Professional Documents
Culture Documents
PVP2008-61746
STRESS CLASSIFICATION LINES STRAIGHT THROUGH SINGULARITIES
Arturs Kalnins
Professor Emeritus of Mechanics
Lehigh University
Bethlehem, Pennsylvania
E-mail: ak01@Lehigh.edu
ABSTRACT
The paper considers geometries of pressure vessels and
components for which the theoretical models contain sharp
corners, representing singularities. The idea is proposed that a
stress classification line passed straight through the singularity
can yield linearized stresses that are applicable in pressure
vessel design. Using elastic finite element analysis, details of
the procedures by which this result can be achieved are given
for two examples. One is a sharp corner at the toe of a fillet
weld. Membrane and bending stresses are calculated directly in
the toe plane, showing little or no dependence on mesh size.
The other is an axisymmetric shell with a flat head and a sharp
corner at the joint. The objective is to determine the primaryplus-secondary stress intensity on a Stress Classification Line
(SCL) through the joint. Two methods are used. One is by
determining the zone of valid SCLs and extrapolating the
linearized stresses to the joint. The other is by calculating the
linearized stresses directly on the SCL through the joint.
Conditions for the use of the SCL through the joint for the
shell/flat head model are established.
NOMENCLATURE
S11, S22, S12 = in-plane stresses in FEA model of 2-D solid
elements in X,Y,Z coordinate system, as shown in Figure 4
S33 = out-of-plane stress in FEA model, in Z direction, hoop
stress for axisymmetric 2-D elements
NFORC1, NFORC2 = nodal forces of 2-D solid elements
m , b = membrane and bending stress, respectively
P+Q = primary plus secondary Tresca stress
m, b = suffixes for membrane and bending stresses
F = force developed by stresses on SCP
M = moment developed by stresses on SCP
SCP = Stress Classification Plane
SCL = Stress Classification Line, an SCP of infinitesimal width
1 INTRODUCTION
For design-by-analysis of pressure vessels, the leap from
shell analysis to finite element analysis (FEA) about forty years
ago brought much benefit but left some details in a more
difficult position. One of such details is a sharp corner* at a
local structural discontinuity. It plays no role in shell analysis
but influences the stresses in FEA. If the corner radius is
unspecified at the design stage, it is common practice to model
the corner with a zero radius. In that case, these corners
represent a singularity.
The problem with singularities is that FEA-calculated
elastic stresses at the singularity increase without bound as the
mesh is refined. These stress values have no physical meaning
for any mesh size. The question is whether the linearized
stresses acting on planes straight through the singularity are
still meaningful for design. This is addressed in the paper.
2 PREVIOUS WORK
Previous work on the classification of FEA-calculated
stresses goes back to more than three decades. The early papers
are by Kroenke1, Kroenke, Addicott, and Hinton2, Hinton and
Hechmer3, and Gordon4. Hechmer and Hollinger5 extended the
concept to three-dimensional geometries. Hollinger and
Hechmer6 gave a summary of this work. More recent papers are
by Ming-Wan Lu, Yong Chen, Jian-Guo Li7, and by Strzelczyk
and Ho8. The focus of these papers is on the membrane and
membrane-plus-bending stress intensities that are appropriate
for assessing the stress intensity limits involving primary and
secondary stresses.
*
For the purposes of this paper, a sharp corner is one for which the
angle through the material between two intersecting boundary planes is greater
than . In a continuum analysis, the stresses at such corners are infinite. For
angles less than , the stresses at the corners are zero.
2005 by ASME
Copyright 2008
Abaqus). FEA models for two mesh sizes were built, consisting
of 1 and 4 elements through thickness, shown in Figure 2. The
deformed shape that results from this loading and boundary
conditions is shown in Figure 3 for the 4-element model.
336
120
12
P
SCL
Figure 1: Geometry of the example
MODEL
This is meant to be a simple example to illustrate a case in
which an SCL passes straight through a singularity. Gordons4
edits are performed to calculate the membrane and bending
stresses for the model shown in Figure 1. The dimensions are in
mm. Both the nodal stress and nodal force methods are used.
