You are on page 1of 12

Speed and accuracy evaluation of additive

manufacturing machines
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek
Faculty of Medical Engineering, University of Maribor, Maribor, Slovenia

Abstract
Purpose – The purpose of this paper is to establish a general method for achievable speed and accuracy evaluation of additive manufacturing (AM)
machines and an objective comparison among them.
Design/methodology/approach – First, a general schematic is defined that enables description of all currently available AM machines. This
schematic is used to define two influential factors describing certain parts’ properties regarding the machines’ yield during manufacturing. A test part is
defined, that will enable testing the influence of these factors on the speed and accuracy of manufacturing. A method for implementing and adapting
test parts is established for individual machine’s testing. This method was used to test four different machines that are predominantly used in Slovenia
at the moment.
Findings – Research has proven that the machine’s yield had a predominant influence on the achievable manufacturing speeds of all the tested
machines. In addition, the results have shown different ranges of achievable manufacturing speeds for individually tested machines. Test parts’
measurement results have shown comparable achievable accuracies for all the tested machines.
Research limitations/implications – Speed evaluation is based on a 2k factorial design that assumes the linearity among individual points of the
experiment. This design was chosen to keep the method as simple and quick as possible, in order to perform testing on those machines otherwise used
in industrial environments. Accuracy evaluation was limited by a rather small sample size of ten fabricated test parts per machine.
Practical implications – The presented evaluation method can be used on any existing or future type of AM machine, and their comparative
placement regarding achievable manufacturing speed and accuracy.
Originality/value – The presented method can be used to evaluate a machine regardless of the AM technology on which it is based.

Keywords Electric machines, Velocity, Accuracy, Manufacturing industries, Production engineering, Slovenia

Paper type Research paper

1. Introduction of a part being processed does not affect the speed of the
process, when the speed is described as a production volume
With an ever increasing number of different additive per a unit of time. Such definition of speed is more general as
manufacturing (AM) machines and technologies available, it compared to a number of layers per a unit of time, which is
has become more difficult for potential users to choose from adequately applicable only in some of the AM systems – 3D
among them. Accurate data about certain machine’s printers mostly. Modern high-end AM systems are
manufacturing speed and accuracy capabilities is necessary in predominantly used for manufacturing end-user parts or in
order to make a correct decision. Often, the only data available service bureaus, where the quantities of produced prototypes
is potentially subjective information provided by a machine’s are rather big. The working spaces of these machines are
manufacturer. This paper presents a method of evaluating usually filled with many, usually different parts with more or
various AM machines’ speed and accuracy capabilities, thus less complex geometry. The geometrical complexity, to some
enabling an objective and neutral comparison among them. extent determines a working space yield, with complex parts
The presented method is also presumed to be general enough to usually containing empty spaces that cannot be filled with
enable evaluation of future AM machines and technologies and other parts in the build job, in order to raise the building
their direct comparison with existing ones. volume’s yield. This means that the speed of the whole job
The presented method was developed under the hypothesis depends on three influencing factors:
that the speed of AM systems depends mostly on the effective 1 geometrical complexity of parts in the build job that
usage of the working space and that the height (Z-direction) influences the yield of the working space in part’s direct
vicinity (“density” of a part);
The current issue and full text archive of this journal is available at 2 yield of the whole working space of the AM system
www.emeraldinsight.com/1355-2546.htm (“density” and number of parts in the job); and
3 the technology used in the specific AM system.

