show that, at a composite strain of 0.5%, the IS S distribution has a maximum
near the fiber break and then decays to zero on either side of the fiber break (Fig. 9b).
FIG. 9--Stress mapping of the fractured fiber at an applied composite strain of 0.5% (full
composite). (a)fiber stress along the selectedfiber gage length, (b) interfacial shear stress distribution
along the selected fiber gage length.
Fiber Stress/GPa
2.9
2.7
2.5
2.3 .*. 9 'l~' '%~.,, . .
2.1
1.9
1.7 f Fiber +IA; Applied Strain 0.50%]

9 Experimental J
1.5 . . * l , . , l l l n . . . . . . . . . . . . . . . . . . . . . . . r,,I .... I .... I
2.9
2.7
2.5
2.3
2.1 **=..o.,,,,,,,..~
1.9
f Fiber +2a.'Applied Strain 0.500/0]
1.7 L 9 Experimental J
1.5 '''~ .... I ......................... ,,,.i .... a .... x
2.9
2.7
2.5
2.3
2.1 ~ 9 9
1.9
1.7 f Fiber +3A: Applied Strain 0.50%1

( 9 Experimental j
1.5 ,..n . . . . ! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . n . . . . I . . . . n
-500 -400 -300 -200 -100 0 100 200 300 400 500
Distance along the FGL / ~ m
FIG. lO~Stress mapping of the adjacent fibers +1, +2 and +3 at location A for an applied
composite strain of 0.5% O~ullcomposite).
Representative stress distributions, along the adjacent fibers of the fractured fiber 0 (location
A) at an applied composite strain of 0.5% are shown in Figs. 10 and l 1. These distributions
clearly verify that the release of stress as a result of fracture in fiber 0 has now been transferred
to the neighboring fibers. The stress magnification is particularly apparent in fibers + 1 (Fig.
10) and - 1 (Fig. 11), which are the nearest neighbors of the broken fiber. The length over
which the stress increases to a maximum value has been termed as the positive affected length
(PAL) and is identical in magnitude ( ~ 200 i~m) with the transfer length It (Fig. 9). The ratio
of the stress in a fiber adjacent to r broken fibers to the far field stress has been termed the
stress concentration factor, Kr [3-6]. The stress-concentration factor K, at the fiber break
(~ = 0), for the experiment presented here, can be calculated from the expression:
Kr = - - (2)
(~apptied
Fiber Stress/GPa
2.9
2.7
2.5
2.3
.
2.1 ..-....'... 9
1.9
1.7
1.5
2.9
2.7
2.5
23 t.u.
9 e 9
21 , 9 -,,~r . . . . ._ v,a~,,s 9 ._
1.9 9
1.7 (Fiber -2A: Apphed Strain 0.50%]

~o Expermwntal J
1.5 .... It i,i I .... I ~. . . . . . . . . . . . . . . . . . Ill ii IIi Ill]'l'I ]
2.9
2.7
25
2.3
1.9
1.7 (Fiber - 3 A Applied Strain 0.50%)

ko Expermaental Jl
1.5 am i l l , , , I,,,,I . . . . . . . . . . . . . . . . . . . . . . . 1 .... 1 .... I
-500 -400 -300 -200 -100 0 100 200 300 400 500
Distance along the FGL / ~m

FIG. 1 l--Stress mapping of the adjacent fibers - 1 , - 2 and - 3 at location A for an applied
composite strain of 0.5% (full composite).
where cr~=~ is the stress of the intact fiber at the plane of fracture (6 = O) and O'applied is its
far field stress. Since the m a x i m u m stress was not always observed at the plane of fracture
(6 = 0), the stress concentration was also recorded from:
o-max
Kr - (3)
O'applied
where crmax is the m a x i m u m stress in the intact fiber regardless of its exact location with
reference to the plane of fiber fracture.
The results for the fiber fracture location A (Figs. 12 and 13) showed that there are no
significant differences between the Kr values calculated from either Eq 2 (Fig. 12) or Eq 3
(Fig. 13) for all strain levels. There is a strong indication that the Kr is related almost
exponentially to the center-to-center interfiber distance. M a x i m u m values of Kr of 1.22 and
1.15 have been measured at center-to-center interfiber distances of + 1 0 . 8 and - 1 1 . 8 Ixm,
1.3
1.2
o~
o= 1.1
(n
4
-35 -30 -25 -20 -15 -10 -5 5 10 15 20 25 30 35
Interfiber Distance I ~tm
FIG. 12--Maximum s t r e s s concentration, Kr, at ~ = O, as a function of interfiber distance for

composite strains of 0.4, 0.45, and 0.5%. For the latter, value of strain results have been collected
from two separate single-fiber fracture locations (A and B) in the composite.
respectively. On the contrary, the results 0f location B are consistently higher (K max = 1.30-1.35
as seen in Figs. 12 and 13) than those of location A and considerable differences between
tr~=~ and ~rmax are observed in Figs. 12 and 13. This clearly indicates that in this location a
second fiber fracture may exist at some distance z beneath fracture B and at a location x other
than ~ = 0.
It is now well established [21] that the tensile strength of composites is determined by the
strength distribution of the reinforcing fibers and the mechanics of load redistribution around
fiber breaks. Since the latter is a function of the strength of the interface, it becomes apparent
that a full understanding of the micromechanics of load transfer in broken, as well as between
E
ii 1.3
b
1.2
1.t,
4
-35 -30 -25 -20 -15 -10 -5 5 10 15 20 25 30 35
Interfiber Distance I ~m
FIG. 13--Maximum stress concentration, Kr, a t cr = max, as a function of interfber distance

for composite strains of 0.4, 0.45, and 0.5%. For the latter, value of strain results have been
collected from two separate single f b e r fracture locations (A and B) in the composite.
broken and unbroken fibers, is a necessary prerequisite for the understanding of fracture
characteristics of fibrous composites. Over the last two decades, a great deal of analytical
work has been done in an attempt to define Kr and hence predict the failure strength of
continuous-fibers composites. Sastry and Phoenix [21] and, much earlier, Hedgepeth [22]
employed shear lag models to predict a value of Kr. For a single-fiber fracture, these approaches
[21,22] predict a value of Kr of 1.33, which agrees reasonably well with the value of approxi-
mately 1.24 measured at an interfiber (center-to-center) distance of 10.8 txm (Fig. 12 for 0.5%
applied composite strain). However, it is worth mentioning that our results show a dependence
of Kr upon interfiber distance or, in other words, volume fraction, which is not accounted for
in these models [21,22]. A more detailed comparison of our experimental results with a variety
of existing analytical models will be presented elsewhere [23].
Conclusions
It has been shown that Remote Raman Microscopy can be employed to determine the stress
transfer profiles as well as the interfacial shear stress distribution in single and standard tensile
full-composite coupons. The values of the maximum ISS developed in a high-modulus carbon
fiber/epoxy system were of the order of 30 MPa for both systems. Stress mapping of full
composites provided information about the stress concentrations in these materials resulting
from fiber fracture(s). The value of maximum stress concentration for a single-fiber fracture
in the high modulus/epoxy system was measured to be 1.24. A dependence between the stress
concentration factor and interfiber distance was also found.
Acknowledgments
We would like to thank the Science and Engineering Research Council (EPSRC), the Defence
Research Agency (DRA) at Farnborough, the Commission of European Communities (Brite-
Euram program), and the British Aerospace (Sowerby, Filton) for financial support. Drs. M.
Pitkethly (DRA) and P. Marshall (BAe) are thanked for advice and encouragement during the
accomplishment of this work.
References
[1] Jones, R. M., Mechanics of Composite Materials, Hemisphere Publications, New York, 1975.
[2] Galiotis, C., Composites Science and Technology, Vol. 42, 1991, pp. 125-150.
[3] Fukuda, H. and Kawata, K., Fibre Science and Technology, Vol. 9, 1976, pp. 189-199.
[4] Eitan, A. and Wagner, H. D., Applied Physics Letters, Vol. 58, 1991, pp. 1033-1035.
[5] Harlow, D. G. and Phoenix, S. L., Journal of Composite Materials, Vol. 12, 1978, pp. 195-205.
[6] Atallah, K. M. and Galiotis, C., Composites, Vol. 24, 1993, pp. 635-642.
[7] Favre, J. P. and Merrine, M. C., International Journal of Adhesion and Adhesives, Vol. 1, 1981,
pp. 311-316.
[8] Miller, B., Gaur, U., and Hirt, D., Composites Science and Technology, Vol. 28, 1987, pp. 17-32.
[9] Piggott, M. R., Composites Science and Technology, Vol. 42, 1991, pp. 57-76.
[10] Kelly, A. and Tyson, W. R. D., Mechanics and Physics of Solids, Vol. 13, 1965, pp. 329-350.
[11] Drzal, L. T., Rich, M. J., and Lloyd, P. F., Journal of Adhesion, Vol. 16, 1982, pp. 1-30.
[12] Mandell, J. E, Grande, D. H., Tsiang, T. H., and McGarry, F. J., Composite Materials: Testing and
Design (7th Conference), ASTM STP 893, J. M. Whitney, Ed., American Society for Testing and
Materials, pp. 87-108.
[13] Chen, E. J. H. and Young, J. C., Composites Science and Technology, Vol. 42, 1991, pp. 189-206.
[14] Pitkethly, M. J. et al., Composites Science and Technology, Vol. 48, 1993, pp. 205-214.
[15] Jahankhani, H. and Galiotis, C., Journal of Composite Materials, Vol. 25, 1991, pp. 609-631.
[16] Melanitis, N., Galiotis, C., Tetlow, P. L., and Davies, C. K. L., Journal of Composite Materials,
Vol. 26, 1992, pp. 574-610.
[17] Galiotis, C., Composites Science and Technology, Vol. 48, 1993, pp. 15-28.
[18] Vlattas, C. and Galiotis, C., Developments in the Science and Technology of Composite Materials,
ECCM-5, A. R. Bunsell, J. F. Jamet, and A. Massiah, Eds., 1992, European Association of Composite
Materials, Bordeaux, France, pp. 415-420.
[19] Guild, E J., Vlattas, C., and Galiotis, C., Composite Science and Technology, Vol. 50, 1994,
pp. 319-332.
[20] Melanitis, N., Galiotis, C., Tetlow, P. L., and Davies, C. K. L., Journal of Materials Science, Vol.
28, 1993, pp. 1648-1654.
[21] Sastry, A. M. and Phoenix, S. L., Society for the Advancement of Materials and Process Engineering
Journal, Vol. 30, No. 4, 1994, pp. 61-67.
[22] Hedgepeth, J. M., "Stress Concentrations in Filamentary Structures," Technical Note D-882, NASA,
Washington D.C., 1961.
[23] Chohan, V. C., Paipetis, A., Marston, C., and Galiotis, C., submitted to Journal of Materials
Science, 1996.
Michael J. Pitkethly I
The Use of Interfacial Test Methods in

Composite Materials Development
REFERENCE: Pitkethly, M. J., "The Use of Interfacial Test Methods in Composite Materials
Development," Fiber, Matrix, and Interface Properties, ASTM STP 1290, C. J. Spragg and
L. T. Drzal, Eds., American Society for Testing and Materials, 1996, pp. 34-46.
ABSTRACT: Interfacial testing generates physical information about the fiber matrix interface
in a variety of forms. In recent years, there has been considerable discussion about the reliability
of the tests and the compatibility of the data generated. One question that has only marginally
been addressed is what is required from interfacial test methods to support the rest of the
composites community in the understanding and development of fiber-reinforced composite
materials. In this paper, the requirements of composite developers and designers are evaluated
and related to the information generated by existing test methods. Data are presented demonstrat-
ing the significant influence that interface has on macroscopic properties. The relationship of
this type of data to interfacial test techniques currently used is also addressed. Those interfacial
test methods in use fall into three basic types and each generates information under different
loading conditions. An attempt has been made to correlate these conditions to failure processes
that occur in full-scale composites. This has clarified where there are deficiencies in both the
test methods used and the type of data required. If interfacial property evaluation in composite
materials is to become an integral part of composite design, then there needs to be an improved
correlation between data generated and end-user requirements
KEYWORDS: composites, interfaces, surface treatments, failure, design, micro-mechanics,

test methods
The interaction between fibers and the matrix has generated curiosity and interest ever since
high-performance composites were first developed. During both fiber and matrix development,
these interactions need to be evaluated to ensure that sufficient bond strength is present to
enable composites to function properly. The interlaminar shear strength test has been used
extensively in this capacity. However, today there are a number of micro-mechanical test
methods that have been developed and continue to be developed that are more sensitive to
changes in interfacial adhesion. As the use of high-performance composites continues to grow
and new improved systems are developed, these evaluation techniques provide a useful research
tool. However, the design capability for composites has also progressed, and large complex
structures are now manufactured .from composites. The design engineer needs a different set
of data than the research scientist, and the designer needs to specify which materials are to
be used in his designs. This paper examines the differing needs of the scientist and engineer
and how information generated by the scientist is reflected in composite properties that the
designer needs to know.
Technology leader, Structural Materials Centre, Defense Research Agency, Farnborough Hants, GUI4
6TD, UK.
34
Downloaded/printed
PITKETHLY ON INTERFACIAL TEST METHODS 35
Information Requirements for Designers and Developers

The needs of the engineer designing composite components differs significantly from the
needs of the scientist developing new and improved materials for use in composite materials.
The engineer's aim is to design a component to undertake the task required as efficiently as
possible with the materials available. The engineer has only a marginal interest, if any, in the
microscopic behavior of the composite. His requirements concern macroscopic behavior and
are related to loading conditions that he has to cope with in his design. Typically, the information
required includes: ultimate strength and modulus of plain material; residual strength and
modulus after impact; open-hole and filled-hole strength under both tensile and compressive
loading and bearing strengths. This information is also required under a number of environmen-
tal conditions: cold (e.g., -25~ 50% RH); temperate (e.g., 20~ 50% RH); and, critically,
hot/wet (e.g., 40~ 96% RH). With this data, the engineer can then design the component.
The scientist, on the other hand, has greater interest in the microscopic behavior of the
composite. His aim is to increase the understanding of materials behavior to improve the
properties of existing materials and aid in the development of superior materials. He has to
respond to the needs of the designer, who knows what improved properties are needed. The
scientist wishes to understand the mechanisms that control and determine the macroscopic
properties of the composite. His interests, therefore, are directed towards the chemical, physical,
thermomechanical, and micro-mechanical properties of the fibers, matrix, and interface between
the two.
The scientist and engineer share the need to understand the failure process in composites.
The scientist has to understand why a material fails and must therefore understand damage
initiation and development at the microscopic level. One technique used to achieve that
understanding is that of micro-mechanical testing. It is essential that the information obtained
by the scientist from the model micro-mechanical tests is directly related to the information
the designer obtains from the macroscopic composite tests.
Characteristics of Composite Behavior Under Load

In this study, only composites reinforced with strong, stiff fibers were examined. However,
there are three different classes of composite materials that have distinctly different behavior
under load, determined by matrix material, The three classes are polymeric (PMC), metallic
(MMC), and ceramic (CMC) matrix composites. The relative properties of the fiber and matrix
play an important part in their subsequent behavior (Table 1). In the case of PMC, the fibers,
typically carbon or glass, have essentially elastic behavior. They are stiff and fail in a brittle
manner. The matrix materials are distinctly less stiff, or softer, than the fibers. They have
elastic behavior initially but can exhibit plastic deformation at high loads. The interface between
the fiber and bulk matrix material is indistinct in that the chemistry of the matrix around the
fiber can be different from that of the bulk. This has been called the interphase, over which
there is a gradient in properties. Even the boundary between the fiber and the interphase can
TABLE l--General differences between the three classes of composite.

Component PCM MMC CMC
Fiber stiff, elastic, brittle stiff, elastic, brittle soft, elastic, tough
Matrix soft, elastic/plastic soft, plastic/elastic stiff, elastic, brittle
Interface indistinct (interphase) indistinct (interphase), distinct
interlayers
be blurred if the matrix interphase material is absorbed into the outer layers of the fibers. In
MMC, the fibers again have essentially elastic behavior; they are stiff and fail in a brittle
manner. The matrix material is less stiff than the fibers and deforms plastically even at low
loads. The interface between the fiber and matrix can vary between a sharp distinction and
an indistinct interphase depending on the system under examination. An additional factor is
that there can also be reaction products formed at the interface producing an interlayer during
fabrication. CMC are the opposite of PMC and MMC; the matrix is usually stiffer and more
brittle than the fibers and has elastic behavior.
Under load, all composites start by behaving elastically, with the fibers and matrix moving
together, and if unloaded the composite reverts to its original shape with no increase in internal
energy. The load range over which this behavior occurs depends on the type of composite,
the volume fraction of fibers, and the orientation of the fibers to the loading axis. If the load
continues to be applied, irreversible changes occur which eventually lead to complete failure
of the composite. Many of these changes are similar in the different types of composite, but
can lead to distinctly different failure modes (Table 2).
The overall strength and stiffness of the composite is determined by the development of
damage at the microscopic level. Control of these damage development processes is key to
improving the performance of composite materials.
Micro-Mechanical Test Methods

In an effort to understand the mechanical behavior of composites at the microscopic level,
a number of test techniques have been and continue to be developed. The principal aim of
these tests has been to evaluate the shear strength of the fiber resin interface and any frictional
effects that may be present. These tests consist of both macro-mechanical and micro-mechanical
test methods, but all try to probe behavior at the individual fiber level. The macro-mechanical
methods consist of the interlaminar or short beam shear test and the 0o/90 ~ tensile test. The
former test aims to promote interracial failure in a shear mode and the latter produces transverse
cracking in the central 90 ~ plies, which has been related to interfacial shear strength. These
tests are not well controlled in that the microscopic loading conditions and failure modes are
not known. Failure can occur through interface failure, matrix yielding, or other undesired
causes. Realizing these drawbacks, a number of micro-mechanical test methods have been
developed. These fall into three generic classes: fragmentation tests [1-4]; fiber pull-out tests
[5-9]; and indentation tests [10] (Fig. 1).
Specimen configuration and loading conditions determine failure modes that may be present.
These may be desirable or undesirable, depending on the property that is being tested. A great
TABLE 2--Failure characteristics of the three classes of composite.

PCM MMC CMC
Microscopic fiber failure fiber failure fiber failure

matrix cracking matrix yield interface failure
interface failure interface failure matrix cracking
---cohesive ---cohesive -----cohesive
--adhesive
matrix yielding
Macroscopic delamination plastic deformation fiber pull-out
fiber pullout fiber pull-out matrix cracking
splitting
transverse ply cracking
Load Load
t t
f ~ Fibre Resin droplet Resin droplet

~ Resin
L _ J Holder ~ / u ~Fibres Fibre
P u l l o u t Test Microbundle Test Microdebond T e s t

$
R Microscope [~--]
Indentor II Objective k____/
Matrix
II1 1III/--I1 ;lllJ
Composite
Fragmentation Test Microindentation Test
FIG. 1--Micro-mechanical test techniques.
deal has been written about the practicality of undertaking these different tests and data
reduction techniques that can be used. Essentially, the principal characteristics of each test
produce different loading conditions (Table 3). Not all tests are appropriate for testing the
properties of the three classes of composites mentioned previously. The primary limitation of
using a particular test is the fabrication of the specimen. Specimens for all the tests can be
TABLE 3--Failure modes in micro-mechanical test methods?

Test Specimen Load Failure Modes Present
Fragmentation single fiber tension progressivedamage accumulation by fiber failure

local matrix yielding
local matrix cracking
fiber-matrix debonding
Pull-out single fiber tension interracial failure
---cohesive or adhesive
matrix yielding
fiber end surface loaded
Micro-droplet single fiber tension interfacial failure
-----cohesiveor adhesive
matrix yielding
nonuniform loading
Micro-bundle single fiber tension interfacial failure
in bundle ---cohesive or adhesive
matrix failure
fiber failure
Indentation single fiber compression interfacial failure
in composite fiber crushing
made using polymer matrix composites. Metallic matrix composites are currently restricted
to fragmentation and indentation tests. Ceramic matrix composites are confined to the indenta-
tion test, a convenient test method because of the low interfacial shear stresses in these systems.
As seen in Table 3, many of the tests evaluate the same failure modes, but the specimen
geometries make subtle but significant differences in what is being measured. We have examined
what data the engineer and scientist require, the characteristics of composites under load, and
the micro-mechanical test methods currently in use. The next step is to examine the data
obtained and assess whether fulfills the needs of both scientist and engineer.
Comparison of Model and Real Loading Conditions

Polymer composites have been in use for many years, but have become increasingly used
in high-technology applications over the past 15 years, particularly in the aerospace industry.
This reflects an increasing volume of information and understanding of these systems. Metal
matrix composites, however, have only very recently found applications suited to their particular
properties: that technology is still young and maturing. Ceramic matrix composites have yet
to find significant applications and still require considerable development before they will be
used in appreciable quantities. Since the majority of the experience and information available
is on polymer matrix composites, of which the fiber-reinforced composite is the highest
performing, it is on these materials that this paper will focus.
In practice, few composites in use today are made from unidirectionally oriented fibers. All
composite structures have fibers running in at least two significantly different directions,
usually three. The design of many composite structures is undertaken in such a way as to
ensure that the properties of the composite laminates used approximate an isotropic material,
when in fact composite laminates are intrinsically anistropic. Hence design data are often
produced using quasi-isotropic laminates. This is primarily due to the fact that laminates with
different lay-ups can have significantly different properties [11], and it is prohibitively expensive
to measure all potential laminate lay-ups.
However, even in composites with complex lay-ups exposed to complex multi-axial loading
conditions, there is often a predominate mode of failure that can be identified and related to
the simple failure mode of a unidirectional laminate under uniaxial loading, the basis for all
fractographic analysis. This simplifying assumption can be used to compare model micro-
mechanical test specimen loading and failure conditions with simple macroscopic composite
loading and failure conditions, and hence gage the relevance of the model tests to real
composites.
Tension
Tension failure is a progressive phenomenon that has been examined and modeled exten-
sively. Individual fibers throughout the composite fail when the applied load exceeds the
strength of the fiber at the weakest point where there is the severest flaw. This means that the
process is statistically based in the initial stages. The process continues with the load being
redistributed over the surrounding fibers until a group of fibers in close proximity have failed
and the redistributed load precipitates catastrophic failure. The size of this critical group varies,
but may be between three and ten fibers [12]. If this process is compared to micro-mechanical
tests, the obvious parallel test is the fragmentation test. In this test, loading conditions are
similarly uniaxial in the fiber direction, and multiple-fiber failure can occur. However, what
the test does not replicate is the influence of redistributed loads or the influence of adjacent
fibers on the uptake of load into the fiber away from the fracture. The other important factor
is that the failure of composite specimens occurs at about 1.5% strain. To obtain complete
PITKETHLY ON INTERFACIALTEST METHODS 39
fragmentation in the model test specimen, applied matrix strains on the order of 4 to 5% are
required. This can lead to the assumption that the tests are not compatible. However, work
using laser Raman to measure fiber strains indicates that the strains in the fibers at full
fragmentation are closer to those experienced in real composites than at first appears [13].
Often full fragmentation can occur by 2% fiber strain, although the apparent matrix strain is
greater. None of the other model test techniques assess multiple-fiber failure. Indeed, if fiber
failure occurs the test is deemed to be invalid.
An additional energy-absorbing mechanism that operates towards the end of the failure
process is that of crack growth and associated fiber pull-out. The three fiber pull-out tests are
the micro-mechanical tests that are obviously relevant to this situation. The loading situations
have both similarities and differences. In the tests the aim is to produce a uniform loading,
while in practice the loading is usually biased to one side as a result of the crack opening
perpendicular to the fiber direction. An attempt to consider the influence of adjacent fibers is
accounted for in one of the test methods (micro-bundle), but the orientation of the fibers is
not typical of a real composite. Therefore, the load distribution is different.
The single-fiber fragmentation test may be considered to be representative of the early
stages of composite tensile failure, but is not applicable to the damage accumulation stage
when fibers fail because of load redistribution. Fiber pull-out tests may be related to the final
stages of composite tensile failure, but lack the influence of surrounding fibers.
Compression
Compressive failure in composites is determined by the stability of the fibers. An instability

leads to buckling, which propagates across the composites as a kink or shear band. Fiber
fracture surfaces have a characteristic appearance with a mixture of tensile and compressive
features. The buckling of the fibers induces a tensile load across the interface on one side of
the fiber and a compressive load on the other. If failure is to occur at the interface, it is more
likely to happen on the tensile side. An important influence on compressive strength is the
shear strength of the matrix material. The greater the shear strength, the greater the compressive
strength. The only micro-mechanical test that applies the load in compression is the indentation
test. However, the gage length used is not sufficient to induce buckling in the fiber. The gage
length is purposely kept short to induce shear failure at the interface. The transverse loads on
the interface are compressive due to expansion of the fiber, and no transverse tensile loads
are induced, assuming that the fibers are aligned down the loading axis.
Shear Loading
Shear loading failures in composites have characteristic appearances depending on whether

they are Mode 1 or Mode 2 failures. Although it is difficult to generalize, Mode 1 failures
are often associated with delamination growth under compressive loading, Mode 2 with flexural
loading. The failure process can be by gradual accumulation of damage or by rapid crack
growth during impact for example. The majority of micro-mechanical model tests apply shear
loading in Mode 2, which is a sliding shear action. The only test that can evaluate interfacial
strength and be classified as a Mode 1 crack-opening mode is the 00/90 ~ transverse ply cracking
test, although this test is not a micro-mechanical test [14].
Although the fiber and matrix are being pulled in opposite directions, the characteristics of
Mode 2 shear failure are not present in the micro-mechanical tests. The composite Mode 2
failures are distinguished by shear cusps [15], while micro-mechanical failures are smooth
and more typical of a transverse tension failure. Therefore, when examining loading conditions
it might appear that there are Mode 2 shear forces at the interface, although the actual mode
of failure may not be a shear failure but a transverse tensile failure.
Effects of Notches
The presence of a notch in composite laminate is to provide a stress concentration which
induces failure at a load lower than that for a plain composite. In compression, the distortion
of the fibers around the hole provides an initiation site for damage to originate. However,
none of the micro-mechanical tests directly replicates the condition around a notch.
Impact and Residual Properties

Impact damage in composites varies depending on a number of factors, including: incident
energy; thickness of the composite; and shape of the indentor. The characteristics of damage
are usually similar, but differ in severity. Impact damage consists of fiber fractures, matrix
cracks, and delamination that spread out and away from the point of impact. Residual strength
of the composite is determined by the level of damage in the laminate. Failure mechanisms
are determined by loading. In compression, damage growth is related to delamination growth
and fiber buckling. In tension, failure is through crack growth and fiber pull-out. The only
mechanism that can be related to a micro-mechanical test is fiber pull-out. The single-fiber
pull-out test can provide information on the strength of the interface and the micro-droplet
test provides information concerning fiber matrix frictional properties.
How Much Interfacial Properties Affect Macroscopic Behavior in Composites

The scientist and engineer may differ in their interpretation of the effects interfacial properties
have on composite behavior. To one a specific change may be significant, while to the other
it is of no relevance at all. A distinction has to be made between fiber-surface treatments and
the application of sizes and coatings. The former attempts to control adhesion between the
fiber and matrix primarily adjusting the surface chemistry of the fiber. There may also be
some etching of the fiber surface. The latter attempts to control the properties by altering the
matrix adjacent to the fibers.
The majority of commercial fiber-surface treatments are applied by an electrochemical
process, but interest in plasma treatments as an alternative treatment is increasing. There has
been a considerable amount of literature published on the effects of varying surface treatments
on micro-mechanical properties. However, this has usually been confined to comparing
untreated fibers with standard commercially treated fibers as an adjunct to examining the test
technique. Interfacial shear strength values using all the tests have been acquired and also
vary depending on the technique used [16]. Effects of surface treatment include increasing
shear strength of the interface (Fig. 2) [17]. The question that needs to be addressed is what
effect the changes in surface treatment have on macroscopic properties and how they can be
related to microscopic changes.
Data on the influence of surface treatment on macroscopic properties are not widely pub-
lished, and the information that is available is often confined to interlaminar shear strength
(ILSS) data. This data usually follow the same trend as that of interfacial shear strength, but
is limited by the bulk shear strength of the matrix. This value is lower than the limiting
interphase shear strength found in micro-mechanical tests (Fig. 2) [17,18]. However, ILSS data
are not very informative about composite component behavior and are not design parameters.
Altering surface treatment affects the tensile strength of a composite depending on the
system under investigation, either reducing it by a small but significant amount [17] or
increasing it [19]. This is not an obvious deduction based on pull-out-type test methods. The
ILSS/IFSS MPa
300
250 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . - ~- ' ~ - -~
200 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
150 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
100~ " ~ -, ............. o- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50 .........................................................
i ~ILSS ~K-IFSS
0
0 1 O0 200 300 400 500 600 700 800
S u r f a c e Treatment Level C / m
FIG. 2--Effect of surface treatment level on interlaminar and interfacial shear strength for HTA
carbon fibers in an epoxy matrix [16].
fragmentation test does give an indication that this is a possibility since fragment length
decreases, indicating that the distance over which the load is transferred is reduced and,
therefore, any stress concentration in adjacent fibers should be greater.
Compressive strength is only marginally affected by an increased level of surface treatment
(Fig. 3). Other authors, however, have differing results using different composite systems [20].
In this case, because the increase in IFSS does not have a significant effect it must be concluded
that the strength of the fiber matrix bond is not the determining factor for compressive strength.
Other work has shown that fiber alignment and matrix shear strength are more dominant,
hence the usefulness of micro-mechanical tests to evaluate changes in compressive properties
are of limited value.
Lee [21] has conducted a series of experiments evaluating the influence of surface treatment
on the fracture toughness of a carbon-fiber epoxy composite system. He found no significant
difference in fracture toughness values between standard treated fibers and fibers treated with
a release agent under both Mode I and Mode II loading. His conclusion was that in these tests
the behavior of crack propagation is assessed, and this occurs mainly in the matrix resin.
Therefore, any interfacial influences are negated. Madhukar and Drzal [22] observed an effect
of surface treatment level at low levels of treatment, but also found that the mode of crack
propagation changes from an interfacial mode to a matrix-cracking mode. Because edge
delamination tests showed that composites with treated fibers had significantly lower strength
and toughness compared to composites with standard treated fibers, he concluded that crack
initiation was greatly influenced by interfacial adhesion and was more likely with a poor
interfacial bond. All micro-mechanical interfacial tests measure initiation and propagation
behavior except the fragmentation test, in which the interface role is activated by the fracture
of a fiber. Until that point shear loads should not appear across the interface unless there are
other influences.
Compressive S t r e n g t h (MPa)
1,400
Q
1,350
D []
[]
1,300
1,250 - . - El- . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1,200
0 10 20 30 40 50 60 70 80 90 O0
Fibre S u r f a c e T r e a t m e n t (%)
FIG. 3--Effect of surface treatment level on the unidirectional compressive strength of 43-750/
9106 carbon fiber epoxy composite.
The other important design criterion is residual strength after impact, particularly compressive
strength. The influence of surface-treatment level has been examined using ENKA HTA
fibers and a model resin system based on DEGBA and TGDDM epoxy resins [23]. Residual
compressive strength, as measured by the Boeing test method, decreases with increasing
impact-energy level, as would be expected (Fig. 4). In addition, composites with untreated
fibers were weaker than composites receiving standard treatment. Composites with overtreated
fibers ranged in strength between the two. These difference are directly related to the amount
of damage in the specimen after impact (Fig. 5). The amount of damage in the composites
with untreated fibers was significantly greater than that of the other composites. Micro-
mechanical tests indicated that the level of adhesion between the fiber and matrix for the
standard treated and overtreated fibers was similar and that damage areas in the impacted
composite were also similar.
Very little investigation has been undertaken on micro-mechanical fatigue tests, although
evaluation of interface characteristics under fatigue loading has been performed using the 0~
90 ~ tension test [24]. Static tests showed that the generation of transverse ply cracks was
significantly affected by the level of surface treatment (Fig. 6). Under fatigue loading significant
differences were also observed (Fig. 7), although the standard and double-treated composites
differed only at high numbers of cycles.
It should be observed that in all these tests the sensitivity of the property under investigation
to the related to the level of surface treatment decreases at higher levels. At high levels the
bond between the fiber and matrix is very good and other failure modes begin to dominate,
such as shear failure in the matrix.
Conclusions
The requirements of design engineers and development scientists differ. The former need
to know how materials behave when incorporated into a component while the latter need to
Strength MPa
500
= Standard
I
9 Untreated
4oo X 0 vertreated
X
300
200 X
X
1 O0
0 ~ _ _ . t ~ _ . _ ~ J ._L ~. L - -
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
Impact Energy J/mm

FIG. 4 Residual compressive strength as a function of impact energy for three different surface
treatments (HTA carbon fibers in a model epoxy resin) [18].
Damage Area mm 2(xl 000)

8
o Standard ]
9 Untreated
X Overtreated
6
0 0.5 1 1.5 2 2~5 3 3.5 4 4.5 5 5.5 6 6.5 7
Impact Energy J/ram

FIG.5 Damage area as a function of impact energy for three different surface treatments (HTA
carbon fibers in a model epoxy resin) [18].
CopyrightbyASTMInt'l(allrightsreserved);MonJan1618:33:32EST2012
Downloaded/printedby
(PDVSALosTeques)pursuanttoLicenseAgreement.Nofurtherreproductionsauthorized.
Transverse Crack Density (1/ram)

1.2
0 . 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Untreated
+ 50%
0.6
100%
* 200%
0.4
0.2 ....
0 ~"~ J"~ ~ J.~ ~.L ,.~ ..~ ~.~ ~ ~.l ~ ~"F ~ "I= t I I
0 100 200 300 400 500 600 700 800 900 1,000
Stress (MPa)
FIG. 6---Effect of surface treatment level on transverse ply crack density under monotonic loading
of T800H/3631 carbon~epoxy composites [19].
Transverse Crack Density (1/mm)

1.4
1.2
i
Untreated
-+- 50%
0.8
100%
-~ 200%
0.6
0.4
0.2
0-~ 9 ~ ~ ~ -~ , , ~ I
10-1 100 101 10 2 10 3 104 10 5 10 6
Cycles
FIG. 7--Effect of surface treatment level on transverse ply crack density under fatigue loading
of T800H/3.63. I carbon/epoxy composites [ 19].
PITKETHLY ON INTERFACIALTEST METHODS 45
understand the mechanisms involved in giving composites the properties they possess. The
micro-mechanical test methods currently being used and developed have been shown by many
workers to be sensitive to differences in interfacial adhesion. These changes can be correlated
with ILSS with reasonable accuracy. Tests on composites have shown that tensile strength,
edge delamination, damage area and residual compressive strength after impact, and transverse
cracking in static and fatigue loading are sensitive to the level of surface treatment. Fracture
toughness tests and compressive strength appear to be less sensitive. However, for all the tests,
both macroscopic and microscopic, sensitivity to changes in the level of surface treatment is
greatest at low levels and decreases as the level of surface treatment increases. At the level
of surface treatment used commercially it appears that the optimum level has been achieved,
since increasing the level has minimal effect. So, while micro-mechanics tests have a role to
play in increasing our understanding of the mechanisms of fiber/matrix adhesion, it is difficult
their use extending beyond the research laboratory to the design of composite structures. If
this step is taken and micro-mechanical information is used in the design process, then either
new tests need to be developed that mimic the situations in composite laminates or improved
modeling capability is required to relate micro-mechanical information to the behavior of
composite laminates.
References
[1] Drzal, L. T., Rich, M. J., Park, W., and Camping, J., "A Single Filament Technique for Determining
the Interfacial Shear Strength and Failure Mode in Composite Materials," Paper 20-C, 35th Annual
Technical Conference, Reinforced Plastics/Composites Institute, SPI, 1980.
[2] Favre, J. P. and Jacques, D., "Stress Transfer by Shear in Carbon Fibre Model Composites, Pt. t"
Journal of Material Science, Vol. 25, 1990, pp. 1373-1380.
[3] Wagner, H. D. and Eitan, A., "Interpretation of the Fracture Phenomenon in Single-Fibre Composite
Experiments," Applied Physics Letters, Vol. 20, 1990, pp. 1965-1967.
[4] Lacroix, T., Tillmans, B., Keunings, R., Desaeger, M., and Verpoest, I., "Modelling of the Critical
Fibre Length and Interfacial Debonding in the Fragmentation Testing of Polymer Composites,"
Composite Science and Technology, Vol. 43, 1992, pp. 379-388.
[5] Miller, B., Gaur, U., and Hirst, D. E., "Measurement and Mechanical Aspects of the Microbond
Pullout Technique for Obtaining Fibre/Resin Interfacial Shear Strength," Composite Science and
Technology, Vol. 42, 1991, pp. 207-219.
[6] Desarmot, G. and Favre, J. P., "Advances in Pullout Tests and Data Analysis," Composites Science
and Technology, Vol. 42, 1991, pp. 151-187.
[7] Favre, J. P. and Merrine, M-C., "Characterisation of Fibre/Resin Bonding in Composites Using a
Pull-Out Test," International Journal of Adhesion and Adhesives, Vol. 1, 1981, pp. 311-316.
[8] Hampe, A., "Grundlegende Untersuchungen an faserverst~'kten Polymeren," Amts-und Mitteilungs-
blatt der BAM, Vot. 18, 1988, pp. 3-7.
[9] Pitkethly, M. J. and Doble, J. B., "Characterising the Fibre/Matrix Interface of Carbon Fibre
Reinforced Composites Using a Single Fiber Pullout Test," Composites, Vol. 21, 1990, pp. 389-395.
[10] Mandell, J. E, Chen, J. H. and McGarry, E J., "A Microdebonding Test for In-Situ Assessment
of Fibre/Matrix Bond Strength in Composite Materials," International Journal of Adhesion and
Adhesives, Vol. 1, 1980, pp. 40-44.
[11] Curtis, P. T. and Baker, L. B., "The Effect of Stacking Sequence and Layer Thickness on the
Compressive Behaviour of Carbon Composite Laminates," Technical Report DRA/AS/MS/930001/
1, DRA Farnborough, Hants, UK, 1993.
[12] Batdorf, S. B., "Tensile Strength of Unidirectionally Reinforced Composites, Part 1," Journal of
Reinjbrced Plastics and Composites, Vol. 13, 1982, pp. 153-164.
[13] Melanitis, N., "Structure of Carbon Fibres, Interface and Properties by the Laser Raman Technique,"
Ph.D. Thesis, Department of Materials, Queen Mary College, University of London, London, 1991.
[14] Peters, P. M. W., "A New Method to Determine the Fibre-Matrix Bond Strength," Proceedings of
Interfacial Phenomena in Composite Materials '89, F. R. Jones, Ed., Butterworths, London, 1989,
pp. 59-62.
[15] Purslow, D., "Matrix Fractography of Fibre-Epoxy Composites," Technical Report 86046, RAE
Farnborough, Hants, UK, 1986.
[16] Pitkethly, M. J. et al., "A Round Robin Programme on Interfacial Test Methods," Composites
Science and Technology, Vol. 48, 1993, pp. 205-214.
[17] Pitkethly, M. J., "The Effect of Surface Treatment Levels on the Interfacial Properties of the Carbon
Fibre/Epoxy Resin Interface," Technical Report TR92073, DRA Farnborough, Hants, UK, 1993.
[18] Madhukar, M. S. and Drzal, L. T., "Fibre-Matrix Adhesion and Its Effect on Composite Properties.
I. In-plane and Interlaminar Shear Behaviour of Graphite/Epoxy Composites," Journal of Composite
Materials, Vol. 25, 1991, pp. 932-957.
II. Longitudinal (0~ and Transverse (90 ~ Tensile and Behaviour of Graphite/Epoxy Composites,"
Journal of Composite Materials, Vol. 25, 1991, pp. 958-991.
IV. Longitudinal (0~ Compressive Properties of Graphite/Epoxy Composites," Journal of Composite
Materials, Vol. 26, 1992, pp. 310-333.
[21] Lee, S. M., "Influence of Fibre/Matrix Interfacial Adhesion on Composite Fracture Behaviour,"
Composites Science and Technology, Vol. 43, 1992, pp. 317-327.
III. Mode I and Mode II Fracture Toughness of Graphite/Epoxy Composites," Journal of Composite
Materials, Vol. 26, 1992, pp. 936-968.
[23] Pitkethly, M. J., "The Correlation Between Chemical, Physico-Cbemical and Mechanical Properties
in Carbon Fibre Reinforced Composites," Technical Report TR92028, DRA Farnborough, Hants,
UK, 1992.
[24] Matsuhisa, Y. and King, J. E., "Effects of Fibre Surface Treatment and Test Temperature on
Monotonic and Fatigue Properties of Carbon Fibre Epoxy Cross Ply Laminates," Proceedings
ICCM-9, Vol. 5, 1993, pp. 145-152.
John A. Nairn, 1 Yung Ching Liu, l and Costas Galiotis 2
Analysis of Stress Transfer from the

Matrix to the Fiber Through an
Imperfect Interface: Application to
Raman Data and the Single-Fiber
Fragmentation Test
REFERENCE: Nairn, J. A., Liu, Y. C., and Galiotis, C., "Analysis of Stress Transfer from
the Matrix to the Fiber Through an Imperfect Interface: Application to Raman Data and
the Single-Fiber Fragmentation Test," Fiber, Matrix, and Interface Properties, ASTM STP
1290, C. J. Spragg and L. T. Drzal, Eds., American Society for Testing and Materials, 1996,
pp. 47-66.
ABSTRACT: A new analysis for stress transfer from the fiber to the matrix through an
imperfect interface was derived using a Bessel-Fourier series. The new analysis is specific to
the fragmentation test. It satisfies equilibrium and compatibility everywhere and satisfies most
boundary conditions exactly. The only approximation is the axial stress boundary condition in
the fiber which is satisfied in an averaged sense rather than exactly. Two important advantages
of the stress transfer analysis are that it can analyze anisotropic fibers and it can include imperfect
interfaces or interphases. Theoretical predictions of stress transfer were compared to experimental
Raman spectroscopy results. The predictions agree well with the experiments and they can be
used to measure interface properties. The stress transfer analysis was coupled with an interfacial
failure criterion to model the fragmentation test. The results of modeling fragmentation data as
a function of applied strain suggested new ways of interpreting fragmentation experiments and
cast doubt on the relevance of the commonly measured interfacial shear strength.
KEYWORDS: composites, stress transfer, Raman spectroscopy, fragmentation test, imperfect

interface, interphase, Bessel functions, axisymmetric stress analysis
In the single-fiber fragmentation test, a single fiber is embedded in a matrix and loaded in
tension until the fiber fractures into fragments [1-6]. The average fragment length and the
distribution of fragment lengths as a function of applied strain tell us something about the
fiber/matrix interface. But the question remains: What do the results tell us about the interface?
Furthermore, which experimental results do we need? How do we interpret those results? This
paper discusses a new stress analysis for stress transfer from the matrix to the fiber in a
fragmentation specimen. We use the analysis to model the fragmentation test and to make
suggestions about interpreting fragmentation test results.
At the first fiber break in a fragmentation test, the axial stress in the fiber is zero. Stress
transfer across the interface permits stress to return to the fiber which permits additional fiber
J Associate professor and graduate student, respectively, Material Science and Engineering, University
of Utah, Salt Lake City, UT 84112.
2 University of London, Queen Mary and Westfield College, Mile End Road, London El 4NS, UK.
47
Downloaded/printed
breaks with continued loading. The fragmentation process ceases when there is insufficient
stress transfer over the length of a fragment for the stress to reach the strength of the fiber.
To interpret the fragmentation test we must analyze the process of stress transfer from the
matrix to the fiber. One analysis method is to assume an elastic-plastic interface [7]. In the
post-yield condition the interface is assumed to have a constant shear stress and therefore a
linear rate of stress transfer into the fiber. With this model, the critical fragment length, or
the length when the fragmentation process ceases, is easily calculated to be [7]
l~ - gf(l~)r, (1)
Tic
where of(l,.) = the strength of the fiber for specimens of length equal to the critical length,
r~ = the fiber radius, and % = the critical interfacial shear stress. This elastic-plastic model
is sometimes incorporated into a two-zone model with a constant shear stress or frictional
stress in the yield zone and elastic stress transfer through a perfect interface in the undamaged
zone [8]. The elastic stress transfer is analyzed by one-dimensional, shear-lag methods. We
suggest that these models are too simplistic. The assumption of a constant interfacial shear
stress is unrealistic for many interfaces. The one-dimensional methods give no information
about radial stress and thus do a poor job of accounting for thermal stresses and frictional
effects. Finally, these models give only qualitative measures of interface properties, such as,
Tic, which are of limited practical use when the goal is to predict the role of the interface in
real laminates.
The first three-dimensional, axisymmetric analyses of stress transfer from a rod to an
elastic medium were done by Muki and Sternberg [9-11]. They treated the matrix as a
three-dimensional elastic continuum. The fiber is reduced to a fictitious one-dimensional
reinforcement over the cross-section of the actual fiber. They reduced the resulting problem
to an integro-differential equation that could be solved numerically. Since their original paper,
the method of replacing the fiber by a one-dimensional reinforced continuum has been widely
used [12-16]. Unfortunately, the method is not suitable for the fragmentation test. Instead of
transfer from a single break to an elastic medium, we want to analyze stress transfer in the
presence of periodic cracks. Instead of isotropic fibers, we want to analyze anisotropic fibers
such as carbon fibers or aramid fibers. Finally, replacing the fiber by a one-dimensional
reinforced continuum blurs the interface. This blurting gives poor results concerning interfacial
stresses [11] and makes it difficult to model an imperfect interface.
In this paper we give a three-dimensional axisymmetric analysis of the fragmentation test
by using Bessel-Fourier series stress functions. The new analysis is nearly exact. It obeys
equilibrium and compatibility everywhere. It obeys all boundary conditions except one. The
single approximation is that instead of the fiber axial stress being exactly zero at the fiber
break, only the average axial fiber stress is equal to zero. Some advantages of the analysis
are that it can handle anisotropic fibers, it can model an imperfect interface, and it gives
explicit results for all components of stress along the interface. After outlining the derivation
of the stress analysis, the predictions were compared to direct experimental results of stress
transfer using Raman spectroscopy. The experiments and predictions agreed well. Finally, we
used the analysis and a simple model for interfacial damage to model the fragmentation tests.
The results of the modeling led to suggestions about conducting and interpreting fragmenta-
tion experiments.
NAIRN ET AL. ON ANALYSIS OF STRESS TRANSFER 49
~0
!I
n
ao (~o ~o do
AT=T ~t.~ -~
~o AT=T AT--O
C. Far-Field D. Perturbation
A. Fragmentation B.One Fragment Stresses Stresses
FIG. I--A. In the fragmentation test, a single f i b e r e m b e d d e d in a large a m o u n t o f matrix
f r a g m e n t s into multiple small fragments. B. The boundary conditions on a cell containing a single
f i b e r f r a g m e n t o f length 1 extending f r o m z = - 1 / 2 to z = +1/2, ~rzz,1 = 0 on the f i b e r ends; ~=.l
= ~=.2 = 0 at z = +_ 1/2; w2 at z = +_ 1/2 is constant ("+_ con. ") or independent o f r; the temperature
differential is T. C. The far-field stresses are the stresses f o r an infinitely long, unbroken f i b e r in
an infinite matrix under an applied stress o f cro and temperature differential T. D. F o r perturbation
stresses, crzz.n = - ~ | on the f i b e r ends; "rrz3 = "rrz,2 = 0 at z = +- 1/2; w2 = 0 at z = +- 1/2; the
temperature differential is O.
Stress Analysis
Stress F u n c t i o n A p p r o a c h
Figure 1 shows the boundary conditions for analysis of the fragmentation specimen. Figure
1A shows a fragmentation specimen. A single fragment from that specimen is magnified in
Fig. lB. Because the fiber is broken at the two ends of the fragment, the axial stress (~rzz) and
shear stress (%0 on the fiber must be zero. By continuity of displacement from one fragment
zone to the next, the axial displacement (w) in the matrix at the fragment ends must be a
constant or independent of r. Symmetry dictates that the shear stress (Xrz) in the matrix at the
fragment ends must be zero. Finally, the entire specimen is subjected to a temperature differential
of T = T, - To where Ts is the specimen temperature and To is the stress-free temperature.
For convenience, we divide the problem into two problems illustrated in Figs. 1C and 1D.
Figure 1C shows the far-field problem or the stresses for an infinitely long transversely isotropic
cylinder embedded in an infinite matrix while the matrix is under an axial load of or0 and
temperature differential of T. The boundary conditions are:
~rzz.2(+-- l/2) = ~ro %z.t(+

- l/2) = "r,~,2(+-- l/2) = 0 w~( +- l/2) = w2(~_ t/2) = - 2 \Era + O~'n
(2)
The form of the far-field stresses is
O'zz,l -~" qloo 0-rr,l = O-co 0"00,1 = 0-~ Trz, l = 0

(3)
r20-~ r20-.
0-zz,2 = 0"0 0-rr,2 -- r2 0"00,2 - - r2 Trz,2 ~--- 0
where subscripts 1 and 2 = the fiber and the matrix, respectively, and r~ = the radius of the
fiber. By equating axial strains and the interfacial radial displacements in the fiber and matrix,
it is easy to solve for the constants t~. and 0-.
2rAVIn 1--Vr l + v m ] EA0-O+(2UA 1--Vr 1 + l,m

"'--~A ET --E~'J Em \EA(etr-etm)+( Er+ Em )(~ EAT
- (4)
~| 2v 2 1 - - V r l+v~
EA Er Em
0-0
--(l'tA - - P m ) ~ -F (IlA(OtA -- am) -Jr (IXT -- Ol.m))T
(5)
0-== 2v ] 1- Vr 1 + Vm
EA Er E,,
where E = modulus, v = Poisson's ratio, and ot = thermal expansion coefficient. The subscripts
A and Trefer to axial and transverse properties of the transversely isotropic fiber. The subscript
m refers to properties of the isotropic matrix.
Figure 1D shows the harder p r o b l e m - - t h e perturbation stresses. Here the matrix ends have
zero shear stress and zero axial displacement. The fiber ends are also under zero shear stress
but have a compressive axial stress of -~| Because temperature differential is included in
the far-field stresses, the stress analysis for the perturbation stresses uses T = 0. The boundary
conditions are:
0-zz, l( + 112) = - t ~ . "r~z,l(---1/2) = "r~z,2(+

- 112) = 0 wt(--- U2) = w2( + l/2) = 0 (6)
We define ~o,~ as the far-field stresses in component i (i = 1 or 2) and ~p.i as the perturbation
stresses in component i due to unit c o m p r e ~ i o n on the fiber; then, by superposit~n, the
stresses in the fragment in Fig. 1B are r = 0-0,i + ~-~p,i. We now proceed to find 0-p,i.
From Lekhnitskii [17], the stresses and displacements for an axisymmetric stress state in a
transversely isotropic fiber can be written as
= O (oz* b__~_~ + aOZXIr~

0-" -~'~r2 + r Or Oz2} (7)
O ( 02~ 1 O~ 02~'~
0-~176 b-~r2 + - - -Or + a--~-zg
(8)
(9)
~ ""~ = ~ \ 77rr~ + -r -Or + a Oz2 )
o (o2. lO. o2.