For the stress method, the nodal stresses are integrated over the
SCL to obtain the forces and moments, while, for the nodal
force method, the internal or reaction forces are summed over
the nodes of the SCL The results are also checked with
Abaqus/Standard11 version 6.7-1 CAE postprocessors stress
linearization option.
The middle node of the left end of the model is restrained
from vertical and horizontal displacement. The middle node at
the right end is restrained from vertical displacement and
subjected to a force of P=1200 Newtons (270 lb). Solutions are
obtained by the Abaqus11 finite element program using 8noded, quadratic, 2-D solid, plane strain elements (CPE8 in
F = 1200 N
1200 12
M=
120 = 5,143N mm
336
(1)
(2)
The term SCL refers to a line that is used to represent an SCP in a 2-D
model. This justifies the term singularity, which refers to a point on the SCL.
It is understood that stresses are applied to a plane (SCP), not to a line (SCL).
1200
= 100 Mpa
12
6
b = 2 5,143 = 214.3Mpa
12
m =
(3)
SCL
(4)
4.2
STACK
Figure 4: Zoom of singularity region
SCL
Node
57
257
457
457
657
857
857
1057
1257
1257
1457
1657
S11
-89.34
-40.93
-0.26
-0.18
42.09
83.75
83.49
112.00
155.60
163.60
272.10
477.70
NFORC1
-43.17
-80.79
-6.97
9.54
83.39
34.62
44.18
244.30
77.79
90.60
465.00
281.50
yc
y
The stress S11 and force NFORC1, both per unit out-ofplane length, are normal to the SCL. In the table, the nodes 457,
857, and 1257 have two values, one received from each
adjoining element. The fact that they are not the same indicates
that they have not been averaged. Even though the stresses
could have been averaged between the elements of the STACK,
the integration scheme for not-averaged stresses, given in
subsection 4.4, is convenient for biased nodal distances.
y2
4.3
OBJECTIVE
Two singularity effects are investigated in this subsection.
The first is the behavior of stresses at a singularity. This is
shown in Figure 6 for the stress component normal to the SCL.
For the 4-element model, the stresses for Figure 6 are taken
from Table 1. It is seen that S11 at node 1657 increases from
333 MPa for 1 element mesh to 478 MPs for a 4 element mesh.
This illustrates the main problem addressed in this paper that
stresses at singularity diverge with refined mesh.
Mk =
y1
(6)
hk
[( y yc )1 S111 + ( y yc ) 2 4 S112 + ( y yc )3 S113 )
6
where y is the coordinate with origin at bottom of SCL (see
Figure 5), k denotes the k-th element, y1 and y2 are the y
coordinates at bottom and top of k-th element, hk is its height,
subscripts 1, 2, and 3 refer to the three nodes on the SCL of the
k-th element, and yc is the y value to the centroid of the SCP
represented by the SCL (node 857 in Figure 5). For the 4element model, the membrane and bending stresses acting on
the SCL are then obtained from equations (7) and (8).
500
400
Stress, MPa
( y y )S11dy
1 k =4
Fk
t k =1
6 k =4
b = 2 Mk
t k =1
300
m =
200
100
0
(7)
(8)
-100
4.5
-200
0
10
12
4-elements
S11dy
y1
hk
( S111 + 4 S112 + S113 )
6
b =
6
t2
m =12
(y
m =1
(9)
yc ) NFORC1m
(10)
NFORCm and
ym are the nodal force and the y coordinate of the m-th node,
respectively, on the SCL.
1 m =12
NFORC1m
t m =1
4.4
Fk =
m =
4.6
(5)
4.7
RESULTS
The results are shown in the tables below.
Elements
Error %
Nodal Stress
Method
103.2
3.2
100.6
0.6
Nodal Force
Method
100.0
0.0
100.0
0.0
Abaqus CAE
Postprocessor
103.2
3.2
100.7
0.7
Exact
n/a
100.0
0.0
5.1
MODEL
The edit is performed for the axisymmetric vessel shown in
Figure 7. To assess divergence with refined mesh, FEA models
with 1, 2, 4, and 8 elements across the shell wall were built.