Rapid Prototyping Journal


17/1 (2011) 64– 75 Received: 27 May 2009
q Emerald Group Publishing Limited [ISSN 1355-2546] Revised: 9 October 2009, 9 March 2010, 24 April 2010
[DOI 10.1108/13552541111098644] Accepted: 30 June 2010

64
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

The paper describes the speed and accuracy evaluation This definition is needed to describe the influence of AM
method, which uses the first two influencing factors to machines’ working space on its achievable overall speed.
determine the suitability of the AM technology (the third Every AM machine builds parts on a build tray over a
factor) for a specific job. The results of the experimental work known surface area. Vertical movement of the machine
presented in this paper, clearly show the differences among defines its workspace volume. Every machine also has some
AM technologies tested in terms of suitability for different kind of actuator unit (laser beam source, jet nozzle, etc.) that
“density” of building jobs. It is shown that part’s geometry actually forms and builds the part (Pahole et al., 2005). For a
and building space yield significantly influences the speed of positional description, a unified coordinate system is defined
some AM systems, making them suitable for high volume as follows. The origin is placed at the centre of that work tray
production. On the other hand, some systems’ speed remains side where the parts are to be built. Z-axis points in the
uninfluenced by the two factors, which make the technology direction of vertical movement. X and Y axes are parallel to
useful for developmental and prototyping tasks only. the tray’s edges. In the case of a machine with a cylindrical
Furthermore, the presented method enables a quick shaped workspace, X and Y axes orientations are
evaluation of accuracy of the AM machines that can be used unimportant, but usually correspond to the orientation of
in everyday praxis. Experience show that contemporary AM the machine’s software coordinate system (Figure 1).
systems’ reliability, in terms of acceptable (accurate) produced A common characteristic of all AM machines is that the
parts, can be very low, sometimes even at 50 percent. This is work tray moves vertically only as much as necessary during
mostly due to a lack of any quality control systems installed into individual part manufacturing. Consequently, two different
the AM machines by their producers. The demand will force the workspace volumes can be established. One is the maximal
producers to install such systems but at the moment users have workspace volume defined by the tray’s dimensions and the
to rely on their own skills and experiences. The accuracy maximal possible vertical movement of the individual
evaluation method presented in this paper was used to establish machine. Additionally, the actual workspace volume for a
the accuracy levels of tested machines, but the testing part is certain build can be defined by tray dimensions, and any
designed in a way that enables quick and effective accuracy necessary vertical movement of its manufacturing (Figure 2).
evaluation of every building job and is therefore used as a quality For the purpose of this research, when a fabricated part is
assurance tool. considered, the fabricated volume contains the volume of all parts
or components manufactured in an individual machine’s run.
2. State of the art Figure 1 General AM machine’s schematics
There are already several papers describing methods and
techniques of evaluating AM machines’ performance. Some
Actuator unit
methods are based on evaluating different AM machine’s Workspace
performances for building an already well defined end-user volume
part (Chuk and Thomson, 1998; Lan et al., 2005). This kind
of approach cannot guarantee a truly objective evaluation and
comparison among various AM machines, due to being
influenced (at least to some degree) by the requirements and
properties of the part. Other researchers focus on particular
type of AM machines (Han and Tseng, 2002; Han et al., z
2003). These methods also include some building parameter Part
y x
optimization procedures and cannot be used for direct
comparison to other machine types. Some papers focus on Coordinate system
accuracy issues and implement a complex test part that also
requires a complex inspection procedure (Mahesh et al., Build tray
2004). Researches focused on manufacturing speed are
usually meant for build time estimation of certain machine
type (Campbell et al., 2008).
Evaluation method in this paper is not meant for build time
Figure 2 Maximal and actual workspace volume
estimation on particular machine, but to establish the
achievable manufacturing speed range of tested machines.
Actual build tray

These established speed ranges can then be used for direct


movement

comparison among different machines. The rather simple and


quick accuracy evaluation is also included, making the
Part
Maximal build tray

presented method also a tool of quality assurance that can


movement

be used in every AM job as a mean of quality control.


ay
Build tr
3. General schematics of an AM machine
Despite a number of different AM technologies and machines
available, they all need to be built rather similarly, due to the
layered manufacturing principle (Drstvensek, 2004). This ay
enables a definition of general AM machine’s schematics that Build tr
can be used to present the properties of all existing machines.