(lO)
"r,Z=~rr~3r2 + r--~r + a--~zZ J
b - 1 0 2 ~ I "r
U -- - - (11)
2Gr 3rOz
1 {02~ 10") (d + 2Vaa) O2~It

w =-~A ~ Or2 + -r--~r + (12)
g A 3;s 2
where u is radial displacement and the constants are
a--
--VA(1 + vr)
(13)
vi Er
t -- - -
Ea
b = (14)
HEr
EA
EA
-- - VA(1 + vT)
GA
C~ (15)
] --
v] Er
_ _
Ea
Ea
--(1 - vr)
2Gr
d- (16)
1 - --
EA
where G is shear modulus. The stress function 9 must satisfy the equation
v~v~,I,= o (17)
where the operators are defined by
32 1 0 1 02
V 2 = Or----
2+- (18)
r-dr + s~
- ~
and the constants, sj and s 2 are
a + c + ~/(a + c) 2 - 4 d
s2= (19)
2d
a + c - ~/(a + c) 2 - 4 d
s~= (20)
2d
The hoop displacement (V) and the unspecified shear stresses ('rr0 and'r0z) are zero in axisymme-
tric stress states. For an isotropic matrix, these equations reduce to the well-known result in
Love [18]:
a( a2•
Crrr = -~Z vmVZx - ~2r2) (21)
a( lax) (22)
~OO = "~Z VmV2X -- 7
(rz.. = ~z (2 - Vm)V=x - -~z2] (23)
a(
~z = ~rr (1 - Vm)V2X - - -~Z2/
o2• (24)
1 02•
u . . . . . (25)
2G Oraz
l[
w = ~-~ 2(1 - Vm)V2• - az2j
o2• (26)
where the equation for • is
V4X = 0 (27)
We begin by stating stress functions for the fiber and matrix that solve the perturbation
stresses in the fragmentation specimen almost exactly. The stress functions in the fiber (W)
and the matrix (X) for a fragment of length l with the origin for z = 0 at the center of the
fragment are
+ ~ sin kiz(blilo(f31ir) + b2ilo(f~2ir)) (28)

i=1
~o
X = A t z In r + ~ sin kiz(aoiKo(kir) + aukirKl(kir)) (29)

i=1
where Au, A i, bji, and aji are undetermined constants,
2i'rr ki
ki - and [3ji = - (30)
l sj
Both the fiber and matrix stress functions contain a Bessel-Fourier series. The fiber stress
function has modified Bessel functions of the first kind (lo(x)); the matrix stress function has
modified Bessel functions of the second kind (Ko(x) and Kl(x)). The fiber has only modified
Bessel functions of the first type, because the second type diverges as r approaches 0; the
matrix has only modified Bessel functions of the second type because the first type diverges
as r approaches ~. The Fourier series include only sin kiz terms due to symmetry about z = 0.
In deriving the stress functions, we started with the Bessel-Fourier series terms. Bessel-
Fourier series are a well-known method for analysis of cylinders with arbitrary transverse
loading [18], but, by themselves they cannot solve the fragmentation problem. If they are used
by themselves, the only possible solution is one that vanishes identically. We traced this
problem to the substitution of Fourier series into equations involving derivatives (as seen in
Eqs 7-12). Unfortunately, the derivative of a Fourier series is not necessarily the Fourier series
of the derivative of that function. In particular, stress functions based only on a Bessel-Fourier
series will never recover components of the stress state that are independent of z. To compensate
for this deficiency, we added more terms to the stress function. The leading terms in Eqs 28
and 29 were selected because they provide the desired z-independent components to the stresses
and they are consistent with the boundary conditions. The leading terms in the fiber are based
on the polynomial stress functions described by Lekhnitskii [17]. We chose polynomial stress
functions including terms up to z5. The final form of the additional terms in the fiber stress
function is dictated by symmetry and by a requirement that the shear stress at z = + I/2
remains zero. The leading terms in the matrix stress function were chosen to give z-independent
normal stresses and zero shear stress.
In the interest of brevity, we will not explicitly state all stresses, strains, and displacements.
Instead, we state only those needed to reproduce the results in this paper--the axial stresses,
shear stresses, radial stresses, axial strains, and radial displacements in the fiber and matrix:
ffzz.l = Bz + B3d~ 2 + cos ki~ bli - d Io([3li~) + b2i -- - d I0(f32i~) (31)

i= 1 $2
[ (~12 )
"rrz.l = i=~l sin ki~ bli - a
I1([~1i~)
.....sl + (~22- ) -I1([32i~)
b2i a - ] s2 J (32)
= Bl -- B3 ['(1 + vr)p2 + m
O'rr, l
I 3
(33)
+b2i((a- ~)I0([32i0 + ( 1 s2- b)/,(132i~)]]~]j
Owj 2va 1 B3 [ 2VAErp 2]

EZZ, 1 - -
a~ EA B ' + E A B 2 + 2 - G ~ (1-vr)~2+ 3EA J
SPa(1 q- 1)7") 1 d +~E21)Aa)

+ ~ cosk i
i=1 B3(-1) i EAk2 q- bli S~GA I~ (34)
+ b2,
1 = I- .40 bli(b - 1) ll(13ti~) +

_ _ b2i(b- -- 1) 1 (35)
+ cos -
9= ~ + Sl~ IS. ~ 13",i J
ee
Orzz,2 : ~ COS ki~[aoiKo(ki~ ) + ali(ki~Kl(ki~) -- 2(2 - vm)Ko(ki~))] (36)
i=l
oo
%z,Z = ~] sin ki~[aoi(-Kl(ki~)) + all(2(1 - vm)Kl(ki~) - ki~go(ki~))] (37)
i=1
O'rr'2 : "~ "1- [(

i=l cos kilo aoi -Ko(ki~) - - -
Kt(ki~)]
+ ali((1 - 2vm)Ko(ki~) - ki~Kt(klgg))] (38)

3
1 ]~ sin ki~
(39)
w2 = ~ ~=~ ~ [ao~Ko(kiO + a~i(ki~Kl(kiO - 4(1 - v,,)Ko(ki~))]
OW2 -- 1
ezz,2 = 0 ~ 2G,, i__~
j= cos ki~[aoiro(ki~) + ali(ki~Kt(ki~) - 4(1 - vm)Ko(ki~))] (40)
u2 = - 2 - - ~ -ff + ~ i=, cos kit ao, ~ + atil~Ko(ki~) (41)
These stresses, strains, and displacements were calculated by substituting the appropriate stress
function into one of Eqs 7 - 1 2 or Eqs 21-26 and then transforming into a dimensionless
coordinate system with ~2 = rlrl and ~ = zlrl. We defined some new constants (Bt, B2, and
B3) and redefined the remaining constants all in terms of the original constants in Eqs 28 and
29. For use in dimensionless equations, we have also redefined ki to be
2rliTr iTr
ki - - - - (42)
l p
where p = ll(2rO is the aspect ratio of the fragment. Finally, we note that the displacements
ui and wi are dimensionless displacements (Uactuat[rl or Wactual]rl).
The radial stress, axial strain, and radial displacement in the fiber all have a term in the
Fourier expansion summation that does not involve a Bessel function. These non-Bessel
function terms arise from the leading terms in the fiber stress function. These leading terms
give stresses and strains that are independent of ~ as well as stresses and strains that vary as
41. Because our analysis involves term-by-term equating of the coefficients of the Fourier
expansion, we must resolve the ~2 terms into a Fourier series. The constant parts of these
series are kept with the X-independent stresses. The Fourier parts are added to the Fourier
expansion summation. These non-Bessel function terms in the Fourier expansion are crucial
to our analysis procedure. Without them, our method would be unable to yield a nonzero
solution to the problem. With them, a solution is possible.
In the dimensionless coordinates, the fiber fragment extends from - 9 to p. Inspection of
the matrix shear stress, the matrix axial displacement, and the fiber shear stress show that
'r,z,2(~p) = 0 w2(-+-p)= 0 "r,z,l(-----p)= 0 (43)
Thus the stress state satisfies most of the boundary conditions exactly. The single remaining
boundary condition is (rzz,l (--+p) = - 1 . Because (rzz.l (---P) in Eq 31 is a function of 6, this
final boundary condition cannot be satisfied exactly, Instead, we satisfy it in the average or
we satisfy (O'zz,m(+P)) = - 1 . Integrating (rzz,t over the cross-section gives the average axial
stress in the fiber:
(trzz3) = Bz + - - ~ + i=l
~ cos ki~ bii ~ - d ~li ~ j (44)
Thus we seek to satisfy
-1 =Bz+T + '~' ( - 1 ) i bu -d ll(~li) +b2i (45)

,=l f~l, ] f32i J
Because the fiber stress function satisfies V 2 V ~ = 0 and the matrix stress function satisfies
V4X = 0, the stress state derived from those stress functions satisfies equilibrium and compatibil-
ity at all locations. From above, we see that the stresses satisfy the matrix boundary conditions
and fiber shear-stress boundary condition exactly. The single approximation concerns satisfying
the fiber axial-stress boundary condition in the average. The solution can be said to be exact
except in regions very near the fiber ends. An exact solution would show stress singularities
where the fiber crack tip meets the matrix. Our Fourier expansion solution cannot have
mathematical singularities, but our solution does correctly show large stress concentrations
near the crack tip. Very local to the crack tip, the stresses increase as the number of terms in
the Bessel-Fourier series increases. Slightly farther away from the crack tip, the stresses rapidly
converge and become independent of the number of terms in the Bessel-Fourier series. Finally,
we note that the fiber stress function does not reduce to the correct result for an isotropic
fiber. Similar results for isotropic fibers, however, can be generated by using the new fiber
stress function of
15vf 4 )
= A30Z3 + A32r2z + Aso (2 - vf)z5 - ---if-- r z - 5(1 - vf)r2z3
oo
+ ~] sin kiz(blflo(kir) + b2ikirIl(kir)) (46)

i-I
where vf is the Poisson's ratio of the isotropic fiber.

The solution for a full fragmentation specimen can be constructed by piecing together
individual solutions for each individual fiber fragment. Formally, such a solution only applies
to specimens with periodic fiber breaks. If the breaks are not periodic, the solution will still
obey boundary conditions, but it will include discontinuities in the matrix axial stress at the
junctions between fiber fragments. Thus, this solution contains an additional approximation
when applied to specimens with nonperiodic breaks.
Imperfect Interface
Most analyses of stress transfer from a matrix to a fiber resort to assuming a perfect interface
[9-14]. A definition of a perfect interface is that trr,, "r,z, u, and w are all continuous at r =
r~ or at ~ = 1. The goal of the fragmentation test and other interface tests, such as the pull-
out test [19,20] or microbond test [21,22], is to measure interface properties. It is self-evident
that analyses that assume a perfect interface will never be helpful in interpreting such tests.
The assumption of a perfect interface predetermines the interface properties in the analysis
and thus cannot be used to study the effect of varying those properties on experimental results.
We must therefore include some model for an imperfect interface into our fragmentation
test analysis.
The mathematician's approach to an imperfect interface is to relax interfacial continuity
conditions and allow discontinuities in ~r,, T~z,U, and w [23]. In linear theories, the discontinuities
are assumed to be linear functions of the interfacial stress state. In static loading conditions,
stress equilibrium requires trrr and "rrzto be continuous regardless of the quality of the interface.
The remaining discontinuities in u and w are functions of the interfacial stresses. Hashin put
this imperfect interface model into physical terms for composites [24,25]. The interface in
real composites is better described as an interface zone of finite dimension or an "interphase."
Within the interphase, the mechanical properties differ from both the fiber and the matrix. If
the interphase plays a role in composite properties, then it must allow the fiber to displace
relative to the matrix. Unfortunately, we are unlikely to have detailed information about the
thickness or the mechanical properties of the interphase. To make matters more complex, there
might be a gradient of mechanical properties across the interphase. Adding an interphase zone
with variable and probably unknown mechanical properties severely complicates our analysis
of the fragmentation test. Hashin proposed collapsing the 3-D interphase into a 2-D interface
[24,25]. The effect of the interphase is modeled by allowing displacement discontinuities at
the 2-D interface that are linearly related to the stress in each displacement direction. Denoting
interface discontinuities with square brackets, such as [u] = u2(1, ~) - ul(1, ~), a fiber interface
reduces to
[u] - O'rr,
~ )l(1, __ O'r~,2(1, ~) (47)
/9, D,
[w] - T~z,l(1,
~ ) _ "r~z,2(l, ~) (48)
D+ D+
IV] - Tr0'l(l' ~) -- Tr0"2(l' ~) (49)

Dt Dt
For axisymmetric stresses, "rr0 = [V] = 0 and we only need to consider [u] and [w]. The
terms D, and D+ are called interface parameters. A perfect interface is described by D, = D~
= ~; a disbonded interface is described by D, = D~ = 0; intermediate values describe an
imperfect interface. It is important to recognize that collapsing the interphase to a 2-D interface
does not mean we are ignoring the reality of an interphase. Instead, we are using a mathematical
trick that lumps the effect of the interface into two interface parameters D,, and D+. In principle,
D,, and D+ could be calculated for an interphase if its mechanical properties and dimensions
were known. Some sample calculations for planar interfaces are given in Ref. 23. For example,
consider an interphase of thickness ti on a fiber. The discontinuity in axial displacement across
the interphase is rl[w] (w here is a dimensionless displacement). A simple one-dimensional
analysis for shear strain in the interphase gives
rl[w]
~.i = ~ (50)
ti
Substituting the imperfect interface model for [w] and assuming the interphase shear stiffness
is Gi we find a physical interpretation for Ds as
rlGi
D~ = - - (51)
ti
Thus D~ has units of a modulus and is related to the effective shear stiffness of the interphase.
This one-dimensional picture probably oversimplifies the physical meaning of Ds. Ds is better
imagined as a measure of the ability of the interphase to transfer stress from the matrix back
into the fiber.
In the fragmentation test, thermal shrinkage of the matrix and differential Poisson's contrac-
tion between the fiber and the matrix both promote compressive radial stresses [26]. Calculation
of (r~(1, ~) from the above stress analysis confirms that Cfrr is compressive over the entire
interface except for extremely small zones near the fiber ends. Underdominantly compressive
radial stresses, the expression for [u] implies a negative discontinuity or implies the matrix
penetrates into the fiber. This situation illustrates a flaw in a linear imperfect interface. While
negative discontinuities are permissible for tangential displacement, they should be forbidden
for normal displacements. In the fragmentation test we have the fortunate simplification that
(rr~ is compressive. We can thus prevent negative discontinuities in normal displacements while
still using a linear theory simply by setting/9, = oo. We are not assuming the interface is
perfect in the radial direction; we are just exploiting the fact that O'rr is compressive and therefore
the quality of the interface in the radial direction should have no effect on fragmentation results.
The boundary conditions for the fragmentation test with an imperfect interface reduce to:
O'rr,,(1 , ~) = errs,2(1, /~) (52)
wrz,,(1, K) = ~'~.2(1, K) (53)
[u] = 0 (54)
f0; "rrz'l(l' ~) (55)

[w] = (Ezz'2 - Ezz'l)d~ - Ds
(crzz,~(--+P)) = - 1 (56)
These conditions are exactly enough to determine all the constants in the stress functions in
Eqs 28 and 29. The need for this result influenced the choice of the leading terms in the stress
functions. The constants are determined by equating, term-by-term, the terms in the Bessel-
Fourier series of each boundary condition. This process involves solving a 4 by 4 linear system
for each term in the Bessel-Fourier series and one 4 by 4 linear system for the constant terms.
These equations can be easily and rapidly solved on a personal computer.
For a sample calculation, we plotted stress transfer from a high-modulus (HM) carbon fiber
to a room-temperature cured epoxy matrix (see fiber and matrix properties in Table 1). Figure
2 plots the average axial stress in the HM carbon fiber for various values of Ds. The stress
has been normalized by dividing by ~=. For a perfect interface (Ds = oo) the stress transfers
back into the fiber in about 30 fiber diameters. Experimental results for stress transfer on the
same fiber and matrix show stress transfer in about 50 fiber diameters [27,28]. Our calculations
for a perfect interface with no parameters are in qualitative agreement with experimental
results. We note that stress transfer into anisotropic, high-modulus carbon fibers is considerably
slower than into isotropic glass fibers. The slower transfer into carbon fibers is mostly a
consequence of the higher modulus ratio between the fiber and the matrix [11,15]. As D~
decreases, the stress transfer zone gets longer. As D~ approaches zero, the axial stress approaches
zero, as it should for a disbonded interface. The specific value of D,, about 500 MPa, gives a
stress transfer zone of 50 fiber diameters which agrees exactly with experimental results [27,28].
Comparison to Raman Experiments

Certain Raman bands in carbon fibers shift when the fiber is under stress [29]. Several
investigators have used this shift to directly measure the stress in a carbon fiber embedded in
a matrix [27-32]. Here we consider a specific set of experiments on a short HM carbon fiber
embedded in a room-temperature cured epoxy [28]. The mechanical properties for the fiber
and matrix are given in Table 1. The short fibers were embedded in the matrix and the stress
in the fiber as a function of distance from the fiber end was measured at several levels of
applied strain. Details about the experimental procedures are given in Ref. 28. In this section
we compare two experimental results to predictions. The two experimental results are stress
transfer at low strain and stress transfer at high strain after evidence of interfacial damage.
The stress analysis in this paper is for a fragment that has broken away from a continuous
fiber. The experimental results are for an isolated end of an embedded short-fiber, which
1.0
0.8
~ \ \V-I
0 50 100 150 200

Distance (Fiber Diameters)
FIG. 2--Sample calculation of the effect of an imperfect interface on the average axial tensile
stress in an HM carbon fiber as a function of distance along the fiber for a fiber fragment that is
200.fiber diameters long. All stresses have been normalized to the far-field axial fiber stress of ~ .
D~ = oo is a perfect interface; D~ = 0 is a disbonded interface; intermediate D~ (in MPa) is an
imperfect interface.
differs from the end of a fragment from a continuous fiber. There is another consideration,
however, that argues in favor of comparing the analysis to experiments on short fibers rather
than experiments on fragmented fibers. To make specimens with fiber fragments, it is necessary
to embed a continuous fiber in a matrix and extend the specimen until the fiber begins to
fragment. The first fragments do not appear until the applied strain exceeds the strain to failure
of the fiber, or around 0.8% strain for HM fibers. Thus, it is impossible to do experiments on
fragments that have been loaded to less than 0.8% strain. If the fragmentation process itself
causes interfacial damage, then it is also impossible to study elastic transfer from the matrix
to the fiber. By elastic transfer, we mean stress transfer through an undamaged, although
perhaps still imperfect, interface. To get some results in the elastic regime, it is preferable to
do experiments on short-fiber specimens.
Figure 3 compares the Raman measurements of stress transfer at an applied strain of 0.6%
to the predictions of the Bessel-Fourier series stress analysis. We began by assuming a perfect
interface. The result in the Ds = ~ curve shows that the predictions agree reasonably well
with experimental results. The experimental stress transfer, however, is slightly slower than
the predictions. By varying D~, we found that Ds = 500 MPa agrees better with experimental
results. Comparison with other experiments in the low-strain regime (~ < 0.8%) [28] showed
that they could all fit the same value of Ds. We claim that D~ = 500 MPa provides a useful
measure of the quality of the interface between HM carbon fibers and the epoxy matrix.
Figure 4 shows experimental results at an applied strain of 1.0%. The stress transfer begins
slowly, but at about 50 fiber diameters from the end undergoes a discontinuous change in
slope. Comparison to Fig. 2 shows that no single value of D, can predict such a change in
stress transfer rate. Instead, we suggest that the high applied strain caused a damaged zone
in the vicinity of the fiber end. The damaged zone could be caused by events such as matrix
cracking, matrix yielding, interfacial debonding, or fiber splitting. Whatever the cause of the
damage, we believe its effect is to change the effective value of D~ near the fiber end. We
thus propose a two-zone model as illustrated in Fig. 5. Within the damaged zone that extends
a distance ra from the fiber end, the interface is characterized by a low value of Ds. The
stresses are found by analysis of a fragment of axial ratio p~ which is equal to the axial ratio
of the entire fragment. In the central portion of the fragment, the Ds value is high. The stresses
are found by analyzing a fragment of length P2 where Pz is chosen such that the average axial
3000
...... ;.., . . . . . . . 6.o, . . . . . . . . .
' s O [] D O 0 EIO0
ElM [] []D
=-- 2000
".~
<
~ 1000
iT. T
[]
100 200 300

Distance (Fiber Diameters)
FIG. 3~-A comparison of Raman measurements q[ stress transfer at an applied strain of 0.6%
to predictions using the Bessel-Fourier series. The Dr = ~z curve is the prediction for a perfect
interface. Setting D~ = 500 MPa gives a more accurate prediction of the stress transfer process.
5000 , , , , , . . . . . . , , , . . . . ~ ~ ~ . . . . . .
4000 ~ 1
g
3000 i
J~0~~2000 / ~ .... I ......... i ,~,I

"3,Cl
i~ 1000
B
OI 1oo 200. . . . . 300
Distance(FiberDiameters)
FIG. 4 - - A comparison of Raman measurements of stress transfer at an applied strain of 1.0%
to predictions using the Bessel-Fourier series. The predictions are for a two zone model with Ds
= 5 near the fiber break and Ds = 500 in the central portion of the fiber.
fiber stress is continuous at the edge of the damaged zone. The two-zone model is only an
approximate model because only the average fiber stress is continuous at the junction between
the two zones. The axial and shear stresses in the matrix and the shear stress at the interface
will be discontinuous. We believe the two-zone model still provides a useful model for stress
transfer in the presence of a damaged interface.
Figure 4 compares predictions of the two-zone model to experimental results at an applied
strain of 1.0%. The D, value in the center of the fiber represents stress transfer across an
undamaged interface. As such, it should be expected to be the same as the D, value measured
~t 92- I~
Pl r~
FIG. 5--A simple two-zone model for predicting stress transfer in the presence of interfacial
damage. The D~ value is low near the fiber ends and high in the middle of the fragment, ra is the
length of the damaged zone. The value of P2 is chosen such that the average axial stress in the
fiber is continuous at the edge of the damaged zone.
over the entire fiber fragment in low-strain experiments. We thus used D, = 500 MPa for the
central zone. The best-fit D~ for the damaged zone was D~ = 5 MPa. Using these two values
of Ds and a damaged zone size of rd = 48 fiber diameters, the analysis agrees well with
experimental results. Note that the high strain experiments have less scatter than the low strain
experiments because the absolute shifts in the Raman bands are larger. The D~ = 500 MPa
result obtained by fitting the central portion of the fiber is perhaps more reliable than the similar
result obtained by comparison to low-strain experiments. We also calculated the interfacial shear
stresses and the octahedral shear stress in the matrix using the two-zone model. The interfacial
shear stress agrees well with the Raman results [28]. The octahedral shear stress results suggest
that the mechanism for interfacial damage is matrix yielding. In other words, the size of the
damage zone used to fit the experiments in Fig. 4 coincides identically with the damage
zone required to prevent the octahedral shear stress from ever exceeding the condition for
matrix yielding.
Fragmentation Test Analysis
In the fragmentation test, a given fiber fragment is typically assumed to break into two
smaller fragments when the stress in the fiber reaches the strength of the fiber at the length
of the fragment. The peak stress in a fragment is always in the middle. The fragmentation
test is thus described by the equation
(crzz,l(i~ = 0)} = or.t,(1) (57)
Inserting the Bessel-Fourier series analysis for (crzz,l(~ = 0)) and inverting gives fragment
length or fiber break density ( = 1//) as a function of applied stress (~ro) and temperature
differential (T). In this section we compare predictions to experimental results. Unfortunately,
many fragmentation results report only the critical length or the fiber break density after the
fragmentation process ceases. To get more insight into the fragmentation process, it is preferable
to have data at sub-critical lengths or data that gives fiber break density as a function of
applied load [5,6,32]. Here we compare our predictions to experimental results by Huang and
Young [32]. They measured fiber break density as a function of applied strain for high-modulus,
PAN-based, T50 carbon fibers in a room-temperature-cured epoxy. The mechanical properties
of the fiber and matrix are given in Table 1.
The fiber length-strength relation was empirically found to be linear on a semi-log plot.
T50 fiber strength in GPa as a function of length in mm was measured to be [32J
TABLE l--Thermal and mechanical properties used for the fiber and the matrix.
Property HM or T50 Carbon Fibers Epoxy Matrix
Diameter, 2rt, jxm 7

Tensile Modulus, Ea or E.,, GPa 390 2.6
Transverse Modulus, Er, GPa 14
Axial Shear Modulus, GA o r Gm, GPa 20 0.97
Axial Poisson's Ratio, Va or v., 0.20 0.34
Transverse Poisson's Ratio, vr 0.25
Axial CTE, cxa or %,, (10-6/~ -0.36 40.0
Transverse CTE, c~r ( l 0-6/~ 18
tr,tt(/) = 3.75 - 0.817 log l (58)
Combining Eqs 57 and 58 and assuming the interface is characterized by a single value of
D~, we can predict the fragmentation process as a function of Ds. The results appear in Fig.
6. At low break density, the experimental results can be predicted with a range in D~. No
single value of Ds, however, can predict all experimental results. The experimental results
bend more quickly than any of the theoretical curves. Analogous to the Raman results, we
suggest that the fragmentation results at high crack density should be modeled with a two-
zone model instead of a single value of D,.
To conduct a two-zone analysis of the fragmentation test we need to know D~ in the central
portion of the fiber, know D~ near the fiber break, and have enough information about the
interfacial damage process to predict ra as a function of applied strain. Ds in the central zone
reflects stress transfer across an undamaged, although possibly imperfect, interface. We find
this value from the results at the lowest break densities. From Fig. 6, the value of D~ that fits
the results best without ever dropping below the results is Ds = 100 MPa. Ds in the damaged
zone influences the rate at which the predictions level off at high strain. For now, we treat it
as an adjustable parameter.
An important part of analyzing fragmentation tests is understanding the interfacial damage
process and predicting rg as a function of applied strain. We did not have specific information
about interfacial damage in these specimens. To guess a damage mechanism, we examined
the interfacial stresses for applied strains at the onset of damage or at the onset of deviations
from the predictions when Ds = 100 MPa. We noticed that the octahedral stress, defined as
1
"ro~, = ~ ,/(tr,r -- q=)2 + (tr= __ treO)2 + (tr= __ trO0)2 + 6,r2z
.)
(59)
predicted no yielding for low strains, but predicted matrix yielding for strains above the onset
of damage. In other words, "ror ~ v/2/3 try, where try = 75 MPa is the tensile yield stress of
the matrix [32], at the onset of damage. We therefore suggest that interfaciai damage in these
specimens was controlled by matrix yielding at the interface. For any given Ds values in the
1.00
J i A l Z J
0.75
E
0.50-
r~
t~
0.25
o I
o 0.25 0.50 0.75 1.00 1.25 1.50
Applied Strain (%)
FIG. 6---Comparison between fragmentation experiments on T50 carbon fibers in a room-tempera-
ture cured epoxy matrix and theoretical predictions assuming the interface can be characterized
by a single value of Ds.
damaged zone and the central zone, it is a simple calculation to plot % , along the interface
for various values of rd. For each applied strain, we selected rd such that the interfacial value
3"roct
o r - - ~ was lower than the tensile yield stress of the matrix. One complication of this analysis
was that for a finite number of terms in the Bessel-Fourier series, there were oscillations in
"roct. To average out the high-frequency oscillations, we looked at the average value of 'roct
over a small region at the tip of the damaged zone. We averaged "ro,.tover a region that was
2% of the fiber length. This region size was selected empirically; it was the smallest region
for which the prediction of rd as a function of applied strain was observed to be independent
of the number of terms used in the Bessel-Fourier series.
The results of predicting fragmentation results using a two-zone model are given in Fig. 7.
We assumed Ds = 100 MPa in the central zone, that the matrix tensile yield stress is 75
MPa, and that the interfacial damage process is controlled by matrix yielding. By fitting the
experimental data, we found D~ = 20-30 MPa in the damaged zone. The predictions agree
well with the experimental results. Although the fit may not be unique, we have used fragmenta-
tion data to extract three pieces of information about the interface. D~ = 100 MPa in the
central zone characterizes stress transfer across an undamaged interface. D~ = 20-30 MPa in
the damaged zone characterizes stress transfer across a damaged interface. The yield stress of
75 MPa characterizes the failure process at the interface.
Discussion
Analysis of both Raman experiments and fragmentation experiments makes it possible to

determine Ds in the "elastic" regime and D,. for a damaged interface. In Raman experiments,
the elastic D~ can be determined from either low-strain experiments or from the central portion
of high-strain experiments. The damage-zone De can be determined from results near the fiber
end in high-strain experiments. In the fragmentation test, the elastic Ds can be determined
from the low break-density results. The damage-zone Ds can be determined from the critical
length results or from the rate at which the break density curve levels off. Both values of Ds
1.00 , , , / i /
Ds, a = 30 MPa ~ / / ~ ~
0.75
E
"c~ 0.50
m 0.25-
0 I
0 0.25 90.50 0.75 1.00 1.25 1.50
AppliedStrain(%)
FIG. 7---Comparison between fragmentation experiments on TSO carbon fibers in a room-tempera-
ture cured epoxy matrix and theoretical predictions using a two-zone model. D~ = 100 MPa in the
central zone; D~.d (or Dr in the damaged zone) equal to various values; the size of the damaged
zone was assumed to be controlled by matrix yielding at a tensile yield stress of 75 MPa.
give information about the interface, but we suggest that the elastic Ds is much more relevant
to real laminates than the damage-zone D , Real laminates are not fragmented into critical
length fragments. Thus stress transfer in real laminates is controlled by the elastic D~. To the
extent that the interface plays a role in laminate properties, its effect should be determined
by the elastic D~ and much less influenced by the damage-zone Ds.
We note that most work using the fragmentation test reports only the critical length or the
corresponding interfacial shear stress. Because critical length is controlled by the damage-
zone D , it is possible that critical length results are not relevant to the role of the interface
in real laminates. The flaw of the fragmentation test is not in the experiment itself, rather in
the idea that only critical length is important. The fragmentation test can be fixed by doing
experiments at sub-critical length or at low break density [5,6,32]. The analysis in this paper
can be used to extract the elastic D~ from such experiments.
Complete modeling of the fragmentation test requires modeling the interfacial damage
process. Here we assumed a matrix yielding process based on the interfacial octahedral shear
stress. An advantage of the Bessel-Fourier series stress analysis is that we can calculate all
the components of interfacial stress and therefore do a realistic interfacial yielding analysis.
Some previous analyses assumed shear yielding and examined only the interfacial shear stress.
Because interfacial shear stress is zero at the fiber break, a simple shear yielding model is
inadequate. It predicts there will be no yielding near the fiber break. In contrast, the octahedral
shear stress it not zero. Because of large radial, hoop, and axial stresses near the fiber break,
the octahedral shear stress is nonzero and yielding is naturally predicted to extend from the
fiber break along the interface. A realistic analysis of interfacial yielding must account for
the important contributions of normal stresses to the yielding process.
We do not claim that matrix yielding is the damage process in all fragmentation experiments.
Other damage modes such as matrix cracking, fiber cohesive failure, frictional sliding, or
interfacial debonding might occur instead. In principle, the Bessel-Fourier series stress analysis
could be used to develop fragmentation models based on other interfacial damage processes.
For example, optical observation of the interface with glass fibers in certain matrices suggests
interfacial debonding as the fragmentation process proceeds [33]. We are in the process of
predicting such interracial damage by calculating the energy release rate for growth of an
interfacial debond. The debonding could be predicted to grow when that energy release rate
exceeds the interfacial toughness. Perhaps the fragmentation test can then be used to measure
interfacial toughness.
In principle, a fragmentation test as a function of applied strain can be used to extract
information about stress transfer across damaged or undamaged interfaces and about the
interfacial damage process. Practically speaking, however, there are too many variables to be
guaranteed of finding accurate information. A preferred approach is to couple fragmentation
experiments with independent experiments on the same specimens. One could, for example,
couple fragmentation experiments with Raman spectroscopy experiments. The Raman experi-
ments could be used to directly measure the elastic Ds and the damage-zone D , The fragmenta-
tion experiments could input these parameters and focus on the interfacial damage process.
Modeling the interfacial damage could be further coupled with optical microscopy to ensure
that the proposed failure models are realistic.
Acknowledgments
This work was supported in part by a grant from the Mechanics of Materials program at
the National Science Foundation (CMS-9401772) and in part by a Fulbright Fellowship for
author J. A. Nairn. We thank R. J. Young (University of Manchester, UK) tbr providing the
fragmentation test experimental results.
References
[1] Wadsworth, N. J. and Spilling, I., "Load Transfer from Broken Fibres in Composite Materials,"
British Journal of Applied Physics (J. Phys. D.), Vol. 1, 1968, pp. 1049-1058.
[2] Fraser, A. A., Ancker, F. H., and DiBenedetto, A. T., "A Computer Modeled Single Filament
Technique for Measuring Coupling and Sizing Effects in Fiber Reinforced Composites," Proceedings
of the 30th Conference SPI Reinforced Plastics Division, Vol. Section 22-A, 1975, pp. 1-13.
[3] Drzal, L. T., Rich, M. J., and Lloyd, P. E, "Adhesion of Graphite Fibers to Epoxy Matrices: I.
The Role of Fiber Surface Treatment," Journal of Adhesion, Vol. 16, 1983, pp. 1-30.
[4] Bascom, W. D. and Jensen, R. M,, "Stress Transfer in Single Fiber/Resin Tensile Tests," Journal
of Adhesion, Vol. 19, 1986, pp. 219-239.
[5] Wagner, H. D. and Eitan, A., "Interpretation of the Fragmentation Process Phenomenon in Single-
Filament Composite Experiments," Applied Physics Letters, Vol. 14, 1990, pp. 1965-1967.
[6] Yavin, B., Gallis, H. E., Scher, J., Eitan, A., and Wagner, H. D., "Continuous Monitoring of the
Fragmentation Phenomenon in Single Fiber Composite Materials," Polymer Composites, Vol. 12,
1991, pp. 436 a.A.6.
[7] Kelly, A. and Tyson, W. R., "Tensile Properties of Fibre-Reinforced Metals: Copper/Tungsten and
Copper/Molybdenum," Journal of Mechanics and Physics of Solids, Vol. 13, 1965, pp. 329-350.
[8] Lacroix, T., Tilmans, B., Keunings, R., Desaeger, M., and Verpoest, I., "Modeling of Critical
Fibre Length and Interracial Debonding in the Fragmentation Testing of Polymer Composites,"
Composites Science and Technology, Vol. 43, 1992, pp. 379-387.
[9] Muki, R. and Sternberg, E., "On the Diffusion of an Axial Load From an Infinite Cylindrical Bar
Embedded in an Elastic Medium," International Journal of Solids and Structures, Vol. 5, 1969,
pp. 587-605.
[10] Muki, R. and Sternberg, E., "Elastostatic Load-Transfer to a Half-Space from a Partially Embedded
Axially Loaded Rod," International Journal of Solids and Structues, Vol. 6, 1970, pp. 69-90.
[11] Muki, R. and Sternberg, E., "Load-Absorption by a Discontinuous Filament in a Fiber-Reinforced
Composite," Journal of Applied Mathematics and Physics, Vol. 22, 1971, pp. 809-824.
[12] Fowler, G. E and Sinclair, G. B., "The Longitudinal Harmonic Excitation of a Circular Bar
Embedded in an Elastic Half-Space," International Journal of Solids and Structures, Vol. 14, 1978,
pp. 999-1012.
[13] Rajapakse, R. K. N. D. and Shah, A. H., "On the Longitudinal Harmonic Motion of An Elastic
Bar Embedded in an Elastic Half Space," International Journal of Solids and Structures, Vol. 23,
1987, pp. 267-285.
[14] Pak, R. Y. S., "On the Flexure of a Partially Embedded Bar Under Lateral Loads," Journal of
Applied Mechanics, VoI. 56, 1989, pp. 263-269.
[15] Mbanefo, U. and Westman, R. A., "Axisymmetric Stress Analysis of a Broken, Debonded Fiber,"
Journal of Applied Mechanics, Vol. 57, 1990, pp. 654-660.
[16] Pak, R. Y. S., and Gobert, A. T., "Axisymmetric Problems of a Partially Embedded Rod with
Radial Deformation," International Journal of Solids and Structures, Vol. 39, 1993, pp. 1745-1759.
[17] Lekhnitski, S. G., Theory of an Anisotropic Body, MIR Publishers, Moscow, 1981.
[18] Love, A. E. H., A Treatise on the Mathematical Theory of Elasticity, Dover Publications, New
York, 1944.
[19] Penn, L. S. and Bowler, E. R., "A New Approach to Surface Energy Characterization for Adhesive
Performance Prediction," Surface and Interface Analysis, Vol. 3, 1981, pp. 161-164.
[20] Piggott, M. R., Chua, P. S., and Andison, D., "The Interface Between Glass and Carbon Fibers
and Thermosetting Polymers," Polymer Composites, Vol. 6, 1985, pp. 242-248.
[21] Miller, B., Muri, P., and Rebenfeld, L., "A Microbond Method for Determination of the Shear
Strength of a Fiber/Resin Interface," Composites Science and Technology, Vol. 28, 1987, pp. 17-32.
[22] Gaur, U. and Miller, B., "Microbond Method for Determination of the Shear Strength of a Fiber/
Resin Interface: Evaluation of Experimental Parameters," Composites Science and Technology,
Vol. 34, 1989, pp. 35-51.
[23] Martin, P. A., "Boundary Integral Equations for the Scattering of Elastic Waves by Elastic Inclusions
with Thin Interface Layers," Journal of Nondestructive Evaluation, Vol. 11, 1992, pp. 167-174.
[24] Hashin, Z., "Thermoelastic Properties of Fiber Composites With Imperfect Interface," Mechanics
of Materials, Vol. 8, 1990, pp. 333-348.
[25] Hashin, Z., "Composite Materials with Interphase: Thermoelastic and Inelastic Effects," Inelastic
Deformation of Composite Materials, G. J. Dvorak, Ed., Springer-Verlag, New York, 1990, pp. 3-34.
[26] Nairn, J. A., "A Variational Mechanics Analysis of the Stresses Around Breaks in Embedded
Fibers," Mechanics of Materials, Vol. 13, 1992, pp. 131-154.
[27] Malanitis, N., Galiotis, C., Tetlow, P. L., and Davies, C. K. L., "lnterfacial Shear Stress Distribution
in Model Composites Part 2: Fragmentation Studies on Carbon Fibre/Epoxy Systems," Journal of
Composite Materials, Vol. 26, 1992, pp. 574--610.
[28] Melanitis, N., Galiotis, C., Tetlow, P. L., and Davies, C. K. L., "Monitoring the Micromechanics
of Reinforcement in Carbon Fibre/Epoxy Resin Systems," Journal of Materials Science, Vol. 28,
1993, pp. 1648-1654.
[29] Robinson, I. M., Zakikhani, M., Day, R. J., Young, R. J., and Galiotis, C., "Strain Dependence of
the Raman Frequencies for Different Types of Carbon Fibres," Journal of Materials Science Letters,
Vol. 6, 1987, pp. 1212-1214.
[30] Schadler, L. S., Laird, C., Melanitis. N., Galiotis, C., and Figueroa, J. C., "Interracial Studies on
Carbon/Thermoplastic Model Composites Using Laser Raman Spectroscopy," Journal of Materials
Science, Vol. 27, 1992, pp. 1663-1671.
[31] Melanitis, N., Galiotis, C., Tetlow, P. L., and Davies, C. K. L., "Interfacial Shear Stress Distribution
in Model Composites: The Effect of Fiber Modulus," Composites, Vol. 24, 1993, pp. 459-466.
[32] Huang, Y, and Young, R. J., "Analysis of the Fragmentation Test for Carbon Fibre/Epoxy Model
Composites Using Raman Spectroscopy," Composites Science and Technology, submitted, 1994.
[33] Wagner, H. D., Nairn, J. A., and Detassis, M., "Toughness of Interfaces from Initial Fiber-Matrix
Debonding in a Single Fiber Composite Fragmentation Test," Applied Composite Materials, Vol.
2, pp. 107-117.
J. Jenny Yuan, 1 John M. Kennedy, 1 and Dan D. Edie 2
Modeling the Dynamic Response of the

Fiber/Matrix Interphase in Continuous
Fiber Composite Materials
REFERENCE: Yuan,J. J., Kennedy,J. M., and Edie, D. D., "Modeling the Dynamic Response
of the Fiber/Matrix Interphase in Continuous Fiber Composite Materials," Fiber, Matrix,
and Interface Properties ASTM STP 1290, C. J. Spragg and L. T. Drzal, Eds., American Society
for Testing and Materials, 1996, pp. 67-83.
ABSTRACT: A mathematical model was developed for the torsional dynamic response of
unidirectional composites with imperfect bonding between the fibers and matrix. The model is
basically an assemblage of homogeneous isotropic cylinders with different dynamic properties.
The interfacial region where the fibers and matrix are chemically and mechanically bonded
was modeled as a nonuniform interphase. The interphase, containing both a stiff region and a
soft region, was incorporated into a three-phase model. To study the effects of the interphase
on the dynamic response of the composites, the dynamic modulus of the interphase was varied
to simulate fiber/matrix bonding conditions. A low storage modulus and high loss modulus of
the interphase, corresponding to poor interracial bonding, were found to result in more damping.
The torsional dynamic response for a pitch-based carbon fiber/epoxy composite system with
different fiber volume fractions was evaluated with the current models. The dynamic response
of the composites predicted by the three-phase model agreed well with values measured by
torsional dynamic mechanical test, while the two-phase model with perfect bonding between
the fibers and matrix not only underestimated the loss modulus but failed to predict the trend
of the loss tangent of the composites. These observations indicated that a viscoelastic interphase
was significant and necessary to obtain good correlation with experimental results.
KEYWORDS: composites, fiber/matrix bonding, interphase, dynamic mechanical analysis

(DMA), micromechanics
The performance of composite materials depends not only on the fiber and matrix properties,
but also on the quality of the interfacial bond where the constituents interact chemically and
mechanically. The effectiveness of load transfer at the interface depends upon the extent of
chemical and mechanical bonding. In addition, the properties of the composite are further
compromised by structural defects such as voids, impurities, and microcracks which tend to
concentrate in the interfacial region. Therefore, an interphase with its own characteristics has
been widely considered in more comprehensive micromechanics models which are evaluating
the thermomechanical behavior of composites [1].
In general, the interphase is defined as the interfacial region extending from the point in
the fiber where the local properties begin to change from the bulk fiber properties to the point
in the matrix where the local properties are those of the bulk matrix [2-4]. The properties of the
i Doctoral student and associate professor, respectively, Department of Mechanical Engineering, Clem-
son University, Clemson, SC 29634-0921.
2 Professor, Department of Chemical Engineering, Clemson University, Clemson, SC 29634-0909.
67
Downloaded/printed
interphase are controlled by factors associated with constituents' properties and the composite
fabrication process. The resultant interphase, therefore, has a very complex chemical structure
and mechanical behavior. Its characteristics are best illustrated schematically as shown in Fig.
1 [1,2,4]. A brief description of the regions follows. 9
Fiber surface layer--The fiber surface layer is a thin layer just below the fiber-free surface.
The outermost surface layer may contain numerous defects because it is exposed to the reactive
environment during fiber processing. Also, the atomic and molecular composition of the fiber
surface can be quite different from the bulk fiber [2]. The surface area of the fiber is much
greater than simply its geometrical value of circumference times length because channels,
pores, and cracks increase the surface area.
Interface layer--This is the layer where the fiber and matrix chemically and mechanically
bond. The quality of the interfacial bond is affected by the fiber surface treatment. Fiber
surface treatment can remove part or all of the original weak fiber surface which may contain
impurities and defects, and it can also create chemically active functional groups which will
bond better to the matrix. The surface activity, however, can be altered during the composite-
forming process because of voids and the absorbed chemical species [2-4]. This layer is highly
stressed due to the property mismatch between the fiber and the matrix [5]. It is usually the
most complex layer in the interphase.
Interlayer--For carbon-fiber-reinforced epoxy composites, the interlayer usually consists
of a thin layer of epoxy applied as sizing. Drzal et al. [2] proposed that the interlayer had a
gradient in amine concentration decreasing from that in the bulk resin to a very low value
near the fiber surface. Since this interlayer has less than the stoichiometric amount of amine
necessary for full cure, the properties of this interlayer are usually stiffer than the bulk matrix.
Williams et al. [6], however, suggested that an interlayer exists which is severely mitigated
by the presence of the adjacent fiber. The matrix material that lies within 250 nm of the fiber
is substantially softer than the bulk matrix, while the matrix material adjacent to the fiber
exhibits an apparent high modulus due to constraining effects of the fiber. The properties of
the interlayer, therefore, appear to be controlled by both chemical effects and fiber constraint.
Constrained matrix layer--Between the interlayer and the bulk matrix is a region of pure
matrix material that is influenced by the fiber. The matrix in this region has a higher stiffness
than the bulk matrix because of reduced polymer chain mobility and changed packing density
of the macromolecules caused by fiber constraints and fiber-matrix interactions [7-9].
Constrained matrix layer