The 4-element model is shown in Figure 7. The axis of
symmetry is marked by the dot-dashed line and the SCL of
interest by the red line at the joint of the shell and the head.
For the cylindrical shell, the inside and outside radii are 10
and 11.5 inches (254 and 292 mm), respectively, and the length
is 15 inches (381 mm). The flat head is 3 inches (76.2 mm)
thick. Uniform internal pressure is applied. The lower end of
the shell is subjected to a symmetry boundary condition in Y
direction. The nodes of the shell and head are defined
separately and tied together at the joint. The modulus of
elasticity is 30,000 ksi (207 GPa), Poissons ratio is 0.3, and the
design stress intensity S m is 17.5 ksi (120.7 MPa).
Elements
Error %
Nodal Stress
Method
214.2
0.0
217.0
1.3
214.3
0.0
214.3
0.0
Abaqus CAE
Postprocessor
214.4
0.1
217.5
1.5
Exact
n/a
214.3
0.0
Nodal Force
Method
4.8
DISCUSSION OF EXAMPLE
The results of Table 2 and Table 3 illustrate the key point
made in this paper that while stresses at singularity diverge
with refined mesh, the membrane and bending stresses do not.
Comparison of the results of the nodal stress and nodal
force method support Gordons4 finding that the nodal
force method is the more accurate of the two, matching, in
fact, the exact results for both the 1 and 4 element models in
this case. These results support the claim that an SCL passed
straight through the singularity of a finite element model can
yield linearized stresses that do not diverge with refined mesh,
at least for the 2-D solid elements used in the example.
The close agreement of the nodal stress method and the
Abaqus11 CAE postprocessors stress linearization option is
expected because the stress integration rules are essentially the
same as those given by equations (5) to (8). However, the
procedure stated in subsection 4.6 had to be followed to
achieve that result.
For the primary and secondary stress categories, the
calculation of stress intensities is required. This is what
Gordons4 Stress Intensity Edit is meant for. It will be discussed
next.
Figure 7: Model for Shell/Flat Head Example
5.2
P+Q INGREDIENTS
S11, S22, S12 are the in-plane stresses in the X,Y directions
shown in Figure 7, and S33 is the out-of-plane hoop stress in Z
direction. The suffixes m and b denote membrane and
bending stresses, except in S11(m) and S12(m), where the m
has nothing to do with physical membrane action.
Critical SCL
5.3
Valid SCLs
S11
S12
25.97
-1.20
-0.26
-0.87
-0.99
-1.04
-1.01
-1.01
-19.13
1.72
-0.08
-0.66
-0.01
-0.01
-0.02
-0.02
30
25
20
15
10
5
0
-5
0.00
0.25
0.50
0.75
1.00
1.25
1.50
5.4
EXTRAPOLATION TO JOINT
The P+Q stress at the joint can be obtained by
extrapolating the stresses from the valid zone to the joint. This
procedure was used in Appendix IV of Hechmer and
Hollingers5 WRC Bulletin No. 429 for a similar geometry.
To see how this would work for the current example, the
linearized stresses on the ID of the shell, which were
determined in subsection 5.3, are plotted in Figure 11 and
Figure 12. In these figures, a valid-SCL limit is assumed at
Y=0.75 from the joint. If the limit were assumed at Y=1.125
inches from the joint, two more points would move from the
valid to the invalid zone.
Figure 11 indicates no problem of extrapolating S22(m+b)
and S12(m) from the valid into the invalid zone and arriving at
the same value at the joint (at Y=0) as that obtained from
placing an SCL straight through the joint. Thus, for the current
example, the validity requirement that the in-plane shear stress
distribution over the SCL appear parabolic and its surface
stresses be close to zero is unnecessary. While the S12 and S22
stresses at the singularity diverge with refined mesh and are
unusable, this is not true for the linearized stresses over the
SCL through the joint.