65
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Fabricated volume does not include the support or carpet In order to describe the workspace yield influence, two
structure volume. The fabricated volume is contained inside influential factors are used and a new geometrical entity is
the machine’s working space defined by the size of the defined called “part’s envelope”. Part’s envelope is a block with
building tray and the height of the fabricated volume, and not edges (a, b, c) parallel to individual coordinate axes. The edges’
the maximal achievable height of the specific machine. By such lengths are equal to the maximal part’s dimensions in individual
definition, the comparison among different AM systems depends axes, according to the part’s orientation in the machine’s
only on the job height and not on potential AM system’s workspace. The first factor is defined as the ratio between
possibilities. the part’s volume and the volume of the part’s envelope. It is
named the “Volume Factor” and is used to describe the
influences of the part’s geometry on the workspace yield
4. Properties of AM machines
(Figure 3):
The research work presented in this paper focused on two
characteristics of individual machines, the speed and the Part’s volume
Volume ratio ¼
accuracy of manufacturing. The general agreement of AM Part’s envelope volume
experts that the production time of AM job depends mostly
The second factor is defined as the ratio between part’s envelope
on its height can be very misleading especially, when one
volume and part’s actual workspace volume. By definition, the
simplifies the meaning of time in the connection to
envelope edge c and the vertical dimension of the actual
production speed. Namely, this theory has roots in the rapid
workspace volume are always the same. Therefore, this factor
prototyping world where production is always one-off, not
can be equivalently described as a ratio between the envelope
only in terms of individualization but also in terms of number
bottom surface and the work tray surface (Figure 4). It is named
of parts per job. With emerging number of service providers
and even factories using AM systems in production processes
this theory became rather useless and not only misleading. Figure 3 Part’s envelope
This work is based on the hypothesis that the speed of AM
systems depends mostly on the effective usage of the working Part
space and that the height (Z-direction) of a part being envelope
processed does not affect the speed of the process when the
speed is described as a production volume per a unit of time.
Therefore, the speed in this work is described by the value of
the average manufacturing speed, calculated from the part’s
volume and the time taken to fabricate: c
 3 z
cm y
Average manufacturing speed b
h x
Machines
Part’s volumeðcm3 Þ coordinate a
¼ Part
Time of manufactureðhÞ directions

This method of evaluating machine’s speed is more universal Part’s volume


and reliable than methods that use a number of layers or Volume ratio =
Part’s envelope volume
millimetres of Z-direction per hour. Therefore, it enables a
more objective comparison of different AM machine types.
Figure 4 Tray ratio
The manufacturing accuracy is described by deviations
between the finished part’s dimensions and the dimensions of
its CAD model. Due to the nature of AM, it is appropriate to
separate deviations according to the coordinate directions of Build tray
the machine:
Part envelope
Deviation X ¼ dim X 2 dim X CAD
(bottom surface)
Deviation Y ¼ dim Y 2 dim Y CAD
Dim Y

Deviation Z ¼ dim Z 2 dim Z CAD


b part
It was established during the research, that the value for the
average manufacturing speed is not constant for an individual
machine, but depends largely on how efficiently the machine’s a
workspace is used during the manufacture of a certain part
(run). Consequentially, average manufacturing speed of a
certain AM machine cannot be described by the exact value of
cm3 per hour, because this value depends on how efficiently
the machine’s workspace is used during particular build.
Instead, a certain range (conditioned by the workspace yield Dim X
influence) of achievable average manufacturing speeds can be Envelope bottom surface
Tray ratio =
established. Build tray surface