, ~ Interlayer
\- -- - ' - ~ Interface layer
~,...,,,~ - ---7 Fiber sudace layer
FIG. 1--Schematic representation of the interphase.
YUAN ET AL. ON THE DYNAMIC RESPONSE 69
The interphase consisting of the layers described is certainly inhomogeneous. It should not
be viewed as a simple surface where the fiber and matrix contact, but an interaction zone with
defects and stress concentration. The interphase, which may be of microscopic or macroscopic
dimensions, has different structure, chemical composition, and mechanical properties compared
to the bulk fiber and the bulk matrix. The mechanical properties of the interphase and thus
the composite are directly related to its physical characteristics and the quality of bonding in
this region.
Fiber-reinforced composites are known to exhibit viscoelastic behavior and thus energy
dissipation or damping. For composite materials, the major sources contributing to the overall
damping response are the fiber, the interphase, the matrix, defects, and damage. A viscoelastic
interphase will dissipate hysteretic energy during dynamic loading. Also, an interphase with
imperfect fiber/matrix bond and detects will have apparent damping due to friction mechanisms.
Torsional dynamic mechanical analysis (DMA) has been used to characterize the interfacial
bond strength for carbon fiber/epoxy composite systems [10-11]. Edie and Kennedy et al.
showed that damping is related to the level of fiber surface treatment. Composites without
fiber surface treatment had lower short-beam shear strengths and higher dynamic-loss moduli
than composites with surface-treated fibers. Their results imply that a decrease in the strength
of the fiber/matrix bond is reflected by an increase in the dynamic loss modulus. In general,
a weak interfacial bond will dissipate more energy than a strong interracial bond. Although
it has been suggested that any change in the fiber/matrix bond or the physical characteristics
of the interphase should be reflected as a change in the dynamic properties of the interphase,
the DMA properties of the composite have not been correlated to microscopic phenomena in
the interphase.
The major objective herein was to study the effect of imperfect fiber/matrix bonding on the
dynamic response of the composite materials with a comprehensive micromechanical model.
A nonuniform interphase was incorporated to model the complex interfacial region where
structure and material properties are influenced by the numerous factors discussed previously.
The dynamic properties of the interphase were varied to simulate different interfacial bond
conditions in the model. The predictions were then compared with torsional DMA experimental
results for a pitch-based carbon fiber/epoxy composite system.
Theoretical Analysis
Torsional Dynamic Mechanical Analysis
The torsional dynamic properties of a viscoelastic material can be determined by subjecting

the material to a low-amplitude, sinusoidal shear strain at frequencies lower than its resonance
frequency [12-13]. Under steady-state conditions, the resulting sinusoidal shear stress lags
behind the applied shear strain by a phase angle ~, because of the viscoelastic nature of the
composites. Usually, it is convenient to express the strain and the stress in complex notation as
~* = ~o exp(itot) (1)
T* = To exp[i(tot + ~)] (2)
where the shear strain and shear stress are taken to be the real parts of ~* and T* respectively;
~/o and To are the amplitudes of the respective strain and stress cycles; to is the angular
frequency; and t is time. The complex shear modulus, G*, a measure of the viscoelastic
behavior of the material, is thus given by
T* To
G* . . . . exp(iS) = I G lexp(cos ~ + i sin ~) (3)
~/* "Yo
where
G* = G' + iG*" (4)
and
G' = IG*lcos ~ (5)

G" = IG* I sin 5 (6)
which are commonly known as the storage modulus and loss modulus, respectively.
The above representation implies an effective resolution of the stress cycle into two compo-
nents, one of which has an amplitude To cos 8 and is in-phase with the strain cycle and the
other which has an amplitude To sin ~ and is 90 ~ out-of-phase with the strain cycle. It follows
that G' is the ratio between the amplitude of the in-phase component of stress to the strain
amplitude, and G" is the ratio of the amplitude of the out-of-phase stress component to the
strain amplitude [12].
During a dynamic shear test, the maximum energy stored in the specimen is achieved in a
quarter cycle while the energy stored over a full cycle is equal to zero [13]. The energy in a
quarter of cycle is given by
W =
f0 T -~ dt (7)
Substituting ~/* = "% sin ~ot and T* = To sin (tot + ~),
/cos B "rr sin 8)

W=To~o~ 2 + 4 (8)
The first term represents the elastic energy stored in the specimen. The second term is the
energy dissipated per quarter cycle. Defining Ws as the maximum elastic stored energy. AW
as the energy dissipated per complete cycle, AWIWs as the specified damping capacity, and
tan ~ as loss tangent.
To'% "Y~ G' (9)

w, = T c ~ = 7
AW = TrTo',/osin ~ = "rr',/EG" (lO)
AWlW~ = 2Tr tan ~ (11)
G~ nonuniform G~ . uniform
r r
FIG. 2--,4 three-phase model with a (a) nonuniform interphase and (b) uniform interphase.
Therefore, loss modulus and toss tangent are both measures of the damping in a viscoelas-
tic material.
Micromechanics Model
The torsional dynamic response of a unidirectional composite with imperfect fiber/matrix
bonding was studied with a multi-phase concentric cylinder model. The effects of imperfect
fiber/matrix bonding and defects in the interfacial region were studied by incorporating an
interphase into a three-phase model. The dynamic properties of the interphase were varied to
model different materials and interfacial bonding conditions. For comparison, the composite
was also modeled as a two-phase material with perfect bonding between the fiber and the matrix.
Although the interphase is inhomogeneous, its properties are often assumed to be uniform
because it is difficult to experimentally determine the material properties of the interphase.
Therefore, to model the complex nature of the interphase, a nonuniform interphase with two
uniform regions was incorporated into a three-phase model (Fig. 2a). The weak fiber surface
layer ~md constrained matrix material (see Fig. 1) in the interphase were modeled as a stiff
region with higher storage modulus but less damping than the matrix. The materials containing
poor fiber/matrix bonding, defects, and damage, which may lead to interfacial slippage, were
modeled as a softer region with lower storage modulus but more damping than the matrix.
The most challenging part of the study was to determine the thickness and the material
properties of the interphase because no experimental data are available for polymer matrix
composites. The dynamic moduli of the interphase can be deducted from the DMA experimental
results by iterating the thickness and dynamic properties of the interphase until the analytical
results converge to the experimental results. Then the dynamic moduli can be varied in the
model to study the influence of the interphase on the dynamic response of the composite.
The micromechanical model is basically an assemblage of homogeneous transversely iso-
tropic cylinders with different dynamic properties. The development follows that of linear
elastic materials [14,15]. The constitutive equations for the unidirectional composite are written
in terms of effective moduli and stresses and strains averaged over a representative volume
element. The bonding between phases is perfect.
The representative volume element for the micromechanics model is presented in Fig. 3.
The model consists of three phases: the fiber; a nonuniform interphase with a stiff region and
a soft region; and the matrix, having outer radii rf, ril, ri2, and rm, respectively. For the three-
phase model with a uniform interphase, the average properties of the stiff region and the soft
region are used for the interphase.
To derive the general elastic solution for an assemblage of composite cylinders, the two-
phase composite cylinder is subjected to homogeneous boundary displacements. In cylindrical
coordinates, these displacements can be written as
u: = w(~ 0) (12)
ur = 7oZ cos 0 (13)
u0 = - % z sin 0 (14)
The elastic solution for this assemblage of composite cylinders [17] is given by the following
equations for Phase 1
w~ O) = (Ar - r)~to cos 0 (15)

(I)(r,
r~z O) = GiA~o cos 0 (16)
"r~)(r, O) = -GiA~o sin 0 (17)
and for Phase 2
w(2)(r, O) = (Br + -C- - r ) ~/ocosO (18)

r
C
~)(r, O) = G 2 ( B - ~)~/o c o s 0 (19)
"t~)(r,
Z 0) = - G 2 B + ~/o sm 0 (20)
Where Gt and G2 are the shear moduli for Phase 1 and Phase 2, respectively. The constants
of integration for each cylinder are determined by the application of the following continuity
and displacement boundary conditions:
w(l)(rl, 0) = w(Z)(rt, 0) (21)

"r~)(rl, 0) = "r~)(rl, 0) (22)
w(r2, 0) = "/0r2 cos 0 (23)
By equating "r,z(r2, 0), calculated from the two-phase composite cylinder model, and "r,z(r2, 0)
for a homogeneous transversely isotropic cylinder, the axial shear modulus of the composite
G12 can be determined from
FIG. 3--The representative volume element of a three-phase model.

Gt2 = ~ G2 B - (24)
In terms of the elastic properties of Phase 1 and Phase 2, the shear modulus of a two-phase
composite cylinder can be written as
r-'/2 )
r2]
Gl2(tWOphase) = G2(1 +
G2
Gl - G2
1[ (rl]
l-
(25)
The three-phase composite cylinder model is analyzed by first modeling the two innermost
cylinders. Such as the fiber and the stiff interphase region, by an equivalent homogeneous
composite cylinder using the two-phase model solutions. The equivalent axial shear modulus
is given by
Gil = Glz(fiber, stiff interphase region) = Gil(l +

rill (26)
Gn
Gf- Gil
The first equivalent homogeneous composite cylinder is then combined with the soft interphase
region to form the second equivalent homogeneous composite cylinder. The equivalent axial
shear modulus for the second equivalent composite cylinder is given by
r,~]
G,q -- Gl2(fiber, i n t e r p h a s e ) = Gi2(1+ (ri, 2 (27)
Giz 1[
Gill - Ga
Finally, modeling the second equivalent composite cylinder and matrix phase by another
equivalent homogeneous composite cylinder, the equivalent axial shear modulus obtained by
the following equation is actually the effective axial shear modulus for the three-phase model
with a nonuniform interphase
+ \rm]
Gc = Gi2(fiber, interphase & matrix) = Gm(1 Gm 1 [- r,2 2 (28)
The dynamic response of the composites is determined by using the correspondence principle
to relate the dynamic properties to the linear elastic properties of the composites. The correspon-
dence principle states that the effectivecomplexmoduliof a viscoelasticheterogeneousspecimen
arefound by replacingthe phase elastic moduli withphase complexmoduli in the expression
for the effective elastic moduli of an associated heterogeneous elastic specimen, with identical
phase geometry [16,17]. Therefore, the dynamic shear modulus Go* is determined by substitut-
ing for the elastic phase moduli Gf, Gi, and G,, phase complex moduli Gf*, Gi*, and Gin*. The
general dynamic solution for the two-phase viscoelastic model is given by
(r,l~ ]
\r2] . (29)
G*2=G, 1+ G* + 1 El /r,'~2]
For the fiber-matrix two-phase composite model, the expression complex shear modulus of
the composite is
G* = G* 1 + \ rm] . (30)
G*~ +~[ l - (rf 2
,~,
c; o.
For the three-phase model with a uniform interphase, the expression for the complex shear
modulus of the composite is given as
G* = G* 1 + \E/ (31)
\ri/
G* 1 + G~ 1[
a~ _ a----7+-~ 1 -
For the three-phase model with a nonuniform interphase, the expression for the complex shear
modulus is given as
G~=
G~[I+ o, (~2
+,[, (~)]]
r[2 2
o,- +~,[l_ ({] -~

G~{l+ G~ r,,/ - G*
G~- G*
(32)
TABLE 1--Material properties used in the micromechanics model

Nonuniform
Interphase
Matrix Stiff Soft Uniform
Fiber Matrix Modified Region Region Interphase
G' GPa 22.0 1.30 1.70 7.0 1.25 4.0

G" GPa 0.0 0.022 0.022 0.02 0.06 0.04
Experimentally Measured Dynamic Properties

In a previous study, the torsional dynamic response of E-105 pitch-based carbon fiber/Epon
828 epoxy unidirectional laminates with different volume fractions were measured using a
Rbeometrics RDS-7700 dynamic mechanical spectrometer [10]. During the loading, torsional
cyclic twist was applied to thin rectangular samples; the torque and angle of twist were
measured during the test. The complex shear modulus of the composite was then determined
based on the theory of linear viscoelasticity. Measurements were made at room temperature
over a frequency range of 0.1 rad/s to 25 rad/s with maximum strain of 0.0005. Detailed
material properties, testing procedures, and results are presented in Ref. 10.

The results of Edie et al. [10] and Kennedy et al. [11] strongly suggest that the interphase
region between the fiber and matrix has a strong influence on the dynamic response of
unidirectional composites under torsional dynamic loading. Their results indicate that dynamic
properties increase with fiber volume fraction. Increase in storage modulus is expected because
the fibers are stiffer than the matrix, but increase in loss modulus is not expected because the
volume of matrix is decreasing. This response indicates that other phenomena such as interphase
response are responsible for the increase in loss modulus. The following results provide
analytical evidence that the interphase plays an important role in the dynamic response of
composites. In the micromechanics models, the fiber is assumed to be elastic and the matrix
and interphase are assumed to be viscoelastic. The determination of the material properties
of the interphase was based on the physical understanding of the interphase and its effect on
the overall dynamic response. A parametric study was conducted to enhance understanding
of the influence of each individual constituents on the overall dynamic mechanical properties,
with an emphasis on the significance of interphase property variations on the overall dynamic
property. The interphase with independent dynamic moduli was incorporated into the multi-
phase composite model. The predicted dynamic moduli of the composite were then compared
with experimental results, and the interphase properties were adjusted to match the experimental
results. The interphase properties, which provided the best correlation between the predictions
and experimental results, are listed in Table 1.
The effects of the interphase properties on the dynamic response of the composite are
evaluated by varying the dynamic moduli of the interphase using a three-phase model with a
uniform interphase (Fig. 2b). First, the normalized storage modulus of the interphase was
varied between 0.85 and 5.0 to simulate an interphase varying from soft to stiff (Figs. 4a,b).
The dynamic moduli of the interphase and the composite were normalized by the corresponding
modulus of the matrix. Compared to a composite with no interphase, the results show that a
relatively low interphase storage modulus results in a low composite storage modulus and a
larger composite loss modulus, although its effect is only profound for large fiber volume
12.0 R=5.0
R=G' i /G' m
! R=2.5
!
G" i = G" m n~
o;
!
J ! R=1.0
R=O.aS
Normalized 8.0
I!
I
Storage ,~'~
Modulus I t/Z~"
j '..."
Is" 9
G'/G' m Z s9 ",",p
j s g..
4.0
Gm
0.0 , , i , ,
0.0 0.2 0.4 0.6 0.8 1.0
Fiber Volume Fraction

FIG. 4--The effect of the interphase storage modulus on the dynamic response of the composite
(a) normalized storage modulus and (b) normalized loss modulus.
fractions. For the given composite system, its dynamic response is insensitive to the interphase
storage modulus except at very high fiber volume fractions (> 0.80), when the interphase
volume fraction is comparable to or even higher than that of the matrix. The normalized loss
modulus of the interphase was varied from 0.75 to 5.0 to simulate an interphase with low to
high damping (Figs. 5a,b). The interphase loss modulus has essentially no effect on the
composite storage modulus, while its effect on the loss modulus of the composite is significant.
The composite loss modulus is dramatically higher when the damping in the interphase is
large or the composite has a high fiber volume fraction (Fig. 5b). Therefore, composite damping
due to poor interfacial bonding is significantly influenced both by increased damping and
degraded stiffness. High fiber volume fractions magnify the effect of the interphase on composite
dynamic response, but this effect is only pronounced at unrealistic high fiber volume fractions.
Consequently, the effect of the interphase on the overall composite dynamic response is in
response to the increased viscous nature of the interphase.
The dynamic responses of composites with different fiber volume fractions were measured
in a previous study [I1]. The results were compared with predictions from the micromechanics
model. The predictions for all three models for the storage modulus of the composites were
low but close to the experimental data (Fig. 6a). The three models predicted a very different
loss modulus and tan ~ (Figs. 6b,c). Based on comparison with experimental data, the three-
phase model with a nonuniform interphase correctly predicted the relationship between dynamic
response and the fiber volume fraction of the composite. The two-phase model, which neglects
the effects of the interphase, not only severely underestimated the loss modulus but also gave
6.0-
R=G' i /G' m
G"i = G m
R=0.85
4.0
Normalized R=1.0
Loss
Modulus
I!
G"/G m I'~
v R=2.0
I
2.0 R=5.0
Gin".
0,0 l I I " I I
0.0 0.2 0.4 0.6 0.8 1.0

FIG. 4---Continued.
a wrong relationship between tan ~ and fiber volume fraction. Interestingly, the two-phase
model predicts that loss modulus increases with fiber volume fraction even though the volume
of viscous matrix is decreasing. This response is due to increased shear strain in the matrix
because of a higher volume of stiff fibers.
The nonuniform interphase appeared to be a more reasonable interphase model based on
the model predictions. The predictions between the three-phase model with a uniform and
a nonuniform interphase were very close for both the storage modulus and the loss modulus
of the composites. This was expected since the average values of the nonuniform interphase
were used for the uniform interphase. However, predictions of loss tangent from models
with the two types of interphase were quite different (Fig. 6c). The nonuniform interphase
consisting of stiff and soft regions was able to correctly predict the ascending trend of tan
with fiber volume fraction, while the uniform interphase could predict either an increase
or decrease in tan ~. The correct prediction could only be achieved for certain values of the
storage modulus and the loss modulus of the uniform interphase. Therefore, a nonuniform
interphase with a stiff region and a soft region is a more reasonable simplified interphase
model to study the effects of the interphase on the dynamic response of the composites. The
model with a nonuniform interphase will only predict a decrease in tan ~ with increasing
fiber volume fraction when both regions in the interphase have higher storage modulus than
the matrix.
Extensive study of the mechanisms of fiber/matrix bonding and the physical nature of the
interphase is necessary to construct a complete interphase model. It was found that discrepancies
between the micromechanics analysis and experimental data were still significant. For the
12.0
m
R=G" i /Gm
G' i = G' m
R=0.75
R=I.0
R=2.5
Normalized s.0
R=5.0
Storage
Modulus
G'/G'm
4.0
Gm
0.0
0.0 0.2 0.4 0.6 0,8 1.0

20.0
G'i = G'm R=5,0

I
15.0 I
I
I
I
Normalized I
Loss I
I
Modulus 10.0 I
I
/ R=2.5
G"/G" m I
/ /
f s ~
/ ,,,S
5.0 P
J
o"
.,
,s ,.,,, R=I,0
7 -"E. . . 5. . . . . - =0
Gin"
.0 '" 9 ! "' . . . . . . . 1' "'" ~ ! '~ '~
1.0 0,2 0.4 0.6 0.8 1,0

FIG. 5--The effect of the interphase loss modulus
o n the dynamic response of the composite (a)
normalized storage modulus and (b) normalize loss modulus.
16.0
o experimental data
--- two-phase
. . . . . three-phase (uniform interphase)
three-phase (nonuniform interphase)
12.0
Storage
Modulus
8.0
G' (GPa)
0 0
0 0 0
4.0
000
0.0
0.0 0.2 0.4 0.6 0.8 1.0

0.6 "] o experimental data
--- two-phase
0.5-
0.4
t " .... three-phase (uniform interphase)
three-phase (nonuniform interphase)
Loss
Modulus 0.3
G" (GPa)
0.2
o o
. 9 "9
O 0 0 ., , o " * '
0.1
0.01 , , ,' ,
0.0 0.2 0.4 0.6 0.8 1.0

FIG. 6--Comparison between experimental data and micromechanical model predictions of the
dynamic response of the composite (a) storage modulus, (b) loss modulus, and (c) loss tangent.
0.05 o experimental data

--- two-phase
. . . . . three-phase (uniform interphase) /
0.04 three-phase (nonuniform i n t e r p h a s ~
Loss
Tangent
tan S
0.03
o
0.02
0.01 ~~
0.00 I , ," , ,
0.0 0.2 0.4 0.6 0.8 1.0

FIG. 6~Continued.
given composite system, all three models underestimated the the dynamic response of the
composites. One possible reason is that the dynamic properties of the matrix are different in
a pure resin sample than in a composite. Studies have shown that the glass transition temperature
of the matrix is higher in a composite than in neat resin [18]. The difference in glass transition
temperature suggests that the dynamic properties of the matrix are different because of the
constraint of the fibers and the interactions between the fibers and matrix. Also, the curing
process for pure matrix is usually different from that of the composite. Based on this speculation,
a higher storage modulus was used for the matrix in the micromechanics model. The results
(Figs. 7a-c) show that the complex modulus predicted by the three-phase model with a
nonuniform interphase was much closer to the experimental results. Interestingly, adjusting
the dynamic properties of the matrix in the two-phase model may give closer predictions for
storage modulus and loss modulus, but it still does not correctly predict the ascending trend
of tan ~ fiber volume fractions. These observations indicated that a viscoelastic interphase
was significant and necessary for obtaining good correlation with experimental results.
Conclusion
A micromechanical model was developed to predict the torsional dynamic response of

composites with imperfect bonding. The imperfect fiber/matrix bond was evaluated by varying
the dynamic properties of an interphase. Based on a comparison of the predictions from

different models with torsional DMA experimental results, the following can be concluded:
(1) The damping of the interphase contributes significantly to the torsional dynamic proper-
ties. The effect of the interphase on the dynamic response of the composite is more significant
in composites with high fiber volume fractions. Increased loss modulus and decreased storage
modulus of the interphase, corresponding to the imperfect fiber/matrix bonding, result in
more damping.
(2) A three-phase model incorporating a nonuniform interphase consisting of a stiff region
and a soft region is reasonable to study the dynamic response of unidirectional composites.
The two-phase model with perfect fiber/matrix bonding cannot correctly predict the dynamic
response of composites. A three-phase model with a uniform interphase sometimes may not
predict the correct trend for loss tangent.
(3) Agreement between experimental data and model predictions was reasonably good.
Discrepancies are expected because the interphase is extremely complicated and thermal
residual stresses were not considered.
8.0 ~ o experimental data

- - - three-phase
. . . . . two-phase (matrix property modified) /.
three-phase (matrix property modified) /,-"
6.0'
0 0~_o'* D
~ps- S/
Storage " ss/
Modulus 0 0 ~' sp
4.0' / js "pS
G' (GPa) ~s S
2.0'
0.0 I I I I I I
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

FIG. 7--Comparison between experimental data and predictions given by a three-phase model
with modified matrix properties of the dynamic response of the composites (a) storage modulus,
(b) loss modulus, and (c) loss tangent.
0.3 o experimentaldata
- - - three-phase
..... two-phase(matrix property modified)
three-phase (matrix property modified)
Loss 0.2 /
Modulus
G" (GPa)
0 0 7 s S/
0.1 ~~~.~.~...."s].
o o~....~_- ~'- ...................
0.0 , I , . , ~' ,
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

0.04 o experimentaldata
- - - three-phase
..... two-phase(matrix property modified) , A t
three-phase (matrix property changed) /

0.03
o ~o .9
Loss 0 0 J ~.~" ~
Tangent
tan 0.02
0.01
0.00 I l 11 I I | 9
0.~ 0.1 0.2 0.3 0.4 0.5 0.6 0.7

FIG. 7--Continued,
References
[1] Jayaraman, K., Reifsnider, K. L., and Swain, R. E., "Elastic and Thermal Effects in the Interphase:
Part I. Comments on the Characterization Methods," Journal of Composites Technology & Research,
Vol. 15, No. 1, Spring 1993, pp. 3-13.
[2] Drzal, L. T., Rich, M. J., Koeing, M. E, and Lloyd, P. E, "Adhesion of Graphite Fibers to Epoxy
Matrices: II. The Effect of Fiber Finish," Journal of Adhesion, Vol. 16, 1982, pp. 133-152.
[3] Drzal, L. T., "The Effect of Polymer Matrix Mechanical Properties on the Fiber-Matrix Interfacial
Shear Strength," Materials Science and Engineering, A126, 1990, pp. 289-293.
[4] Hughes, J. D. H., "The Carbon Fiber/Epoxy Interface--A Review," Composites Science and
[5] Adams, D. F., "Micromechanical Predictions/Experimental Correlations of the Influ.ence of the
Interface on the Mechanical and Physical Properties of a Unidirectional Composite," Composite
Interfaces, H. Ishida and J. L. Koenig, Eds., North Holland, NY, 1986, pp. 351-363.
[6] Williams, J. G., Donnellau, M. E., James, M. R., and Morris, W. L., "Properties of the Interphase
in Organic Matrix Composites," Materials Science and Engineering, A126, 1990, pp. 305-312.
[7] Schultz, J. and Nardin, M., "Interfacial Adhesion, Interphase Formation and Mechanical Properties
of Single Fiber Polymer Based Composites," Controlled Interphases in Composite Materials, H.
Ishida, Ed., 1990, pp. 561-567.
[8] Thomason, J. L., "Investigation of Composite Interphase Using Dynamic Mechanical Analysis:
Artifacts and Reality," Polymer Composites, April 1990, Vol. 11, No. 2, pp. 105-113.
[9] Thomason, J. L. and Morsink, J. B. W., "Investigation of the Interphase in Glass Fiber Reinforced
Epoxy Composites," Interfaces in Polymer, Ceramic, and Metal Matrix Composites, H. Ishida,
Ed., 1988, pp. 503-512.
[10] Edie, D. D., Kennedy, J. M., Cano, R. J., and Ross, R. A., "Evaluating Surface Treatment Effects
on Interracial Bond Strength Using Dynamic Mechanical Analysis," Composite Materials: Fatigue
and Fracture, STP 1156, W. W. Stinchcomb and N. E. Ashbaugh, Eds., American Society for
Testing and Materials, Philadelphia, 1993, pp. 419-429.
[11] Kennedy, J. M., Edie, D. D., Banerjee, A., and Cano, R. J., "Characterization of InterfaciaI-Bond
Strength by Dynamic Analysis," Journal of Composite Materials, Vol. 26, No. 6, 1992, pp. 869-882.
[12] Read• B. E. and Dean• G. D.• The Determinati•n •f Dynamic Properties •f P•lymers and C•mp•sites•
John Wiley & Sons, New York, 1987.
[13] Ferry, J. D., Viscoelastic Properties of Polymers, John Wiley & Sons, New York, 1970.
[14] Hashin, Z. and Rosen, B. W., "The Elastic Moduli of Fiber-Reinforced Materials," Journal of
Applied Mechanics, Vol. 31, 1964, pp. 223-232.
[15] Hashin, Z., "Theory of Fiber-Reinforced Materials," Final Report, Contract NAS1-8818, Nov.
1970; NASA CR 1974, 1972.
[16] Hashin, Z., "Complex Moduli of Viscoelastic Composites--II. Fiber Reinforced Materials," Interna-
tional Journal of Solids and Structures, Vol. 6, 1970a, pp. 797-807.
[17] Hashin, Z., "Complex Moduli of Viscoelastic Composites--I. General Theory and Application to
Particulate Composites," International Journal of Solids and Structures, Vol. 6, 1970h, pp. 539-552.
[18] Gerard, J. E, Perret, P., and Chabert, B., "Study of Carbon/Epoxy Interface: Effect of Surface
Treatment of Carbon Fibers on the Dynamical Behavior of Carbon/Epoxy Unidirectional Compos-
ites," Controlled lnterphases in Composite Materials, H. Ishida, Ed., Elsevier, NY, 1990, pp.
449-456.
M. R. Piggott t and Yu (June) Xiong 1
Direct Observation of Debonding in

Fiber Pull-Out Specimens
REFERENCE: Piggott, M. R. and Xiong, Y., "Direct Observation of Debonding in Fiber
Pull-Out Specimens," Fiber, Matrix, and Interface Properties, ASTM STP 1290, C. J. Spragg
and L. T. Drzal, Eds., American Society for Testing and Materials, 1996, pp. 84-91.
ABSTRACT: Single-fiber pull-out experiments under the microscope were compared with
results from Instron pull-out experiments using the same glass fiber-room-temperature-cured
epoxy system. It was found that when the loading system was stiff the pull-out curve obtained
with the Instron had a transition which could be identified with the initiation of debonding, as
seen under the microscope. Under these conditions debonding appeared to be a controlled
fracture process. The work of fracture was about 70 • 20 Jm -2, and was governed by an
equivalent debonding shear stress of 44 _ 8 MPa. The yielding process was initiated at a shear
stress of 25 • 6 MPa. These results could be obtained using a very simple analysis because
they were not affected by friction.
KEYWORDS: interface, debonding, failure, pull-out
The fiber-polymer interface, or interphase, while being at the heart of the fiber-reinforced
polymer and much-studied, is much misunderstood. The results of measurements using different
techniques give results which can differ widely, and sometimes do not even correlate [1].
Sweeping assumptions are made in the derivation of interface strengths from the measurements,
yet such results are seldom challenged. This is particularly a problem with the fragmentation
test, which relies on an extrapolation of fiber strength, which has been shown to be totally
invalid [2] and assumes a constant interfacial stress along the fiber surface, which has also
been invalidated [3-5].
In any case it is almost certainly inappropriate to assign a failure stress to the interface:
there is evidence that the interface is brittle [6, 7] even when highly ductile polymers, such as
LDPE, are used [8]. Consequently we should seek to develop techniques which ensure that
stable crack propagation can be sustained. The fiber pull-out method appears to be such a
technique. The slow development of cracks can be observed under the microscope and related
to features on the force-distance plot during pull-out (Fig. l) [9]. This can only be done if
the free length is extremely short and the embedded length is moderately long. (Some of the
benefits of keeping the free length short have been appreciated for a long time [10].)
In this paper we describe studies of force-distance plots with glass fiber-epoxy pull-out
specimens in which the free length was kept down to ~ 0.2 to 0.5 mm (9 to 23 fiber diameters).
Experimental Method
E glass fibers obtained from Fiberglas Canada were used. They had a diameter of 22 Ixm
and were coated with an epoxy-compatible sizing. They were partially embedded in Shell
1Professor Emeritus and graduate student, respectively, Advanced Composites Physics & Chemistry
Group, Department of Chemical Engineering and Applied Chemistry, University of Toronto, Toronto,
Canada M5S 3ES.
84
Downloaded/printed
PIGGOTT AND XIONG ON DEBONDING 85
I
Location A B C Location A B C D E
Force (N) 0.7 0.9 0.3 Force (N) 0.8 1.4 1.5 0.5 0.5
1.6- BC~
1.O
1,2,
g o
~ 0,8-
0,5"
O
~ o.4.
I
0 0.2 0.4
0.2 0,4 0.6
Displacement (ram) Displacement (ram}
FIG. 1--Top, appearance of pull-out specimens during debonding under increasing force, and
below, the force-distance plot obtained concurrently. Fiber" embedded length 0.36 mm at left and
0.64 mm at right.
EPON 815 epoxy cured with trietlaylene tetramine. Figure 2 shows schematically the mold
used and the specimen produced.
The mold was made with silicone rubber, and a slit was cut in the side to hold the fiber.
The slit did not admit any polymer, so there was no meniscus where the fiber entered the
polymer. In addition, the specimen was quite thin (2 mm) and so the fiber could be observed
at moderate magnification under the microscope (X 50). Room-temperature curing was used,
and this gave very clear resin specimens.
The specimens were held in a special pulling device that could be mounted in the microscope
(Fig. 3). To ensure the smallest possible free length, the plate to which the fiber was to be
attached was brought up very close to the polymer surface while being watched through the
microscope. A thin layer of cyanoacrylate glue was used to secure the fiber once the plate
was in position. The free length was between 0.2 and 0.3 mm in these experiments.
The micrometer was turned while the specimen was observed in the microscope using
polarized light, and the forces required to initiate the various stages of failure were recorded.
Since the speed of testing was not controlled, some confirmatory tests with the 20~
and the high temperature-cured polymers were carried out at a constant cross head speed in
FIG. 2--Left, mold used to partly embed a fiber without producing a meniscus at the fiber entry
point; right, the pull-out specimen.
FIG. 3--Experimental set up for the pull-out test under the microscope.
the Instron. For this, the cylindrical type of specimens used in previous work were used. They
were made using the carousel arrangement described previously [8]. In these tests the free
length was 0.3 to 0.5 mm.
Compression tests were carried out on small cylinders of the resin to determine the yield,
ultimate strength, and the Young's modulus. The tests were carried out at a cross head speed
of 1 mndmin. The yield point was taken as the first noticeable deviation from linearity in the
force-distance plot, and the ultimate strength was taken as the maximum stress.
Experimental Results
Under the microscope, the debonding initiation could easily be seen and the force required
recorded. The results so obtained, shown in Fig. 4, indicate that embedded length in the range
0.4 to 1.7 mm had no significant effect on the force. These specimens were cured at 20~
and had 12.3% of hardener. Higher temperature cures gave higher values of the initiation force
<
E
z
-2 § ,--r .+_ §
o
0.5 I 1.5
EMBEDDED I.ENG'TH (ram)
FIG. 4--Debonding initiation force versus embedded length. Tests carried out under the micro-
scope, displacement rate irregular. 20~ cured specimens.
z~
n~
z
o
r~
2 4 6 8 10 12 14
CURING TIME (h)
FIG. 5--Debonding initiation force (mean for a number of embedded lengths) versus curing time
for cure at 60~ (solid rectangles) and 70~ (open rectangles). Tests carried out under the microscope.
(Fig. 5) and although it was not affected by embedded length, it did increase with increasing
cure time. In these experiments the displacement rate was irregular and not recorded.
Experiments carried out in the Instron at a constant displacement rate (0.5 mm/min), with
short free lengths (0.3 to 0.5 mm) had at least two transitions before ultimate bond failure.
Figure 6 shows a typical plot with slope changes at A (force FA) and B (for FB) with final
failure at F,,~.
The force for the first change in slope (FA) was only --0.3 N, and was substantially
independent of embedded length (Fig. 7). Fn ranged from ~3.5 to 6.5 N and again appeared
to be independent of embedded length (Fig. 8). The maximum force, although highly scattered,
did appear to increase monotonically with embedded length (Fig. 9).
After debonding, further pull-out was resisted by friction, and the maximum frictional force
was approximately proportional to embedded length (Fig. 10). The line plotted on the figure
corresponds to an average frictional shear stress of 10.9 _+ 0.2 MPa. The properties of the
polymers are given in Table 1.
Discussion
The details of the force-distance plots in the fiber pull-out experiment have hitherto been
used primarily for the determination of the maximum force and the frictional parameterS. The
maximum force gives the bond strength, and under certain circumstances (i.e., fibers not too
stiff and moderately long embedded lengths, so that fiber Poisson's shrinkage effects are
significant) the coefficient of friction and interfacial pressure can be separately evaluated [11].
1..C
0
0.2 o4 06 o8
D I S P L A C E M E N T (ram)
FIG. 6---A typical force-distance tracing obtained from a pull out test on the Instron machine.
88 FIBER, MATRIX,AND INTERFACE PROPERTIES
TABLE l--Compressive properties of polymers.

Cure Yield, MPa Ultimate, MPa Modulus, GPa
72h20~ a, 53 + 4 93 + 2 2.3 - 0.1

2h60~ 57 - 4 86--- 2 2.1 -+ 0.1
2h70~ 54_ 2 87 • 2 1.9 • 0.1
2h 80~ 56 • 2 84 • 2 2.0 • 0.1
a12.3% triethylene tetramine. The other resins contained 9.1%.
The frictional values are obtained from the variation of applied force after debonding is
complete (i.e., after the maximum force). A short free length is important for these observations.
This appears to be even more important in the early part of the pull-out curve. When the
free length is long (--50 mm), even systems with very fast response appear to reveal very
little here [12]. A careful look at earlier reports, however, indicates that with very short free
lengths (--0.5 mm) non-linearitiesare noticeable in the cases of glass and Kevlar in polyethylene
[8] and carbon in epoxy [13]. These were not remarked upon at the time.
However, a change in slope was predicted in a treatment of debonding involving friction
[4]. In the experiments described here, two changes in slope were observed with the tests in
the Instron. The first one, at a force FA of ~ 0.3 N (Fig. 7), appears to be too low to correlate
with the results obtained by observation under the microscope. The second slope change, at
FB (Fig. 8), while still a little low at 0.52 +-- 0.09 N as compared with --0.60 +_ 0.07 N for
N
N ~ II 9
05 I 1.5 2
E M B E D D E D L E N G T H (ram)
FIG. 7--The force for the first change in slope (FA) versus embedded length.
. . B B
0 0,5 I 15
EMBEDDED LENGTH (ram)
FIG. 8--The force for the second change in slope (F~) versus embedded length.
1.
o,
0 0.5 1 1.5
EMBEDDED LENGTH [mm)
FIG. 9--The maximum force versus embedded length.
the microscope observations, is identified tentatively as the debonding initiation force. It is

noteable that both of these are more or less independent of embedded length, while the
maximum force (Fig. 9) increases monotonically with increasing embedded length.
It seems likely that Fa involves some type of yielding process. We expect yielding to start
[14] when
FA = 2~rrZriy tanh(ns)/n (1)
where
2r is the fiber diameter,
Xly is the polymer shear strength,
s is the embedded length, L, divided by r, and
n 2 = EmlEf(1 + vm)ln(R/r) (2)
Here Em and v,, are the Young's modulus and Poisson's ratio of the polymer and 2R its
thickness. Ef is the Young's modulus of the fiber. For the room-temperature cured resin n =
0.0652 and the line drawn on Fig. 7 is given by
FA = Fa tanh(ns) (3)
with ffa being the mean value of FA. Comparing Eqs 1 and 3, we can calculate % with Fa =
1.5-
0.5 1 15
EMBEDDED LENGTH (mm~
FIG. 10---The maximum frictional force versus embedded length.
0.29 - 0.07 N, r = 11 I~m, R = 3 mm, Ef = 72 GPa, E,, = 2.3 GPa (Table 1) and v,, =
0.34. This gives "riy = 25 _+ 6 MPa. This is quite close to the yield stress x/~, i.e., 31 MPa
(Table 1). Since the yon Mises yielding criterion does not apply very well to polymers [15]
we conclude that the fit to Eq 1 is moderately good, and the identification of FA with yielding
is highly plausible. (Note that the triaxial stress state at the interface can and should also be
allowed for [4], at least with materials which do obey von Mises.)
We can also be reasonably confident that Fs marks the onset of debonding. Firstly, the
microscope results (0.60 + 0.07 N) and Instron results (0.52 ___ 0.09 N) overlap, so that the
difference is not significant, and secondly, there may be some delay in observing the debonding
force under the microscope, especially as it required hand-eye coordination. Debonding appears
to be a brittle process, so we use the equation [4]:
FB = 2qrr~/EfGir tanh(ns) (4)
where Gi is the work of fracture of the interface. Since in Fig. 8 the curve given by
F8 = F8 tanh(ns) (5)
appears to be quite a good fit, with fiB, the mean value of F 8, i.e., equal to 0.52 --- 0.09 N,
we can estimate Gi. This comes to about 70 --- 20 Jm -2.
The shear debonding stress, equivalent to this work of fracture, Xd, is obtained by comparing 9
Eq 5 and Eq 1 with "riyreplaced by "rd [4]. This gives rd = 44 ___ 8 MPa. Clearly the bond in
this case is quite strong. However, a brittle fracture process appears to be indicated by the
sudden drop in force at failure and the scatter of results.
The maximum force increases monotonically with increasing embedded length so the results
are contaminated by friction. The results for FB shown in Fig. 8 suggest that Fm~ need not
be used and, furthermore, our analysis is much simplified. It is evident that actual pull-out
may not even be necessary. FA and FB precede complete interface failure. In principle, therefore,
it is not necessary to embed a short length of fiber. Any length longer than ~ 1.32/n diameters
will have the result within ___ 1%. (This gives ns > 2.64 and tanh(ns) = 1.00 within 1%, so
Eq 1 reduces to
FA = 2"nr2"riyln (6)
which is independent of embedded length.)
Conclusions
When carried out under appropriate conditions, the pull-out method appears to give results
for interface yielding stress and interface work of fracture without the values having to be
corrected for friction. For the test, it is important to have a very short fiber free length and a
stiff testing machine. Changes in the slope indicate yielding first and then fracture. The average
force for each change in slope gives the corresponding values.
Glass embedded in 20~ epoxy had an interface which yielded at a shear stress of
25 + 6 MPa, and fractured with a work of fracture of 70 - 20 Jm -2. The equivalent debonding
shear stress was 44 +-- 8 MPa.
Acknowledgment
The authors would like to thank NSERC (Canada) for financial support of this work.
References
[1] Jacques, D. and Favre, J. P., Proceedings of the International Conference on Composite Materials
VI and the European Conference on Composite Materials 2, Elsevier, London, Vol. 5, 1987,
pp. 471--480.
[2] Dai, S. R. and Piggott, M. R., Composites Science and Technology, Vol. 49, 1993, pp. 81-87.
[3J Galiotis, C., Composites Science and Technology, Vol. 42, 1991, pp. 25-50.
[4] Piggott, M. R., Composites Science and Technology, Vol. 42, 1991, pp. 59-78.
[5] Figueroa, J. C., Carney, T. E., Schadler, L. S., and Laird, C., Composites Science and Technology,
Vol. 42, 1990, pp. 77-102.
[6] Penn, L. S. and Lee, S. M., Journal of Composites Technology, Vol. 30, i989, pp, 23-30.
[7] Di Benedetto, A. T, Composites Science, 1991, pp. 103-124.
[8] Piggott, M. R. and Dai, S. R., Polymer Engineering and Science, Vol. 31, 1991, pp. 1246-1249.
[9] Xiong, Y. and Piggott, M. R., Composites Science and Technology, 1994, in press.
[10] Chua, P. S. and Piggott, M. R., Composites Science and Technology, Vol. 22, 1985, pp. 33--42.
[11] Piggott, M. R. and Reboredo, M. M., Proceedings of the 34th International SAMPE Symposium,
1989, pp. 1913-1923.
[12] Miller,B., Gaur, U., and Hirt, D. E., Composites Science and Technology, Vol. 42, 1991,pp. 207-220.
[13} Wang, Z. N. and Piggott, M. R., Proceedings of American Society of Composites, 1991, pp. 725-731.
[14] Piggott, M. R., Advances in Composites: Physical and Physicochemical Effects, T. L. Vigo and B.
J. Kinzig, Eds., VCH Publishers, New York, 1992, pp. 221-266.
[15] Liu, K. and Piggott, M. R., Composites, accepted December 1994.
James D. Miller, 1 Gerry W. Zajac, 2 and Thanh Nguyen 2
Interlaboratory Study of Adhesion Using

Voltage Contrast XPS
REFERENCE: Miller, J. D., Zajac, G. W., and Nguyen, T., "Interlaboratory Study of Fiber/
Matrix Adhesion Using Voltage Contrast XPS," Fiber, Matrix, and Interface Properties,
ASTM STP 1290. C. J. Spragg and L. T. Drzal, Eds., American Society for Testing and Materials,
1996, pp. 92-102.
ABSTRACT: An intedaboratory study (ILS) was conducted to determine the utility and precision
of a test method for measuring fiber/matrix adhesion of carbon fiber-reinforced polymer matrix
composites. The test method is based on voltage contrast X-ray photoelectron spectroscopy
(VCXPS). The technique provides a quantitative measure of the relative amount of exposed fiber
and exposed matrix at the fracture surface of a failed composite specimen. Eleven laboratories
participated in two rounds of intedaboratory testing. The first round or pilot run was conducted
to familiarize laboratories with the technique. Each laboratory collected data on one of eight
material variants to compare with data from the host laboratory. In the second round each
laboratory analyzed different coupons from the same composite laminate. Repeatability and
reproducibility statistics were calculated and demonstrated that the method has interlaboratory
utility. Precision statistics were influenced by the heterogeneity of samples generated from a
single-carbon fiber-reinforced polymer matrix composite laminate.
KEYWORDS: adhesion, composite, carbon fibers, voltage contrast, X-ray photoelectron spec-
troscopy (VCXPS), interlaboratory study, test method
Voltage contrast X-ray photoelectron spectroscopy (VCXPS) has been used to characterize
interfacial bonding in carbon fiber-reinforced polymer matrix composites [1,2]. The electrically
conductive (fiber) and electrically insulating (matrix) portions of a composite fracture surface
give separate signals in the XPS spectrum when the sample is directly or indirectly biased.
Integration of the C Is peaks originating from the fiber and the matrix results in a quantitative
measure of the relative amount of each component exposed at the fracture surface. A parameter
C//C,, is calculated as a ratio of the exposed fiber to exposed matrix and represents the relative
amount of cohesive versus adhesive failure. Cf/Cm approaches zero for complete cohesive
failure in the matrix. CflC,, can be as high as five for adhesive failure in very poorly bonding
fiber/resin.
The fiber and matrix portions of a carbon-fiber reinforced polymer are likely to have different
atomic compositions. Therefore, it is useful to normalize the Cf/C,, value by a ratio of the relative
carbon compositions. The normalization factor is typically around 0.9 and the normalized value
is represented a s (Cf/Cm) n.
A VCXPS test method was written and published by the Suppliers of Advanced Composite
Materials Association (SACMA) as Recommended Method 26-94 [3]. SACMA does not
consider these recommended methods as standards but is coordinating its efforts with ASTM
Amoco Performance Products, Inc., 4500 McGinnis Ferry Rd., Alpharetta, GA 30202.
2 Amoco Corp., Amoco Research Center, 150 W. Warrenville Rd., Naperville, IL 60566.
92
Downloaded/printed
MILLER ET AL. ON ADHESION 93
to obtain standard status [4]. To determine precision statistics and the utility of the test method,
an inteflaboratory study was planned and conducted based on the guidelines of the ASTM
Standard Practice for Conducting an Interlaboratory Study to Determine the Precision of a
Test Method (E 691). Interlaboratory test participants were supplied with the test method, a
fractured Round 1 test specimen, and a worksheet to record and calculate data. Round 1 testing
began in June 1993 and was designed to familiarize users with the technique and to allow
comparison of results with host laboratory data. Two laboratories determined they did not
have sufficient equipment to participate. An attempt was made to obtain diversity in the types
of laboratories included in the study. Twelve external laboratories agreed to participate in the
interlaboratory study and ten external laboratories completed Round 1 testing. As listed in
Table 1, the participants represented five material suppliers, four universities, two aerospace
primes, and one independent analytical laboratory.
In Round 2 testing machined 90 ~ flexure coupons from one composite laminate were failed
by the host laboratory. One failed coupon was supplied to each of nine laboratories beginning
in December 1993. Both sides of the fracture surface were analyzed by the participant laboratory
and the host laboratory. Eight laboratories completed testing by August 1994.
Experimental
The host (laboratOry VCXPS) was performed on a Surface Science Laboratory model SSX-
100 spectrometer. The system was recently upgraded with an M-probe option with line-focus
X-ray source. The X-ray source was monochromatized aluminum K~ (1486.6 eV) radiation.
Normal operating parameters employed a spot size of 400 by 1000 ~tm and a pass energy of
156 eV resulting in a resolution of 1.33 eV full-width at half-maximum on a Au 4fvz core line.
The composite fracture surface was mounted vertically, directly on the sample stage with
silver paste grounding on both exposed end faces. The stage was externally connected to
ground via a jumper from the multipin feedthru. A number of other grounding schemes were
employed including: a gold foil mask held against the sample surface with metal clips;
conductive adhesive tape used to secure the back of the sample to the stage; and metal clips
used to clamp the sample directly to the stage. All of the various grounding schemes produced
comparable results. The silver paste method was the standard procedure used by the host
laboratory. Interlaboratory study participants chose whatever method best suited their individual
set-ups.
Typically a sample was grounded as described above and a 1-3 eV flood gun electron
energy was applied to separate the matrix-related carbon photoelectron contribution from the
grounded carbon-fiber photoelectron peak. The flood gun electron energy setting was varied
TABLE 1--1nterlaboratory study participants.