This behavior of S22(m+b) and S12(m) is attributed to the
fact that these stress components arise from the meridional
force and moment and shear force that are necessary to satisfy
the continuity and equilibrium of the structure at the joint. This
is in line with the results obtained in the Force and Moment
Edit of section 4, confirming the expectation that finite element
stresses satisfy equilibrium of structural elements.
On the other hand, Figure 12 indicates a more uncertain
extrapolation of S33(m+b) and S11(m) to the joint.
5
0
-5
-10
-15
-20
0.00
0.25
0.50
0.75
1.00
1.25
1.50
20
S22(m+b)-valid
15
S22(m+b)-invalid
Stress, ksi
S12(m)-valid
S12(m)-invalid
10
8
7
6
5
Stress, ksi
-5
0.000 0.375 0.750 1.125 1.500 1.875 2.250 2.625 3.000
S33(m+b)-valid
S33(m+b)-invalid
S11(m)-valid
S11(m)-invalid
4
3
2
1
0
-1
0.000 0.375 0.750 1.125 1.500 1.875 2.250 2.625 3.000
5.5
0.2
0
-0.2
-0.4
-0.6
-0.8
0.75 in. from Joint
-1
-1.2
0.00
0.25
0.50
0.75
1.00
1.25
1.50
S11(m)
0.7
0.0
4.4
5.5.3
6 CONCLUSIONS
5.6
DISCUSSION
The two methods considered for calculating P+Q stresses
at a joint with a discontinuity were the extrapolation and the
direct calculation over an SCL through the joint. The former
involved calculation of stresses over many SCLs, a search for
the valid SCL zone, and then extrapolation of the stresses to the
joint. The latter involved the stress calculation over a single
SCL. By any measure, the latter appears preferable. The only
obstacle in its application was the uncertainty in the calculation
of the average through-thickness normal stress.
The question can be raised whether this obstacle is real. It
is not immediately obvious what role, if any, the throughthickness normal stress plays in the P+Q stress. It is not
primary because it does not equilibrate applied loading on
structural elements, and it is not secondary because it does not
participate in satisfying continuity of the shell/flat head
structure. It is neither P nor Q.
Even if this argument is not accepted and convincing
support for its inclusion can be formulated, the results of
subsection 5.5.2 showed minimal effect on the P+Q stress.
Most important, the FEA-calculated S11(m) gave a P+Q stress
20% less than that obtained from the extrapolated values,
rendering it unusable. It also showed that its extrapolated
magnitude, depending on the extrapolation technique, could
vary from zero to 3.4 % of the meridional membrane plus
bending stress, which is the main actor in ensuring continuity
and equilibrium of the joint.
Two other options could be considered for the throughthickness normal stress contribution to the P+Q stress at the
joint, which use its value in the valid SCL zone. Gordons
paper4 sets it equal to the applied pressure on the shell I.D.,
which for this case would be -1.0 ksi. The other is to use the
FEA-calculated -0.5 ksi for its average that is shown in Figure
12. Referring to Table 5, this would increase P+Q to 19.1 ksi
and 18.6 ksi, respectively.
As shown in subsection 5.5.2, the average stress is bound
to be greater than these values, which does not justify the
increased P+Q stress. Setting it to zero seems like a valid
compromise for using an SCL straight through the singularity.
4. For the axisymmetric shell/flat head example, an FEAcalculated average through-thickness normal stress over an
SCL containing a singularity is affected by a numerical
disturbance of the singularity. If used in the Tresca P+Q
stress intensity, it is shown to be unconservative.
5. The results suggest that neither the average throughthickness normal stress nor its bending component (i.e., the
linear part) should be included in the P+Q stress at a joint
with a discontinuity. If accepted, this could result in a simple,
conservative, and justifiable compromise for using an SCL
straight through the singularity.
6. The values of Tresca and Mises P+Q stresses, output by
the Abaqus/Standard11 version 6.7-1 CAE postprocessor
linearization routine should not be used on an SCL passing
through a singularity.
REFERENCES
1
10