The Aerospace Corporation
Amoco Performance Products, Inc.
DuPont
ICI Fiberite Inc.
Kansas State University
Michigan State University
Northrop Corporation
Rocky Mountain Laboratories, Inc.
Shell Development Company
3M
University of Cincinnati
Virginia Tech
until baseline separation of the two major carbon Is photoelectron peaks was achieved. In the
host laboratory good resolution was obtained at a 3 eV flood gun setting, although the actual
electron energy was several electron volts higher. A spectrum was then acquired which usually
had excellent signal-to-noise ratio since the samples were composed predominantly of carbon.
In general, five random spots per sample surface were analyzed. In Round 2 testing, five
random spots (approximately I mm 2) on each side of the fracture surface (approximately 27
mm 2) were analyzed and averaged.
The areas of the separate contributions were measured by curve fitting with 80% Gaussian
and 20% Lorentzian peak shapes after Shirley background correction. The value Cf/Cm was
calculated as the ratio of the integrated areas. The final value (Cf/Cm), was calculated by
multiplying CJCm by a parameter A representing the relative amounts of carbon in the fiber
and matrix. The parameter A was determined from atomic composition measurements made
on the composite fracture surface during the VCXPS experiment. Generalized guidelines for
sample handling, test procedures, instrument conditions, data treatment, and calculations were
provided to the interlaboratory study participants and are described in the test method [3].
The XPS spectrometers used by the interlaboratory study participants included hemispherical
electron energy analyzers (VG ESCA lab MKII, HP 5950A, PE 5000 series SSI M-Probe,
and Kratos AEI ES 200 B) and cylindrical mirror analyzers (PHI 548). Only some of these
systems employed monochromatized X-ray radiation with either AI or Mg sources. Power
settings ranged from 400-600 W. In most cases electron flood guns were utilized in order to
differentially bias the fractured sample surfaces. However, in one case a simple DC voltage
bias was applied between sample and ground to change the surface potential. The direct bias
produced an inverted order of the carbon fiber and matrix C Is lines with the fiber peak
appearing at lower apparent binding energy. Analysis areas ranged from approximately 0.25
mm z to approximately 20 mm 2. The excitation and collection geometries varied with each of
the spectrometers employed.
Round 1 Testing
Round 1 interlaboratory test data are summarized in Table 2. Mean, standard deviation,
number of spots analyzed, and coefficient of variation are listed for each interlaboratory
participant and for data collected on the same sample by the host laboratory. The eight
material variants examined included four different intermediate-modulus carbon fibers in two
thermosetting epoxies and three thermoplastic polyethersulfones. The individual combinations
of fiber, resin, processing history, and failure mode produced different fiber/matrix interfacial
adhesion and values of the measured parameter (Cf/Cm),. Values ranged from 0.033 (Laboratory
1), representing excellent interfacial bonding, to 0.98 (Laboratory 10) representing only moder-
ate interfacial bonding. Example spectra collected during Round 1 testing by the host laboratory
are shown in Fig. 1. The spectra depict the relative intensities of the fiber and matrix peaks
for different values of Cf/C,,.
All the Round I interlaboratory data compared favorably with the host laboratory data. A
statistically significant difference in the means at 95% confidence existed only for Laboratory
8. Examination of the spectra showed that the peak resolution for Laboratory 8 was poor,
which can cause substantial errors in curve-fitting and (Cf/C,,), calculation.
All laboratories that attempted the VCXPS experiment were able to obtain appropriate
spectra and calculate C/Cm or (C/C,,,),,. Two laboratories determined they did not have sufficient
equipment after agreeing to participate and receiving samples. During Round 1 testing the
sample examined by Laboratory 1 had to be substituted because it was severely damaged by
TABLE 2--Round 1 interlaboratory (Cf/Cm)n data.
Participant Host
Laboratory Laboratory
Laboratory 1
Fiber/Resin: T-650/42/ERL 1982 avg 0.0331 0.0302

Lay-up: quasi-isotropic std dev 0.0039 0.013
Failure: CAI delamination number 6 5
Laminate: B022091 S 1 CV 12% 4 1%
Laboratory 2
Fiber/Resin: T-650/42/RADEL | 8320 avg 0.25 0.29

Lay-up: (0)L6 std dev 0.0092 0.042
Failure: 90 ~ flexure number 2 5
Laminate: T061712 S13A/B CV 4% 14%
Laboratory 3
Fiber/Resin: T-650/35/RADEL 8320 avg 0.303 0.29 I

Lay-up: (0)s std dev 0.048 0.13
Laminate: T070692 CV 16% 44%
Laboratory 4
Fiber/Resin: T-650/35/ERL 1902 x-bar 0.353 0.35 t'4

Lay-up: (90)5 std dev 0.042
Laminate: C01249-2 S1 CV 12%
Laboratory 5
Fiber/Resin: T-650/42/RADEL C x-bar 0.40 0.41

Lay-up: (0h5 std dev 0.028 0.053
Failure: compressive interlaminar shear number 2 5
Laminate: 89-03-09-02T CV 7% 13%
Laboratory 6
Fiber/Resin: T-650/42/RADEL C avg 0.46 0.43

Lay-up: (0)15 std dev 0.033 0.041
Laminate: 89-03-09-02T $2 CV 7% 10%
Laboratory 7
Fiber/Resin: T-650/42/HTA (ICI) avg 0.49 0.44

Lay-up: (0)15 std dev 0.089 0.081
Laminate: 89-05-02-1T SI CV 18% 18%
Laboratory 8
Fiber/Resin: T-650/42/RADEL C avg 0.55 0.33

Lay-up: (0)15 std dev 0.019 0.030
Laminate: 89-03-09-02T $3 CV 3% 9%
TABLE 2--Continued
Participant Host
Laboratory Laboratory
Laboratory 9
Fiber/Resin: HITEX 45-9B/RADEL C avg 0.61 0.57

Lay-up: (0)15 std dev 0.12 0.13
Laminate: 89-06-14-1T S1 CV 19% 22%
Laboratory 10
Fiber/Resin: IM-8/RADEL C avg 0.983 0.841

Lay-up: (0)15 std dev 0.18 0.22
Laminate: 89-03-17-01A CV 19% 26%
tCt/Cm data for comparison.
2Cf/Cm data. Fiber atomic composition data too weak (low S/N).
3Cf/Cm data. Atomic composition data not recorded.
4Cf/Cm data. Different fracture surface. Initial surface contaminated.
an imaging electron beam. The electron beam apparently caused sufficient carbonization of
the sample surface to drastically change the CflCm value. Additionally, the sample received
from Laboratory 4 appeared to be contaminated by silver paste and gave an atypical Cf/C,,
value in host laboratory testing. Therefore, the specimen was refractured and retested by the
host laboratory.
Round 2 Testing
Round 2 testing was conducted on failed 90 ~ flexure coupon specimens from a single-
composite laminate. The Round 2 sample description is listed in Table 3 and the testing
results and intermediate statistics are summarized in Table 4. Interlaboratory data as well as
measurements made on all samples by the host laboratory are shown. Multiple spots on each
surface (opposing sides of the fracture plane) were recorded and averaged together for reporting
purposes. Example spectra recorded on Round 2 specimens by interlaboratory participants are
shown in Fig. 2. Two laboratories that had fair to poor peak resolution during Round 1
testing had near baseline resolution during Round 2 testing indicating that familiarity with the
experiment aided in the analysis. Eight labs completed Round 2 testing. Of the remaining
three labs which participated in Round 1, two labs did not start Round 1 testing in time to
participate in Round 2 and one lab was significantly time-delayed by a sample substitution.
Between-laboratory (h) and within-laboratory (k) consistency statistics were calculated and
plotted in Figs. 3 and 4. Based on their critical values for this interlaboratory study there was
not a strong need to eliminate any laboratory test data or require retesting. High values of the
within-laboratory consistency statistic k reflect in part differences in the average (Cf/Cm),,
values for the two opposing surfaces of the fracture plane.
Precision statistics were calculated for Round 2 data and are shown in Table 5 along with
the intermediate statistics..They include the repeatability (st) and reproducibility (sR) standard
deviations and the 95% repeatability (r) and reproducibility (R) limits. In this case the "differ-
ence two standard deviation limit" (2.8) was used to calculate r and R. The values of r and
R represent 60-70% of the mean value of (Cf/C,,),. Since only one material was analyzed it
was not possible to comment on the variation of precision statistics with property level.
c-
~ B
~ A
0
O_
I I I
29~ 280 270
Binding Energy (eV)
FIG. 1--VCXPS C Is spectra of composite specimens from Round I testing recorded by the host
laboratory: (A) Ce/C~ = 0.053; (B) Ct/Cm = 0.299; (C) CJCm = 0.457; (D) C~/C,, = 0.707.
Experience in analyzing numerous failed composites by VCXPS suggested that the high
values of r and R were partially dependent on laminate heterogeneity. Additional statistical
calculations on within-laboratory data were used to determine the degree of sample heterogene-
ity. A frequency histogram of all Round 2 data is shown in Fig. 5. The histogram above the
bar represents all the individual spot measurements collected by the interlaboratory participants.
There are a total of 80 individual measurements which are not equally weighted between the
eight participants. The histogram below the bar represents all the individual spot measurements
collected by the host laboratory. There are a total of 90 individual measurements which are
equally weighted between 9 specimens. The histograms indicate only slightly greater variability
TABLE 3--Round 2 sample description.

Fiber THORNEL | T-650/42
Matrix RADEL 8320
Lay-up ( 0 ) 16
Failure 90~ flexure
Lami nate T061712
90~ Flex Strength 10.9 ksi
TABLE 4---Round 2 interlaboratory (CflCm), data,

Participant Laboratory Data Host Laboratory Data
std
Laboratory Coupon # avg dev number CV, % avg std dev number CV, %
1 $7,$9 0.47 0.057 12 12 0.38 0.062 20 16

3 $5 0.49 0,060 12 12 0.38 0.048 10 12
6 S19 0.36 0,043 10 12 0.42 0.056 10 13
7 $3 0.32 0,070 12 21 0.32 0.028 10 9
8 S15 0.45 O,l l 4 24 0.42 0.066 10 13
9 S17 0.31 0.045 10 14 0.34 0.040 10 12
10 $7 0.43 0,16 10 37 0.38 0.074 10 19
11 S1 0.39 0.040 10 10 0.35 0.032 10 9
Avg. of lab avgs. 0.40 0.38
St. dev. of lab avgs. 0.068 0.038
Number of labs 8 8
CV of lab avgs. 17 10
when eight laboratories did the testing versus testing by a single laboratory. Precision statistics
were recalculated in Table 6 where s,' and r' represent the within-specimen statistics and sR'
and R' represent the between-specimen statistics when only one laboratory conducted the
testing. The values are smaller than those found in interlaboratory testing but still represent
40-50% of the mean value.
Multiple measurements on one specimen were made by one of the interlaboratory test
participants to indicate the distribution of values within the specimen. Figure 6 shows the
frequency histogram of 72 individual measurements made on two sides (A,B) of specimen
$3 by Laboratory 7. Only the CJC,,, value was measured in order to save time. The distribution
is similar to that shown for the interspecimen analysis conducted by the host laboratory shown
in Fig. 5 (below bar). The statistics for both sides of the specimen and the specimen as a
whole are shown in Table 7. They indicate that the two sides of the specimen are different
with the confidence level for A versus B exceeding 99.9%. However, even a single side of a
specimen shows as much variability as the interspecimen comparison.
Multiple measurements on a single spot of one specimen were made by the host laboratory
to provide an indication of test repeatability that might be achieved if the samples were
homogeneous and identical. Ten separate measurements were made on specimen S3A and the
statistics are summarized in Table 8. Only the CflCm value was measured in order to save
time. The variability is significantly reduced but may be underrepresented since the measure-
ments were made one after another. The standard deviation is less than I/t0 its value for
measurements made along the same surface (Table 7) indicating that much of the interlaboratory
test variability can be attributed to specimen heterogeneity. The measured interspecimen and
intraspecimen variability is a disadvantage in the interlaboratory study but can be used to
characterize the interfacial bonding in carbon fiber/polymer matrix composites.
Conclusion
The study demonstrated interlaboratory utility of the test method for measuring fiber/matrix
adhesion of carbon-fiber reinforced polymer matrix composites using VCXPS. Despite the
large variations in system geometries, instrument parameters, and other variables which might
influence the differential charging phenomena and calculation of (CflC,,),,, reasonable interlabo-
ratory data were obtained. The majority of test variability appeared to be sample-related.
C
C
I
0
r-
n
I
i i i i i
295 280 265

Binding Energy (eV)
FIG. 2--VCXPS C Is spectra of composite specimens from Round 2 testing recorded by interlabo-
ratory participants: (A) Laboratory 1, (B) Laboratory 3, (C) Laboratory 8, (D) Laboratory 7, and
(E) Laboratory 2.
2.5 T- 2,15
. . . . . . . . . . . . . . . . . . . . . . .
'~ 0.5
=~ -o 5
-15
-2.15
. . . . . . . . . . . . . . . . . . . . . . .
=25 ~
1 3 6 7 8 9 10 11
Laboratory
FIG. 3--Between-laboratory consistency statistic (h) for Round 2 interlaboratory testing:

h(critical) at the 0.5% significance level equals 2.15.
i .55
1 3 6 7 8 9 10 11
Labot~tory
FIG. 4--Within-laboratory consistency statistic k for Round 2 interlaboratory testing: k (critical)

at the 0.5% significance level equals 1.55.
TABLE 5--Round 2 (Cf]Cm) n precision statistics--interlaboratory.
Average of lab avgs. 0.4034

St. dev. of lab avgs. 0.06828
sr 0.08236
sR 0.1029
r 0.231
R 0.288
FIG. 5--Frequency histogram of Round 2 data. Histogram above bar represents interlaboratory
data (80 individual measurements). Histogram below bar represents host laboratory data (90
individual measurements).
TABLE 6--Round 2 (Cr/Cm). precision statistics--interspecimen.
Average of specimen avgs. 0.3758

St. dev. of specimen avgs. 0.03814
st' 0.05310
SR' 0.06263
r' 0.149
R' 0.175
FIG. 6---Frequency histogram of data collected on Round 2 sample $3 by Laboratory 7 (72

individual measurements).
TABLE 7--Round 2 Ct/Cm statistics--intraspecimen.

Side A Side B Combined (A,B)
Average 0.2588 0.3535 0.303 5

Standard deviation 0.03345 0.05608 0.065 66
Number 38 34 72
CV 12.9% 15.9% 21.6%
TABLE 8--Round 2 CflCm statistics, single spot.

Average 0.323 9
Standard deviation 0.002 550
Number 10
CV 0.787%
Measuring composite interfacial-bonding heterogeneity through statistical analysis of VCXPS

data may provide an additional way to characterize carbon-fiber reinforced polymer matrix
composites.
Acknowledgments
The authors wish to expression their appreciation to the participating individuals and labora-
tories: C. S. Hemminger and H. A. Katzman (Aerospace Corporation); G. Blackman and R.
Cornelia (DuPont); C. Bowland (ICI Fiberite Inc.); P. M. A. Sherwood and C. L. Weitzsacker
(Kansas Street University); L. T. Drzal and D. J. Hook (Michigan State University); T. Steelman,
J. Wahlquist, and A. Yen (Northrop Corporation); C. D. Butler (Rocky Mountain Laboratories);
R. C. Yeates (Shell Development Company); D. Goetz and J. H. Thomas, III (3M); F. J.
Boerio, R. A. Gray, and W. A. Pofahl (University of Cincinnati); F. Cromer and J, Dillard
(Virginia Tech),
References
[lJ Miller, J. D., Harris, W. C., and Zajac, G. W., "Composite Interface Analysis Using Voltage Contrast
XPS," Surface and Interface Analysis, Vol. 20, 1993, pp, 977-983.
[2] Boland, J. J., "Voltage Contrast XPS--A Novel Scheme for Spatially Resolved XPS Studies,"
Surface and InterfaciaIAnalysis, Vol. 10, 1987, pp. 149-152.
[3] SRM 26-94, "Fiber/Matrix Adhesion of Carbon Fiber Reinforced Polymer Matrix Composites,"
SACMA Recommended Methods, Suppliers of Advanced Composite Materials Association, Arlington,
VA, 1994.
[4] Preface, SACMA Recommended Methods, Suppliers of Advanced Composite Materials Association,
Arlington, VA, 1994.
Carl R. Schultheisz, 1 Carol L. Schutte, 1 Walter G. McDonough, 1
Kenneth S. Macturk, 2 Mavyn McAuliffe, 2 Srikanth Kondagunta, 2 and
Donald L. Huntson 2
Effect of Temperature and Fiber Coating

on the Strength of E-Glass Fibers and the
E-Glass/Epoxy Interface for Single-Fiber
Fragmentation Samples Immersed in
Water
REFERENCE: Schultheisz, C. R., Schutte, C. L., McDonough, W. G., Macturk, K. S., McAu-
liffe, M., Kondagunta, S., and Huntson, D. L., "Effect of Temperature and Fiber Coating
on the Strength of E-Glass Fibers and the E-Glass/Epoxy Interface for Single-Fiber Frag-
mentation Samples Immersed in Water," Fiber, Matrix, and Interface Properties, ASTM STP
1290, C. J. Spragg and L. T. Drzal, Eds., American Society for Testing and Materials, 1996,
pp. 103-131.
ABSTRACT: The effect of absorbed moisture on the strengths of fibers and the fiber/matrix
interface for an epoxy reinforced with continuous fibers of E-glass is under investigation. Single-
fiber fragmentation tests of glass/epoxy model composites have shown degradation of the
strengths of both the fiber and the interface after immersion in water. The fragmentation
specimens were tested as-fabricated and after immersion in distilled water at 25 and 75~ for
more than 4000 h. Two coatings were applied to the fibers, one epoxy-compatible and the other
vinylester-compatible, in an effort to include the initial interfacial shear strength as a variable.
Analyses of the fragmentation test results adapting the approach of Wagner and coworkers were
used to determine moisture-induced changes in the fiber strength, making it possible to also
evaluate changes in the interfacial strength.
KEYWORDS: durability, E-glass fibers, epoxy, interfacial strength, moisture, temperature,

surface coatings
Structural applications of polymer composite materials are increasingly being studied for
use in potentially large civilian markets such as the automotive industry, infrastructure construc-
tion and repair, and deep water oil production. Several barriers remain, however, including a
need to characterize the long-term durability of these materials in order to ensure the acceptable
performance of structures throughout their expected lifetimes [1]. Because of the cost of raw
materials, both the automotive and oil industries have concentrated on using glass fiber
reinforcement for a number of applications. Here we direct our durability research toward
Research engineers, Polymer Composites Group National Institute of Standards and Technology,
Gaithersburg, MD 20899.
2 Post-doctoral fellow, guest researchers, and group leader, respectively, Polymer Composites Group,
National Institute of Standards and Technology, Gaithersburg, MD 20899.
103
Downloaded/printed
assessing moisture-induced degradation of the interface between glass fibers and polymer
matrix materials, and the degradation of the glass fibers themselves.
As with earlier studies in this area [2,3], the single-fiber fragmentation test [4-7] has been
employed to evaluate the interracial shear strength. In the fragmentation test, an epoxy coupon
containing a single fiber is loaded in tension, also loading the fiber in tension through shear
transfer across the interface. With increasing load, the fiber fractures into smaller and smaller
fragments, until the shear transfer across the interface is insufficient to load any of the remaining
fiber fragments beyond their individual tensile strengths; this state will be referred to as
mechanical saturation. The resulting fragment length distribution is then used to relate the
interracial shear strength to the tensile strength of the fiber [8-11]. The length of a fiber
fragment that is just long enough to allow the shear stress to build up a tensile stress equal
to the strength of a fragment is called the critical length Lc; since any fragment longer than
the critical length will fracture, the average fragment length at mechanical saturation should
lie between LJ2 and Lc, so that the critical length is taken as 4/3 of the average fragment
length at mechanical saturation.
Water is expected to attack the interface, decreasing the interfacial shear strength, resulting
in longer fragments. However, water also significantly attacks the surface of the glass fibers
themselves [12-14] (to a much greater extent than is found with carbon fibers [15,16]), leading
to lower tensile strengths, which results in shorter fragment lengths. Thus, there is a problem
in separating the competing effects of interfacial degradation and fiber strength degradation
on the fiber fragment distribution. This problem has been addressed by adapting a model of
Wagner and coworkers [17-19] to determine both the fiber strength and the interfacial strength
from the same test.
This paper covers results of an effort to expand our program to study the effects of immersion
in water at 25~ as well as at 75~ (as employed by Schutte, et al. [2]), and to study the influence
of different fiber coatings. While elevated temperatures should accelerate the absorption of
water and any chemical processes that lead to degradation of fiber and interface, the elevated
temperature may also lead to chemical changes that would not take place at temperatures
ordinarily experienced in service. Thus, it is important to correlate results from longer-term
testing at lower temperatures to results from elevated temperature conditions in order to
evaluate the potential application of elevated temperatures for accelerated tests. The earlier
study [2] indicated that exposing the specimens to dry conditions at 75~ resulted in no
measurably significant changes when compared to the initial dry specimens.
In addition to the temperature variation, E-glass fibers having two different coatings were
employed: an epoxy-compatible coating (intended to provide a strong interface with the epoxy
matrix) and a vinylester-compatible coating (intended to provide a weaker interface with the
epoxy matrix). Although the exact nature of the coatings is proprietary, a typical coating (or
size) contains a lubricant, antistat, binder and coupling agent [20]. The first three components
aid in processing and protect the fibers when handled. The coupling agent (typically a silane)
serves to bridge the gap between the inorganic glass surface and the organic matrix material;
silanes are produced with a number of organofunctional groups tailored to the chemical nature
of the matrix and having varying levels of hydrophobicity [1,20]. The different coatings were
used in an effort to systematically vary the parameters that affect the behavior of the composite,
and it was hoped that the two coatings would result in significant differences in tests on
otherwise identical composite materials, providing further information on the role of the
interface. Also, it was expected that the different chemical structures of the coatings might
change at different rates, providing different levels of protection for the fiber, and influencing
the rate at which the fibers degrade. Plueddemann [20] provides ample evidence that coupling
agents promote adhesion at the fiber/matrix interface in both the dry state and after hydrothermal
conditioning. He also cites a reference [21] and suggests that it implies that the coupling agent
SCHULTHEISZ ET AL. ON EFFECTS OF MOISTURE ON FIBER STRENGTH 105
may protect the glass surface against stress corrosion by water, but the stress corrosion in that
paper refers to interfacial failure at a bond between epoxy and aluminum. Some additional
work at NIST indicates that the organic functionality of silane coupling agents can affect the
amount to which the glass is attacked [22], although it may also be influenced by the strength
of the bond between the coupling agent and the matrix material. Presumably, if water diffuses
to the interface through the matrix, the nature of the organic end of the coupling agent could
serve to protect the fiber. However, if water moves along the interface attacking the fiber
surface [12-14] and the bond between the glass and the coupling agent, the organic functionality
of the coupling agent may not affect the strength of the glass; in this case, the relative
degradation of the glass as compared to the interface may depend mainly on the pH of the
water and the composition of the glass itself.
Experimental I
Fragmentation Test Preparation

The single-fiber fragmentation test employs single filaments of glass encapsulated in dog-
bone-shaped epoxy resin specimens loaded in tension along the axis of the fiber. The samples
were cast in an eight-cavity silicone mold made out of RTV-664 from General Electric,
following the procedure described by Drzal, et al. [23]; further details can also be found in
the paper by Schutte, et al. [2]. The dogbones measure 65-mm in overall length by 1-mm
thick, with a test section approximately 27-mm long and 4-mm wide in the center. The grip
sections are 10.5-mm wide and transition smoothly into the test section.
Single filaments of E-glass fiber (Owens-Coming) were separated from spooled fiber tows.
Two sets of fibers were used. The first set was coated with an epoxy-compatible size (type
158B) and the second set was coated with a vinylester-compatible size (type 366), in an effort
to introduce some initial variability in the strength of the interface. The average diameters of
the epoxy-compatible fibers ranged between 17.6 and 23.9 ~m, with a mean of 20.3 ~m; the
average diameters of the vinylester-compatible fibers ranged between 12.5 and 18.0 p~m, with
a mean of 15.0 p~m. Once a filament was separated from the bundle, the filament was placed
in a cavity and aligned via the sprue slots in the center of each end of the cavity. After the
fibers were in position, they were fixed in place by pressing them onto double-stick tape lying
outside the molds. Finally, strips of a high temperature tape were placed over the double-stick
tape to keep the fibers from moving.
The resin and curing agent were prepared by weighing out 100 g of a di-functional epoxy,
diglycidyl ether of bisphenol-A (DGEBA, Epon 828 from Shell Chemical Co.) and 14.5 g of
meta-phenylene diamine (mPDA, Fluka Chemical Company) in separate beakers. Subsequently,
both beakers were placed in a vacuum oven (Fisher Scientific Isotemp Vacuum Oven, model
281 A) that was set at 75~ to let the mPDA crystals melt and to lower the viscosity of the
epoxy. The silicone mold containing the fibers was also placed in an oven (Blue M Stabiltherm,
model OV-560A-2) set at 75~ to help to eliminate bubbles that normally are formed when
warm resin is poured into a room temperature mold. After the mPDA had melted, it was
mixed with the epoxy and the mixture subjected to vacuum for seven minutes. Finally, the
resin was poured into the mold cavities containing the fibers, and any air bubbles removed
with a syringe. The mold was placed into a programmable oven (Blue M, General Signal,
l Certain commercial materials and equipment are identified in this paper m order to specify adequately
the experimental procedure. In no case does such information imply recommendation or endorsement by
the National Institute of Standards and Technology, nor does it imply necessarily that the items are the
best available for the purpose.
model MP-256-1, GOP), and the resin cured by heating at 2.8~ to 75~ holding at 75~
for 2 h, heating at 2.8~ to 125~ holding at 125~ for 2 h, then turning the oven off.
After the samples had cooled slowly overnight in the oven to room temperature, they were
examined and any specimens with fibers that had broken or had not remained aligned along
the axis of the dogbone were discarded. Finally, the ends of the samples were covered with
a room-temperature-cure epoxy in order to reduce the amount of water having direct access
to fiber/matrix interface from the exposed end of the fiber.
Fragmentation Test Protocol

A schematic diagram of the experimental apparatus is shown in Fig. 1. The hand operated
loading jig [7] is mounted on an X-Y traverse rig (Newport Corporation 431 series) and
8 I
3
t i\ 5
1 TV Camera 5 Microscope
2 TV Monitor 6 Single Fiber Specimen
3 Loading Device 7 Linear Variable Differential Transformer
4 XYTranslation Stage 8 Data Acquisition System
9 Displacement Gage
FIG. 1--Schematic of the single-fiber fragmentation apparatus.
attached to the Nikon Optiphot Pol microscope. The traverse is used to translate the loading
jig back and forth under the microscope. The position of the fiber under the microscope is
measured using an LVDT (Trans-Tek 1002-0012), and the diameter of the fiber is measured
at 5 positions before the test using video calipers (Boeckeler VIA-100). Two marks are made
on the surface of the epoxy specimen, approximately 10-mm apart, and the positions of
identifiable microscopic features within those marks are monitored to measure the strain during
loading. A dial gage is attached to the jig to monitor the displacement between the grips.
All tests were performed at room temperature in dry conditions. Specimens that were
immersed in 75~ water were removed from the oven and placed in room temperature water
the day before testing. Although some specimens failed before reaching mechanical saturation,
all of the data in this paper comes from samples that were successfully tested to mechani-
cal saturation.
In testing, the grip displacement is increased by 75 I~m (a strain increment of approximately
0.2%), followed by a waiting time of 10 min, after which the number of fiber breaks in 15
mm is recorded and the strain is determined from the positions of the gage marks. The waiting
time at each loading step was found necessary to prevent crack propagation into the matrix
causing failure of the specimen before mechanical saturation. The average rate of grip displace-
ment is therefore of the order of 0.05 to 0.1 ixrn/s.
Water Uptake, Diffusion Coefficient, Swelling Strains

The rate of absorption of water by dogbone samples of the epoxy containing no fibers is
shown in Fig. 2 for immersion at 25 and 75~ as determined by repeated weighings of
individual dogbones. Results from another investigation [2] indicate that the equilibrium content
of water at 75~ is approximately 4% after immersion for over 12 000 h, while it appears
from Fig. 2 that the equilibrium content of water at 25~ is only about 1.6%, but is achieved
much sooner. However, the 25~ data is only available for a shorter time scale at the moment.
In contrast to these results, Shen and Springer [24] suggest that the equilibrium content of
water for immersed samples is independent of temperature.
One-dimensional diffusion coefficients for the matrix material have been estimated using
the model of Shen and Springer [24] on epoxy resin blocks measuring approximately 6.8 by
31.8 by 30.6 ram. This model applies an assumption of Fickian diffusion, using the slope of
3.5 L w i i i i
3.0
2.5
ooOOOoe~
2.0
1.5
1.O
0.5
0.0
-0.5 L I I J.---. r t
0 1 0 0 0 2000 ,3000 4000 5000 6000
lime of ~rnmersion (hours)
FIG. 2--Weight-percent of water absorbed by dogbone samples of DGEBA/mPDA epoxy resin
immersed in water at 25 and 75~
3.0
, , , I // irr e
2.5 1/2
/...
0.082 ~lhour ~/.o lir 75~
o
m 2.0 ;O"~
#o
<
1.5
,~ 1.0
em
0.5
I/2
0.0
0 10 20 30 40 50
Time (hours l/z)
FIG. 3--Weight-percent of water absorbed by rectangular blocks of DGEBA/mPDA epoxy resin

immersed in water at 25 and 75~ Water absorbed plotted versus square root of time to evaluate
Fickian diffusion behavior and calculate diffusion coefficients.
the water uptake versus the square root of time as shown in Fig. 3. As with the dogbone
samples, the weight of water absorbed was determined from repeated weighings of the epoxy
blocks. With the estimates of the equilibrium water content given above, and including a
correction to account for diffusion into all six sides of the blocks, the one-dimensional diffusion
coefficients are calculated to be 1.83 • 10 -3 mm2,/hat 75~ and 5.61 • 10 -4 mmZ/h at 25~
The linear swelling strain of the epoxy as a function of water uptake has been measured
on a single dogbone specimen immersed in water at 75~ This data compares well with data
of Wong and Broutman [25] for an identical epoxy immersed at 90~ as shown in Fig. 4.
Wong and Broutman suggest that for water uptake to about 1%, the water is primarily filling
open volume already available in the sample, while above 1% water uptake the increase in
the volume of the sample is approximately equal to the volume of absorbed water. We have
u i i i
1.0 I ~ 90~ (WongondBroutrno~ ,o

~.~ 0.6 0 75"C /~//otS~/o
c
Lo
j:
F~ o.s
3
ir 0.4
9
L
o
~ o.2
0.0
I I I I
0 I 2 3 4
Weight ~r Woter Absorbed
FIG. 4--Linear swelling strain as afunction of absorbed waterfor DGEBA/mPDA epoxy immersed
in water at 90~ (data from Wong and Broutman [25]) and 75~ Jbr this study.
assumed that the relation between swelling strain and water uptake is independent of temperature
for the purposes of correcting strain values measured in the single-fiber fragmentation test.
We have also assumed that the thermal strains caused by heating the samples are reversible,
so that experiments at room temperature all have comparable thermal strains. However, there
may be some relaxation effects that need to be included.
Weibull and Poisson Characterizations of the Strength of Glass Fibers

Since water attacks the flaws in glass [12-141, particularly in the presence of stresses, it is
important to be able to evaluate the change in the distribution of flaws in the glass fibers after
immersion of the samples. This section describes the method used to assess the changes in
the strength of the fibers by monitoring the number of breaks in the fiber as a function of
applied strain. In addition, evaluating the strength of the fiber in this manner provides a
reasonably robust method for evaluating the strength of a fragment at the critical length, which
is used to determine the strength of the fiber/matrix interface.
The strength of glass fibers depends strongly on the distribution of flaws on the fiber stu'face,
and thus the strength of glass is often described using a Weibull probability model, which is
determined from a large number of tests on identical fibers of a given length. The (two-
parameter) Weibull probability of survival at stress ~r for a fiber of length L is given as
L o- ~
where ot0 is the scale parameter associated with length Lo, and 13 is a length-independent
shape parameter.
Wagner and coworkers [17-19] have employed a Poisson probability model describing the
spatial frequency of flaws to investigate the strength of fibers measured in the single-fiber
fragmentation test. In this approach, the strength of a fiber is characterized using an assumed
Poisson distribution of flaws of different magnitude. For a Poisson distribution, the probability
of survival at stress cr for a fiber of length L is given as
Ps = exp(-k(~)L) (2)
where k(cr) is the cumulative flaw density encountered up to the stress cr (number of flaws
per unit length for all stress levels up to or) [26].
Equating the Poisson and Weibull probabilities of survival, and calculating k(cr) from the
number of breaks n(cr) in length L0 at stress ~r, one obtains
exp[-~ n(cr)] = e x p [ - ~ 0 (~)1~1 (3)
Further, choosing L = L0 = 15 mm for the single-fiber fragmentation test as it is now

performed, and taking logarithms of both sides of the equation, the number of breaks is related
to the Weibull parameters as
ln(n(o')) = 13 ln(o') - 13 ln(c~0) (4)
where n(cr) and ct0 are parameters that depend on the reference length ~ . The fiber stress cr
can be calculated from the strain applied to the composite ~.cthrough the fiber Young's modulus
Ef as ~r = Efec. Thus, the parameters describing the strength of the fiber can be estimated
from a straight line fit through a plot of ln(n(tr)) versus ln(cr). An example of this method of
evaluating the data is shown in Fig. 5, indicating the changes in the fragmentation process
after immersion in water.
Equation 4 should be valid over some range of stress where the number of breaks is
statistically significant, but where most of the length of the fiber is strained to the same extent
as the matrix (so that the fiber stress or = Efec). However, the tensile stress in the fiber ends
near the breaks is less than the intact fiber stress of Erec, as the tensile stress is built up by
shear transfer near the ends of the fragment. As the load and the number of breaks increase,
the cumulative ineffectively stressed length (denoted Ix(or)) increases.
There are several approaches to interpreting the influence of the ineffectively stressed length
of the fiber. The method employed in this paper is to replace L by L0 - tx(cr) on both sides
of Eq 3, in which case Eq 4 will hold until L = Lo - p,(cr) -- 0, when the probability
descriptions will have no meaning. At that point, no further fragmentation can take place, so
this model would predict a bilinear behavior of ln(n(cr)) versus ln(cr), increasing at first and
then remaining constant.
A second model of the influence of the ineffectively stressed length of the fiber was used
by Shioya, et al. [3] (a similar approach has also been employed by Baillie and Bader [27]).
This model may be somewhat more robust, as it provides for three parameters in fitting the
data, which allows the model to capture more of the behavior beyond the onset of mechanical
saturation. In this model, it is recognized that the flaw density in the Poisson description
should really reflect the number of flaws per effectively stressed length (as suggested by
Wagner and Eitan [17]), so h(~r) = n(~r)/(Lo-lx(cr)). In the Weibull distribution, however, the
length Lo is just a reference length for the parameter ao. In this case, Eq 3 becomes
exp[-LL~o - P~(~ -] exp[

~(or) n(o')J = --(~oo/]
L~ ~L~la'(~ ~ (5)
2.5 J i i i
Or, I/
o 366 25*C 1800 hours[ /
2.0 9 366 75~ 1872 h o u ~
1.5
.a 1.0
E
z
.n 0.5
0.0
I
0.0 0.1 0,2 0.3 0.4 0.5 0.6
Log Fiber Stress (GPo)
FIG. 5--Effect of immersion in water on fragmentation process for samples with the vinylester-
compatible coating (designated 366). An example of the curve fit to determine the Weibull parameters
for 3 samples, one tested dry, one after 1800 h immersion in water at 25~ and one after 1872 h
immersion in water at 75~
SCHULTHEISZ ET AL. ON EFFECTS OF MOISTURE ON FIBER STRENGTH 1 11
Note that ~@r) can be written as n(~r)8(cr), where 8(cr)/2 is the shear stress transfer length at
each end of a fragment (with 2 fragment ends per break). With this substitution, the number
of breaks is related to the fiber stress as
L0ff~
n(~) =L0a~ + ~(~)g~ (6)
Since 8(or) < L0, then at very low stresses this equation reduces to Eq 4, again giving a straight
line region for a plot of ln(n0r)) versus ln(cr), while at high stresses, the second term in the
denominator becomes important. If 8(~r) is taken as a constant, Eq 6 provides three parameters
(e~, 13, 8) that can be used to fit the data, with g equivalent to the asymptotic fragment length
at mechanical saturation, g = L0/nsoo, where ns~ is the (extrapolated) number of breaks at
mechanical saturation at infinite stress. Alternatively, using the shear lag model of Kelly and
Tyson [8] (discussed more fully below), g(cr) would be a linear function of ~r, with g(cr) =
D~/(4"r), where D is the fiber diameter and r is the constant shear stress across the interface.
In this case there are again three parameters (~x, 13, T) that can be used to fit the data, and the
interfacial shear strength "r is extracted directly. However, choosing ~(cr) = Dcr/(4~) leads to
the result that n(cr) will begin to decrease at high stresses (because the assumptions inherent
in the derivation break down).
The analysis presented previously appears able to determine the degradation of the strength
of the fibers, which makes it possible to discriminate between the degradation mechanisms
that affect the fibers and those that affect the fiber/matrix interface. The approach does seem
to have some difficulties, however; it is possible that additional insight can be gained through
methods that incorporate computational simulations [26,28,29].
Tensile Tests of the Epoxy

Two of the models of interfacial strength examined below employ the elastic properties of
the matrix in the calculations. Thus, molded dogbone samples of the resin were also loaded
in tension to determine the modulus and yield behavior as a function of water absorption at
75~ These dogbones have the same in-plane dimensions as those used for the single-fiber
fragmentation test, but had thicknesses of approximately 1.7 ram. The as-molded dogbones
were ground to provide a uniform thickness in the test section to within 0.025 ram. The ground
surfaces were polished under water with 2400 grade paper, and in order to remove any water
absorbed during grinding and polishing, the specimens were placed under vacuum at room
temperature for 532 h, leading to an average weight loss of 0.57%. Five samples were tested
at that time, while the remaining samples were placed in de-ionized water in an oven at 75~
Five samples were removed and tested after 1, 4, 14, and 50 days of immersion in an effort
to test at equal intervals of absorbed water. Water uptake is referenced to the weight of the
samples after drying.
The tests were performed at room temperature in dry conditions under displacement control
at a crosshead speed of 5.1 Ixm/s in order to compare with some earlier tests. Load, displacement
and strain data were taken at 0.5 s intervals. Figure 6 shows representative stress/strain behavior,
and Figs. 7 and 8 indicate the change in Young's modulus and maximum tensile stress
(effectively the yield stress in tension assuming an elastic/perfectly plastic model) as being
approximately linear with absorbed water. The average strain at failure remained approximately
constant at 6.9%. The stress shown in the figures is calculated from the load and the initial
area, while the strain is the engineering strain (change in length divided by initial length). No
evidence of necking is visible in the failed specimens; the failure is generally along planes
approximately perpendicular to the loading, and often in the vicinity of the transition between
100 i i i i i , u ,
Dry
80 ays
"~ 60
4o
20
I I i I I I I I
0 1 2 3 4 5 6 7 8
Strain ( ~ )
FIG. 6--Representative tensile stress-strain curves for DGEBAImPDA epoxy tested dry and after
4 and 50 days' immersion in water at 75~
the grip section and the test section, indicating that the radius of curvature in that region may
be too small.
Four additional tests were also performed on dry dogbone samples, two at a crosshead speed
of 5.1 p~m/s and two at 0.27 ~rn/s to more closely match the single-fiber fragmentation test.
These specimens were tested as-fabricated, without grinding to make a uniform test section,
so the measured stress is somewhat less well determined. These tests indicate that the slower
test speed reduces the modulus by approximately 15%, and leads to similar reduction of the
maximal tensile stress. As a first approximation, it would seem reasonable to estimate the
matrix modulus and ultimate tensile stress found in the single-fiber fragmentation test as being
80% of the quantities measured at the faster rate shown in Figs. 7 and 8.
Thus, for calculation purposes, Young's modulus was assumed to be linearly related to the
absorbed moisture, with an initial value of E,, = 3.06 GPa and a slope of - 0 . 3 3 GPa/%HzO.
5.0 v t i i v i i
g
4.5
~ 4.0
?,
~ 3.5
~ 3.0
OO Oo 9
2.5
2,0 I I q I I I I
0 0.5 1 .0 1 ,5 2.0 2.5 3.0 3.5
Weight ~= W a t e r A b s o r b e d
FIG. 7--Young's modulus of DGEBA/mPDA epoxy resin as a function of wt% water absorbed
after immersion at 75~ Straight line fit through data has initial value of 3.83 GPa with slope of
-0.41 GPaI%tt20.
SCHULTHEISZ ET AL. ON EFFECTS OF MOISTURE ON FIBER STRENGTH 1 13
120
g
100
~a
"~
e~ ~0
9
"~ , 60
40 I r I I I I I
0.0 O. 5 1.0 1.5 2.0 2.5 ,3.0 3.5
Weigh~ Z W a ~ e r AbsQrbed
FIG. 8--Maximum tensile stress of DGEBA/mPDA epoxy resin as a function of wt% water
absorbed after immersion at 75~ Straight line fit through data has initial value of 90.6 MPa with
slope o f - 6 . 9 MPa/%HzO.
The Poisson's ratio and coefficient of thermal expansion of the matrix are taken as vm : 0.35
and o~m : 6.8 • 10-5/~ [11], and are assumed to remain constant.
Fracture Toughness of the Epoxy

Another property of the matrix that influences the single-fiber fragmentation test (and will
he important for evaluating the results of composite delamination tests) is the fracture toughness.
The fracture toughness of the matrix material has also been evaluated as a function of immersion
in water at room temperature and at 75~ Compact tension specimens (with nominal overall
dimensions of 32 by 30 by 7 ram) were tested at room temperature in dry conditions under
displacement control at a rate of 83 I~m/s, following the ASTM Standard Test Method for
Plane-Strain Fracture Toughness and Strain Energy Release Rate of Plastic Materials (D 5045)
[30]. Results are shown in Fig. 9. The samples immersed at 25~ show a much greater increase
in fracture toughness at a lower absorbed water content than the samples immersed at 75~
but it must also be recognized that the 25~ samples were immersed for considerably longer
than the 75~ samples. The amount of absorbed water in the first set of immersion tests
corresponds roughly to the amount of water in the first fragmentation tests performed after
immersion, while the second set of tests took place at approximately the same water content
as the longest term fragmentation tests. However, since the dogbone samples are much thinner
and have a higher surface area to volume ratio, the dogbones achieve these levels much faster:
dogbones immersed at 75~ reach 1.3% absorbed water after 1 day, and dogbones immersed
at 25~ reach 0.65% absorbed water after 4 days.
Although one would expect the fracture toughness to increase with increasing water content,
it may be that the elevated temperature also leads to further chemical changes (possibly
affecting any remaining uncured components), offsetting somewhat the plasticizing effect of
the absorbed water. If the absorption of water increases the molecular mobility sufficiently,
the elevated temperature may allow for further crosslinking to take place. Long-term immersion
in 75~ water causes significant darkening of the epoxy (although the water itself does not
change color significantly), and seems to result in a somewhat brittle surface layer on the
fragmentation specimens, which may be a factor in the fracture results. Neither the dry samples
1.6 i ~ i i i J
524.5 0
.~ 1.4 hours 3116
E 0 A ~ hours
o 1.2
115
_~ 1.0
~ o.s
c
x:
la3.S 1341
hours hours
2 Dry
0.4
ta
o 0 25"C I m m e r s i o n j
0.2
9 75"C I m m e r s i o n
0.0 I I I I i i I
0.0 0.5 1.0 1.5 2 0 2 5 3.0
WeighL % W a t e r Absorbed
FIG. 9--Plane strain fracture toughness of DGEBA/mPDA epoxy resin tested dry and as a
function of wt% water absorbed after immersion at 25 and 75~ tested in compact tension geometry.
The percentage of water in the first data sets correspond roughly to the earliest fragmentation tests
after immersion; the second data sets have twice as high a water content and are near saturation.
Times of immersion are indicated on the figure. Lines connect the mean values of the data sets.
nor those immersed at room temperature showed a strong tendency toward crack-arrest once
fracture initiated, but the specimens immersed at 75~ generally took two or three steps to
propagate the crack to the end of the specimen. This result may indicate that the room
temperature immersion only created local toughening in the vicinity of the crack tip, while at
75~ the moisture penetrated much further, causing toughening throughout the specimen. For
all of the tests, the specimens were pre-cracked before placing them in water (ensuring that
the specimens could be tested before waiting for the planned levels of moisture absorption),
placing the crack-tip region in direct contact with the water. The test results might change if
the specimens were pre-cracked after immersion.
Estimation of the Strength of the Fiber at the Critical Length
The strength of the fiber at the critical length is needed to calculate the interfacial shear
strength. This strength has been estimated from the procedure described above, by fitting a
straight line through initial points of a plot of ln(n(~r)) versus ln(tr) to evaluate ot0 and 13 in
the WeibuU distribution. The fiber stress cr is taken as Efec (Ef has been measured as 67.5
GPa [31]), where ec is the strain in the composite, which has been corrected to include the
effects of swelling as estimated from the data in Figs. 2 and 4, with ec = s + s + s163
where e0 is the measured applied strain (change in length between specimen gage marks
divided by the initial length between marks) and s is the estimated swelling strain. A value
for fiber strength suitable for use in estimating the interfacial strength has been determined
by assuming that the bilinear model is appropriate, meaning that the fiber strength at the
critical length is given by the intersection between the straight line fit to ln(n(~)) versus ln(~)
and the line n = constant = nc (number of breaks corresponding to the critical length). The
critical length is calculated from the average fragment length at mechanical saturation using
the factor of 4/3 discussed in the introduction, so Lc = 4L J3. L, can be calculated from the
reference length and the number of breaks at mechanical saturation (n, which is a parameter
SCHULTHEISZ El" AL. ON EFFECTS OF MOISTURE ON FIBER STRENGTH 115
determined from the experiment) as L~ = Lo/ns, so nc = 3nil4. These assumptions yield a fiber
strength of
crf = OLo(nc) till

{z.o'~"~'
= elot~-~c) (7)
A common alternative method used to estimate the strength of the fibers makes use of the
Weibull parameters to calculate the average strength of a fragment with a length equal to the
critical length. This strength is denoted by (tr}, and is calculated as [19]:
(O') = Ot0 I" 1 q- (8)
Note the similarity between Eqs 7 and 8. When the Weibull shape parameter is large, say for
13 > 0.8, the F function in Eq 8 is a slowly varying function approximately equal to unity,
so these estimates lead to similar values and the trends are the same. However, for very small
13, the F function becomes very large, so that the average strength calculated from the Weibull
distribution is very much larger than any stress actually applied to the fiber.
This difficulty when 13 is small has effectively forced the use of Eq 7 rather than Eq 8. In
the attack by water on the fiber, the spectrum of flaw sizes is broadened a great deal, which
is reflected in a very low value of 13. For a Weibull distribution with a large 13, the average
strength (of a fiber of length Lo) is close to the scale parameter ~Xo.When 13 is small, however,
much of the distribution is moved into the high-stress tail, and the calculated average strength
is therefore very much higher than or0, leading to unrealistic estimates of the fragment strength.
This problem indicates that while the strength of the fragments may be modeled locally with
a Weibull distribution, the Weibull distribution does not describe the full spectrum of flaws
after immersion in water, particularly those that fail at higher stresses.
This inadequacy of the Weibull distribution to describe the spectrum of flaws after immersion
means that some other description of the flaws will be needed to model the degradation of
the glass from a fracture mechanics viewpoint. Also, it appears that the evaluation of the
Weibull parameters is not strictly necessary to estimate the strength of the fragments at the
critical length, which is needed to estimate an interfacial strength. What is necessary is a
protocol that will consistently provide a measure of the stress in the region where the number
of breaks stops increasing, and a consistent estimate of the critical length.
Interfacial Shear Strength Calculations

The calculated fiber strength has been used to estimate an interfacial shear strength using
several analyses. The simplest estimate employs the shear lag model suggested by Kelly and
Tyson [8] for a metal matrix composite. In this model, it is assumed that the load is transferred
through a constant shear stress "r at the fiber ends, with
1D
Z Lc
where D is the fiber diameter, Lc is the critical fiber length (assumed equal to four thirds of
the average fiber fragment length at mechanical saturation, L, = 4Lf13), and o~f is the fiber
strength calculated in Eq 7. This model effectively leads to an estimate of the average shear
stress acting on the fiber ends.
A second estimate of the shear strength of the interface can be calculated using the elastic
model of Cox [9,10]. This model has the advantage of including the effects of changes in the
elastic properties of the matrix caused by moisture absorption. Using this model, the interfacial
shear strength will be estimated by evaluating the maximum shear stress at the fiber end,
which is given as [10]:
Tmax
Dk(\COsinh(kLc/2))
: 1"o'f (10)
where k is a parameter involving the elastic properties of the fiber and matrix:
(11)
k = ~ (1 + Vm)(Ef---Em)ln(ZrmlD)
Em is the matrix Young's modulus, evaluated from the tensile tests described previously, Vm
is the matrix Poisson's ratio, taken to be 0.35 [11], and Ef is the fiber Young's modulus, taken
to be 67.5 GPa as before [31]. The parameter rm in the denominator represents the radius of
the cylinder of matrix assumed to surround the fiber, and has been taken as 0.5 mm; the
function is relatively independent of the value of r,,. Asloun, et al. [10] cite a suggested relation
with which to estimate the critical length to diameter ratio LflD based on this model (by
assuming that k = 2/Lc in Eq 11):
Lc /E-E xml- /2rm)] u2

(12)
Another analysis of the stress state surrounding a broken fiber has been provided by Whitney
and Drzal [11]. This model is also an elastic analysis, and can also include the effects of
moisture on the matrix material as well as incorporate swelling separately. The stress-free
condition of the broken fiber end is enforced, so that the shear stress is found to reach a
maximum away from the fiber end, while in the Cox model the maximum shear stress is
found at the fiber end. As was done with the Cox model, we will characterize the interfacial
shear stress by the maximum shear stress on the fiber. The axial tensile stress in the fiber
(uniform across the fiber cross-section) is given by
ax = [1 - (4.75E + l)e-475X]AiEo (13)
where E is the length from the end of the fiber divided by one half the critical length, Y =
2x/Lc (Whitney and Drzal define their critical length as one-half of the L~ as used in this
paper), and the parameter 4.75 is chosen by defining one-half the critical length as the point
where the stress reaches 95% of its far-field value, which is A~e0. Choosing a different
percentage would change the value of that parameter. A j is a function of the elastic properties
of the fiber and matrix that includes the effects of swelling or thermal strains, which are
denoted with overbars (g:f = o~f~T, ~m = O~m AT + es):
4KrG.,Vr
AIEO = El(E~ - -El) + K[ + G,-------~
[(vf - Vm)EO + (1 + v..)g,,, - (1 + vj)gf] (14)
For the isotropic glass fiber in this study, the quantity Kf is given by
E~ (15)
K / = 2 ( 1 + v:)(1 - 2vf)
The shear stress on the surface of the fiber near the end is given by
Wxr = --4.75btAiE0~e -4-75x (16)
where
Gm ] 1/2
(17)
The maximum magnitude of the shear stress is found at X = 1/4.75, but is independent of the
numerical parameter (chosen to be 4.75), with 'rm~ = laAl eo/e. For this model, Tmax has been
evaluated by recalculating the far-field tensile stress in the fiber from Eq 14, then refitting
In(n) to the new In(or), and determining a fiber strength (using Eq 7) that corresponds to
A l ~ - This procedure leads to a somewhat lower estimate of the fiber strength, because the
second term in Eq 14 is very small in comparison with the first term, so that An e0 is very
closely approximated by Ef~; the fiber stress in this model is therefore approximately equal
to the stress calculated using the applied strain only, with no correction for swelling. (An
alternative would be to set the far-field stress Aleo equal to the fiber strength calculated
with a swelling correction, but that choice seems incompatible with the model. On the other
hand, incorporating es into ~,, apparently leads to a discontinuity in axial strain across the
fiber/matrix interface.)
Whitney and Drzal also calculate a value for the critical length to diameter ratio (using the
definition that one-half the critical length is the point where the stress reaches 95% of its far-
field value). With this definition, the critical length is given by
Lc 2.375
--=-- (18)
D
while a different definition would lead to a different number in this equation. Note that equations
(12) and (18) lead to a similar dependency of LclD on the change in the matrix modulus.
The properties of the E-glass fibers used in the calculations are: E/= 67.5 GPa [31], vy =
0.22 and cxf = 5 • 10-6]~ [32], and are assumed to be independent of moisture content or
time of exposure.
Radial Stresses
Another interesting aspect of Whitney and Drzal's paper is a prediction for the radial stress
distribution between the fiber and the matrix. On cooling the composite after curing, the
difference in the coefficients of thermal expansion between the fiber and the matrix will lead
to a compressive stress across the fiber matrix interface. However, as moisture swells the
matrix, the stress across the interface will eventually change from compression to tension,
leading to significant weakening of the interface. The stress acting on the interface is given by
o-r = [A2 - Adx2(l - 4.75~)e-4.75x]s 0 (19)
where A2 is also a function of the elastic properties of the fiber and matrix and the swelling
and thermal strains:
_ 2KfG,.
A2eo Kf @ a"~m[(p'f" - Pro)F-"0~- (1 + Vm)E,~ -- (1 + U/)Ef] (20)
This functional representation of the stress across the interface has extreme values at E = 0
and at E = 2/4.75 , the latter being the first location where the stress would become tensile. In
order to investigate the effects of swelling, the swelling strains in Fig. 4 have been approximated
by a polynomial, with Es (in percent) = 0.152w + 0.0345w 2, where w is the wt% of
absorbed water.

The first result that must be mentioned is that it was difficult to bring dry single-fiber
fragmentation samples containing fibers with the epoxy-compatible coating (158B) to mechani-
cal saturation, making it difficult to calculate an accurate value of the interfacial shear strength
for those samples. Only two dry 158B samples (of ten or so tested) were successfully brought
to mechanical saturation; these two samples must therefore have weaker interfaces than the
average dry 158B samples. In the samples that could not be brought to mechanical saturation,
fiber fractures resulted in large cracks penetrating into the matrix, which eventually caused
failure of the specimen. In samples with weaker interfaces, the energy released from the fiber
fractures will cause debonding along the interface rather than cracking into the matrix.
Immersion in water initially made it possible to strain both the epoxy-compatible and the
vinylester-compatible samples to achieve mechanical saturation; the uptake of water results
in a combination of interfacial degradation and increased matrix fracture toughness. Interest-
ingly, after immersion at 25~ for 3000 h, both the epoxy-compatible and the vinylester-
compatible samples again displayed failure as a result of cracks originating from fiber fractures.
However, after further immersion to 4000 h at 25~ the samples could again be tested
successfully. This phenomenon seems very odd in light of the fracture toughness results in
Fig. 9, but it may only reflect some statistical anomaly in the choice of samples.
The strength of a fiber fragment at a length equal to the critical length is an important
parameter for this study, since all three of the models of interfacial strength are linearly
dependent on that measure of the fiber strength. In addition, that fiber strength represents the
strength at the onset of mechanical saturation, which should be a good indicator of the tensile
strength of a uniaxial composite. The degradation of the strength of the E-glass fiber fragments
at the critical length is clear from the data displayed in Figs. 10 and 11. Figure 10 shows the
strength of the fibers at the critical length as a function of absorbed water at both 25 and 75~
for both types of coatings; Fig. 1 l shows the same data as a function of the time of immersion.
The type of fiber coating does not seem to influence the strength of the E-glass very much.
The lines in the figures are least squares fits to the data to indicate the trends.
Figure 10 seems to indicate that the use of elevated temperatures can provide a means of
accelerating the effects of immersion at lower temperatures. However, when plotted against
time as in Fig. 11, the data seems more ambiguous. The apparent increase in strength at 75~
after 4000 h reflects the competition between degradation of the fiber and degradation of the
SCHULTHEISZ ET A L ON EFFECTS OF MOISTURE ON FIBER STRENGTH 119
4.0 u ~ n i i ~ g
I O 25~ Epoxy-CompotibFe
B 9 75~ Epoxy-Compotible
3,5 Q 25~ Vinylester-Cornpat]ble
s
B 9 75~ Vinylester-Compat[ble
[3
~, 3,0
.o
B ~ 9 9 9
C, 2,s
0 @ D 9 9 9
s 2.0
1.5 Q 9 :
iS
1.0
,,,,o
0.0 0.5 1.0 1.5
,,-:,1
2.0 2.5 3,0 3.5
Weight % Woter Absorbed
FIG. lO--Effect of water uptake on the strength of E-glass fiber fragments at the critical length.
Samples immersed in water at 25 and 75~ Fibers had either an epoxy-compatible coating (158B)
or a vinylester-compatible coating (366). Dots indicate dry samples. Line is least-squares fit through
all data points. These estimates of the fiber strength include the swelling strain as part of the
mechanically applied strain.
interface. In general, samples immersed at 75~ for more than 1500 h had 15 to 25 initial
fiber breaks (caused by swelling and/or an increased rate of chemical attack), while the room
temperature samples had no more than one or two initial breaks, However, some of the 75~
samples had very few initial breaks, indicating that the interface had been degraded to the
extent that the swelling strain was not affecting the fiber, thus possibly invalidating the crucial
assumption that the strain in the fiber (far enough away from any breaks) was identical to the
4.0
%-
(ZL
0
3.5
3,0
o
u
c.) 2.5 ~5oC Io 0
2.1D
1.5
x} 9 750C
"tZ 0
1.0
0 1000 2000 ,3000 4000 5000
T{me of IrnmeFsion (hours)
FIG. t l--Effect of time of immersion on the strength of E-glass fiber fragments at the critical
length. Samples immersed in water at 25 and 75 ~ Fibers had either an epoxy-compatible coating
(158B) or a vinylester-compatible coating (366). Dots indicate dry samples. Lines are least-squares
fit through all data points at each temperature. These estimates of the fiber strength include the
swelling strain as part of the mechanically applied strain.
strain in the matrix during the test. If the strain in the fiber was actually less than in the matrix,
the estimate of both the fiber strength at mechanical saturation and the interfacial strength
would decrease. Samples with few initial breaks showed clear evidence of long regions of
debonding adjacent to the breaks, but the fiber did not appear to have been completely separated
from the matrix throughout the specimen.
The fiber strengths calculated for the model of Whitney and Drzal [11] using the far-field
fiber stress from Eq 14 were similar, but somewhat lower, since the swelling strains were not
included as part of the mechanically applied strains. In Fig. 10, the average fiber strength
decreases from about 2.75 GPa for dry fibers to about 1.6 GPa for samples with 3.5 % absorbed
water; the fiber strength calculated using Eq 14 decreases from 2.75 GPa to 1.1 GPa at 3.5
% absorbed water.
In general, the strength of the fiber fragments at mechanical saturation after a given time
of immersion does not appear to vary much with the temperature of the water, as may be seen
in Fig. 11. However, a strong effect of water temperature is reflected in the change in the
Weibull parameters n0 and 13. Both of these parameters show a much stronger decrease for
samples immersed at 75~ than for those at 25~ as can be observed in Figs. 12 and 13. The
changes in n0 and 13couple together to result in similar fiber strengths at the onset of mechanical
saturation for the two different temperatures, as is apparent in the example shown in Fig. 5.
The decrease in 13 represents a broadening of the distribution of the sizes of the flaws, while
the decrease in e<0 represents a shift toward a larger average flaw size (leading to a lower
average fiber strength). The parameter c~0 is effectively a measure of the strength of a fiber
with length Lo = 15 mm; it is clear that this strength decreases fairly rapidly during immersion
at 75~ and eventually drops below the level of stress caused by the swelling, which is why
the 75~ samples start with initial breaks. The strength of the fiber fragments at the critical
length is less temperature dependent since this strength reflects those flaws of smaller size
present in the fibers, and these flaws are apparently attacked at a similar rate at both
temperatures.
3.0 i i i i
0 25"C Epoxy-Compatible
9 75"C Epoxy-Compatible
2.5 a 25"C V~nylester-Compatlble
a Q i 7beC Vinylester-Compatlbl~
c~
2.0
1.5 ~ /25~ a
~
E
u
1.0
=.
"~, 0.5 ~/75"C
0.0
I I 1 I i
0 1000 2000 .3000 4000 000
Time ~hours)
FIG. 12--Change in Weibull scale parameter c~o calculated from the fragmentation process as
a function of time of immersion in water at 25 and 75~ with an epoxy-compatible coating
(158B) and fibers with a vinylester-compatible coating (366). The swelling strain is included as
part of the mechanically applied strain. The lines are least squares fit to c~o = (1.75 GPa)exp[-t/
t~(T)], with t~(25~ = 9110 h and t~(75~ = 1100 h.
2.5 ,
25"c Epo~-compodb,e I
75"C E p o x y - C o m p a t i b l e II
2O 25"C Vinylester-CompatTbte I'~
75*C Vinylester-Compatible J~
E e 0
o
o Q
o~ ~o
I I I I I
0 I000 2000 3000 4-000 5000
Time (hours)
FIG. 13--Change in Weibull shape parameter ~ calculated from the fragmentation process as
a function of time o f immersion in water at 25 and 75~ with an epoxy-compatible coating
(158B) and fibers with a vinylester-compatible coating (366). The swelling strain is included as
part of the mechanically applied strain. The lines are least squares fit to ~ = 9.10 exp[-t/t~(T)],
with tB(25~ = 19 700 h and t~(75~ = 1090 h.
Ratips of the critical length to diameter of the fiber fragments (LID) are shown in Figs.
14 and 15 for the vinylester-compatible (366) and epoxy-compatible (158B) fiber coatings.
The values of LiD for both types of coating are very similar, although the epoxy-compatible
fibers generally result in somewhat lower critical length to diameter ratios, corresponding to
slightly higher interfacial shear strengths. For both coatings, immersion at 25~ results in very
little change in LiD. However, after 3000 h of immersion, some of the samples at 75~ display
100 25"C Vinylester-CompatTble 1

75"C V ~ n y l e s t e r - C o r n p a t l b l e
Resceled ModeJ Predictions
80
75"C
s [ 9
40 o
Q
20
0 1 I i I I
0 1000 2000 3000 4 - 0 0 0 5000
Time of Immersion (hours)
FIG. 14--Change in the critical length-to-diameter ratio of the E-glass fibers as a function of
time of immersion in water at 25 and 75~ for fibers with a vinylester-compatible coating (366).
The curves represent the analytical predictions of both the Cox model/9,10] and the model of
Whitney and Drzal [11], but with parameters in the models rescaled to match the dry samples at
zero time.
i i J J i
1 O0
b 2s*c~p;xy-~ompoub4e t
9 7 5 " C Epoxy-Compatible 9
- - Rescoled Model P'ed ct ons~
80
o ,60 75"C
\
40
9 2 .c
.o~O~ o~,
20
0 I I ! 30100 .t-
O 1000 2000 4000 5000
Time of immersion (hours)
FIG. 15--Change in the critical length-to-diameter ratio of the E-glass fibers as a function of
time of immersion in water at 25 and 75~ for fibers with an epoxy-compatible coating (158B).
The curves represent the analytical predictions of both the Cox model/9,10] and the model of
Whitney and Drzal [11], but with parameters in the models rescaled to match the dry samples at
zero time.
a large increase in L i D corresponding to the weak interfaces (with few initial breaks) discussed
previously. Also shown in these figures are the predictions of the change in L i D from the
models of Whitney and Drzal [11] and Cox [9,10]. These models depend only on the change
in matrix modulus with absorbed water, and do not explicitly include the degradation of the
fiber strength in the calculations. The curves shown actually depict the predictions of both
models, since their functional dependence on the matrix modulus is very similar; however, in
both cases the model predictions have been rescaled so that the LJD at zero time matches the
L i D of the dry samples (averaged over all samples including both coatings). For the model
of Cox [9,10], it was suggested that the parameter k in Eq 11 was equal to 2/Lc: rescaling the
prediction to match the experiments indicates that k for these materials is closer to 7.5/Lr
Similarly, Whitney and Drzal [11] assume that the critical length corresponds to a fragment
that achieves 95% of the expected far-field stress; matching the model to the experiments
indicates that a fragment at the critical length reaches 99.96% of the expected far-field stress
(changing the parameter 2.375 in Eq 18 to 5.10).
The interfacial shear strengths estimated using the model of Kelly and Tyson [8] are shown
in Figs. 16 and 17 for the vinylester-compatible (366) fiber coating; the data is shown as a
function of absorbed water in Fig. 16 and as a function of time of immersion in Fig. 17.
Figures 18 and 19 show similar data for the epoxy-compatible (158B) fiber coating. Both
coatings apparently result in similar dry interfacial shear strengths of approximately 35 to 40
MPa. The scatter in the data in these figures arises from scatter in both the measured strength
of the fibers and the critical length to diameter ratios.
As with the fiber strength data in Fig. 10, Fig. 16 indicates that the immersion at elevated
temperatures could serve as an accelerated test method for the vinylester-compatible coating.
However, Fig. 17 also indicates that the degradation of the interface might be modeled as a
linear function of time of immersion, with different rates for different temperatures.
In contrast, the data for the epoxy-compatible coating in Fig. 18 has too much scatter to
draw convincing conclusions as to the use of elevated temperature to accelerate tests of this
coating. One difficulty is that the two data points from dry samples only indicate the lowest
interfacial strengths of the dry specimens, as the other samples fractured, as mentioned pre-
60
0 25"C Vinylester-Competible ]
9 7 5 * 8 Vinylester-Compatlble
s0
o
J::
o o
40
03
o
o~ 3 0
o
o
,~ 2 0
c
10 I I I I _t i I
0.0 0.5 1.0 1.5 2.0 2.5 3,0 3.5
Weight ~ Weter A b s o r b e d
FIG. 16~Interfacial shear strength calculated using the model of Kelly and Tyson [8] as a
function of wt% water absorbed after immersion in water at 25 and 75~ forfibers with a vinylester-
compatible coating (366). Line is least-squares fit through all data points. Dots indicate dry samples.
The swelling strain is included as part of the mechanically applied strain.
viously. Also, it appears that the elevated temperature may actually increase the epoxy-
compatible interface initially; this trend may indicate that the higher temperature leads to
additional chemical bonding at the interface that outweighs the adverse effects of the absorbed
water. The data in Fig. 19 seem to indicate that modeling the interfacial degradation as a linear
function of time would be more appropriate, but the scatter in the data is still not reassuring.
Interfacial shear strengths calculated using the Cox model [9,10] with the fiber strength
determined including the swelling strains are shown in Figs. 20 through 23. These values
follow the same trends as discussed above for the model of Kelly and Tyson, although these
60
I 13 25~ V i n y l e s t e r - C o m p o t i b l e
9 75*C V [ n y l e s t e r - C o m p o t i b l e
50
v o
o o
B
~ 40 o
~ I ~ e 25"C
~ 30
o
~ 20
c 9 75"C -m~...~m
t0 I I I __1 I
0 1000 2000 3000 4000 000
Time of ; m m e r s l o n (hours)
FIG. I7--1nterfacial shear strength calculated using the model of Kelly and Tyson [8] as a
function of time of immersion in water at 25 and 75~ for fibers with a vinylester-compatible
coating (366). Lines are least-squares fit through all data points at each temperature. Dots indicate
dry samples. The swelling strain is included as part of the mechanically applied strain.
60
' _' ~- L ' '
1': 25;C Epoxy-Compat;ble]
75~ Epoxy-Compatible
5o
40
30
o ~ ~
3
20
%
10 I I l I I I I
0.0 0.5 1,0 1.5 2.0 2,5 3.0 3.5
Weight 9 WaterAbsorbed
FIG. 18--lnterfacial shear strength calculated using the model of Kelly and Tyson [8] as a
function of wt% water absorbed after immersion in water at 25 and 75~ for fibers with an epoxy-
compatible coating (158B). Line is least-squares fit through all data points. Dots indicate dry
samples. The swelling strain is included as part of the mechanically applied strain.
strengths are considerably higher than those calculated using the Kelly-Tyson model (as they
should be, since the Kelly-Tyson model represents an average stress, while the Cox model
indicates the peak stress). In fact, the calculated strengths are higher than the matrix yield
stress, indicating that the elastic analysis may need modification in a region close to the end
of the fiber. This model follows the trend of the decrease in fiber strength more closely, since
the dependence on L i D is much less than in the Kelly-Tyson model. Again, the immersed
samples indicate very little dependence on temperature or coating type.
60
50
~-: 25~ Epoxy-Compatible
75~ Epoxy-Compolible
v
I ~
4-0
0 o
9
~ 3o
u ) 0 ~ 0 * 90
~ 20
I~ I I I 9
,o o' 1000 2000 3000 4000 5000
T~me of I m m e r s i o n ( h o u r s )
FIG. 19--1nterfacial shear strength calculated using the model of Kelly and Tyson [8] as a
function of time of immersion in water at 25 and 75~ for fibers with an epoxy-compatible coating
(158B). Lines are least-squares fit through all data points at each temperature. Dots indicate dry
200
o 25"C Vinylester-Compot~ble
180
o 9 75"C Viny]ester-Compatible
160
x:
[]
~4o
120
D 0 9
o~ 100
'~ 80
c 60
40
0.0 0,5 1 0 1 5 2 0 2,5 3.0 3.5
Weight ~ Water A b s o r b e d
FIG. 20~Interfacial shear strength calculated using the model of Cox/9,101 as a function of
wt% water absorbed after immersion in water at 25 and 75~ for fibers with a vinylester-compatible
coating (366). Line is least-squares fit through all data points. Dots indicate dry samples. The
swelling strain is included as part of the mechanically applied strain.
Interfacial shear strengths calculated using the model of Whitney and Drzal [11] are shown
in Figs. 24 and 25. These strengths are very similar to the results from the Cox model, but
somewhat lower, since the calculated fiber strengths are lower because the swelling strain is
not included as part of the mechanically applied strain. The function Ix changes rapidly upon
immersion, after which it remains fairly constant and the results mirror exactly the decrease
in the fiber strength. These strengths also represent peak stresses, and as with those found
using the Cox model, these stresses also exceed the yield stress of the matrix, indicating a
need for an elastic/plastic analysis.
200 , , , .... ,
E] / 0 25~ VTnylester-Compotible
L9 75~ Vinylester-Compat{ble
150 o
x: o
J 2 5=C o
.g
{ 50
0 I I I 1 I
0 1000 2000 3000 4000 5000
Time of I m m e r s i o n (hours)
FIG. 21--lnterfacial shear strength calculated using the model of Cox/9,10] as a function of
time of immersion in water at 25 and 75~ for fibers with a vinylester-compatible coating (366).
Lines are least-squares fit through all data points at each temperature. Dots indicate dry samples.
200 = p J , ,
180
lo 2 .o E~
75"C E p o x y - C o m p a U b l e
~ 160
~ 140
m
120
oo .;. :..
100
g
't
80
m
c
40 ,, ,~
0.0 0.5 I .0 1.5 2.0 2.5 3.0 3.5
Weight ~ Water Absorbed
FIG. 22--1nterfacial shear strength calculated using the model of Cox [9,10] as a function of
wt% water absorbed after immersion in water at 25 and 75~ for fibers with an epoxy-compatible
coating (158B). Line is least-squares fit through all data points. Dots indicate dry samples. The
swelling strain is included as part o f the mechanically applied strain.
The influence of absorbed moisture on the radial stresses at the interface is shown in Fig.
26; the expansional strains consist of a thermal component assuming a 75~ quench plus the
strain caused by swelling. When there is no applied strain and no moisture has been absorbed,
the thermal strains induce a contraction across the interface everywhere along the fiber, as
shown by the dashed lines. At approximately 2% absorbed water, the radial stresses at the
interface change from compression to tension, which should tend to promote debonding at
the interface (note that the samples at 75~ have reached 2.5% absorbed moisture after 1800
h of immersion, while the room temperature samples are near 1.5%). The addition of an
200
~ - 25~ E p o x y - C o m p a t i b l e
180 L 9 75=c E p o x y - C o m p a t i b l e
g.
v 160
oa 14o I
U~
120 o 9
100 / 9 C
"G
E 60 o 9
40 I I I I I
0 1000 2000 3000 4000 5000
Time of Immersion (hours)
FIG. 23--1nterfacial shear strength calculated using the model of Cox [9,10] as a function of
time of immersion in water at 25 and 75~ for fibers with an epoxy-compatible coating (158B).
Lines are least-squares fit through all data points at each temperature. Dots indicate dry samples.
SCHULTHEISZ E T AL. O N E F F E C T S OF MOISTURE ON FIBER STRENGTH 127
200
25"C Vinylest e r - C o m patible

7 5 " C V i n y l e s t e r - C o m potible
[3
150
s
100
~n
-6
a ~
o 50
I I I I I I I
0.0 0.5 1.0 1,5 2.0 2.5 3.0 3.5
Weight ~; Woter Absorbed
FIG. 24---Interfacial shear strength calculated using the model of Whitney and Drzal [11] as a
function of wt% water absorbed after immersion in water at 25 and 75~ for fibers with a vinylester-
compatible coating (366). Line is least-squares fit through all data points. Dots indicate dry samples.
applied strain of 1% causes a significant increase in the maximum radial stress near the end
of a fiber fragment, so that the radial stress is tensile for all levels of absorbed water; the far
field stress on the fiber actually changes to be more compressive, however, delaying the
transition to a slightly higher level of absorbed moisture, which could significantly increase
the time for that transition, as the rate of uptake of water decreases considerably at higher
levels of water content, as shown in Fig. 2.
200 ' ' L _' ' '

I I0 25;C Epoxy-Compatible l
u 75"C Epoxy-Compatible
o-
150
m 0
m G 0 9~ :
100
vl
u
a 50 0 9 0:0
E
0.0 0.5 ~.0 1.5 2,0 2.5 3.0 3.5
Weight % Water Absorbed

FIG. 25--1nterfacial shear strength calculated using the model of Whitney and Drzal [11] as a
function of wt% water absorbed after immersion in water at 25 and 75~ for fibers with an epoxy-
compatible coating (158B). Line is least-squares fit through all data points. Dots indicate dry
I I I I I
8O
maximum
/" stress
60
4o
s .... E0 = 10
CO
20
maximum _.~-2272:
stress
-20
I I I I I
0 1 2 3 4
Weight N Water Absorbed
FIG. 26--Effect of absorbed water on radial tensile stresses at the interface calculated using
the model of Whitney and Drzal [11]. Curves indicate the maximum and far-field radial tensile
stresses after a 75~ quench, for zero applied mechanical strain and for an applied mechanical
strain of 1%. Maximum stresses occur near the fiber ends; the far-field stresses act over most of
the fiber surface away from the ends.
Conclusions
The method used in this paper to analyze the data from the single-fiber fragmentation test
is intended to provide an estimate of both the strength of the fiber and the strength of the
interface from the same test. Some of the effects of immersion appear to occur quite rapidly,
and although this work is aimed at long-term durability, it may be useful to look more closely
at the initial changes in order to investigate degradation mechanisms. It is clear that degradation
of the glass fibers will affect the tensile strength of a composite; degradation of the fiber/
matrix interface will also reduce the strength of a composite in a number of ways [33-35].
From the analysis, the strengths of E-glass fibers immersed in water seem to show a fairly
constant rate of decrease with the level of absorbed water, and the influence of temperature
can be seen in the changing Weibull parameters calculated from the fragmentation process.
However, it appears that initial variability in the interface can influence the relative rates of
degradation of fiber and interface. An initially strong interface should transfer the swelling
strain to the fiber, leading to increased static fatigue and stress corrosion [12-14], resulting
in a weaker fiber. An initially weak interface, however, may experience considerably more
debonding and possibly insufficient stress transfer during testing, making the fiber appear
stronger than it actually is, leading to possibly anomalous results. The assumption that the
strain in the fiber is the same as the strain in the matrix is crucial in calculating accurately
the fiber strength; some independent method of measuring the glass fiber stress would be very
useful, as has been shown in Raman spectroscopic investigations employing carbon fibers [36].
The two coatings were intended to provide an initial difference in interfacial strength and
possibly show different rates of change. However, in general, the two coatings led to similar
interfacial shear strengths, regardless of the model used to evaluate those strengths. The tensile
and fracture tests indicate significant changes in the matrix properties with absorbed water;
the models of Cox [9,10] and Whitney and Drzal [11] do include these effects to some extent,
but the interfacial shear strengths calculated using those models exceed the measured matrix
yield stress, indicating a need for some form of elastic/plastic analysis. The model of Kelly
and Tyson [8] effectively provides a measure of the average shear strength of the interface,
and so it is appropriate to include debonded regions of the fiber in measuring the fragment
length; however, for the elastic analyses [9-11[, regions of debonding would violate the
assumptions underlying the models, so the lengths of those regions should be removed when
evaluating'the average fragment length. Debonded regions can be identified to some extent
by the patterns of birefringence at the fiber/matrix interface, but we have not yet incorporated
a consistent method of measuring any debonded regions in our procedure as yet.
The fibers with the epoxy-compatible coating (type 158B) showed considerably greater
variability in the interfacial strength when compared to the fibers with the vinylester-compatible
coating (type 366). This result may be an artifact arising from the processing of the fibers or
the construction of the fragmentation samples.
The dry samples of the epoxy-compatible (158B) fibers are assumed to have stronger
interfaces than the vinylester-compatible (366) fibers on average, as most of the dry epoxy-
compatible samples could not be taken to mechanical saturation because of matrix crack-
ing originating from fiber fractures. Such strong interfaces might actually be a problem in
samples with cracks across the fibers, if it suppresses the toughening mechanism of crack
blunting through debonding along fibers. Oddly, the samples with both coatings immersed at
25~ for 3000 h also failed from matrix cracks initiated by fiber fractures, but this result does
not seem compatible with the consistently decreasing interfacial strengths and increasing matrix
fracture toughness with absorbed water. In any event, further immersion to 4000 h then allowed
for successful tests.
Most of the samples used in these tests had their ends capped using a 5-min epoxy to isolate
the end of the fiber (and the interface) from direct contact with the water, but the capped and
uncapped samples appear to show no significant differences. Assuming that the water travels
preferentially along the interface, it may be that the end caps are thin enough or not sufficiently
well-bonded to the matrix to allow water to reach the end of the fiber fairly quickly, after
which the moisture travels along the interface and attacks the fiber very quickly.
The effect of end-caps is also in question because these tests indicate a much faster rate of
degradation of the E-glass fiber compared to the interface, which is opposite to results from
earlier tests which did not isolate the fiber ends [2]. This study employed different coatings
from those in [2], which may be the cause for this difference, or it may be that a less restricted
flow of water along the interface will affect the chemical bonds at the interface more than the
structure of the glass itself.
In a composite structure, it is expected that the higher content of fibers will constrain the
amount of strain caused by swelling in the direction parallel to the fibers. However, there
will be much less of a constraint in the directions perpendicular to the fibers, so that the
effect of swelling strains on the radial stresses at the interface (leading to debonding) may
be more important in structural applications than the swelling-induced fragmentation and
degradation of the strength of the fibers. In addition, constraint along the fiber direction
may actually increase the radial tension, if the volume of the swollen matrix is the same as
in an unconstrained sample.
Finally, the considerable scatter in the data makes it clear that further testing is necessary
to determine the repeatability and reliability of the results. It is certainly much easier to
fabricate and control the quality and environment of these microcomposites than for full-
scale composite test samples. However, the test as it is now performed at NIST is tedious
and time-consuming, and some of the scatter may result from the fact that at least six different
operators have collected data for this study. It is expected that these difficulties will be
overcome through automation of the test using an acoustic emission detector or image
processing system in place of the current manual loading system and detection of fractures
using the microscope. One issue that still needs to be addressed, however, is how many
samples need to be tested, and from how many different locations within a spool of fiber,
in order to obtain a representative sample size with which accurate predictions can be made
about composite structures.
Acknowledgments
The authors gratefully acknowledge the contribution of the glass fibers from Owens-Coming,
with considerable help from David Dwight. We also thank Martin Chiang for consultations
regarding modeling of the results.
References
[1] Schutte, C. L., "Environmental Durability of Glass-Fiber Composites," Materials Science and
Engineering, Vol. R13, No. 7, 1994, pp. 265-324.
[2] Schutte, C. L., McDonough, W., Shioya, M., McAuliffe, M., and Greenwood, M., "The Use of a
Single-Fibre Fragmentation Test to Study Environmental Durability of Interfaces/[nterphases
Between DGEBA/mPDA Epoxy and Glass Fibre: The Effect of Moisture," Composites, Vol. 25,
1994, pp. 617-624.
[3] Shioya, M., McDonough, W. G., Schutte, C. L., and Hunston, D. L., "Test Procedure for Durability
Studies of the Fiber-Matrix Interface," Proceedings of the 17th Annual Meeting and Symposium
on Particle Adhesion, The Adhesion Society, Orlando, FL, 20-23 February 1993, pp. 248-251.
[4] Narkis, M., Chen, J. H., and Pipes, R. B., "Review of Methods for Characterization of Interfacial
Fibre-Matrix Interactions," Polymer Composites, Vol. 9, 1988, pp. 245-251.
[5] Herrera-Franco, P. J. and Drzal, L. T., "Comparison of Methods for the Measurement of Fibre/
Matrix Adhesion in Composites," Composites, Vol. 23, 1992, pp. 2-27.
[6] McDonough, W. G., Herrera-Franco, P. J., Wu, W. L., Drzal, L. T., and Hunston, D. L., "Fibre-
Matrix Bond Tests in Composite Materials," 23rd International SAMPE Technical Conference,
21-24 October 199I, pp. 247-258.
[7] Drzal, L. T. and Herrera-Franco, P. J., "Composite Fiber-Matrix Bond Tests," Engineered Materials
Handbook, Adhesives and Sealants 3, ASTM International, Metals Park, OH, 1990, pp. 391-405.
[8] Kelly, A. and Tyson, W. R., "Tensile Properties of Fibre-Reinforced Metal: Copper/Tungsten and
Copper/Molybdenum," Journal of the Mechanics and Physics of Solids, Vol. 13, 1965, pp. 329-350.
[9] Cox, H. L., "The Elasticity and Strength of Paper and Other Fibrous Materials," British Journal
of Applied Physics, Vol. 3, 1952, pp. 72-79.
[10] Asloun, El. M., Nardin, M., and Schultz, J., "Stress Transfer in Single-Fibre Composites: Effect
of Adhesion, Elastic Modulus of Fibre and Matrix, and Polymer Chain Mobility," Journal of
Materials Science, Vol. 24, 1989, pp. 1835-1844.
[11] Whitney, J. M. and Drzal, L. T., "Axisymmetric Stress Distribution Around an Isolated Fiber
Fragment," Toughened Composites, ASTM STP 937, N. J. Johnston, Ed., American Society for
Testing and Materials, Philadelphia, 1987, pp. 179-196.
[12] Charles, R. J., "Static Fatigue of Glass. I," Journal of Applied Physics, Vol. 29, 1958, pp. 1549-1553.
[13] Charles, R. J., "Static Fatigue of Glass. II,"Journal of Applied Physics, Vol. 29, 1958, pp. 1554--1560.
[14] Metcalfe, A. G. and Schmitz, G. K., "Mechanism of Stress Corrosion in E Glass Filaments," Glass
[15] Jones, C. J., Dickson, R. F., Adam, T., Reiter, H., and Harris, B., "Environmental Fatigue of
Reinforced Plastics," Composites, Vol. 14, 1983, pp. 288-293,
[16] Jones, C. J., Dickson, R. E, Adam, T., Reiter, H., and Harris, B., "The Environmental Fatigue
Behaviour of Reinforced Plastics," Proceedings of the Royal Society of London, Series A, Vol.
396, 1984, pp. 315-338.
[17] Wagner, H. D. and Eitan, A., "Interpretation of the Fragmentation Phenomenon in Single-Filament
Composite Experiments," Applied Physics Letters, Vol. 56, 1990, pp. 1965-1967.
[18] Yavin, B., Gallis, H. E., Scherf, J., Eitan, A., and Wagner, H. D., "Continuous Monitoring of the
Fragmentation Phenomenon in Single Fiber Composite Materials," Polymer Composites, Vol. 12,
1991, pp. 436--446.
[19] Wagner, H. D., Wood, J. R., and Marom, G., "Clarifying the Application of Weibull Statistics for
Determining the Stress State of a Fibre From Fragmentation Tests," Advanced Composites Letters,
Vol. 2, 1993, pp. 173-176.
[20] Plueddemann, E. P., Silane Coupling Agents, Second Edition, Plenum Press, New York, 1991.
[21] Patrick, R. L., Gehman, W. G., Dunbar, L. and Brown, J. A., "Scanning Electron. Microscopy of
Fracture Surfaces," Journal of Adhesion, Vol 3, 1971, pp. 165-175.
[22] Macturk, K., Schutte, C., Schultheisz, C., Hunston, D., and Tarlov, M., "The Role of Coupling
Agents in Composite Interface Strength and Durability," Materials Research Society Symposium,
San Francisco, CA, April 1995.
[23] Drzal, L. T., Rich, M. J., Camping, J. D., and Park, W. D., 'qnterfacial Shear Strength and Failure
Mechanisms in Graphite Fiber Composites," Proceedings of the 35th Annual Technical Conference,
Reinforced Plastics/Composites Institute, Paper 20C-1, 1980, pp. 1-5.
[24] Shen, C. -H. and Springer, G. S., "Moisture Absorption and Desorption of Composite Materials,"
Journal of Composite Materials, Vol 10, 1976, pp. 2-20.
[25] Wong, T. C. and Broutman, L. J., "Water in Epoxy Resins Part II. Diffusion Mechanism," Polymer
Engineering and Science, Vol. 25, 1985, pp. 529-534.
[26] Henstenburg, R. B. and Phoenix, S. L., "Interfacial Shear Strength Studies Using the Single-
Filament-Composite Test. Part II: A Probability Model and Monte Carlo Simulation," Polymer
Composites, Vol. 10, 1989, pp. 389-408.
[27] BaiUie, C. A. and Bader, M. G., "Strength Studies of Single Carbon Fibres in Model Composite
Fragmentation Tests," Composites, Vol. 25, 1994, pp. 401-406.
[28] Curtin, W. A., "Exact Theory of Fibre Fragmentation in a Single-Filament Composite," Journal
of Materials Science, Vol. 26, 1991, pp. 5239-5253.
[29] Waterbury, M. C. and Drzal, L. T., "On the Determination of Fiber Strengths by In-Situ Fiber
Strength Testing," Journal of Composites Technology and Research, JCTRER, Vol. 13, 1991,
pp. 22-28.
[30] ASTM D 5045-93, "Standard Test Methods for Plane-Strain Fracture Toughness and Strain Energy
Release Rate of Plastic Materials," Annual Book of ASTM Standards, Vol. 08.03, 1994, pp. 310-318.
[31] Shioya, M., Tokyo Institute of Technology, private communication, 1993.
[32] Encyclopedia of Materials Science and Engineering, Supplementary Vol. 2, MIT Press, Cambridge,
MA, 1990, p. 838.
[33] Madhukar, M. S. and Drzal, L. T., "Fiber-Matrix Adhesion and Its Effect on Composite Mechanical
Properties. I. Inplane and Interlaminar Shear Behavior of Graphite/Epoxy Composites," Journal
of Composite Materials, Vol. 25, 1991, pp. 932-957.
[34] Madhukar, M. S. and Drzal, L. T., "Fiber-Matrix Adhesion and Its Effect on Composite Mechanical
Properties. II. Tensile and Flexural Behavior of Graphite/Epoxy Composites," Journal of Composite
Materials, Vol. 25, 1991, pp. 958-991.
[35] Madhukar, M, S. and Drzal, L. T., "Fiber-Matrix Adhesion and Its Effect on Composite Mechanical
Properties. III. Longitudinal (0~ Compressive Properties of Graphite/Epoxy Composites," Journal
of Composite Materials, Vol. 26, 1992, pp. 310-333.
[36] Schadler, L. S., Laird, C., Melanitis, N., Galiotis, C. and Figueroa, J. C., "Interfacial Studies on
Carbon/Thermoplastic Model Composites Using Laser Raman Spectroscopy," Journal of Materials
Science, Vol. 27, 1992, pp. 1663-1671.
Catherine W o o d I a n d Walter B r a d l e y 1
A New Technique to Study the Interfacial

Strength and Transverse Cracking
Scenario in Composite Materials
REFERENCE: Wood, C. and Bradley, W., "A New Technique to Study the Interfacial
Strength and Transverse Cracking Scenario in Composite Materials," Fiber, Matrix, and
Interface Properties, ASTM STP 1290, C. J. Spragg and L. T. Drzal, Eds., American Society
for Testing and Materials, 1996, pp. 132-151.
ABSTRACT: A new technique has been developed to study interfacial debonding and subse-
quent coalescence into transverse cracks using a loading stage in an environmental scanning
electron microscope. This technique has been used to study the effect of moisture on the
interfacial strength and transverse cracking behavior of unaged composite specimens (~ 1.1%
of matrix weight is moisture) compared to composite specimens saturated by soaking in seawater
(~ 2.2% of matrix weight is moisture). Specimens experienced both initiation controlled failure
and propagation controlled failure. The type of failure did not appear to depend upon the
moisture content; rather it depended upon the microstructure of the composite. Analysis has
also been performed, aiding in the interpretation of the experimental results. The mathematical
models show that the interfacial residual stresses in the radial direction due to thermal cooldown
are compressive in a homogeneous microstructure; while in an inhomogeneous microstructure,
the stresses are tensile at the boundary of the inhomogeneity. In both cases the presence of
moisture relaxes the stresses.
KEYWORDS: transverse tensile strength, ESEM, moisture, debonding, carbon fiber composite
Interfacial strength is recognized as a critical property which can significantly affect many
different mechanical properties in a composite material system. At least four tests have been
developed in recent years to measure the interfacial strength in composite materials: (1)
fiber pull-out [1], (2) single-fiber critical length method [2], (3) microdebonding [3], and (4)
microbead [4]. These tests have recently been reviewed by Narkis, Chen, and Pipes [5]. It is
worth noting that for each of these tests, the loading produces mechanical stressing that is
primarily shear, though secondary tensile stresses may also be induced. Furthermore, residual
stresses which may be both normal to the interface and parallel to the interface result from
differential thermal contraction between the matrix and the fiber.
Transverse cracking in multi-axial laminates is generally the result of stresses applied
perpendicular to the fiber orientation in plies with a 90 ~ orientation. Prediction of this very
common initial failure event for tension and fatigue loading would presumably require knowl-
edge of the interfacial strength for a state-of-stress in which the mechanical stress is primarily
normal rather than shear. Furthermore, one might reasonably expect the interfacial strength
to vary with the state-of-stress in as much as normal stresses would test principally chemical
and physical adhesion while shear loading could test some combination of chemical and physical
~Research assistant and professor, respectively,Center for Mechanics of Composites and the Department
of Mechanical Engineering, Texas A&M University, College Station, TX 77843-3123.
132
Downloaded/printed
WOOD AND BRADLEY ON INTERFACIAL STRENGTH 133
adhesion and mechanical interlocking, particularly when the residual stresses perpendicular to
the interface (in the radial direction) are compressive, which is usually the case.
Our immediate interest is to (1) develop a better technique to assess the interracial strength
of a composite when the mechanical loading produces primarily stresses which are normal
rather than shear at the fiber/matrix interface and (2) to better define the relationship between
interracial strength and transverse tensile strength in the 90 ~ plies of muki-axial composite
laminates. Our longer-term goal is to use this new technique to study the effect that moisture
has on the inteffacial strength of composite materials, which then influences the transverse
strength and other mechanical properties which might be important for offshore applications
of composite materials [6].
In this paper we present a new experimental technique to directly observe interfacial failures
in composite materials where the principal mechanical stresses are applied perpendicular to
the fiber. The use of this technique in combination with finite element analysis will allow the
determination of the interracial strength for a state-of-stress which is principally normal to
the interface (rather than shear, or parallel to the interface). This technique will also allow the
determination of the failure scenario which leads to transverse cracking and delamination in
composite materials.
Experimental Procedure
A tensile specimen is used which has a region of the gage section with a reduced cross-
sectional area, as shown in Fig. 1, and contains some (but not all) plies perpendicular to the
tensile axis of the gage section. This specimen is carefully polished on one of its edges
(thickness/length plane) with sufficient care to reveal the fibers and matrix deafly. For this
initial work, we have used a TACTIX 556 thermosetting resin with IM-7 fibers, laid up with
a (+45/0/-45/90)~ stacking sequence, Polishing is accomplished with specially developed
techniques which use grinding and polishing materials attached to metal doweling which is
rotated using a drill. This polishing tool used in combination with templates has allowed us
to polish specimens with a uniform thickness to have a gradually varied gage width along the
length of the gage section, which causes the initial damage in the composite as it is loaded
to be localized in a region near the center of the gage where the specimen width is minimum,
and therefore, the stresses in the composite are maximum.
The specimens are subsequently tested in an ElectroScan environmental scanning electron
microscope (ESEM) which obviates the need for coating the specimen to avoid charging. An
uncoated specimen is desirable because with coated specimens it is difficult to discern whether
it is the coating or the specimen which has cracked. Also, the environment inside the ESEM
chamber can be adjusted to 100% humidity so that the saturated specimens will not outgas
moisture. Testing is conducted using a mechanical stage manufactured by Ernest Fullham,
Edge polished for viewing in ESEM
,qw
m 9 0 ~ plies
FIG. 1--Tensile specimen.
Inc. (4450 N capacity) with a mechanical drive, linear variable displacement transducer (LVDT),
and a load cell, with the specimen gage section loaded in nominally uniaxial tension. The
displacement is increased incrementally with the specimen carefully examined at each displace-
ment level to determine whether local interracial failure has begun, or possible transverse
cracking, which may sometimes develop immediately upon local interfacial failure. Since the
test is run in displacement control, the stresses within the specimen will relax, causing the
specimen to remain stable, during the ten to fifteen minutes it takes to thoroughly examine
the surface. This process of step by step loading of the specimen with careful examination of
the specimen at each displacement step is continued until transverse cracking completely across
the 90 ~ ply is noted.
Wet and nominally dry specimens were studied in this program. The dry specimens were
tested as received without either artificial drying or soaking; however, one piece of this laminate
was artificially dried in a furnace and was found to contain approximately I. 1% of the matrix
weight in water. These specimens will be called unaged hereafter to distinguish them from
specimens aged (or soaked) by immersion in seawater until they approached a saturated level
of seawater absorption.
The aged specimens were soaked in simulated seawater (2.3% NaC1 by weight) prepared
from distilled water and "Instant Ocean," a mix of salt and trace elements that can be purchased
at pet stores to provide a simulated ocean environment for tropical fish. The specimens were
weighed periodically and it was determined that the eight-ply laminate reached saturation for
moisture absorption in approximately four months (Fig. 2) after absorbing an additional 1. 1%
of the matrix weight in water, for a total of 2.2 wt%. The saturated specimens were then
separated into two groups to see if time dependent degradation of the interfacial strength in
this composite might result. One group, which will be called short-term aged, was maintained
for an additional five months, for a total of nine months. The other specimens, which will be
called long-term aged, were conditioned for a total of two years. Since both sets of aged
specimens began soaking before a polishing procedure had been established, the specimens
were polished approximately one week before testing and then reimmersed in seawater. It was
determined that the specimens lost approximately 0.04% of the saturation moisture content
during the polishing procedure, and the specimens had almost completely regained this lost
moisture after three days of reimmersion.
As the neat resin was not available for investigation, it was not possible to determine how
moisture affects the properties of the resin. However, we expect that the usual reduction in
modulus occurred with the accompanying increase in toughness. If the reduction in matrix
properties with moisture absorption is the primary factor in moisture induced composite
degradation, then failure will begin with matrix cracking at either incipient flaws in the matrix
or at the three or nine o'clock locations around the fibers in the 90 ~ ply. If interfacial failures
0 40~4
~ 0 4042
0 2 4 6 g I0 12
TLmr (square root days)
FIG. 2--Saturation curve of carbon~epoxy composite.
dominate due to moisture absorption, debonding at six and twelve o'clock positions around
the fibers will be seen.
The laminate stresses measured using this new technique were transformed to 90 ~ ply stresses
using a FORTRAN program based upon classical laminated plate theory. The stresses reported
in the experimental results are the 90 ~ ply stresses.
A lower magnification picture of the unaged composite specimen studied in this work is
seen in Fig. 3. The center four plies ( - 4 5 / 9 0 / 9 0 / - 4 5 ) are seen bounded on either side by 0 ~
plies. Not seen in this figure are the +45 ~ plies on each surface. Only one small incipient
interfacial failure was noted in the specimen before testing began, as seen in Fig. 4, and this
incipient failure did not cause the major transverse crack which eventually developed in
the specimen.
The first sign of applied load induced interfacial failure is found at a ply stress of 15 MPa
in this unaged specimen, as seen in Fig. 5. At a slightly higher ply stress of 17 MPa, additional
interfacial failures are noted, as seen in Fig. 6. Surprisingly, these local interfacial failures
were not found to grow as the stress was increased incrementally from 17 MPa to 21 MPa.
Finally, at a ply stress of 22 MPa, one of the local debonds began to grow, as seen in Fig. 7.
With the load continuing to be increased monotonically, growth of this interfacial flaw is still
quite localized up to a stress of 38 MPa, as seen in Fig. 8. At a 90 ~ ply stress of 38 MPa, a
second transverse crack which apparently grew from the incipient interfacial failure previously
noted in Fig. 4 is seen growing toward the primary transverse crack, as seen in Fig. 9. The
two coalesce at a ply stress of 39 MPa, and subsequently, this transverse crack grows rapidly
FIG. 3--Low magnification view of polished edge of unaged composite with 0/-45/90/90/-45/0
plies clearly seen in ESEM.
FIG. 4---Low and high magnification ESEM view of 90~ ply with a small incipient interfacial
debond prior to loading.
FIG. 5--First indication of interfacial Jailure in fibers that bound a resin rich region of the
composite microstructure; ply stress = 15 MPa.
FIG. 6----Second indication of debonding, 17 MPa ply stress.
FIG. 7--Coalescence and growth at a ply stress of 22 MPa of initial debonds which initiated at
a ply stress of 15 MPa), as seen in Fig. 5.
FIG. 8--Subsequent growth of debonding previously seen in Figs. 5 and 7, but only after ply
stress is raised to 38 MPa.
FIG. 9--Second transverse crack growing from incipient flaw seen in Fig. 4 toward primary
transverse crack as seen in Figs. 4, 7-9, at ply stress = 38 MPa.
across the entire 90 ~ ply and part way across the adjacent 90 ~ ply, as seen in Fig. 10. The
several interfacial debonds which occurred at 17 MPa (see Fig. 5) are still seen in Figs. 9
(upper, center) and 10 (upper, right) to have not grown, much less coalesced. This region is
shown in better detail in Fig. 11, which should be compared with Fig. 6.
The observations of the short-term aged specimen proved to be somewhat more difficult,
even though the ElectroScan was being operated with a moist atmosphere in the chamber.
The picture lacked the clarity of the unaged specimen, probably due to some slight outgasing
of moisture from the specimen since the specimen chamber was not initially set to give 100%
humidity. Furthermore, no local debonding was noted prior to the first appearance of transverse
cracking at a 90 ~ ply stress of 19 MPa. For example, the short-term aged specimen with 6
MPa ply stress is seen in Fig. 12. It is most likely the case that interfacial failure preceded
transverse cracking, but the loading steps being used in the experiment were too large to see
the interfacial failures prior to their propagation into the full-blown transverse crack, as seen
in Fig. 13.
With the ESEM's chamber adjusted to 100% humidity, ESEM imaging of the long-term
aged specimens was much better. The first interfacial debonds, as seen Fig. 14, were observed
in two different locations at a ply level stress of 0.7 MPa. By 0.9 MPa the debonds had begun
to coalesce as seen in Figs. 15 and 16. The first site (Fig. 14a) did not change much from
0.9 MPa ply stress to 8 MPa ply stress; however, as Fig. 17 shows, the second site (Fig. 14b)
was beginning to grow into a crack. As the stress increased from 8 MPa to 35 MPa neither
site continued to grow into a transverse crack. However, at a ply stress of 35 MPa a transverse
crack was observed below the site in Fig. 14, as shown in Fig. 18. It is likely that this crack,
like that of the short-term aged specimen, was preceded by local interfacial failures which
were unstable, coalescing into a small, unstable transverse crack which grew across the 90 ~
ply due to the large displacement and associated stress increment.
FIG. lO---Coalescence of cracks seen in Fig. 9 with subsequent growth.
FIG. 1 l--Original debonds in high fiber density region, still not coalesced, showing no growth
at ply stress = 39 MPa).
FIG. 12--Short-term aged specimen loaded to 6 MPa ply stress with no indication of interfa-
cial failures.
FIG. 13--Short-term aged specimen with transverse crack across both 90~ plies at a ply stress
of 19 MPa.
The three specimens discussed previously are representative of three different failure scenar-
ios which were observed with the ESEM: (1) debonds form and coalesce, eventually forming
a transverse crack (i.e., propagation controlled failure), (2) there is no sign of debonds, the
first sign of damage is a full-blown transverse crack (i.e., initiation controlled failure), and
(3) a combination of the previous two scenarios where observed debonds may coalesce but
do not form a transverse crack; the transverse crack forms somewhere else without any prior
indications. The 90 ~ ply stress levels reported above, however, are not representative as Tables
1 and 2 show. For the unaged specimens, the ply stress at which a transverse crack was
observed ranges from 22 MPa to 39 MPa (mean = 31 MPa, standard deviation -= 6 MPa).
The stress in the 90 ~ plies of the long-term aged specimens ranged from 23 MPa to 43 MPa
(mean = 35 MPa, standard deviation --- 7 MPa). The tables also show that failure Scenario
l occurs only in the unaged specimens, while scenarios 2 and 3 are not exclusive to a specific
set of specimens. The short-term aged specimen and one of the long-term aged specimens
failed by scenario 2, and three of the unaged specimens and four of the long-term aged
specimens failed by scenario 3.
Another interesting occurrence was noted while scanning the surface of the long-term aged
specimens prior to testing. In five of tile seven composite specimens tested using this new
technique, chunks of the matrix had been removed during the polishing process. This was
surprising as the procedure used to polish these specimens was gentler than the procedure
used for the unaged specimens. Only two of the specimens had such extensive damage that
there was a transverse crack present before a load was applied; Fig. 19 shows one such
incidence. These specimens were not included on Table 2. In the remainder of the specimens,
the transverse crack did not occur near polish damaged regions.
FIG. 14---Initial debonds at two different sites in long-term aged specimen, ply stress = 0.7 MPa.
FIG. 15--Debonds seen in Fig. 14a beginning to coalesce at a ply stress o f O.9 MPa.
FIG. 16--Debonds seen in Fig. 14b beginning to coalesce at a ply stress o f O.9 MPa.
FIG. 17--At ply stress of 8 MPa debonds seen in Figs. 14b and 16 have begun to form a crack.
FIG. 18--Transverse crack at ply stress of 35 MPa in long-term aged specimen where no previous
indications had been noted.
WOOD AND BRADLEY ON INTERFACIALSTRENGTH 145
TABLE l--Failure scenario and stress at which transverse crack formed in unaged specimens.
Specimen Scenario 1 2 3 4 5 6
Laminate Stress, MPa 365 260 209 311 280 328

Ply Stress, MPa 39 28 22 33 30 35
Failure Scenario 1 2 1 1 2 3
TABLE 2--Failure scenario and stress at which transverse crack formed in short-term aged (S)
and long-term aged (L) specimens.
Scenario
Specimen S- 1 L- 1 L-2 L-3 L-4 L-5
Laminate Stress,
MPa 179 324 354 349 214 400
Ply Stress, MPa 19 35 38 37 23 43
Failure Scenario 2 3 2 3 3 2
FIG. 19--Transverse crack and polishing damage prior to loading in long4erm aged specimen.
In addition to comparing the specimens based upon the degree of aging, insight can also
be gained by comparing the specimens in terms of failure scenarios. The average stress at
which the first signs of damage (debonding or small cracks) occurred can be seen in Fig. 20a.
It can be seen that the unaged specimens required a higher stress to initiate damage than the
long-term aged specimens, 7.0 -+ 5.9 MPa versus 3.1 -+ 1.6 MPa. From Fig. 20b, one can
see that an average ply stress of 31.0 _+ 3.6 MPa was required to form a transverse crack in
FIG. 20---Average failure ply stresses of unaged and aged specimens: (a) ply stress at first sign
of failure, failure scenarios 1 and 3, (b) ply stress at transverse crack when failure is propagation
controlled, failure scenarios 2 and 3.
unaged specimens; while in both sets of aged specimens, transverse cracks formed at an
average ply stress of 32.5 + 9.4 MPa.
Analysis
Elasticity calculations were performed to aid in the interpretation of the experimental results.
From looking at Fig. 5, one can see" that debonding initiated on the boundary of a resin rich
region. However, in the short-term aged specimen, Fig. 12, the microstructure contained no
inhomogeneities even though both specimens were from the same panel. To understand the
differences in residual stresses between a homogeneous and an inhomogeneous microstructure,
two different models were used. The model in Fig. 21 represents a homogeneous microstructure
where the composite is modeled as a single fiber surrounded by a cylinder of matrix with
radius b. In this model, the residual stress is determined as a function of the center-to-center
fiber spacing (2b). Figure 22 depicts an inhomogeneous-microstructure model where the
composite is modeled as a cylinder of matrix surrounded by a cylinder of material with
composite properties. In this model, the residual stress is a function of the size of the resin
rich region.
The solution to the equation for force equilibrium between two concentric cylinders cooled
from the glass transition temperature of the matrix to room temperature, but with different
thermal coefficients of expansion only (13 = 0) and moisture induced swelling (1~ > 0) is:
FIG. 21--Homogeneous microstructure-model (cr.(b) = 0).
FIG. 22--1nhomogeneous-microstructure model.
E EctAT EfSAm
O'rr (1 -- V)(I -- 2V) [(1 -- V)err + ~(~00 + eZz)] 1 -- 2V 1 -- 2V (1)
This solution was obtained by assuming plane strain, satisfying compatibility, and applying
two different boundary conditions. The boundary conditions were a traction-free boundary at
b, where b is the radius of the outer cylinder, and a displacement-free boundary at b. Table
3 lists the constants used to determine the residual interfacial stresses. Unfortunately, some
of the required properties of the fibers, matrix, and composite could not be obtained from the
literature, so values for similar materials were used or the values were calculated. Also, the
room temperature modulus was used, neglecting viscoelastic relaxation during cooldown,
causing the residual stresses to be overestimated.
The two boundary conditions should yield upper and lower bound estimates of the actual
residual stress. In the homogeneous model the "displacement free" boundary condition was
unrealistic; it yielded positive thermal residual stresses of the order of l07 Pa for both unaged
and aged specimens. Thus, only the results for the traction free boundary condition are presented
in Fig. 21. In the inhomogeneous model, the boundary conditions produced a more realistic
upper and lower bound, as seen in Fig. 22. Both models show that the presence of moisture
TABLE 3--Data used to determine residual interfacial stresses of an IM-7/TACTIX 556 composite
(Vt = 0.5).
Manufacturer Data Calculated Values Values for Similar
[7,8] Experimental Measurements [9,10] Materials [10,11]
Em = 2.6 GPa Am,,a = 1.1% weight gain v<.12 = 0.355 v,, = 0.41
Eril = 275.6 GPa Am,. I = 0.5% weight gain Ecil = 137.8 GPa am = 102.6 txe/~
Tg = 210~ acli = 41.4 I~ed~ Eta 2 = 6.1 GPa Ut12 = 0.3
a~2z = 1.2 laePC ~3,, = 4000 }~cd% et./la = - 1 . 2 I.LeJ~
weight gain
[~,.z2 = 3t00 Izcd% weight af22 = 27 I~ePC
gain
Ejz2 = 13.8 GPa
"Percent weight gains are from unaged to aged conditions.
relaxes the interfacial residual stresses. It is worth emphasizing that in the homogeneous case,
the residual stress normal to the interfaces are relaxed to a less compressive value whereas
in the inhomogeneous case the residual stress at the boundary of the resin rich region relaxes
from more tensile to less tensile with moisture absorption.
Determining residual stresses at the fiber/matrix interface and calculating the strengths from
the measured laminate stress at debonding using finite element analysis is in progress and
will be published in the future. However, for the results obtained in this preliminary study,
the failure initiated at a inhomogeneity in the microstructure, which makes the modeling
somewhat more complicated than models developed to date [12,13]. The analysis will be
further refined by accounting for the viscoelasticity on cooldown. Additional measurements
in the SEM are also planned to be done in conjunction with the finite element analysis.
Discussion
It has generally been assumed that interfacial debonding would lead almost immediately to
transverse cracking, with at least four recent studies making this very reasonable assumption
[12-15]. However, our results clearly show for the dry specimen pictured in Figs. 3 through
11 that the failure event seems to be propagation controlled rather than initiation controlled.
As previously mentioned, the microstructure of this dry specimen had resin rich regions which
were not noted in the short-term aged specimen. It is in these regions that debonding initiated
because, as shown in Fig. 22, the residual stresses at the boundary of a resin rich region are
tensile, due to the greater thermal coefficient of expansion (or contraction) for the resin versus
the surrounding fiber/matrix region. This means that less applied load is needed to initiate
failure on the boundaries of resin rich regions than in areas where the microstructure is
homogeneous and the residual stresses at the fiber/matrix interface are compressive. As the
crack grows into a more homogeneous microstructure, more stress is required to propagate
the crack due to the compressive thermal residual stresses (Fig. 21) at the fiber/matrix interface.
In homogeneous microstructures, such as that of the short-term aged specimen, failure
is initiation controlled; whereas in an inhomogeneous microstructure failure is propagation
controlled. It is hypothesized that when a transverse crack occurred in a region where no
previous signs of damage were noted that a stress level was first reached where one or more
debonds formed. Then, if the interfacial strength of the surrounding fibers and the local residual
compressive stresses at the fiber/matrix interface are very close to that of the fibers which
failed initially, the local redistribution of load causes adjacent interfacial failures which quickly
coalesce to form a transverse crack. This also occurred in the long-term aged specimen seen
in Figs. 14 through 19; however, debonds initially formed and coalesced along resin rich
regions which were nowhere near the transverse crack. In order for the initial debonds along
the resin rich regions to continue growing into a transverse crack, a higher stress level was
needed to cause propagation into the surrounding homogeneous microstructure. However,
before this stress level could be reached, debonds occurred in regions with a more homogeneous
microstructure, and a transverse crack was formed.
Absorbed moisture can potentially affect the interfacial strength in at least three ways. First,
the swelling stresses that occur with moisture absorption reduce the size of the residual
compressive stresses at the interface, or even make them tensile if sufficient moisture is
absorbed, reducing the necessary mechanical loading needed to cause interracial failure [13].
Second, moisture reduces the modulus, which will reduce the residual stresses at the interface.
Third, the absorbed moisture may collect preferentially at the interface, reducing the chemical
bonding between the matrix and the fiber sizing or between the fiber sizing and the fiber. Wu
et al. [16] have recently shown that moisture concentrations as high as 17% in the absence
of a sizing and as high as 12% in the presence of a sizing can accumulate within 30 angstroms
of a polYmer/silicon interface in a polymer, with only 3% moisture in the bulk polymer.

Obviously, the concentration of such amounts of water at polymer/fiber interfaces, if it occurs,
would be expected to significantly degrade the interfacial strength.
The microstructural differences between the aged and unaged specimens (they were cut
from the same panel) greatly complicates the comparisons. In a composite specimen which
experiences initiation controlled failure (i.e., one with a homogeneous microstructure), one
would expect an aged specimen to have a lower interracial strength than an unaged specimen.
Figure 21 shows that the moisture contained in an aged specimen will also cause the thermal
residual stresses to be less compressive. In specimens where the microstructure is inhomoge-
neous there will be damage adjacent to the inhomogeneity coalescing into a small transverse
crack which can only grow into the homogeneous portion of the microstructure with a substantial
increase in the ply stress. In such cases, Fig. 21 leads one to believe that the aged specimens
will have a higher interfacial strength than the unaged specimens since the residual thermal
stresses become less tensile for the aged specimens. However, it should be remembered that
moisture may reduce the interracial bonding strength which would decrease the interfacial
strength of the aged specimens relative to the unaged specimens. There is so much scatter in
the data, as Tables 1 and 2 and Fig. 20 show, that no conclusions can be made concerning
the effect of moisture on the stress at which damage initiates or at which transverse cracking
occurs. However, as the long-term aged specimens were damaged by polishing, it appears that
long-term exposure to seawater does have some effect on the composite, but principally
in reducing the matrix strength. Also, the aged specimens did not experience propagation
controlled failure.
Comparing our results to those of other researchers, it is obvious that moisture is not the
only variable which affects the transverse strength of a composite. Pomies and Carlsson [13]
have found moisture induced reductions up to 66% in the stress at which transverse cracking
occurred in two graphite/epoxy systems while Grant and Bradley [6], who investigated three
graphite/thermosetting resin composites, reported at most a 17% reduction in transverse tensile
strength. Since all six systems investigated in the three studies (including this one) are somewhat
different, it is possible that the degree of degradation of the interfacial strength depends on
the composite system sizing, processing, and microstructure, and/or the weight percentage of
water absorbed. This is a most important question that needs further study.
Finally, the question of whether the observations at the surface are somehow indicative of
the behavior in the bulk of the composites should be addressed. One might expect that the
free edge effects on a micromechanics level (fibers and matrix) as well as on a ply level
(interlaminar shear stresses between plies) could perturb the results. However, Crews et al.
[12] have demonstrated experimentally that such is not the case for transversely loaded
specimens with varying degrees of off-angle fiber orientation. Furthermore, we did not note
any tendency for our transverse cracks to originate at the ply boundaries as one might expect
when considering thermal residual stresses. If one considers Poisson's stresses as well as
thermal residual stresses, it is not intuitively clear where in the ply one would expect crack
initiation. In these particular specimens the transverse cracks began in the middle of the ply,
further suggesting that the interlaminar shear stresses, even at the free edge, do not seem to
play a dominate role in transverse cracking. This might be interpreted as implying that the
failure criteria for interfacial strength is much more sensitive to normal stresses at the fiber/
matrix interface than to shear stresses.
Concluding Comments
This new technique of direct observation of a transverse tensile test using an ESEM has
provided some interesting insights concerning interracial debonding and transverse crack
growth. Preliminary work to date utilizing this technique has yielded the following observations.
First, as seen experimentally, debonding occurs preferentially at the boundary of inhomogeneit-
ies. This is easily explained by the analytical models which predict tensile residual stresses
at the boundary of a resin rich region and compressive stresses at the fiber/matrix interface
in a homogeneous microstructure. Second, transverse cracking was seen to be propagation
controlled in half of the unaged specimens, and initiation controlled (which was also dominated
by interfacial failures) in half of the unaged specimens and all of the aged specimens. The
fact that all of the specimens did not experience initiation controlled transverse cracking is
contrary to what is generally assumed in modeling of composites, where a homogeneous fiber
distribution is assumed. These differences in the failure scenarios are attributed to the variations
in the microstructures of the test specimens. Third, the analysis indicates that the ply stress
to give debonding leading to transverse cracking should be lower in the seawater saturated
specimens than in the unaged specimens, when both specimens have a homogeneous microstruc-
ture. However, when the microstructure is inhomogeneous, the analysis is not so straightforward.
Absorbed moisture will reduce the tensile stresses around the inhomogeneity, thus increasing
the interfacial strength. At the same time, moisture may also reduce the interfacial bonding
strength. The differences in the microstructure from sample to sample and the variety of failure
scenarios observed in the limited tests (12) to date makes it impossible to confirm this prediction.
Acknowledgments
Financial support for this research program has been provided from the National Science
Foundation through the Offshore Technology Research Center at Texas A & M University and
the University of Texas. The support of the Texas A&M University Electron Microscopy
Center, and especially Randy Scott, is gratefully acknowledged. Finally, the support of the
National Science Foundation to purchase the environmental scanning electron microscope used
in this study is appreciated.
References
[1] Broutman, L. J., Interfaces in Composites, ASTM STP 452, 1969, p. 27.
[2] Drzal, L. T., Rich, M. J., and Camping, J. D., "Interfacial Shear Strength and Failure Mechanisms
in Graphite Fiber Composites," Proceedings of 35th Annual Technical Conference of SPI, 1980,
p. 22-A.
[3] Mandell, J. E, Chen, J. H., and McGarry, E J., International Journal of Adhesion and Adhesives,
Vol. 1, 1980, p. 40.
[4] Miller, B., Muni, P., and Rebenfeld, L., Composite Science and Technology, Vol. 8, 1987, p. 17.
[5] Narkis, M., Chen, J. H., and Pipes, R. B., Polymer Composites, Vol. 9, 1988, p. 245.
[6] Grant, T. S. and Bradley, W. L., Journal of Composite Materials, Vol. 29, 1995, pp. 852-867.
[7] "TACTIX Performance Polymers for Advanced Composites and Adhesives," The Dow Chemical
Company, Freeport, TX, 1994.
[8] "Hercules | Carbon Fibers Product Data Sheets," Number 868-2, Hercules Advanced Materials
and Systems Company, Hercules, UT, 1994.
[9] Chamis, C. C., "Simplified Composite Micromechanics Equations for Hygral, Thermal, and
Mechanical Properties," SAMPE Quarterly, Vol. 115, 1984, pp. 14-23.
[10] Daniel, 1. M. and Ishai, O., Engineering Mechanics of Composite Materials, Oxford University
Press, New York, 1994.
[11] McCrum, N. G., Buckley, C. P., and Bucknall, C. B., Principles of Polymer Engineering, Oxford
University Press, New York, 1992.
[12] Crews, J. H., Naik, R. A., and Lubowinski, S. J., An Analysis of Fiber-Matrix Interface Failure
Stresses for a Range of Ply Stress States, NASA Technical Memorandum 108999, NASA Langley
Research Center, Hampton, Virginia, July 1993, p. 1.
[13] Pomies, E and Carlsson, L. A., Journal of Composite Materials, Vol. 28, 1994, p. 22.
[14] Karlak, R. E, "Failure Modes in Composites II," Proceedings of 6th Annual Spring Meeting of
Metallurgical Society of ACME, University of Pittsburgh, Pittsburgh, PA, May 1979.
[15] Pomies, F. and Carlsson, L. A., Proceedings of American Society for Composites 6th Technical
Conference, Albany, N.Y., Technomic Publishing Co., Lancaster, PA, 1991, p. 558.
[16] Wu, W., Orts, W. L, Majkrzak, C. J., and Hunston, D. L., "Water Absorption at a Polyimide/Silicon
Wafer Interface," to appear in Polymer Engineering and Science.
A. Toshimitsu Yokobori Jr., 1 Hidetoshi Takeda, 2 Takeshi Adachi, 3
J. C. Ha, 3 and Takeo Yokobori 4
Characteristics of Fatigue Life and

Damage Accumulation of Short Fiber-
Reinforced Polymer Composites
REFERENCE: Yokobori, A. T., Jr., Hidetoshi, T., Adachi, T., Ha, J. C., and Yokobori, T.,
"Characteristics of Fatigue Life and Damage Accumulation of Short Fiber-Reinforced
Polymer Composites," Fiber, Matrix, and Interface Properties, ASTM STP 1290, C. J. Spragg
and L. T. Drzal, Eds., American Society for Testing and Materials, 1996, pp. 152-167.
ABSTRACT: The relation between fatigue life and damage accumulation of fiber-reinforced
polymer composite (FRP) is not yet clarified.
For practical use of FRP, it is necessary to relate the fatigue life to the mechanism of damage
accumulation. Damage formation is controlled by the mechanical behavior of the interface
between the matrix and fiber. We used short glass fiber-reinforced polycarbonate composite in
our experiments. By using an in situ (real time) observational fatigue testing machine, we
investigated the relationship between fatigue life and damage accumulation.
From these results, the fatigue life of this material was found to be dominated by damage
accumulation which results from microfracture at the interface between the matrix and fiber.
This microfracture is controlled by a cycle-dependent mechanism.
KEYWORDS: short fiber-reinforced polymer composite, glass fiber, polycarbonate, fatigue

life, damage formulation, cycle-dependent mechanism, interface, discontinuous fiber, chopped
fiber, Injection molding compound
Fracture behavior of discontinuous fiber-reinforced composites (FRP) is considered to come

from the microscopic fracture at the interface between the matrix and fiber.
The major mechanisms that control such microscopic fracture are interracial debonding and
pull-out, fiber fracture, delamination, and matrix cracking. Not only tensile stress but also
shear stress affect these fracture mechanisms [1-5].
When these polymer composites are applied to practical use, it is important to estimate
their mechanical behavior under cyclic fatigue conditions, and mechanisms of microscopic
fatigue failure were therefore investigated [6-8]. Concerning fatigue life, we clarified the
mechanical characteristics of the effect of load frequency on fatigue life for metals under high-
temperature or corrosive conditions [9-12] and for ceramics [13]. These characteristics enable
us to perform unified classifiable estimation of their fatigue failure mechanisms with relation
to cycle-dependent (CD) and time-dependent (TD) characteristics and their interaction mecha-
~Associate professor, Department of Mechatronics and Precision Engineering, Tohoku University, Aoba
Aramaki Aobaku, Sendai, #980-77, Japan.
2Idemitsu Petrochemical Co., 1-1 Anezaki Kaigan, Ichihara, Chiba, #299-01, Japan.
31shinomaki Senshu University, Ishinomaki, Miyagi, #986, Japan.
4Dean and professor, School of Science and Engineering, Teikyo University, l-1 Toyosatodai
Utsunomiya City, Tochigi, #320, Japan.
152
Downloaded/printed
YOKOBORI ET AL. ON FATIGUE LIFE AND DAMAGE 153
TABLE 1--Mechanical properties and composition of the composite material.

Matrix Polycarbonate
Reinforcement Glass fiber

Fiber Content, wt% 30
Specific Gravity 1.42
Yield Strength
Tensile Strength, MPa "1'18
Modulus of Elasticity, MPa 5880
Elongation, % 3
Bending Strength, MPa 136
nisms. However, these time-dependent mechanisms are different from each other, for example,
void diffusion mechanisms under high-temperature conditions, chemical reaction mechanisms
under corrosives environment conditions, and visco-elastic properties at the glass phase for
ceramics where CD and TD mean cycle and load application time effects under fatigue
conditions.
It has not yet been clarified whether the fatigue life of fiber-reinforced composites is
controlled by a cycle or a time-dependent mechanism. In this paper, using a short glass
fiber-reinforced polycarbonate composite material, we investigated the effect of varying load
frequency and load wave form on fatigue life and determined what mechanism controls the
fatigue life characteristic. Furthermore, to clarify its mechanical controlling factor, we carried
out real-time observation of the accumulated damage of the specimen as a result of fatigue
loading, fractographic observation of fracture surfaces, and finite element stress analysis to
clarify the damage formation.
Material, Specimen, and Test Method

The material used was a short fiber-reinforced polymer composite consisting of glass fibers
(30 wt%) and polycarbonate matrix. The diameter of the glass fiber was 13 tLm. The average
length of fiber was 0.28 ram. This matrix is general purpose type bisphenol-A polycarbonate.
The molecular weight (weight average Mw) was 27 800 and molecular weight distribution was
2.78. Fiber was made of E glass and manufactured by winding melted glass through a bushing
and then chopping it. Its surface was treated with a ulethan-type coupling agent and aminosilane-
type binder. Mechanical properties and composition of this material are shown in Table 1.
Mechanical properties of the matrix and fiber are shown in Table 2. Materials were supplied
as injection-molded plaques with dimensions of 70 by 70 by 2 mm. A thick molding has a
preferred alignment of fibers in the flow direction of the surface layers and with fibers aligned
transverse to the flow direction in the central layer [7]. However, the plaques supplied for
these experiments were thin (2 ram). Therefore, in this case, the molding had a preferred
TABLE 2--Mechanical properties of the matrix and fiber.

Matrix Fiber (Glass)
Specific Gravity 1.2 2.54

Yield Strength, MPa 62.0 2940.0
Tensile Strength, MPa 71.0 3430.0
Modulus of Elasticity, MPa 2160.0 73 500.0
Elongation, % 101.0 4.0
154 FIBER,
MATRAIXN, DN
I TERFAC
PE
ROPERTE
IS
90 DEG
/ " / ...,......,.....,.~.. . ,.f')"[):Z,-'~k,..............

,S'.--<~!
"", .-'L................................
45"DE;G
/" Z/ /,." ....................... - ~ . . . . . . . . . . . . . . II
/- (/= 0 DEG
FIG.1--Cut-out method of a specimen from the plaque.

alignment of fibers in the flow direction both on the surface and in the central layers. Further-
more, to avoid variation of the fiber alignment, each specimen was cut out from the same
central site of each plaque, as shown in Fig. 1. We used double-edged V-shaped notched
specimens with a 30 ~ notch and a thickness of 2 mm (Fig. 2).
To investigate the effect of anisotropy on the fatigue behavior of this material, we used
three types of specimens that were cut from the plaques in the 0 (parallel), 45, and 90 ~
68
~.~o~
3 +o.oz~ notch
.\,
~-4~
I I
...... ~!
6 9 9 9
....... '- ~ notch
Width of specimen,W=4 Thickness of snecirnen,t=2
30 =
notch part
V
p =0.05
cq
c5
I
FIG. 2--Double-edged V-shaped notched specimen.
(transverse) direction to the injection-molded flow direction. We called these specimens 0, 45,
and 90 ~ specimens. Cyclic fatigue tests were carried out using a fatigue testing machine that
enables real-time observational experiments. This machine system was designed to enable
automatic observation of the mechanical behavior of deformation and damage accumulation
around the crack tip during fatigue loading under computer control. A schematic illustration
of this system is shown in Fig. 3. The effects of stress frequency, stress wave form, and the
direction of fiber alignment on the fatigue life characteristics were investigated.
Fatigue stress amplitude, Act, was set up at Act = 32 MPa and minimum gross stress, (rm~n,
was zero, that is, stress ratio R (crmi,/~rmax) = 0 for all tests. Tests were performed under
atmospheric conditions, although the test temperature was held at 47 +- I~ to avoid the effect
of room temperature variations on the mechanical behavior of the test pieces. Fatigue tests
were performed under load frequencies of 0.06, 0.2, and 0.7 Hz to investigate the effect of
frequency on fatigue life. Furthermore, the effect of stress wave form on fatigue life was
investigated using slow-fast (TR = 15 s, TO = 5 s), fast-slow (TR = 5 s, To = 15 s), and
symmetrical (TR = To = 10 s) stress wave form conditions. TR is the rising time to the peak
stress and To is the decreasing time to the minimum stress of the fatigue test.
Effects of Load Frequency and Stress Wave Form on Fatigue Life
The characteristics of the effect of the load frequency (/3 on fatigue life are shown in Fig.
4 for 0, 45, and 90 ~ and the matrix specimens. These results were analyzed by referring to
the characteristic curve shown in Fig. 5 [14]. Fatigue life is defined as the time or number of
stress cycles to fracture from the instant of fatigue stress application. When the inverse value
of fatigue life time, 1/tf, is constant when f varies, it shows fatigue fife is controlled by the
time-dependent mechanism, TD. This characteristic will be shown in the lower range of f.
When lltf is proportional to f, fatigue life is controlled by a cycle-dependent mechanism, CD.
(Fatigue cycle life, Nf, takes the same value.) This characteristic is shown in the higher range
o f f In the intermediate region, it is thought that fatigue life was affected by both TD and
Tensile Directi0n
Test Piece
I
I I I X axis
Video Controller
Camera Magnification
lens
I I I
VT ll X,Y,Z axis controller
I Controller
I
Personal
Monitor ] Computer
FIG. 3---Schematic illustration of the system.
FIG. 4---The characteristics of the effect of the load frequency f on the fatigue life for the
composite material in each fiber aligned direction and matrix.
CD mechanisms as the interaction mechanism [9-14]. The right-hand downward characteristic

of 1/tIis sometimes shown in the highest region off, for example, under an ultrasonic frequency
of fatigue [14]. The detailed mechanism in this region remains to be clarified in the future.
From the results shown in Figs. 4 and 5, characteristics on the fatigue life of the 0 ~ specimen
(alignment parallel to the load axis) shows a cycle-dependent mechanism. Those of 45 and
90 ~ specimens also show a cycle-dependent mechanism in the region o f f < 10 -~ Hz. In the
region o f f > 10 -~ Hz, with an increase o f f parallel and right-hand downward characteristics
of fatigue life emerged for 45 and 90 ~ specimens followed by that of a cycle-dependent
mechanism. These are the same characteristics as those in the high-frequency region shown
in Fig. 5. This mechanism remains to be clarified in the future. The matrix, marked by a solid
circle symbol in Fig. 4, has a smaller gradient to the horizontal axis than that of a cycle-
dependent mechanism. Therefore, the fatigue life of the matrix is considered to be involved
in a time-dependent mechanism.
Experimental results of the effect of the stress wave form on the fatigue life for 0, 45, and
90 ~ specimens are shown in Fig. 6. It was found that the fatigue life of the 0 ~ specimen is
two orders of time longer than that of the 90 ~ specimen. This demonstrates conspicuous
anisotropic properties of the fatigue life of this material. Systematic dependence on the stress
wave form of the fatigue life (slow-fast, fast-slow, and symmetrical stress wave form) were
not found for specimens in any other fiber alignment directions.
From the results shown in Figs. 4 and 6, the fatigue cycle life, Nt of short glass-fiber-
reinforced polycarbonate composites is found to be dominated by a cycle-dependent mechanism
and was not affected by frequency and stress wave form.
a>l
),=1 x
.. ! I
o
'-- \
"-,,... s"-"~ ~j~" ,,~' X= 0
"--" Intermediate Region

a=l
log f
FIG. 5--The master curve on the characteristics of the effect of frequency on fatigue life.
Fractography
Fracture behavior of 0, 45, and 90 ~ specimens was observed by scanning electron microscopy
(SEM) observation (HITACHI Ltd. $415) and is shown in Fig. 7(a-c) and Fig. 8(a-c). The
0 ~ specimen shows an irregular fracture surface (Fig. 7a) and pull-out of glass fiber (Fig. 8a).
The 90 ~ specimen shows a comparatively flat fracture surface (Fig. 7c) and delamination traces
of the glass fiber (Fig. 7c). The fracture behavior of the 45 ~ specimen is different from the
others and has a fracture line inclined at 20 -- 30 ~ to the horizontal axis (Fig. 7b). The fracture
surface shows both pull-out and delamination of glass fiber (Fig. 8 b).
Damage Formation Around the Crack
Progressive behavior of the damage around the notch tip under fatigue testing for 0, 45,
and 90 ~ specimens was obtained by real-time observation using a video microscope (Mitsubishi
Electric Ltd.) as shown in Fig. 9(a-c). These photographs were obtained by transmitting a
ray through the specimen thickness from a source of light at the opposite side of the specimen
surface. Therefore, the damage region is observed as a dark region because of the refracted
light in this area. However, the damage configurations of the 0 and 90 ~ specimens extend
perpendicular to the applied load axis; the aspect ratio of the 0 ~ specimen is smaller than that
of the 90 ~ specimen (thick layer region in 0 ~ specimen and thin layer region in 90 ~ specimen).
The damage configuration of the 45 ~ specimen is affected by the direction of glass fiber
alignment. It extends in an oblique direction to the load axis from the notch tip.
From these results, damage extension behavior for each fiber-alignment direction corresponds
well with features of the fracture surface and its direction. Furthermore, final rapid fracture
occurs when the damage regions link through the specimen width. For the 0 ~ specimens, we
measured the progressive behavior of the damage area, D, for various frequencies. Values of
158 FIBER, MATRIX,AND INTERFACEPROPERTIES
' ' ' I ' ' ' , I
I "~ (sec}(sec) ~orm

[ o J 10 10 T./~T~
]" 10 10 / \
lu115 5 T,//~To
1"115 5 / \
I A ] 5 1 5 TR/~TD
10 4
(D
O
>., Q
O
10 3
i I I
0 45 90
Direction,degree
FIG. 6--The effect of stress wave form on the fatigue life for O, 45, and 90~ specimens.
D were plotted against nondimensional fatigue cycles, NINf as shown in Fig. 10, where Nf is
each fatigue life cycle. The damage area, D, is found to be independent of frequency, f, and
is characterized by a unique relation with N/Nj:
Since fatigue life of this material was found to be dominated by a cycle-dependent mechanism,
values of Nf for each frequency condition take almost the same value except for scattering of
the data. Therefore, Fig. 10 shows that D does not depend on the frequency but on the number
of fatigue cycles, N. The same results were obtained for 45 and 90 ~ specimens.
From this result we can determine that the progressive behavior of a damage region is also
dominated by cycle-dependent mechanisms.
Mechanical Analysis Between the Notch Tip and Fiber

Damage formation of metal at a high temperature is believed to be due to microcrack
initiation or slip by shear stress [15]. Analytically, these phenomena are considered to be
induced by the high equivalent stress. Analysis of equivalent stress is used to predict damage
FIG. 7--Fracture behavior by SEM observation: (a) 0 ~ specimen, (b) 45 ~ specimen, and (c)
90 ~ specimen.
FIG. 8--Fracture surface by SEM observation: (a) 0 ~ specimen, (b) 45 ~ specimen, and (c)
90~ specimen.
FIG. 9--Real-time observation on the progressive behavior of damage around the notch tip: (a)
0 ~ specimen, (b) 45 ~ specimen, and (c) 90 ~ specimen.
1.2 I! 1 I I-- I I I I It
- ~ DEG00'06HzI ~
DEG0 0.2Hz
" 0.8 DEG0 0.7Hz
E
E
o 0.4
jDOl[]O~ iq3qO
0 0.5 1
N/Nf
FIG. lO---Characteristics of the extension of the damage area plotted against nondimensional
fatigue cycles.
formation [16-19]. We determined that the damage region, observed as the dark region in
Figs. 9a--c, corresponds to the concentrated region of the equivalent stress and performed a
finite element analysis (FEM) to investigate the equivalent stress distribution. Since composite
material can be regarded as a nonlinear elastic body, in this analysis an elastic-perfect plastic
stress-strain constitutive law was adopted as the effect of local inelastic deformation. This
was due to the debonding and pull-out of fiber, microcrack initiation, delamination, and slip
at the interface between the matrix and fiber.
Simple mechanical models for the short fiber-alignment near the notch tip as shown in Fig.
11 (a-c) were analyzed. The shape of the model was the same as the actual specimen. The
diameter and length of this glass fiber model were 0.05 and 0.3 ram. The distance between
the notch tip and the center of glass fiber was 0.75 mm. Because of the symmetry of this
mechanical model, we analyzed one-quarter of this model for 0 and 90 ~ specimens and half
for the 45 ~ specimen. The number of elements and node points are shown in Table 3 and the
mechanical properties used for FEM analysis are shown in Table 4. We used FEM software
on the elastic stress analysis by Y. Yamada (EPIC) [20]. The high equivalent stress region
(HESR) predicted by FEM is shown in Fig. 12(a-c). In the 0 ~ specimen, HESR originated
initially around both ends of the short-fiber and spread over the wide region in the load axis
direction. Therefore, stress concentration around the notch tip was reduced. On the other hand,
in the 90 ~ specimen, stress concentration around the notch tip was magnified by the interaction
at the end of the fiber. This caused the link of the HESR between them. Consequently, the
HESR extended along the fiber length direction. In the 45 ~ specimen, initially the link of the
HESR between the notch tip and the end of the fiber occurred and extended in the inclined
direction along the fiber. Configurations of the HESR for 0, 90, and 45 ~ specimens were
similar to those of the observed damage region for each case as shown in Figs. 9(a-c).
Therefore, this local HESR was considered to induce microfracture such as debonding, pull-
out and delamination at the interface and form the damage region.
Discussion
We can consider the following physical phenomenon in the fatigue failure of short fiber-
reinforced composite materials. Experimentally, fatigue life was shown to be dominated by a
cycle-dependent mechanism. Experimentally and analytically, it was found that the damage
~a) 0 deg S~ecimen ~D) q5 deg Soecimen
l 90*
I a=0.75mm
(C) 90 aeg Specimen
FIG. 11--Mechanical .models on the short-fiber alignment near the notch tip used for FEM
analysis: (a) 0~ specimen, (b) 45 ~ specimen, and (c) 90 ~ specimen.
TABLE3--The number ofelemen~ and node poin~ usedfor FEManalysis.

No. of Elements No. of Nodes
0~ 888 482
45~ 1840 975
90~ 903 490
164 FIBER, MATRIX,AND INTERFACE PROPERTIES
TABLE 4--Mechanical properties used for FEM analysis.

Young's Modulus, MPa Poisson's Ratio Yield Strength, MPa
Po|ycarbonate 2 160 0.3 62

Glass 73 500 0.22 2940
region originates in the high equivalent stress region around the notch tip. The progressive
behavior of the damage region was also dominated by the cycle-dependent mechanism. By
fractographical observation, the following results were obtained. For the 0 ~ specimen (fiber
direction was parallel to the load axis), pull-out of fiber due to debonding at the interface was
observed. By FEM analysis, debonding was found to initiate around the end of the short fiber
where equivalent stress concentrates. For the 90 ~ specimen (fiber direction was perpendicular
to the load axis), many sites of delamination occurred at the interface along the fiber length
direction as equivalent stress concentrated along this direction.
From the results mentioned above, fatigue life of this composite material closely related to
the progressive behavior of the damage region that was dominated by a cycle-dependent
mechanism. This dependence was induced by microfracture such as debonding, pull-out, and
delamination at the interface in the HESR region. Since a macroscopic main crack was not
observed, to estimate fatigue life of this material it was important to clarify the mechanical
behavior of the damage region by relating microfracture mechanisms.
There are many reports that discuss damage formation and classification and the damage
extension behavior that affect fatigue failure life [21-24]. Our results clarify the relationship
between fatigue failure and damage accumulation of this material. It is hoped that clarification
of unified microscopic fracture mechanism under the fatigue condition will be developed
through all these research efforts. However, this paper is not concerned with the effect of
anisotropy on strength, rather with clarification of the fatigue failure mechanism. The following
strength data were obtained: for the 0 ~ specimen, 85 MPa; 45 ~ specimen, 65 MPa; 90 ~ specimen,
55 MPa; and matrix, 65 MPa. These results also show the anisotropic properties of strength
and correspond well to the effect of anisotropy on fatigue failure life as shown in Fig. 6.
Conclusions
1. When fatigue load was applied in the direction of fiber alignment, pull-out of fiber
from the matrix was observed by SEM and fatigue life was found to be dominated by a cycle-
dependent mechanism.
2. When fatigue load was applied in the 45 ~ direction or perpendicular to the reinforced
direction of the fiber, fatigue life was also dominated by a cycle-dependent mechanism in the
lower load frequency region, although different frequency characteristics of fatigue life emerged
in the higher frequency region. Furthermore, delamination between fiber and matrix was
observed by SEM.
3. Real-time observation of the mechanical behavior of damage accumulation showed that
damage extension behavior was also dominated by a cycle-dependent mechanism. Final rapid
fracture occurs when the damage regions link through the specimen width. Furthermore, by
perforrfiing FEM analysis, this damage was found to be formed in the higher equivalent stress
field region.
4. From these results, fatigue life of this material was found to be dominated by damage
accumulation which resulted from microfracture at the interface between the matrix and fiber.
This microfracture was controlled by a cycle-dependent mechanism.
FIG. 12--The high equivalent stress region obtained by FEM analysis." (a) 0 ~ specimen, (b) 4 5 ~
specimen, and (c) 90 ~ specimen.
References
[1] Friedrich, K., "Controls Fatigue Resistance and Fracture Toughness of Short and Long Fiber
Reinforced Thermoplastics," in Benibana International Symposium on How to Improve the Tough-
ness of Polymers and Composites, organized by I. Narisawa, Yamagata Univ., 1990, pp. 201-220.
[2] Choi, N. S. and Takahashi, K., "Effect of Fiber Diameter on Strengthening Mechanisms in Short
Fiber Reinforced Thermoplastics," in Benibana International Symposium on How to Improve the
Toughness of Polymers and Composites, pp. 249-254.
[3] Choi, N. S. and Takahashi, K., "Short Glass Fiber Reinforced PET Composites." Journal of the
Japanese Society of Composite Materials, Vol. 15, No. 5, 1989, pp. 222-228.
[4] Wang, S. J., Baptiste, D., Bompard, P. H., and Francois, D., "Microscopic Failure Mechanisms of
a Unidirectional Glass Fiber Composite," in Journal of Fatigue Fracture Engineering and Materials
Structure, Vol. 14, No. 4, 1991, pp. 391-403.
[5] Hull, D., An Introduction to Composite Materials, Cambridge University Press, England, translated
by H. Miyairi, K. Ikegami, and I. Kimbara, Baifuukan Pub., Japan, 1982.
[6] Kim, H. C. and Ebert, L. J., "Axial Fatigue Failure Sequence and Mechanisms in Unidirectional
Fiberglass Composites," Journal of Composite Materials, Vol. 12, 1978, pp. 139-152.
[7] Lang, R. W., Manson, J. A., and Hertzberg, R. W., "Mechanisms of Fatigue Fracture in Short Glass
Fiber Reinforced Polymers," Journal of Materials Science, Vol. 22, 1987, pp. 4015--4030.
[8] Sato, N., Kurauchi, T., and Kamigaito, O., "Relationship Between Fatigue Property and Microfailure
of Interface in Composite Materials," in "'Benibana International Symposium on How to Improve
the Toughness of Polymers and Composites, "' Organized by I. Narisawa, Yamagata Univ., 1990,
pp. 277-284.
[9] Yokobori, A. T., Jr., Kaji, Y., Kuriyama, T., and Yokobori, T., "Cyclic and Time Dependent High
Temperature Fatigue Crack Growth Rate and Damage Mechanics," in Transactions of the Japanese
Society of Mechanical Engineers, Vol. 54, No. 503, 1988, pp. 1304-13l 1.
[10] Yokobori, A. T., Jr, Kaji, Y., Kuriyama, T., and Yokobori, T., "'The Effect of Damage on Fatigue
Crack Growth Rate Under High-Temperature Creep-Fatigue Multiplication," in Transactions of
the Japanese Society of Mechanical Engineers, Vol. 57, No. 539, 1991, pp. 2349-2354.
[11] Yokobori, A. T., Jr., Yokobori, T., and Takasu, N., "Effects of Frequency, Stress Rising Time, and
Stress Hold Time on Corrosion Fatigue Crack Growth Behavior of Low Alloy Cr-Mo Steel," in
Corrosion Cracking, V. S. Goel, Ed., American Society of Metals, 1986, pp. I-9.
[12] Yokobori, A. T., Jr., Chiba, N., Yamada, T., Kosumi, T., and Yokobori, T., "Effect of the Stress
Rising, Holding, and Descending Time on Fatigue Crack Growth in a Corrosive Environment," in
Current Japanese Materials Research, H. Okamura and K. Ogura, Eds., Elsevier Applied Science
Publishers, U.K. Vol. 8, 1991, pp. 125-143.
[13] Yokobori, A. T., Jr., Adachi, T., Abe, H., Takahashi, H., Nakayama, J., and Fujita, H. "Time
Dependent and Cyclic Dependent Effects of Alumina Ceramics Under Cyclic Fatigue Conditions,"
Journal of the Ceramic Society of Japan, International Edition, Vol. 98, 1990, pp. 961-967.
[14] Yokobori, T., "Approaches to Critical Issues and Questions in Fracture," Journal of the Japanese
Society for Strength and Fracture of Materials, Vol. 24, No. 4, 1990, pp. 95-116.
[15] Yokobori, A. T., Jr. and Yokobori, T., "The Crack Initiation and Growth Under High Temperature
Creep, Fatigue and Creep-Fatigue Multiplication," in Engineering Fracture Mechanics, Vol. 31,
No. 6, 1988, pp. 931-945.
[16] Kachanov, L. M., Introduction to Continuum Damage Mechanics, Martinus Nijhoff Publishers, the
Netherlands, 1986.
[17] Hult, J., Materials and Engineering Design the Next Decade, B. E Dyson and D. R. Hayhurst,
Eds., The Institute of Metals, London, 1989, p. 29.
[18] Bassani, J. L. and Hawk, D. E., "Damage Equations and Influence of Damage on Crack Tip Fields
Under Creep Conditions," in International Seminar on High Temperature Fracture Mechanisms
and Mechanics, Mecamat, M. B. Danbrme, Ed., Dourdan, France, 1987, pp. 1, 19-40.
[19] Levy, A. and Papazian, J. M., "Elastoplastic Finite Element Analysis of Short-Fiber-Reinforced
SiC/AI Composites: Effects of Thermal Treatment," Acta Metallurgia Materials, Vol. 39, No. 10,
1991, pp. 2255-2266.
[201 Yamada, Y., Plasticity and Viscoelasticity, Baifukan Publishers, 1972 (In Japanese.)
[21] Jamison, R. D., Schulte, K., Reifsnider. K. L., and Stinchcomb, W. W., EffFcts of Defects in
Composite Materials, ASTM STP 836, American Society for Testing and Materials, Philadelphia,
1984, pp. 21-55.
[22] Jamison, R. D., Composite Materials, Fatigue and Fracture, ASTM STP 907, 1986, pp. 252-273.
[23] Aoki, T., Kubo, T., Koyama, K., Kondo, K., and Kobayashi, S., "Effect of Damage on Fatigue
Strength of CFRP Laminates," Journal of the Japanese Society of Composite Materials, Vol. 15,
No. 3, 1989, pp. 122-131.
[24] Poursatip, A. and Beaumont, P. W. R., Mechanics of Composite Materials, Z. Hasin and C. T.
Herahovich, Eds., Pergamon Press, 1983, pp. 449-456.
J. P a u l A r m i s t e a d I a n d A r t h u r W. S n o w 1
Fiber/Matrix Interface Studies Using

Fragmentation Test
REFERENCE: Armistead, J. P. and Snow, A. W., "Fiber/Matrix Interface Studies Using
Fragmentation Test," Fiber, Matrix, and Interface Studies, ASTM STP 1290, C. J. Spragg and
L. T. Drzal, Eds., American Society for Testing and Materials, 1996, pp. 169-181.
ABSTRACT: Some fiber-resin composite interface work done at the Naval Research Laboratory
is reviewed beginning with several applied studies and progressing to more fundamental work.
Butadiyne (diacetylene) was evaluated as a novel carbon-fiber surface treatment applied via
vapor deposition polymerization. Results of its application to high-strength carbon fibers, high-
modulus carbon fibers, and aramid fibers are summarized. Reduction of carbon-fiber surface
polarity by annealing under hydrogen was evaluated as a means to decrease the sensitivity of
the interface to moisture.
In the above experiments the fragmentation test was used to determine relative levels of
adhesion and the mechanisms of failure at the fiber/matrix interface. Comparisons of the adhesion
of high-modulus and high-strength carbon fibers to epoxy and cyanate matrices were used to
evaluate models for the fragmentation test. For high-modulus fibers the adhesion was poor and
a friction factor approach may be suitable for modeling this type of interfacial failure. For high-
strength fibers the adhesion was much higher and possibly limited by the shear properties of
the matrix.
The simple mechanics models discussed previously do not explain adhesion changes where
chemistry is the only variable. In a study of three matrix resins with comparable mechanical
properties a linear dependence was shown between measured "good" adhesion and relative
resin model compound basicity.
KEYWORDS: adhesion, fiber/resin interface, fragmentation test, epoxy resin, cyanate resin,
high-modulus fiber, high-strength fiber, aramid fiber, hydrogen bond-strength scale
Adhesion at the Fiber/Matrix Interface
Twenty-five years ago Zisman summarized the conditions necessary for good adhesion in
polymer/glass fiber composites beginning on the molecular level with surface chemistry, the
work of adhesion, and the importance of complete wetting of the fibers by the liquid matrix
resin [1]. He also noted that composites are processed materials and that the properties can
be dominated by the nonidealities of the interface, such as the presence of voids and occlusions,
which are stress concentrators and potential crack initiators. These can arise from incomplete
resin penetration into fiber tows or be induced in regions where adhesion is poor as a result
of high interfacial stresses due to property mismatch. Such regions might arise from the
presence of adsorbed contaminants on the fiber surface or the presence of a fiber surface layer
of low cohesive strength. Improved processing techniques could minimize void formation but
JChemical engineer and research chemist, respectively, U. S. Naval Research Laboratory, Code 6120,
Washington, DC 20375.
168
Downloaded/printed
ARMISTEAD AND SNOW ON FRAGMENTATION TEST 169
design changes and possibly the inclusion of an elastic interface would be necessary to reduce
stresses at the interface due to property mismatch.
Today, processing techniques are much improved, void contents are routinely lower, but
the concerns remain the same. Carbon fiber/organic matrix composites have even greater
property mismatches and are being pushed to higher temperatures resulting in materials with
large internal stresses. To this end, we began our work with a means to alter the fiber/matrix
interface and evaluate changes in stress transfer across the interface. We ended up working
towards wetting and specific chemical interactions with the functional groups of the cured
and uncured matrix resins. In this paper, several applied projects aimed at improving the
carbon fiber/epoxy matrix interface are reviewed and some of the more fundamental work
that followed these initial studies is presented.
Fragmentation Test as an Interfaeial Adhesion Diagnostic Tool

A Novel Fiber Surface Treatment
Butadiyne (diacetylene), can undergo thermal vapor deposition polymerization on and within
inert polymer substrates such as polyethylene and teflon [2]. The more compatible the hydrocar-
bon monomer is with the substrate, the further it penetrates before it polymerizes. The resultant
polymer is intimately bound to the substrate and has a complex structure consisting of polyene
and polyacene structures. Depending on thermal treatment, pendant terminal acetylenic groups
are found. If butadiyne polymerizes similarly when carbon fibers are used as a substrate, we
have a novel surface treatment which can functionalize the carbon-fiber surface. The terminal
acetylenic groups can react directly with matrix resins or be used to functionalize the fiber
surface in any desired fashion or to tailor the fiber/matrix interface. The vapor deposition
polymerization technique should allow the monomer to access and penetrate all the fiber
surfaces in a tow. The resulting highly aromatic surface treatment should also have good
thermal stability. To evaluate fiber/matrix adhesion as a function of the hutadiyne fiber-surface
treatment we chose the single-fiber fragmentation test [3]. The reasons for this choice were
the small scale of our operation, which coated only a few short lengths of tow at a time and
precluded composite fabrication, and the expected critical lengths of less than 0.5 mm which
would make a pull-out technique difficult.
Carbon fibers (Hercules | AS4) were treated to 0.2, 1.1, and 9.4 wt% polybutadiyne and
tested using the fragmentation test [4]. The results are presented in Table 1 along with
a comment on the observed birefringence pattern. "Good adhesion" corresponds to bright
birefringence patterns which emanate from the fragment ends along the fiber/matrix interface
TABLE l--Fragmentation results for butadiyne-treated AS4 fibers in epoxy-mPDA matrix.

Treatment Coating" Mean Fragment Adhesion b, Birefringence
wt% Thickness, nm Length, mm MPa Observation
-- -- 0.40 42 good
02 3.0 0.30 67 good
1. l 16.4 0.29 69 good
9.4 t 38.0 0.29 69 poor
~Estimated from wt% applied, density, and fiber diameter.
bCalculated from Kelly-Tyson equation using Ir = 4/3 mean fragment length.
=Hercules Incorporated, Wilmington, DE 19894.

and which maintain a significant intensity when the matrix strain is removed, indicating plastic
deformation. "Poor adhesion" corresponds to fainter birefringence patterns which all but
disappear when the matrix strain is removed. The butadiyne surface treatment results in a
decreased mean fragment length at all deposition levels. When the coating is thickest, the
birefringence observation changes to "poor."
The Kelly-Tyson equation was used to estimate the interfacial adhesion from the fragmenta-
tion results [5].
, = (cr~/(2l~) (1)
In the above equation, d is the fiber diameter, ar is the fiber failure stress at the critical
fragment length, and lc is the critical fragment length. When fiber properties are constant, the
interfacial adhesion, T, is inversely proportional to the critical fragment length. In these studies
the critical fragment length was taken to be 4/3 the mean fragment length.
The fiber-failure stress dependence on gage length was determined by evaluating the strengths
of control and 0.5 wt% treated fibers at three gage lengths and using a two-parameter Weibull
distribution to extrapolate to shorter gage lengths. At the same gage length the coated fibers
have a 15% higher strength indicating that the surface treatment also serves to protect or heal
surface flaws. Similar testing of epoxy-sized and unsized IM6 fibers resulted in no change in
fiber tensile strength indicating that just covering the fiber surface (lowering surface energy)
is not enough to increase tensile stress [6]. This indicates that a true flaw-healing mechanism
is at work.
When both the decease in fragmentation lengths and the increase in fiber tensile strength
are taken into account, the polybutadiyne surface treatment results in nearly a 60% increase
in adhesion (Table 1). The change in birefringence pattern from good to poor at high surface
treatment is probably due to a change in failure mechanism-to-interfacial crack propagation
through the thickest and more brittle coating.
Fibers with Inherent Adhesion Problems
The success with AS4 fibers, which are high-strength PAN-based fibers, led to work with
fibers that are inherently more difficult to bond to: high-modulus and aramid fibers.
High Modulus Fibers--HMS4 is a PAN-based carbon fiber from Hercules that has been
heat treated to 2600~ and has a modulus of roughly 360 GPa. The AS4 fibers discussed
previously have been heat-treated to 1500~ and have a modulus of roughly 245 GPa. The
higher heat-treatment temperature increases the width, thickness, and alignment of the graphitic
ribbons that comprise the carbon fiber. In fragmentation tests with an epoxy-mPDA matrix
resin, Drzal and coworkers compared the adhesion of these two types of fibers [11]. After
performing single-fiber fragmentation tests, samples were microtomed and the interface were
viewed using transmission electron microscopy (TEM). Though cracks propagated down the
interface after fragmentation for both fiber systems, the AS-fiber interfacial failure was adhesive
and the HMS-fiber interfacial failure was cohesive within the outer layers of the fiber. The
butadiyne surface-treatment may be a means to penetrate the crevices and pores in the HMS4
fiber surface and reduce the cohesive failure observed when the interface is sheared.
The butadiyne vapor deposition polymerization proceeded more slowly with HMS4 fibers
and a roughly 0.1 wt% treatment was achieved. Fragmentation tests showed that the treatment
decreased the adhesion at the interface (increase mean fragment length) and there was no
qualitative change in the birefringence patterns (Table 2). The birefringence patterns for both
control and treated fibers were typical of poor adhesion: the interfacial crack propagates down
the fragment length without inducing much plastic deformation in the matrix. The coating
was not resolvable using scanning electron microscopy (SEM).
Aramid Fibers--Kevlar ~3- 49 is an aramid fiber with high strength and high modulus
compared to other organic fibers and is nonconductive and less dense than carbon fibers.
Drawbacks of the fiber include low compressive strength and apparent difficulties in adhering
to matrix resins [7]. The fiber has a skin-core morphology and when loaded in shear, failure
can occur at the skin-core interface [8]. It was hypothesized that if the butadiyne surface
treatment could penetrate the skin of the Kevlar-49 fiber and reinforce the skin-core interface,
the observed adhesion to matrix resins would improve.
Kevlar-49 fibers were surface-treated with 0.5, 0.84, 1.39, and 2.2 wt% of the butadiyne
vapor deposition polymerization surface treatment. Extensive work was carried out on the
0.84% treated fibers [9,10]. Adhesion to an epoxy/mPDA matrix was evaluated using the
fragmentation test. Optical microscopy, SEM and TEM, were used to study fiber fracture
surfaces and to determine the penetration of the surface treatment. Differential scanning
calorimetry (DSC) was used to look for changes in the melting behavior of the fibers observed
when the butadiyne treatment was applied to, and penetrated into, other polymer substrates [2].
The transparent yellow-tinted fibers were darkened in proportion to the surface treatment.
Fragmentation tests on control samples were influenced by the failure mechanism that arises
from the fibrillar morphology of the fiber. In carbon fibers the tensile failure is a sharp
transverse fracture and the fragmentation birefringence patterns begin at the well-defined
broken fiber ends and move away from the break. Total extinction is observed between the
fragment ends. During these tests the Kevlar-49 fiber-break traversed 10 to 20 fiber diameters
and was most easily seen in brightfield by a split, apparently in the fiber skin at a low angle
to the fiber axis. Using crossed polarizers the bulbous regions moving away from the break
were similar to those observed with carbon fibers, but there were varying levels of extinction
around the fiber throughout the broad fracture region. The fracture regions for the 0.84%
treated fiber looked like the control Kevlar in brightfield, but under crossed polarizers there
were numerous irregular bright regions indicative of stress concentrations. The bulbous regions
away from the break looked the same as those in the untreated fiber. No change was observed
in mean fragmentation length with fiber surface-treatment (Table 2).
TABLE 2--Fragmentation results for butadiyne-treated aramid and high-modulus fibers.

Fiber Treatment wt% Mean Fragment Length, mm Adhesion, MPa
HMS4 a -- 1.60 ...

0.1 J .97 ...
Kevlar-49 b -- 1.27 15.7
0.5 1.20 16.6
0.84 1.23 16.1
1.39 1.27 15.7
"See [91.
bSee [10].
3E. I. DuPont de Nemours & Company, Inc., Wilmington, DE 19898.
DSC was used to check for changes in the melting behavior of the treated fibers as an
indicator that the butadiyne had penetrated the fiber skin. The fiber-melting endotherms were
unperturbed although the weak polybutadiyne exotherm was observed. Treated and untreated
fibers broken on an Instron 4206 were observed under the optical microscope and the SEM.
No differences were observed in the highly fibrillated fracture surfaces. Thin sections of epoxy-
embedded treated and untreated fibers were viewed under the TEM [10]. The skin-core fiber
morphology was not visible nor were coatings below 2.2 wt%. At 2.2 wt%, a thin nodular
coating was seen on the fiber surface. The process of microtoming created both tension and
compression in the epoxy/fiber interface and caused some debonding. At 2.2%, coating debond-
ing was at both the epoxy-polybutadiyne and polybutadiyne-fiber interfaces. At lower levels
of treatment, where the coating was not resolvable, there were more fibrillar connects between
the fiber and matrix in regions debonded during sectioning relative to the untreated fibers.
The butadiyne surface-treatment was not effective in increasing the interfacial adhesion as
measured by the fragmentation tests (Table 2). The stress concentrations observed in the long-
fiber break region suggest that the coating changed the nature of the fiber surface, and possibly
improved adhesion in an irregular and spotty manner. The DSC analysis indicates that the
coating did not penetrate the fiber surface to the extent that it would alter the crystallization.
When the coating was heavily applied it appeared as a nodular surface coating in electron
micrographs indicating incompatibility with the fiber. Debonds caused by sectioning appeared
on both the resin and fiber side of the polybutadiyne coating, again indicating that the coating
did not penetrate the fiber surface. TEM micrographs of the more lightly treated fibers showed
more fiber/matrix fibrillation across the debonded regions than appeared in untreated fibers,
suggesting that fiber skin/matrix adhesion may have been improved. However, adhesion mea-
surements did not indicate improvement, possibly because of the inability of the surface
treatment to penetrate the skin of the fiber and improve the skin-core cohesion.
Influence of Fiber Surface Chemistry on Moisture Sensitivity of Fiber/Matrix Interface
It was postulated that hydrogenating the fiber surface to reduce surface polarity, although
it may slightly reduce dry fiber-matrix adhesion properties, might greatly improve humid-aged
properties. The postulate follows on the work of Drzal, et al., who showed that a surface-
treated AS1 fiber experienced good adhesion with an epoxy matrix, as determined by the
single-fiber fragmentation test, and that the surface concentration of oxygen on these fibers
was roughly 20% [11]. Heating the fibers in a hydrogen atmosphere to 750~ reduced the
surface oxygen content from 20 to 3%, yet reduced the interfacial adhesion by less than 10%.
In later work it was shown that the interfacial adhesion of AS 1 fibers in moisture-saturated
specimens was reduced by 27% when compared to dry specimens [12]. This work is therefore
a natural extension of the previous citation in which the fiber-surface oxygen concentration
is reduced to create a fiber-matrix interface with lower polarity in an attempt to reduce
hygrothermal deterioration of the fiber-matrix interface [13].
AS4 carbon fibers were tested with three different surface treatments. A control sample was
prepared by heating fibers to 105~ in vacuum. A second sample was prepared by heating at
660~ in a slowly flowing hydrogen atmosphere. A third sample was prepared by heating at
660~ in a slowly flowing argon atmosphere. After treatment the fibers were sealed in glass
tubes. The fiber-surface composition was determined using ESCA and adhesion to wet and
dry matrices was measured using the fragmentation test. Fragmentation specimens were made
by embedding single filaments of the treated tows along the axis of .0625-in.-thick tensile
coupons of a bisphenol-A-based epoxy resin cured with a stoichiometric amount of metapheny-
lenediamine. Dry coupons were aged three months under desiccant at room temperature. Wet
TABLE 3--ESCA analyses of AS4 fiber samples, a

Sample %C %0 %N %Na
Control 85.6 10.5 3.9 < 1.0

H2/660~ 85.9 7.35 2.4 4.35
Ar/660~ 84.5 9.0 3.0 3.5
aLaboratory of E J. Boerio, University of Cincinnati.
coupons were aged 82 days at 50~ and 75% relative humidity using a saturated NaC1 bath
for humidity control.
The surface oxygen content by ESCA measurement for the control AS4 fibers was 10.5%,
which is lower than the 20% previously reported for AS 1 fibers. The surface oxygen concentra-
tion decreased with annealing under argon and hydrogen atmospheres as shown in Table 3.
The concentration of sodium on the fiber surface was-increased during both the hydrogen and
argon annealing treatments. Sodium was present as an impurity in the poly(acrylonitrile) fiber
precursor and is apparently brought to the surface by these treatments.
After 82 days of humid-aging, moisture absorption had not completely stopped, but was
proceeding slowly, indicating that the samples were near equilibrium. The samples had absorbed
1.70 wt% moisture when tested. The mechanical properties of the matrix resin are shown in
Table 4. In general, the modulus and failure properties of the wet matrix were 10% below
those of the dry matrix.
The fragmentation data are presented in Table 5. For the dry specimens, the control had a
mean fragment length of 0.36 mm. The argon-annealed fibers had an unexpectedly shorter
mean fragment length of 0.30 mm and the hydrogenated fibers had a mean fragment length
roughly 10% higher than the control at 0.39 mm, as expected. The difference in fragment
lengths is attributed to adhesion changes as predicted by the Kelly-Tyson equation. The
differences could also be due to changes in fiber tensile strength with annealing treatment.
These were not measured although 600-700~ heat treatments have reportedly decreased fiber
tensile strengths [11]. However, this is of no consequence for comparisons of dry-aged and
humid-aged specimens for particular fiber treatments. For the humid-aged samples, the control
TABLE 4---Mechanical properties of matrix resin.

Property Dry Wet % Decrease
Tensile Modulus, MPaa 2360 2160 8.5

Break Stress, MPa 94 82 13
Break Strain, % 7.4 7.1 4
aGrip separation was used as gage length and for strain resulting in low modulus and high failure
strain values.
TABLE 5--Fragmentation data.

Fiber Dry Mean Fragment Length, mm Wet Mean Fragment Length, mm
AS4 Control 0,36 -+ 0.04 0.39 -+ 0.04

Argon Annealed 0,30 -+ 0.03 0.32 -4- 0.03
Hydrogen Annealed 0.39 -+ 0.04 0.39 + 0.04
and argon-annealed fibers have mean fragment lengths roughly 10% greater than the respective
dry samples, indicating a decrease in adhesion. The hydrogenated fiber showed no change in
mean fragment length indicating no change in adhesion and a less moisture-sensitive interface.
It was originally believed that the decrease in the fiber-matrix adhesion of the control sample
with humid-aging would be 25% or more, as was observed for AS1 fibers [12]. Perhaps
because the AS4 fiber has an inherently lower surface-oxygen content, its adhesion was more
stable under humid environments and only a 10% decrease was observed. Though the trends
observed (i.e., reduction in adhesion to control and argon-annealed fibers with humid-aging
and no reduction for hydrogenated fibers) make sense from an interface chemistry perspective,
the changes observed are dependent upon the precision of the experiment and no definite
conclusions will be stated.
Fundamental Look at Fiber/Matrix Adhesion
The butadiyne surface treatment was successful at improving the adhesion of AS4 fibers,
but was not effective in overcoming inherent morphological barriers to good adhesion in
HMS4 and Kevlar-49 fibers. Hydrogenation of AS4 fibers reduced the concentration of
oxygen on the fiber surface. Though the trends were small, the hydrogenated fibers had
slightly poorer adhesion than control fibers but were less sensitive to humid-aging. In the
previous studies, the fragmentation test proved to be a useful tool in comparing levels of
adhesion and providing clues to the mechanism of interfacial failure. However, the absolute
value of the interfacial adhesion estimated using the Kelly-Tyson equation must be considered
suspect because of the assumptions involved and the multitude of failure mechanisms
observed.
The Kelly-Tyson equation is derived from a force balance equating the fiber-failure strength
at the critical gage length and the tensile stress in the fiber as a result of shear stresses
imposed through the fiber/matrix interface. The only parameters involved are the fiber
diameter and tensile failure stress. The calculated interfacial adhesion is an average value
of the shear stress along the length of the fragment. The equation was derived for brittle
fibers in a ductile matrix where, if there is good adhesion and the neat matrix properties
persist up to the interface, the interface will fail in shear limited by the shear yield stress
of the neat resin.
Polymer matrix resins display nonlinear viscoelastic behavior which, depending on the
particular resin and the form of loading, can sometimes be approximated by linearly elastic
or, ideally, plastic behavior. The Kelly-Tyson model is based on ideally plastic behavior. An
analogous 1-dimensional model assuming elastic behavior was developed by Cox using a
shear lag approach [14]. Whereas in the plastic model the measured adhesion is limited by
the shear yield stress of the matrix resin, in the elastic model the measured adhesion is generally
proportional to the square root of the shear modulus [15,16]. These models are not complicated
and do not take into account normal forces, residual stresses, fragment end effects, Poisson
forces, and so on. Also not taken into account are the statistics involved in failure of the brittle
carbon fibers.
During the fragmentation test, the fiber fracture initiates a crack which propagates normal
to the fiber or along the fiber matrix interface. In the case of HMS4/epoxy-mPDA discussed
earlier, with "poor" adhesion, the crack propagates along the fiber/matrix interface for long
distances with little resistance and induces very little plastic deformation in the matrix resin.
The fragment lengths obtained under these circumstances would be expected to contain contri-
butions from the energy of crack propagation and the friction adhesion of the debonded
interface. In the case of AS4/epoxy-mPDA, with "good" adhesion, the crack forms along the
fiber but does not propagate significantly before fragmentation stops (fragments reach critical
TABLE 6--Mechanical properties of neat epoxy and cyanate resins.~

Epoxy Oxocy
Tensile Modulus, MPa 3400.0 3350.0

Tensile Strength, MPa 90:0 111.0
Failure Strain, % 6.3 5.4
Poisson's Ratio 0.4 0.4
Shear Modulus, MPa 1210.0 1200.0
Shear Strength, MPa 62.0 79.0
Glass Transition, ~ 144.0 240.0
Expanded Coefficient, ~ • 10 -6 79.0 55.0
Normal Stress, MPab 13.0 15.0
"According to ASTM D638 Tensile Properties of Plastics, using type I
specimens with bonded gages.
9As estimated from thermal stresses [19].
length). It is in cases such as this that the models under discussion are most applicable.
Needless to say, the observed fragmentation lengths and birefringence patterns are often found
between these two extremes.
General Study o f Factors Affecting Adhesion

In this study, we evaluated interfacial adhesion using the fragmentation test for two matrix
resins to AS4 and HMS4 fibers. We interpreted the data in terms of the elastic and plastic
models, the mechanical properties of the neat resins, the normal force on the interface, and
the wettability of the fibers by the uncured resins [17]. The two resins were mPDA-epoxy
and bis(4-cyanatophenyl)ether, which is a thermosetting cyanate ester resin and will be referred
to as oxocy. The cyanate resin was chosen because of reported expansion during cure [18].
Dilatometry was used to quantify this effect [19,20].
AS4 Fibers--The mechanical properties of the two resins and fragmentation results are
given in Tables 6 and 7~ For the AS4 fibers, adhesion is 28% higher with the cyanate as
compared to the epoxy resin. In this simple study, a possible chemical predictor for adhesion
is the cured-resin surface energy obtained through contact angle measurements. This was found
to be the same for the two resins. A possible mechanical predictor for adhesion is the thermally-
induced normal compressive stresses resulting from cooldown from the cure temperature.
Although the cyanate resin has a much higher glass transition than the epoxy, it also has a
lower thermal expansion coefficient and the calculated normal stresses are comparable. The
discriminator responsible for adhesion differences (as measured by the fragmentation test)
may therefore be the resin properties. Adhesion is 28% greater for the cyanate/AS4 system
TABLE 7--Fragmentation results.

HMS4 Fiber AS4 Fiber
Mean Fragment Adhesion," Mean Fragment Adhesion,"

Resin Length, mm MPa Length, mm MPa
Epoxy 1.03 13 0.37 46

Oxocy 0.97 14 0.30 59
aCalculated from Kelly-Tyson equation using Ic = 4/3 mean fragment length.
relative to epoxy/AS4. Shear strength is 27% higher for the cyanate resin and the plastic model
predicts a direct proportionality. The shear modulus is 1.5% lower for the cyanate resin and
the elastic model predicts a direct proportion to the square root of the shear modulus. For
these resin fiber systems, the observed trend is a far better fit by the plastic deformation model
(Eq 1) [17,19].
HMS4 Fibers--Adhesion is much poorer with the HMS4 fibers and is roughly the same
for the two matrix resins (Table 7). This is understandable in terms of an interfacial failure
mechanism that is cohesive in the outer layers of the fiber. When the fragmentation specimen
is viewed under crossed polarizers, little deformation is induced in the matrix by crack
propagation. In brightfield, it is obvious that the ends of the fragment move further apart with
increased matrix strain. It is as though the fiber ends are sliding in a debonded tube of resin.
If the crack can propagate without much resistance and if the debonded fiber can slide in the
resin tube, what is controlling the fragmentation length? The thermally induced normal stresses
in Table 6 are nearly equivalent to the average adhesive force estimated from the fragmentation
lengths in Table 7. If a frictional coefficient of 1 is assumed, the coincidence of the normal
compressive force and the measured adhesion suggests that a frictional mechanism may be
the dominant contributor to adhesion in HMS4 fiber/matrix systems as measured by the
fragmentation technique [17,19].
Resin Cure Expansion--The cyanate resin was chosen for this study because of reported
expansion of the resin during cure. It was believed that a thermoset with this property would
contribute to adhesion because of mechanical interlocking. When we monitored the specific
volume of the resin versus cure at room temperature (cure was isothermal at 200~ there was
a minimum at about 40% conversion and a slight increase in specific volume with further
curing. When monitored at cure temperature, the specific volume decreased continuously but
not linearly through full cure. This is to be expected for these systems as the functional groups
of the monomers move from a Van der Waals separation from neighboring molecules to
covalent bond distances with cure. The anomaly of the cure expansion as measured at room
temperature is a result of a much smaller glassy-state thermal expansion coefficient as compared
to the rubbery state and the advancement of the glass transition temperature with cure [19,20].
(There are thermosets that do actually expand on cure. Most often ring opening reactions to
expand the monomer are involved with the cure to counter the specific volume reduction
caused by bringing monomers to covalent distances [21].) Although at the cure temperature
we observed a continuous decrease in specific volume with cure, most of the volume change
occurred before gelation, and nearly all of it occurred before vitrification. When this happens,
the thermal stresses induced from cure cooldown are expected to be much greater than those
induced by resin curing.
Specific Chemical Interactions

It is tempting to look at the poor adhesion in the HMS4 case and the correlation between
the compressive normal force and the average interfacial adhesion from fragmentation data
and say that the measured adhesion is primarily a result of friction. It is also tempting to look
at the relative shear strengths of the cyanate and epoxy matrices and the relative levels of
adhesion from fragmentation data and say that adhesion to AS4 fibers is good in these systems,
that it approaches and is limited by the properties of the resin at the interface. However, in
reality interfacial adhesion is much more complicated. Hydrogenation of AS4 fibers decreased
adhesion by 10% and annealing under argon increased adhesion by 17% relative to vacuum-
annealed fibers. In all cases the matrix was epoxy-mPDA and the fragmentation birefringence
patterns did not show a change in failure mechanism. The only variable involved was the
chemistry of the interface. The control system in our studies was AS4/epoxy-mPDA. For one
batch of samples, made with control AS4 fibers, epoxy, and freshly distilled mPDA, the mean
fragment length increased 30%. No impurity could be found in either the mPDA or cured
epoxy using IR and GC. The neat resin mechanical and thermal properties were the same as
those of other batches of amine. The only variable was, again, chemistry.
In the previous study, the critical surface tensions of the cured mPDA-epoxy and the ether-
linked dicyanate resin discussed earlier were equal, and chemistry was considered as a variable
to explain adhesion differences in these systems. In light of the changes in measured adhesion
observed in these cases for which chemistry was the only variable, we decided to study
chemistry more closely. Schultz et al., have had success correlating fragmentation results with
an acid-base interaction parameter from inverse GC [22]. In our study, we chose to examine
the relative hydrogen bond donor and acceptor strengths of epoxy and cyanate resin structures
and precursor monomers [23]. The resin chemistries investigated included a bisphenol-A
dicyanate cured thermally and a bisphenol-A epoxy cured with a weakly basic aromatic diamine
(metaphenylenediamine) or with the structurally related but strongly basic aliphatic diamine
(1,3-diaminocyclohexane). Cured resin interactions were determined from model compounds
made, essentially, by reacting monomer monofunctional analogues (Fig. 1). The fragmentation
test was again used to measure interfacial adhesion.
Measurement of donor and acceptor strengths involved determination of the hydrogen-
bonded complex formation constant between the resin model compound or monomer and a
reference donor or acceptor probe molecule and then converting this constant to a respective
13-scale (hydrogen bond acceptor strength) or a-scale (hydrogen bond donor strength) parameter
by way of correlation equations [24,25]. For the B-scale measurement, the 19F NMR chemical
shift method using p-fluorophenol as the donor probe was selected [26]. For the a-scale
measurement, the UV spectroscopic method using pyridine N-oxide as the acceptor probe was
selected [27]. The c~- and [3-scales, which are independent of the probe molecule selected,
have been defined as ranging from 0 to 1, where 1 is the strongest respective hydrogen bond
donor or acceptor [24,25].
The results of the hydrogen bond acceptor measurements are shown in Fig. 1. For the resin
model compounds, the order of hydrogen-bond acceptor strength is aliphatic amine-epoxy >
bisphenol-A dicyanate > aromatic amine-epoxy. Within resin systems, the cyanate resin model
compound is a stronger hydrogen bond acceptor (base) than the cyanate monomer. For the
aromatic amine system, the amine and the epoxide monomers have roughly equivalent basicities
and are less basic than the resin model compound. For the aliphatic amine system, the amine
monomer is the strongest base followed by the resin model compound and the epoxide monomer.
The resin model compound hydrogen-bond donor (acid) measurements are also presented
in Fig. 1, and the order is aromatic amine-epoxy > aliphatic amine-epoxy > > bisphenol-A
dicyanate. This order differs significantly from the [3-scale measurements. Also, as expected,
the cyanate resin has an c~-scale value of 0 as a consequence of no active hydrogen.
The mechanical properties of the resins are shown in Table 8. All three resins have essentially
the same tensile modulus and strength, with the bisphenol-A dicyanate having a significantly
higher glass transition and lower failure strain relative to the two epoxy resins. Adhesion to
AS4 fibers was measured using the fragmentation technique. The mode of interfacial failure
as determined from fragmentation-birefringence patterns was similar for all three resins, typical
of good adhesion. The adhesion is plotted versus the relative hydrogen bond acceptor strength
in Fig. 2.
Within the limits of the B-scale values of the three resins and the range of the 'r values
measured, a linear correlation is seen between interfacial adhesion and resin model compound
hydrogen-bond acceptor strength (basicity). There is no correlation with the hydrogen-bond
AROMATIC AMINE EPOXY
0.37
~ -- NH2
0.37
OH.,J,~ OH ~ I
0.43 0.56
9
ALIPHATIC AMINE EPOXY
O -NH
0.71
O ~ OH ~ I 0.51 0.44
CYANATE
~ -O-C'N
0.41
~ O'eN , ' I ' O - x / ~ - ~ 0.47 0.00

N~N
FIG. 1--~- and oL-scale hydrogen bond acceptor and doner strength measurements on epoxy
and cyanate model systems.
donor strength (acidity). The improved adhesion with resin basicity is not surprising as the
standard manufacturers' fiber-surface treatments tend to place acidic functional groups on the
fiber surface. The influence of mechanical properties on the measured adhesion should be
minimal because of the similarity in properties between resins (same modulus and strength).
The bisphenol-A dicyanate resin has the most extreme properties (higher Tg and lower break
strain) and yet the measured adhesion is between that of the two epoxy resins. The two epoxy
resins have nearly identical properties and yet show the largest difference in interfacial adhesion
corresponding to the largest difference in model compound basicity. The linear dependence
TABLE 8--Mechanical properties, a
Tensile Modulus, Failure Stress, Break Strain, Glass Transition,
Resin MPa MPa % ~
Epoxy/Aromatic Amine 2970 89 7.4 166

Epoxy/Aliphatic Amine 3010 88 6.7 144
Bisphenol-A Dicyanate 3070 90 3.7 280
~According to ASTM D638 with clamp-on extensometer.
of interfacial adhesion on model compound basicity was observed over a limited range of
both basicity and 'good' adhesion. The limits of this linear dependence need to be determined
by further investigation.
The model compound hydrogen bonding studies have been carried out in dilute solution
and even though the model compounds contain the functional groups of the cured resin, they
are isolated and free to move into a favorable orientation with the probe molecule. The model
compounds are also polyfunctional and interacting with monofunctional probe molecules.
These subtleties as well as relative hydrogen bond donor measurements are discussed in more
detail in Ref. 23.
Another interesting aspect of this work is the measurement of the relative basicities of the
monomers as well as the model compounds representing the cured resin (Fig. 1). In the cyanate
system, the cured cyanate structure is a stronger base than the monomer and in the aromatic
amine-epoxy system the same is true; the cured resin model compound is more basic than the
80 w I
==
~iphotic ornine epoxy
40
1
_J
_< oromotic ornine epoxy
lad
20 ' ' '
Z
-- 0.35 0.45 0.55 0.65
RESIN HYDROGENBONDACCEPTORSTRENGTH(p)
FIG. 2--1nterfacial shear strength of single AS4 fiber/resin composites versus the relative hydrogen
bond acceptor strengths of the corresponding resin model compounds (error in ~5-scale values is
++_0.01).
two monomers which have roughly equivalent basicities. The aliphatic amine-epoxy system's,
amine is much more basic than either the cured resin model compound or the epoxide monomer.
It is difficult to predict how the relative basicities of reactants and monomers would affect
interfacial adhesion. If monomers interact with the fiber surface and do not bond to the matrix
network, adhesion would be hindered. If monomers interact with both the fiber surface and
matrix, adhesion would be enhanced. If only one monomer is favorably adsorbed on the fiber
surface, a gradient in the chemistry of the matrix could occur in the interface region. The
extent and importance of these phenomena are currently not known.
Summary
Vapor deposition polymerization of butadiyne on a high-strength carbon fiber has been
shown to both improve adhesion to an epoxy-mPDA matrix and to increase the tensile strength
of the fiber. However, the treatment was ineffective at improving the inherently poor adhesion
for high-modulus fiber/epoxy-mPDA and aramid/epoxy-mPDA systems. Reducing the surface
functionality of a high-strength carbon fiber, by annealing at 660~ in a hydrogen atmosphere,
decreased dry adhesion by about 10% relative to an untreated fiber but also decreased the
moisture sensitivity of the interface. These applied studies illustrate the usefulness of the
fragmentation technique as a relative indicator of interfacial adhesion and as a tool to determine
the mechanism of failure at the interface.
The adhesion of HMS4 fibers to cyanate and epoxy-mPDA matrix resins is poor and
independent of resin chemistry. During the fragmentation test an interfacial crack propagates
easily without inducing much plastic deformation in the matrices. The adhesion estimated
from fragmentation results is roughly equivalent to the compressive normal force resulting
from cooldown from cure, indicating that a friction factor approach may be suitable for
modeling this type of interfacial failure. The adhesion of AS4 fibers to cyanate and epoxy-
mPDA matrices is good, involves crack propagation of only short distances along the fiber/
matrix interface, and is strongest for the cyanate matrix as opposed to the epoxy-mPDA matrix.
In this case, adhesion as measured using the fragmentation technique may be good enough to
be limited by the shear properties of the matrix.
The simple mechanics models used to describe the fragmentation test do not do explain
adhesion changes where chemistry is the only variable. In a study of three matrix resins
with comparable mechanical properties (aliphatic amine-epoxy, aromatic amine-epoxy, and
bisphenol-A dicyanate) and good adhesion to AS4 fibers, a linear dependence was shown
between the measured adhesion and relative resin model compound basicity that could not be
explained by current mechanical models.
Acknowledgments
The authors wish to acknowledge the Office of Naval Research for financial support and
thank L. T. Drzal, W. D. Bascom, E J. Boerio, and L. Peebles for advice and assistance.
References
[1] Zisman, W. A., "Surface Chemistry of Plastics Reinforced by Strong Fibers," I&EC Product
Research and Development, Vol. 8, June 1969, pp. 97-111.
[2] Snow, A. W., "Polymerization of Butadiyne: Polymer Characterization and Properties," New Mono-
mers and Polymers, B. M. Culbertson and C. U. Pittman, Eds., Plenum Press, New York, 1984,
pp. 399--414.
[3] Drzal, L. T., Rich, M. J., Camping, J. D., and Park, W. J., "Interracial Shear Strength and Failure
Mechanisms in Graphite Fiber Composites," 35th RP/C Proceedings, SP1 Inc., Section 20-C, 1980,
pp. 1-7.
[4] Armistead, J. E, Snow, A. W., and Bascom, W. D., "Butadiyne Vapor Deposition Polymerization
on Carbon Fibers," International SAMPE Technical Conference Series, Vol. 19, October 1987,
pp. 644-652.
[5] Kelly, A. and Tyson, W. R., "Tensile Properties of Fiber Reinforced Metals. Copper/Tungsten and
Copper/Molybdenum," Journal of the Mechanics and Physics of Solids, Vol. 13, 1965, pp. 329-350.
[6] Arrnistead, J. E, unpublished data.
[7] Chatzi, E. G. and Koenig, J. L., "Morphology and Structure of Kevlar Fibers: A Review," Polymer
and Plastics Technology and Engineering, Vol. 26, 1987, pp. 229-270.
[8] Morgan, R. J., Pruneda, C. O., and Steele, W. J., "The Relationship Between the Physical Structure
and the Microscopic Deformation and Failure Processes of Poly(p-Phenylene Terephthalamide)
Fibers," Journal of Polymer Science: Polymer Physics Edition, Vol. 21, 1983, pp. 1757-1783.
[9] Snow, A. W., Armistead, J. E, and Glison, A., "Effect of Butadiyne Deposition Polymerization on
Kevlar-49 Fibers as a Surface Treatment to Promote Epoxy Resin Adhesion," NRL Memorandum
6120-724:AWS, November 1988.
[10] Kalantar, J. and Drzal, L. T., "Adhesion of Butadiyne Treated Kevlar-49 Fibers to Epoxy Matrices,
A Preliminary Report," June 1988.
[11] Drzal, L. T., Rich, M. J., and Lloyd, P. E, "Adhesion of Graphite Fibers to Epoxy Matrices: I. The
Role of Fiber Surface Treatment," Journal of Adhesion, Vol. 16, 1982, pp. 1-30.
[12] Drzal, L. T., Rich, M. J., and Koenig, M. E, "Adhesion of Graphite Fibers to Epoxy Matrices: III.
The Effect of Hygrothermal Exposure," Journal of Adhesion, Vol. 18, 1985, pp. 49-72.
[13] Armistead, J. P., Snow, A. W., and Peebles, L., "Humid Aging Study of AS4 Fibers," NRL Ltr
Ser 6120-202, May 1989.
[14] Cox, H. L., "The Elasticity and Strength of Paper and Other Fibrous Materials," British Journal
of Applied Physics, Vol. 3, 1952, pp. 72-79.
[15] Rao, V. and Drzal, L. T., "The Dependence of Interfacial Shear Strength on Matrix and Interphase
Properties," Polymer Composites, Vol. 12, 1991, pp. 48-56.
[16] Asloun, Ei, M., Nardin, M., and Schultz, J., "Stress Transfer in Single-Fiber Composites: Effect
of Adhesion, Elastic Modulus of Fiber and Matrix, and Polymer Chain Mobility," Journal of
Materials Science, Vol. 24, 1989, pp. 1835-1844.
[17] Armistead, J. E and Snow, A. W., "Fiber/Matrix Adhesion in Cyanate Resin Systems," Polymer
Preprints, Vol. 31, 1990, pp. 537-538.
[18] Shimp, D. A., "The Translation of Dicyanate Structure and Cyclotrimerization Efficiency to Polycya-
nate Properties," Polymeric Materials Science and Engineering Preprints, Vol. 54, 1986, pp.
107-I 12.
[19] Armistead, J. P. and Snow, A. W., "Fiber/Matrix Load Transfer in Cyanate Resin Carbon Fiber
Systems," Polymer Composites, Vol. 15(6), 1994, pp. 385-392.
[20] Snow, A. W. and Armistead, J. P., "A Simple Dilatometer for Thermoset Cure Shrinkage and Thermal
Expansion Measurements," Journal of Applied Polymer Science, Vol. 52, 1994, pp. 401-411.
[21] Bailey, W. J., Chou, J. L., Feng, P. -Z., Issari, B., Kuruganti, V., and Zhou, L. L., "Recent Advances
in Free-Radical Ring-Opening Polymerization," Journal of Macromolecular Science, Chemistry
Ed., Vol. 25, 1988, pp. 781-798.
[22] Schultz, J., Lavielle, L , and Martin, C., "The Role of the Interface in Carbon Fiber-Epoxy
Composites," Journal of Adhesion, Vol. 23, 1987, pp. 45-60.
[23] Snow, A. W. and Armistead, J. P., "Comparative Epoxy and Cyanate Resin Hydrogen Bonding
and Fiber Interface Study," Journal of Adhesion, Vol. 52, 1995, pp. 223-250.
[24] Kamlet, M. J. and Taft, R. W., "The Solvatochromic Comparison Method. I. The 13-Scale of Solvent
Hydrogen-Bond Acceptor (HBA) Basicities," Journal of the American Chemical Society, Vol. 98,
1976, pp. 377-383.
[25] Taft, R. W. and Kamlet, M. J., "The Solvochromic Comparison Method. II. The a-Scale of Solvent
Hydrogen-Bond Acceptor (HBA) Basicities," Journal of the American Chemical Society, Vol. 98,
1976, pp. 2886-2894.
[26] Gurka, D. and Taft, R. W., "Studies of Hydrogen-Bonded Complex Formation with p-Fluorophenol.
VI. The Fluorine Nuclear Magnetic Resonance Method," Journal of the American Chemical Society,
Vol. 91, 1969, pp. 4794-480l.
[27] Abboud, J. L., Sraidi, K., Abraham, M. H., and Taft, R. W., "Studies on Amphiprotic Compounds.
4. Application of the a u2 Scale to Complexation between Pyridine N-Oxide and Monomeric
Hydrogen-Bond Donors in Cyclohexane," Journal of Organic Chemistry, Vol. 55, 1990, pp.
2230-2232.
Bgatrice Large-Toumi, 1 Michelle Salvia, 2 a n d Ldo Vincent 2
Fiber/Matrix Interface Effect on

Monotonic and Fatigue Behavior of
Unidirectional Carbon/Epoxy
Composites
REFERENCE: Large-Toumi, B., Salvia, M., and Vincent, L., "Fiber/Matrix Interface Effect
on Monotonic and Fatigue Behavior of Unidirectional Carbon/Epoxy Composites," Fiber,
Matrix, and Interface Properties, ASTM STP 1290, C. J. Spragg and L. T. Drzal, Eds., American
Society for Testing and Materials, 1996, pp. 182-200.
ABSTRACT: This paper deals with the study of the influence of fiber/matrix interface on the
fatigue behavior of unidirectional carbon/epoxy composites. These materials are based on a
T300/DGEBA/DDM system and the sizing nature, fiber oxidation, and degree of cure are varied
independently. The interfacial properties were evaluated by the microdebonding test. The results
show the good adhesion developed in our systems. The epoxy specific sizing increases the
interfacial strength and the lack of oxidation treatment slightly decreases it.
In three-point bending, the failures of the composites all appear on the compressive side,
due to undesirable fiber kinking phenomena. To overcome these difficulties, which are linked
to the compressive effects of the loading nose, a compression bending test was adapted to
perform monotonic as well as fatigue tests. Under this configuration, progressive failure initiates
on the tensile side and slowly propagates through the thickness of the sample. The W/Shler's
curves can be plotted with moderate scattering. They are very flat, in comparison to the GFRP
curves, and thus show the excellent fatigue behavior of these materials.
The effect of the interface on both the monotonic and cyclic behaviors is clearly pointed out.
The lack of fiber oxidation increases the monotonic properties without changing the fatigue
resistance significantly. On the other hand, a slight matrix under-cure leads to a better intrinsic
fatigue resistance but to smaller properties-to-failure. The analysis of both monotonic and cyclic
properties reveals the existence of different optima depending on the interfacial properties. The
"best" interface does not exist; it is a function of both the fiber/matrix system properties and
the mechanical performance studied.
KEYWORDS: carbon/epoxy composites, unidirectional composites, fiber/matrix interface, drop

test, acoustic emission, compression bending test, fatigue, S-N curve, damage
Because of their good specific properties advanced composites, especially carbon fiber
composites, are used increasingly in the aeronautics industry as primary structures. This industry
requires materials exhibiting enhanced reliability and durability (damage-tolerance). For this
purpose many studies have been performed to understand and model the behavior of the
materials. This research has shown that composite behavior does not only involve the properties
IEngineer, L'OrEal, 90 rue G6nrral Roguet, 92583 Clichy crdex.

2Assistant professor and professor, head of laboratory, respectively, D6partement MMP, UMR CNRS
5621, Ecole Centrale de Lyon B.P. 163, 69131 Ecully crdex.
182
Downloaded/printed
LARGE-TOUMI ET AL. ON INTERFACE EFFECT 183
of its basic constituents, but that the fiber/matrix interface also plays a leading role in the
global behavior of the material.
Much research has been devoted to the development of specific fiber surface treatments
allowing for improvement of the adhesion interface, often considered to be the "weak point"
of the composite. It is therefore necessary to define the optimum interface properties for a
given fiber/matrix system. This optimum can not be the same for all loading conditions, however.
The purpose o f this work is to study the effects of interface modification on the fatigue
behavior of unidirectional (UD) carbon/epoxy composites loaded in the direction of the rein-
forcement. Many review papers [1-7] have dealt with the behavior of a UD composite submitted
to cyclic loading. However, the influence of the interface has rarely been studied specifically [8]
and such studies do not appear, as of yet, to have been carried out on UD carbon fiber composites.
In order to point out the contribution of the interface, particular materials and techniques
were selected. The flexural loading test was chosen because it is easier to perform and led to
less scattered results than the tensile test. The characterization of the interface was investigated,
first through the global behavior of the composite (fractographic analysis and classic interlami-
nar shear strength (ILSS) tests), then by a micromechanical test involving the interface directly
(the drop test). This characterization allows for classification of materials according to their
interface properties and thus permits a better understanding of the interactions existing between
fiber-matrix bonding and the macroscopic properties of the composite.
Materials
In order to point out the effect of the fiber-matrix interface, three very similar material
models, named S, DS, and NO (Table 1) were used, the only difference being the nature of
the interface. These materials are made from Toray High Resistance fiber (T300-6K) and Ciba-
Geigy DGEBA/DDM (LY1808RF/HT972 similar to the system LY556/HT972). Manufacture
was achieved through lay-up of unidirectional tapes (14 plies) with subsequent curing in the
autoclave. The cure cycle is supposed to lead to a nearly complete crosslinking of the system
(Fig. l - - l o n g cycle). The fiber volume contents is about 60%. The following features vary
successively around a system of reference (material S):
TABLE 1--Definition and glass transition temperature f o r Torsion Dynamic Spectrometry (TDS):"
Frequency 1 HZ; temperature rate equal to 2~ Differential Scanner Calorimeter (DSC): b
temperature rate equal to lO~ for the five materials studied.
Material Cycle of
Reference Matrix Cure Fiber Sizing Oxidation Tg, ~
Second
First Sweep Sweep
S DGEBA/DDM long T300-6K 5 yes 163.0" 167.0a

145.0 b 153.0b
DS DGEBA/DDM long T300-6K 4 yes 166.0a 168.0"
150.00 161.0 b
NO DGEBA/DDM long T300-6K 5 no 162.0" 167.0~
145.0 b 154.~
SC DGEBA/DDM short T300-6K 5 yes 154.5" 164.5~
13ZOb 153.0b
914 914 standard T300-6K 5 yes -- 190.0a
"Performed by TDS.
bperformed by DSC.
1. The type of sizing: Compatibility with the polymer may be enhanced by surface treatment
with sizing. The sizings used are Toray commercial sizings 5 and 4; sizing 5 (material S and
NO), the standard sizing for T300 fiber, is an epoxy- and polyester-compatible sizing while
sizing 4 (material DS) is only epoxy-compatible. The glass transition temperatures of sizings
5 and 4 measured by DMA on T300 fiber before processing are (3 Hz) 16 and 45~ respectively
[9]. At room temperature, sizing 4 (rather the interface system created in the composite material
by the fiber + sizing + resin association) will probably have a glassy behavior while sizing
5 (or the associated system) will lead to a softer interface zone.
2. The oxidative treatment of the fiber (material NO): improved interfacial bonding is
usually obtained by oxidation. T300 fibers present a relatively disordered structure up to the
surface and a high surface reactivity (for a carbon fiber), even without oxidative treatment.
In this case, oxidation does not lead to superficial structural rearrangements, only to the
removal of contaminating species and improvements in chemical reactivity (adding of surface
oxygen groups and increase in the superficial nitrogen rate). The effects are therefore less
marked than in the case of other carbon fibers (e.g., XAU fiber). Nevertheless, it is expected
that fiber-matrix adhesion can be improved. With standard sizing (5) the fibers used for the
material NO were not oxidized.
Thus, only one parameter varies in comparison with the system of reference. A study
concerning the influence of characteristics of the matrix on dynamic behavior was carried out
concurrently. For this purpose two industrial materials were chosen. For both materials the
carbon fibers (T300) had a treatment similar to those for reference material S. In the first
case, retaining the matrix of the previous materials, the degree of crosslinking was modified
by the use of an industrial cycle (Fig. 1--short cycle) without post-cure, leading to an incomplete
crosslinking (material SC) as shown in Table 1 by dynamic spectrometry and Differential
Scanner Calorimeter (DSC) measurements (two successive sweeps). In the second case the
system was made from commercial resin (epoxy resin filled with thermoplastic particles of
Polyethersulfone (PES) (material 914). The curing cycle was the standard cycle for this type
of resin. In both cases the composite was processed using the same technique used for previous
materials, with similar fiber volume contents. Table 1 defines the five composites studied and
gives their mechanical glass transition temperatures determined by DSC and torsion dynamic
spectrometry around the fiber axis.
Jt--Temperature ',~'~ 3 hours Pressure

~'-',
125oc _1_ 1. L I /
\
/
I
I X "7 bars
80~
Time
i D
-0,45 bar short cure cycle ~. [
lona: cure cycle
FIG. 1--Cure cycles for the materials made for DGEBA/DDM.
Experimental Procedures
Since the first research by Boller [10], Owen [11], and Dharan [12], behavior of composites
under tensile cyclic loading has been studied by several other authors. However, the specific
influence of the interface was rarely studied. Recently, some studies have shown the role of
interface modification on the tensile behavior of carbon/epoxy laminates [13,14]. To our
knowledge such studies have not yet been carried out on UD carbon fiber composites (under
cyclic loading in tension or bending).
Concerning a UD glass fiber composite, only tests under flexural cyclic loading appear to
highlight the influence of the interface on fatigue behavior [8,15]. This method is easier to
conduct than a tensile test and the absence of clamping jaws limits scattering of the results.
It is expected that the same result can be obtained in the case of a UD carbon fiber.
However, three-point bending fatigue tests (as monotonic ones) induce local stress concentra-
tions beneath the loading nose. This may promote premature failure on the compressive side
of the sample. This type of failure has already been referred to by several authors [16-18].
The failure always results from a microbuckling of the fiber (kink band) due to local stress
concentration. Such tests do not constitute a characterization, even in compression, of the
material because of the complex stress state in the failure zone on the compressive side of
the sample [19-21].
To overcome these difficulties, a test developed by Fukuda [18] in monotonic loading
conditions was adapted recently [22] to perform fatigue tests. The test consists of applying
an axial compression to a sample of rectangular cross-section. The specimen is gripped at
each end and can accommodate local deformation at these points by in-plane rotation. Axial
compression induces Euler-buckling of the bar at the critical load Pc (Pc = L~ ]]" In a
post-buckling configuration, the sample is submitted to a bending strain as shown in Fig. 2

where the parameters P, L0, L, g, ~, 0, and f are defined. In this post-critical configuration
calculations are carried out with the assumption that "great displacements--small local strains."
In these conditions the expression for the curvature of the beam can be put in the form:
EId20
- - - P sin 0 (1)
ds 2
where s is the curvilinear abscissa.
~ :P
IL x
t.& a :
Lo
FIG. 2--Compression bending configuration; definition of the parameters P, Lo, L, ~, e~, O, .f
The integration of (1) leads to expressions (2a) and (2b) giving, respectively, the relative
longitudinal displacement (~/L0) and the relative deflection f/L0 as a function of the parameter ct:
= 2 1
[ E(sin(a/2))]
F(sin(ed2))_l
(2a)
f = sin(od2) (2b)
L0 F(sin(ed2))
where F (sin(od2)) and E (sin(ed2)) are Legendre canonical functions (an elliptic integral
without an analytic expression).
Stresses and strains induced in the material subjected to the load P can be then calculated
using the following relationships [22]:
in bending:
6Pf
= ~ and e/=
~r2hfP (3)(4)
2/-~Pc
in compression:
P P
~rc = ~-~ and ec = b h E (5)(6)
where b and h are the width and the thickness of the sample, respectively.
In order to secure a quasi-pure bending strain it is necessary that the contributions of both
the shear strain and the pure compressive strain become negligible. For the strain range
considered this condition is deemed to be fulfilled if the span-to-depth ratio (Lo/h) is greater
than 52.
The determination of the bending stress and strain requires knowledge of both the load
(easily measured by means of a force transducer), P, and the deflection at point of midspan,
f. The direct measurement o f f (method adopted by Fukuda [18] and Grandsire-Vinqon, et al.,
[23]) requires the use of a controlled displacement transducer (only one of the sample extremities
moves, thus the point of midspan location moves along the loading axis during the test). This
solution is difficult especially in the case of fatigue tests which are performed under relatively
high frequency. The use of strain gages attached to the middle of a sample is not appropriate.
It was decided to calculateffrom the measured longitudinal displacement ~ using Eqs 2a and 2b.
Tests were performed in a conditioned room (T = 23 -+ 2~ RH = 50 • 5%). The span-
to-depth ratio was chosen to be 52 in order to avoid shear stress. In the case of monotonic
tests the loading rate was 2 mm/min. Fatigue tests were carried out under imposed quasi-sine-
shaped strain. The ratio of the minimum-to-maximum strain during a fatigue cycle was 0.3
in order to remain in a post-buckling configuration throughout the test cycle. The frequency
chosen (25 Hz) involved only a local evolution of the temperature (less than 10~ which
stabilized rapidly. Damage growth is monitored by the stiffness loss.
In both cases the samples were about 100-mm-long, 10-mm-wide, and 1.7-mm-thick. During
monotonic testing monitoring acoustic emission allowed for the detection of the very first
damages, the observation of the initial compression phase of the sample, in particular. Further-
more, a fractographic analysis at two levels (optical microscopy and scanning electron micros-
copy) facilitated observation of the damage (location and type of failures) and provided
information on the quality of the fiber-matrix bonding.
In order to completely understand micromechanical interface properties two techniques were
used in parallel:
1. The three-point bending test is often used with short samples to evaluate the interlaminar
shear strength (ILSS) of a composite. To obtain this, the span-to-depth ratio L/h must be about
5. This test is often considered an excellent way to compare the quality of fiber-matrix adhesion
in different composites [8]. However, shear tests also characterize all types of interfaces,
especially interply interfaces. Moreover, the result can be influenced by the presence of defects
such as voids or fiber misalignments that reduce ILSS values. The passage of a macroscopic
result to a microscopic interpretation is complex and has to be carried out with prudence.
2. The drop test [24] is a method specially designed to directly obtain interface shear
strength. This test is a variant of the pull-out test [25-26]. It consists of performing a tension
test on a fiber fractionally embedded in a matrix microdroplet. During the test, the droplet is
held by means of two blades (Fig. 3). At the other end the fiber is subjected to tension at a
constant strain rate until the fiber is pulled out of the matrix. The test provides the force at
which the matrix droplet/fiber debonding occurs, Fd. An important parameter of the test is
the embedded fiber length, L. In the case of systems for which interface adhesion is strong,
this length has to be sufficiently small so that the required force for debonding remains less
than the ultimate tensile force of the fiber. Indeed, simple analysis shows that there is competition
between the two phenomena [24]. A critical embedded length can be thus defined: Systems
for which interface adhesion is strong (which is the case of the systems studied) need critical
length of less than 100 ixm. Unlike the pull-out test, the drop test technique allows a very
small embedded length (as small as 30 ~m).
Tests are performed with an Adamel (DY22) tensometer supplied with a force transducer
0-5N as shown Fig. 3. The loading rate is 0.10 mm/min. An optical control of the position
of the blade was made before as well as during the test. After each test, the debonded drop
length, L, and its diameter, 2R, are measured by optical microscopy or SEM (Fig. 4). Drop
lengths are typically within the range 20 to 100 ~m. In all cases the failure takes place at the
interface and not in the matrix (adhesive failure). From the knowledge of the debonding force
and the drop geometry, the shear strength at the interface is determined within the frame of
"average stress" modeling.
Monotonic loading behavior
Table 2 provides the values of strength and elongation (er~) at failure obtained by the
compression-bending test, the ILSS values, and the shear strength at the interface given by
I
FIG. 3--Apparatus for the drop test.
FIG. 4---Scanning electron micrograph of the drop test sample after experimentation (that is
after the fiber has been pulled out of the matrix); the residual meniscus (arrow) shows the initial
location of the drop.
the drop test. For the drop tests, only the four systems made from the matrix DGEBA/DDM
were studied. Indeed, the resin 914 is too viscous to allow application of a small drop. This
table clearly points out the good coherence between the measured values of ILSS and the
results gained by the drop test. These two techniques provide the same classification in interface
shear strength terms.
Figure 5 allows the comparison of the nature of failure obtained by classical three-point
bending and compression bending. By compression bending, a "broom-like" failure with
important debonding lengths is obtained; this is a characteristic feature of a tensile failure.
Compressive failures obtained in three-point bending [27] masked the intrinsic behavior of
TABLE 2--Mean static mechanical properties (coefficients of variation in parentheses).

Interlaminar
Flexural Strength Elongation Shear Interface Shear
Modulus, at Break, at Break, Strength, Strength,
Material GPa MPa % MPa MPa
S 111 (2.3%) 1785 (4.0%) 1.66 (3.0%) 76 (5.0%) 72 __+8

DS 106 (2.0%) 1734 (3.0%) 1.67 (4.2%) 79 (3.5%) 78 +__10
NO 110 (1.2%) 1902 (2.2%) 1.82 (1.0%) 67 (6.0%) 60 + 11
SC 107 (1.5%) 1729 (3.9%) 1.63 (1.6%) 75 (3.0%) 70 +- 7
914 112 (2.0%) 1940 (2.8%) 1.87 (3.9%) 99 (3.0%) --
FIG. 5---Carbon~epoxy (material NO)failure features after three-point bending test (top) and
compression bending test (bottom).
the material. This led to the conclusion that in bending loading conditions, the weak point of
the carbon epoxy composites studied is not the behavior in compression but, on the contrary,
is the strength in tension. Failures are progressive; the composite damages gradually by the
failure of fibers (or bundles) followed by delaminations on the tensile side of the sample. The
damage propagates through the width and depth of the sample. The first failures generally
occur because of macroscopic defects of the sample: These can be defects from cutting at the
edge of the sample or defects due to fiber alignment on the tensile face.
Figures 6a and 6b, respectively, provide a representative example of the experimental curve
P = f(~) and the corresponding curve tr = fie). The latter figure (Fig. 6b) shows the great
linearity of the behavior in bending CFRP; the first deviation from linearity coincides with
the appearance of the macroscopic first failures of bundles. Acoustic emission remains negligi-
ble (Fig. 6a) during the entire phase of compression (until the point 1 of curves P = fl~) and
tr = J(e)). Therefore, in this type of test the stage of compression does not involve damage.
Acoustic emission is recorded during the loading plateau (between point 1 and point 3) even
if no macroscopic damage is subsequently detected. Despite the diversity of interpretations
given to acoustic emission results [28-31] it seems that failure of fibers could be associated
with events of large amplitudes and weak energies [29]. Therefore, in the linear part of the
curve ~ = f(e), even if the modulus does not seem weakened microscopic damage associated
with isolated fiber failures (short events ~ 50 p~s) occurs. These failures induce further failures
as well as delaminations of fiber bundles (longer events ~ 200 p~s) and the emergence of a
deviation from linearity. Effectively, from point 3, the emission becomes intense and both
FIG. 6a--Variations of load and acoustic emission counts with displacement (material S).
2000
1800
1600 i I
J
nS
t200
1000
800
,i
6O0 ,./"
4OO /I
jm
f Strain 1%)
I I I I
0,5 l 1,5 2
FIG. 6b--Stresslstrain curve corresponding to load~displacement curve given Fig. 6a.
macroscopic damage (failure of bundles followed by delamination) and the occurrence of the
linear deviation on the stress-strain curve are observed.
The direct observation of failure features were made on materials tested in a monotonic
compression bending test. This type of observation shows the existence of a "failure zone"
Z1 (typically within the range 20 to 30 mm) looking like two interlocked combs (Fig. 7). This
zone can be characterized by the existence of a quasi-constant bending moment My. It is within
this zone that randomly distributed failures of fibers intervene. The zone noted Z2 is that on
which the delamination of broken fibers (or bundles of fibers) spreads; this zone spreads over
4/5 of the sample length. From one material to another zone Z1 exhibits a difference that can
be described, as a first approximation, in terms of a modification in the width of the teeth of
the comb (Table 3). The material NO presents a very specific broom-like feature where the
teeth of the comb are particularly thin (<0.5 mm) and zone ZI is larger than in other cases.
For the other materials, this zone presents larger strips (1 mm), up to 2 mm for the UC material.
In this case, the damage takes place in successive stages failure of large flat bundles followed
by a momentary pause in damage growth until the failure of a new bundle of fibers.
To better understand these observations, a more detailed analysis by scanning electron
microscopy (SEM) was undertaken on failures obtained in both monotonic and fatigue loading
~ omeflt, M f
~ " /(' --
I IZI: Failure zone/ I
I I
I Z2: delamination zone I
FIG. 7--Schematic description of the failure feature observed in the monotonic compression
bending condition.
TABLE 3--Features of the zone ZL

Material Comb "teeth" Width, mm Comments
NO <0.5 ZI zone very large fiber

S, DS 1 failures by bundles
SC 2 "
914 2 "
conditions. This observation points out the similarity between failure features whatever the
loading conditions, i.e., monotonic or dynamic. In the both cases, two types of materials can
be observed:
1. Strongly bonded composites: S, DS, SC, 914. For these materials, the process of failure
involves mainly cohesive failure phenomena by shear of the matrix (Fig. 8 (DS) and Fig. 9
(914)). This is indicated by the presence of strips characteristic of this type of damage. This
mechanism happens after the failure of fibers and controls the delamination stage that follows.
2. Weakly bonded composites: NO. Figure 10 shows that for this material the failure is
widely controlled by interface debonding. Broken fibers often exhibit great lengths of debonding
at their extremities: The feature displays many bare fibers, as well as matrix sheaths emptied
of their fiber where only the imprint remains.
All of these observations are consistent in pointing out the specific behavior of the non-
oxidized fiber composite (NO) in contrast to the other materials. For this material, the damage
process following fiber breakage is governed by interface debonding (specific to a weak
interface). For materials S, SC, DS, and 914, these processes result from cohesive failure,
primarily by shear of the matrix (specific to a strong interface). The great lengths of debonding
observed (zone Z1) as a feature of the material NO are without doubt due to relaxation of
local stress concentrations after isolated fiber failures. This adaptation does not typically occur
in the other systems. It seems, therefore, that a strong fiber-matrix bonding could be harmful
for monotonic mechanical properties.
Behavior Under Cyclic Loading

As the tests are performed under strain control, the applied load P necessary to keep a given
deflection or strain decreases with time or the number of cycles from the initial value P0 and
FIG. 8--Scanning electron micrograph of failure feature observed on a strongly bonded material
(DS) subjected to monotonic compression bending loading.
damage can be followed by variations in samp!.e stiffness. The fatigue life criterion chosen
(and the most common) is the number of cycles Nt0 corresponding to a 10% loss of stiffness.
Unlike the three-point bending tests, the results obtained clearly display the progressive nature
of the damage of CFRP loaded in post-buckling conditions. Figure 11 shows the evolution of
the load ratio P/Po, for the two types of fatigue tests at the same level of maximum strain:
by three-point bending, a sudden failure was observed due to local contact overstresses; by
compression bending the sample undergoes more progressive damage where the mechanisms
are comparable to those obtained in monotonic conditions. The first damage appears near the
macroscopic defects of the sample and, if the level of loading is high enough, a phase of
propagation follows.
Using the criterion Nl0, Fig. 12 shows the S (or e ) - - N curves for the five materials. As in
the case of GFRP studied in three-point bending [32] or compression-bending [15] with
controlled strain, the S-N curves can be described by a linear relationship:
Emax = A -- B log NI0 (7)
where
Emax is maximum imposed strain,

-----
A = is extrapolated strain value inducing 10% stiffness loss after only one cycle, and
B = is relative admissible strain decrease by decade of cycles.
Table 4 gives values of parameters A and B, such as those defined in the formula (7). The
ratio A/e,-fb keeps a quasi-constant value for the five materials. This shows that the A value
FIG. 9--Scanning electron micrograph of failure feature observed on a strongly bonded material
(914) subjected to monotonic compression bending loading.
is strongly linked to failure properties in monotonic conditions. There is, therefore, as in the
case of GFRP [15,32-34] an analogy between the monotonic behavior and the fatigue properties.
In fact, all the results plotted on Fig. 12, except for the case of the material SC, are in agreement
with the SLERA (strength life equal rank assumption) concept. Indeed, this concept states
that if a sample could be tested in both monotonic and fatigue loadings, each individual
member would occupy the same rank in both the strength and fatigue life data series [35].
Very similar values obtained for A and e ~ indicate that the effect of the strain rate is weak
for UD CFRP unlike UD GFRP. In the latter case the importance of this effect can be attributed
essentially to glass reinforcement [36]. The influence of the strain rate is nevertheless more
significant for systems based on DGEBA/DDM than for the material made from industrial
resin. This result is consistent with the materials' relative glass transition temperature positions.
Figure 13 presents normalized S-N curves given by the relationship:
elA =- 1 - B/A log Nl0 (8)
These curves allow better visualization of the real resistance in fatigue independent of initial
strength level.
Discussion of Results
Analysis of the fatigue behavior for a particular material is based on the knowledge of both
parameters A and B of the relationship (7). For all the materials studied, the value of parameter
FIG. 1O---Scanning electron micrograph of failure feature observed on a weakly bonded material
(NO) subjected to monotonic compression bending loading.
100
90
8O
a.*
70
6O
50
i0 ~ 10~- 103 t04 10s
Numbe~ of cycles
FIG. 11--Effect of the loading conditions on the evolution of load ratio PIPo with the number
of cycles (material S; E,,~ = 1.66%).
A is very close to the elongation at break obtained in monotonic loading conditions (Table 4).
Thus, for this analysis, A is considered to be equivalent to the classical failure strain. This
discussion will be, at first, limited to the three isomatrix materials S, NO, and DS varying
only (from a micromechanical point of view) by the level of interface shear strength. For these
three materials, post mortem examination clearly showed two modes of damage: interface
debonding (material NO) and matrix shear (materials S and DS).
Comparison between mechanical properties of the interface zone and matrix explains this
difference in behavior. In Fig. 14 the elongation-to-failure was plotted against the interface
1,8 ~ ~"
1,5 .~ ~.
1 2 3 4 5 6 7
log(number of cycles)
FIG. 1 2 - - S - N curves for the five materials.
TABLE 4---Values of parameters of the S-N curve A and B.

(r = elongation at failure obtained by monotonic loading.)
Material A B (10 -2) A/s %
S 1.77 3.18 107

DS 1.75 4.32 105
NO 1.93 4.24 106
SC 1.69 2.14 104
914 1.84 3.36 98
1,0.
0,8
2 3 4 " 5 6 7
log(number of cycles)
FIG. 13 Normalized S - N curves for the five materials.
shear strength (values are those calculated from drop-test results (Table 2)). The shear strength
of the matrix "rm can be estimated to be about 70 MPa using the previous observations (Figs.
8 to 10). For the material NO the interface shear strength would be less than that of the matrix
(Zone I of Fig. 14); for materials S and DS the situation is reversed (Zone II). It must be
noted that this transition in the local damage mode, that is interface debonding to matrix shear,
leads to a decrease in the elongation at break of the composite. These results can be compared
to those obtained by Bader et al. [37]. As in the present work, the failure properties in monotonic
conditions decrease with the improvement of interface quality. On the other hand, if the
ELongationat break
e,~(%)
I,e. //- ~'\
\
~ Failure feature :
L8-
I .... 1
t.
I
1.7.
1,6- i.llerfacel .... I I' |

I
Zone I ~- 14 ~e 2 Interface $]l~ar
1,5 , ,,--r ~ ~ s~renglh
' ' ' x, (MPa)
50 55 60 65 70 75 80
FIG. 14--Elongation at break (compression bending) versus shear interface strength (drop test).
interface shear strength vanishes (i.e., case of a loose bundle) the elongation at break becomes
very low (er ~ 1% (R'mili et al. [38]); Er ~ 0.6% Bader [37]). This allows the supposition
that an optimal value of the interface shear strength "riopexists (Fig. 14). In the present work
this value is probably not reached, even in the case of material NO, because carbon T300
fiber sizings have a high surface reactivity with the matrix even without oxidative treatTnent.
This decrease in mechanical properties with the improvement of inteffacial strength is
unexpected in the framework of the classic statistical approach to fracture in composites. It
is well known, since the pioneering papers of Rosen [39] and Zweben ]40], that the fracture
of composites is a complex phenomenon. In order to describe such a process it is necessary
to take into account both the isolated failures of fibers and the effect of stress concentrations
in the vicinity of these primary breaks or "singlets." In effect, the n adjacent surviving fibers
in the vicinity of a singlet are subjected to stress concentration kcr on a length h which may
be related to the stress transfer length introduced initially as the ineffective length by Rosen
[39] or to the positively affected length proposed by Barry [41]. The stress concentration in
the bridging fibers induces new failures in the vicinity of the singlet to form a "multiplet"
which may be regarded as a Griffith crack growing in Mode I and leading above a critical
size to the failure of the sample. In a sample containing n/fibers of length 1, the number of
multiplets containing "i" broken fibers or "i-plets" is according to the approximation of Batdorf
and Ghaffarian [42], given by the following relationship:
(9)
where cr is the applied stress, a and m the Weibull distribution parameters for a single fiber,
with n:, hi, kj the previously defined parameters which control the conditions of formation of
new fiber failures around the growing i-plets. If, as in Batdorf's assumption, the failure of the
sample is due to the appearance of a single critical iplet with icr fiber failures, the breakage
occurs when the conditions of Eq 10 are met.
Q,cr(~r) = 1 (10)
The failure conditions depend on all the successive values of hi. These values globally
decrease (as does the ineffective length) with the increase of the interface shear strength, "r,.
An increase in ri results from the (Eq 10) higher stress or strain at break and therefore in an
improvement in the mechanical behavior. On the other hand, the experimental results plotted
in Fig. 14 show that the elongation-to-failure decreases with the growth of "ri, at least within
the range of interface shear strength values measured in the present work. To understand these
results, which are contrary to the previous approach, it can be seen with Bader [37,43] that
the value of the interface strength controls the failure mode. For a very strong interface the
fracture initiates from the first critical iplet as described in the preceding approach and
propagates in a brittle manner across the entire cross-section. On the other hand, if the interface
is weak, (i.e., "ri is low), a fiber matrix debonding occurs as soon as a fiber breaks [44]. For
intermediate values of "r~,this debonding happens at the interface between subcritical iplet (i
< i,.,) and the surrounding matrix. This debonding leads to a local splitting and to the typical
broom like fracture as observed on the tensile side during the compression bending tests (Fig.
4). This debonding uncouples the growing iplet from the rest of the sample and its growth is
stopped. Further iplets therefore have to form until the final failure. This successive iplet
initiation extends the failure process and increases the strength or elongation-to-failure in
rendering the composite more damage tolerant and less brittle. Thus taking into account the
splitting due to debonding around the subcritical iplets, it appears that a decrease in the interface
shear strength can increase the fracture strain or stress as can be seen experimentally. The
observations of the nature of microscopic failures are in agreement with this analysis: The
material NO, having the lowest value of "ri, presents a broom-like or comb-like failure feature
with very fine teeth (Fig. 7) which corresponds to relatively easy debonding conditions. On
the other hand, when a'~ increases the debonded bundles of fibers become thicker (materials
S or DS) and the composite then tends towards a quasi-brittle behavior leading to a decrease
in its failure strain.
The parameter B, the gradient of the S-N line in the diagram (~, log N), describes the
damage tolerance in fatigue conditions. B has been plotted against the interface shear strength
9i in Fig. 15. If at first the discussion is limited to the three isomatrix materials (S, NO, DS),
it is the reference material (S) that exhibits the best damage tolerance in cyclic loading
conditions. Regarding this property, materials (NO) and (DS) are approximately equivalent
and inferior to the reference material. In fact, the damage is depicted in this study by a loss
in stiffness of the sample when loaded under bending conditions. This loss in stiffness is
essentially due to the delamination of fiber bundles broken in the vicinity of the median section
and their subsequent growth towards the ends of the sample [36]. The kinetics of the damage
growth are controlled by the resistance to delamination. When passing from the material NO
B (~0,01)
|
, |
3- I
i!
Q [~ InterfaceLshear(MPa)Strength
2--~ ~
50 60 70 80
FIG. 15--Parameter B of the S-N curve versus shear interface strength (drop test); the lower
B is, the better the fatigue behavior of the material.
198 FIBER,MATRIX, AND INTERFACEPROPERTIES
to the material S the failure mode changes from a quasi-brittle debonding at the fiber-matrix
interface to a cohesive failure in the surrounding matrix. This kind of failure is without doubt
more resistant to crack growth in Mode II. Therefore, it appears reasonable to find a lower
value for B in the case of the reference material S. Nevertheless, the relatively high value of
B associated with the material DS is not clearly understood because a cohesive rupture takes
place in this material, too. It is important to remember that the interfacial zone of the material
DS is probably in the glassy state under test conditions while in the material S it is either in
the rubbery state or in the transition domain.
Moreover, the influence of matrix properties is illustrated by the position of the material
SC in Fig. 15. In the case of material SC, subcrosslinking of the matrix allows for improvement
in the tolerance to fatigue damage by lowering parameter B. It appears that the value of B is
controlled by micromechanical characteristics of the interface zone as well as those of the matrix.
Conclusion
An original and dynamic bending test, the compression bending test was chosen for this
work which deals with the influence of interface shear strength on the fatigue behavior of
unidirectional carbon epoxy composites. Experimental results obtained by monotonic or cyclic
tests show that the weakest point of the tested materials (in bending conditions) is not compres-
sive strength, but tensile strength.
Micromechanical interface properties were measured by two techniques; an indirect global
technique, the ILLS test, and a direct technique, the drop test. These two methods provided
the same classification of the different interfaces.
The fatigue behavior can be described in the plane (emax log N) by S-N lines defined by
two parameters; an extrapolated strain value giving 10% stiffness loss after one cycle and B,
the slope of the line. This behavior depends in a complex manner on all the characteristics
of the interface region and on the surrounding matrix. Relative to the reference material (S)
the results show the following:
1. The lack of oxidative treatment on fibers T300 has little influence on fatigue behavior
but essentially enhances the elongation-at-failure and the parameter A.
2. The use of a specific epoxy sizing (sizing 4) leads to a reduction in the damage tolerance
(increase of the B parameter) with no effect on the elongation-at-failure and the parameter A.
3. The moderate subcrosslinking of the matrix results in a slight decrease of parameters
A and B. The lowering of B has a positive effect on the damage rate.
It is necessary to emphasize optimization of the matrix and interface properties. For example,
although in the case of material 914 interface shear strength is high, its damage tolerance
remains sufficiently high. This fact is probably related to its complex structure (thermoplastic
particles dispersed in the epoxy matrix) leading to improved resistance to the growth of cracks
in Mode II, which in turn slows down the delamination process of broken bundles and loss
in stiffness.
Analysis of the results and comparison between monotonic and cyclic properties show that
a "best" interface does not exist. It is a function of both fiber-matrix system properties and
performance studied. The final interpretation of results depends on the performance required.
Two criteria can be used: the intrinsic resistance to cyclic loading (ratio B/A as seen in Fig.
15) which, in terms of endurance has the classification SC > S ~- 914 > NO > DS, and the
fatigue life expectancy for a given applied strain (Fig. 14). According to this very widespread
criterion, material NO presents the best behavior and the classification becomes (within the
range of 10 3 to 10 7 cycles) NO > 914 > S > SC > DS.
Acknowledgments
The authors are grateful to the Direction de la Recherche et des Etudes Techniques (DRET)
for supporting this work and to Aerospatiale (Louis B16riot Joint Research Center) and Brochier
for the samples manufacturing. The authors also wish to B. Chabert and J. P. Soulier of
Universit6 Claude Bernard Lyon I for their assistance and valuable advice concerning the use
of the drop test.
References
[1] Harris, B., "Fatigue and Accumulation of Damage in Reinforced Plastics," Composites, Vol. 8,
1977, pp. 214-220.
[2] Talreja, R., "Fatigue Behaviour Composite Materials: Damage Mechanisms and Fatigue Life
Diagrams," Proceedings of The Royal Society of London, A 378, 1981, pp. 461475.
[3] Konur, O. and Matthews, E L+, "Effect of the Properties of the Constituents on the Fatigue
Performances of Composites: A Review," Composites, Vol. 20, 1989, pp. 317-328.
[4] Curtis, P. T., "The Fatigue Behaviour of Fibrous Composite Materials," Journal of Strain Analysis,
Vol. 24, No. 4, 1989, pp. 235-244.
[5] Bathias, C., "La Fatigue des Mattriaux Composites ~ Hautes Performances," Mat~riau~r et Tech-
niques, May 1990, pp. 11-17.
[6] Hahn, H. T., "Fatigue of Composites," section 1.4, Delaware Composites Design Encyclopedia,
Vol. I- Mechanical Behavior and Properties of Composite Materials, Technomic Publishing Com-
pany, 1989.
[7] Reifsnider, K. L., "Damage and Damage Mechanics," in Fatigue of Composite Materials, Chap.
2, K. L. Reifsnider, Elsevier Science Publishers, 1990, pp. 11-77.
[8] Shih, G. C. and Ebert, L. J., "The Effect of the Fibre/Matrix Interface on the Flexural Fatigue
Performance of Unidirectional Fibreglass Composites," Composites Science and Technology, Vol.
28, 1987, pp. 137-161.
[9] Toumi, B., "Etude du Comportement en Fatigue des Composites Carbone/t~poxy: Rtle de l'Inter-
face," Doctoral Thesis, No. 94-16, Ecully, 28 March 1994.
[10] Boller, K. H., "Fatigue Characteristics of RP Laminates Subjected to Axial Loading," Modern
Plastics, Vol. 41, 1964, pp. 148-150, 180.
[11] Owen, M. J., "Fatigue of Carbon Fibre Reinforced Plastics," in Composite Materials, J. Broutman
and R. H+ Broutman, Eds., Academic Press, Vol. 5, 1974, pp. 341-369.
[12] Dharan, C. K. H., "Fatigue Failure in Graphite and Glass Fibres Polymer Composites," Journal
[13] lvens, J., Wevers, M. and Verpoest, I., "The Effect of the Carbon/Epoxy Interface on Damage
Accumulation During Fatigue," 1CCM 9, A. Miravete, Ed., Woodhead, Madrid, Vol. 2, 12-16 July
1993, pp. 724-731.
[14] Matsuhisa, Y., Withers, P. J., and King, J. E., "Effects of Fibre Surface Treatment and Test
Temperature on Monotonic and Fatigue Properties of Carbon Fibre Epoxy Cross Ply Laminates,"
Proceedings of ECCM6, 20-24 September 1993, WP Limited, Bordeaux, France, pp. 275-280.
[15] Fournier, P., "Comportement sous Sollicitations Cycliques des Composites Unidirectionnels Verre/
Epoxy. Cas des Rtsines h Tenacit6 Amtliorte," doctoral thesis, ECL, 1992.
[16] Parry, T. V. and Wronski, A. S. "Kinking and Tensile, Compressive and Intedaminar Shear Failure
in Carbon-Fibre-Reinforced Plastic Beam Tested in Flexure," Journal of Materials Science, Vol.
16, 1981, pp. 439--450.
[17] Yurgatis, S. W. and Sternstein, S. S. "A Micrographic Study of Bending Failure On Five
Thermoplastic/Carbon Fibre Composite Laminates," Semi-annual progress report, NASA Grant
NAG-I-253, May 1987.
[18] Fukuda, H. "A New Bending Test Method of Advanced Composites," Advanced Composite Materi-
als, C. Bathias and M. Uemura, Eds., SIRPE Publishers, Paris, 1990, pp. 171-176.
[19] Greszczuk, L. B., "Theoretical Studies of the Mechanics of the Fibre-Matrix Interface in Compos-
ites," Interfaces in Composites, ASTM STP 452, 1969, pp. 42-58.
[20] Binienda, W. K., Roberts, G. D., and Papadopoulos, D. S., "Effect of Contact Stresses in Four-
Point Bend Testing of Graphite/Epoxy and Graphite/PMR-15 Composite Beams," USA, SAMPE
q., Vol. 23, No. 3, 1992 pp. 20-28.
[2l] Uemura, M. and Iwai, H., "Flexural Testing and Evaluation Methods of Advanced Composite
Materials," Advanced Composite Materials, C. Bathias and M. Uemura, Eds., SIRPE Publishers,
Paris, 1990, pp. 134-139.
[22] Vincent, L., Chateauminois, A., Fournier, E, Pelissou, O., and Large-Toumi, B., "Post-Buckling
Method Applied to the Static and Fatigue Characterization of Composites Materials," EACM,
ECCM-CTS, Amsterdam, t992, pp. 235-244.
[23] Grandsire-Vincon, I., Auvray, M. H., and Sigety, P., "Comportement en Compression des Composites
Unidirectionnels," ONERA, rapport technique No. 53/7086 M, June 1992.
[24] Miller, B., Muri, P., and Rebenfeld, L., "A Microbond Method for Determination of the Shear
Strength of a Resin/Fibre Interface," Composites Science and Technology, No. 28, 1987, pp. 17-32.
[25] Piggott, M. R., Sanadi, A., Chua, P. S., and Andison, D., "Mechanical Interactions in the Interphasial
Region of Fibre Reinforced Thermosets," Composite Interfaces, H. Ishida and J. L. Koenig, Eds.,
Elsevier Science Publishing Co., Inc., 1986, pp. 109-121.
[26] Favre, J. P. and Pitkethly, M. J. "Round Robin on Interfacial Test Methods," draft report, 1991.
[27] Toumi, B., Salvia, M., and Vincent, L., "Compression Bending Fatigue Test for CFRP," Matdriaux
et Techniques, No. 6-7, 1994, pp. 45~-9.
[28] Valentin, D., Bonniau, P., and Bunsell, A. R., "Failure Mechanism Discrimination in Carbon Fibre-
Reinforced Epoxy Composites," Composites, Vol. 14, No. 4, October 1983.
[29] Berthelot, J. M., "Relation Between Amplitudes and Rupture Mechanisms in Composite Materials,"
Journal of Reinforced Plastics and Composites, Vol. 7, May 1988.
[30] Rochat, N., Foug~res, and Fleischman, P., "Delayed Acoustic Emission: a Rheological Approach,"
Journal of Acoustic Emission, Vol. 9, No. 2, 1990, pp. 91-96.
[31] Bhai, M. R, and Murthy, C. R. L., "Fatigue Damage Stages in Unidirectional Glass Fibre Epoxy
Composite: Identification Through Acoustic Emission Technique," International Journal of Fatigue,
Vol. 15, No. 5, 1993, pp. 401-405.
[32] Fiore, L., "Contribution h l'l~tude du Comportement en Fatigue des Matrriaux Composites h Renfort
Verre Unidirectionnel," doctoral thesis, ECL, 1988.
[33] Sims, G. D. and Gladman, D. G,, "A Framework for Specifying the Fatigue Performance of Glass
Fibre Reinforced Plastics," NPL report DMA (a) 59, 1982, pp. 1-24.
[34] Mandell, J. F., "Fatigue Behaviour of Fibre-Resin Composites," Applied Science Publ., G. Pritchard,
Ed., London, New York, 1982, pp. 67-107.
[35] Chou, P. C. and Croman, R., "Residual Strength in Fatigue Based on Strength Life Equal Rank
Assumption," Journal of Composite Materials, Vol. 12, 1978, pp. 171-189.
[36] Salvia, M. and Vincent, L., "Contribution h la Modrlisation du Comportement Sous Sollicitations
Cycliques en Flexion Trois Points de Composites UD Renforcrs par des Fibres de Verre," J. P.
Favre and A. Vautfin, Eds., Compte-Rendus JNC9, Saint Etienne, France, Vol. 2, November 1994,
pp. 657--669.
[37] Bader, M. G., Picketing, K. L., Buxton, A., Rezaifard, A., and Smith, P. A., "Failure Micromechan-
isms in Continuous Carbon-Fibre/Epoxy-Resin Composite," Composites Science and Technology,
Vol. 48, 1993, pp. 135-145.
[38] R'mili, M., Mallet, S., Cardinal, S., Merle, P., and Gobin, P. F., "Effect of Coating Process on
Weibull Parameters of Carbon Fibre Bundles for MMC," International Symposium on Advanced
Materials, Proceedings of Japan-Europe Symposium on Composite Materials, Nagoya 1-2 June
1993.
[39] Rosen, B. W., "Tensile Failure of Fibrous Composites," AIAA Journal Vol. 2, No. 11, 1964,
pp. 1985-1991.
[40] Zweben, C., "Tensile Failure of Fibre Composites," AIAA Journal Vol. 6, No. 12, 1968, pp.
2325-2331.
'[41] Barry, P. W., "The Longitudinal Tensile Strength of Unidirectional Fibrous Composites," Journal
[42] Batdorf, S. B. and Ghaffarian, R., "Tensile Strength of Unidirectionally Reinforced Composites-
II," Journal of Reinforced Plastics and Composites, Vol. 1, 1982, pp. 165-176.
[43] Bader, M. G., "Tensile Strength of Uniaxial Composites," Science and Engineering of Composite
Materials, Vol. l, 1988, pp. 1-11.
[44] Piggott, M. R., Load-Bearing Fibre Composites, Pergamon Press, 1980, pp. 83-87.

STP1290 Eb.1415051 1 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

STP1290 Eb.1415051 1 PDF

Uploaded by

Copyright:

Available Formats

Copyright by ASTM Int'l (all rights reserved); Mon Jan 16 18:33:32 EST 2012

Fiber, Matrix, and

Christopher J. Spragg and Lawrence T. Drzal, Editors

ASTM STP Publication Code Number (PCN):

Fiber, matrix, and interface properties / Christopher J. Spragg and

Authorization to photocopy items for internal, personal or educational classroom use, or

Peer Review Policy

Interlaboratory Study of Adhesion Using Voltage Contrast XPS--J. D. MILLER,

A New Technique to Study the Interfacial Strength and Transverse Cracking

Nonlinear Elastic Characterization of

KEYWORDS: graphite, crystal, modulus, X-ray diffraction, uniform stress, crystallite

High-performance carbon fibers are important in many applications, particularly in aerospace

Materials Engineer, NASA Goddard Space Flight Center, Greenbelt, MD 20771.

E--- pv2 (1)

Results and Discussion

F = (f I(3 sin2• - l)cos • d• f l cos x d• (2)

F= [3<cos2r) - l]/2 (3)

TABLE 1--Comparison of stress with average orientation angle.

P25 12.0 13.86 72.82

o') #00 " ~ . . ~ ,,_ P55

TABLE 2--1/sH (Gpa).

Y = 0.0882 doSpl(mDN) (4)

lIE = $33 q- (2s13 - 2s33 + s44)I3[11 + ( s l l - 2s13 + $33 - s44)/5]/1 (5)

Calculated Modulus (GPa)

600 ..................................... ~ ............................................. o~7 .................................. :O. . . . .

400 ........... ~ ................................... . . * ................................................................... '~ ........................

300 ................................. ~......................................................................................... ~ ......

20 40 60 BO 100 120 140 160

FIG. 4---Calculated modulus by Uniform Stress model versus applied stress.

L = size = ICMB cos 0 (7)

Methods of Determining the

Many materials exhibit time-dependent deformation, known as creep, when subjected to a

Methods of Determining Temperature Dependencies

2 Manufactured by Textron Specialty Materials, Inc., Lowell, MA.

5.75 6.00 6.25 6.50 6.75 7.00

Application and Comparison of the Techniques

TABLE 1--Temperature dependence determined by the cross-cut method.

0.0027 10-100 680

0.00 ' ' ' '

Temperature Compensatea Time (minutes x 1019)

TABLE 2--Calculated values of the temperature dependencefor creep of SCS-6.

Arrhenius 0.0027 624

Method A, min -p p Q, kJ/mol

Sherby-Dorn 31.6 0.24 550

~176176176 ,o;o o'oo +o'oo .o'..'oo +ooo

Interfacial Measurements and Fracture

ABSTRACT: The micromechanics of stress transfer in single-fiber as well as multi-fiber

Fundamental to the discussion of composite micromechanics is the representative volume

Remote Laser Raman Microscopy (ReRaM)

1Sol]car (Toray) Industries, France.

Results and Discussion

The Raman Frequency Stress-Gage

(a) Distance along the FGL / mm

Average Fiber Stress/GPa

-0.25 0 0.25 0.5 0.75

(b) Applied Composite Strain / %

24.5 25 25.5 26 26.5 27 27.5 28 28.5

(b) Distance along the FGL / m m

24.5 25 25.5 26 26.5 27 27.5 28 28 5

(b) Distance along the FGL / mm

Average Max. ISS / MPa

1.7 f Fiber +IA; Applied Strain 0.50%]

1.7 f Fiber +3A: Applied Strain 0.50%1