66
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

the “Tray Ratio” and is used to describe the part’s relative size normally distanced above the test part’s faces for D/4. The
regarding the machine’s workspace: centres of the spheres lie on corresponding planes (Figure 6).
Finally, all spheres are subtracted from the main part resulting
Envelope bottom surface in six partial spherical segments formed on the part. For easier
Tray ratio ¼
Build tray surface manipulation of the test parts, the X and Y directional markings
are applied on the part’s upper surface (Figure 7).
According to their definitions, both factors can be calculated for
an individual part and their values are always between 0 and 1. 5.1 Application of test part for manufacturing speed
evaluation
5. Test part configuration The main purpose of this part of the experiment is to evaluate
the influence of previously defined volume and tray ratio
The main purpose of a test part is to enable the testing of
on the tested machine’s manufacturing speed. This experiment
certain AM machines’ speed and accuracy. Additionally, the
is based on a simple 2k factorial design principle requiring the
test part’s configuration must allow some adaptation to
test to be performed at combinations of high and low levels of
certain AM technologies and machines (Ficko et al., 2005).
both influential factors. The defined test part’s setup can be
There are already several existing test parts for AM machines.
used to conduct this experiment by simulating conditions at
Most parts are designed for accuracy evaluation only as, for
various combinations of low and high levels of both factors.
example, the “SLA user group part”. Some researchers use
The variation in volume ratio is achieved by changing the test
more complex test parts that require a relatively long measuring
part’s wall thickness to between the recommended limits of a/10
procedure for accuracy evaluation, and are usually adapted to a
and a/3, consequently designing two test parts with volume
specific machine (Stopp et al., 2008; Mahesh et al., 2004). Other
ratios of 0.25 (low level) and 0.69 (high level). The chosen test
researchers use simple parts, but conduct the testing of a single
part’s edge length for this experiment was 30 mm. The variation
machine on a large sample, with a purpose of optimizing
in tray ratio is carried out by placing different numbers of
manufacturing position and other parameters (Dimitrov et al.,
test parts on the work tray in accordance with its surface area.
2006). These facts make the existing test parts rather unsuitable
for the purpose of this research, which is to establish a relatively
simple and quick method for evaluating and comparing various Figure 6 Definition of spheres’ positions
AM machines. Therefore, it was decided, that a new test part
setup should be defined for this research, based on the following
requirements. The test part must enable accuracy as well as
manufacturing speed evaluation. Second, the measuring
procedure must be relatively quick and adapted for coordinate
measuring machine as well as a 3D optical scanner. Also, the test Z2 Y1
results must be as simple as possible, enabling the potential user Y2
to quickly compare different machines. z Z1
The basis of test part’s definition is a cube with an edge
length a. The cube’s edges are parallel to coordinate system’s X1
directions and one of the corners concourses with coordinate D/4
X2
origin. From this cube a smaller cube volume (with parallel
edges) is subtracted in order to obtain a part with resulting x y
Coordinate
wall thickness, in limits of a/10 and a/3 (Figure 5).
axes directions
Both cubes have a common corner at point (a, a, a).
Additionally, six spheres of diameter D ¼ a/3 are defined. The
spheres are arranged in three pairs. The distance direction Figure 7 Final test part’s configuration
between two spheres centres in individual pairs corresponds
exactly with the individual coordinate axis. The centre distance
length is equal to diameter D. The spheres are labelled X1 and
X2, Y1 and Y2 and Z1 and Z2. The letters stand for the coordinate
axis directions and the lower number marks the sphere in
individual pair that is closer to the coordinate system’s origin.
Y1
Additionally, the planes (XY, XZ and ZY) are defined as
Z2 Y2
Figure 5 Basic and subtraction cubes

Base cube Z1

Substraction cube
X1
z
X2
Wall thickness
y (between a/10 and a/3)
Coordinate
system x
a

67
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

In other words, a low level of tray ratio is simulated by placing Based on ANOVA analysis of the results, the following
sufficient number of test parts on the tray, so that a ratio value assessments can be made:
of approximately 0.1 is achieved. Accordingly, a higher level of .
the model of average manufacturing speed response is
tray ratio (0.9) is achieved by placing a greater number of test statistically significant;
parts. .
volume ratio factor (A) has a significant influence on the
According to the influential factor’s definition, it was model’s response;
predicted that the part’s Z-height does not significantly .
tray ratio factor (B) has a significant influence on the
influence the average manufacturing speed. In order to model’s response; and
confirm this, three repetitions of 22 experiments were planned .
there is also a significant interaction effect (AB) influence
by vertical assembly of identical test parts. By definition, parts on the model’s response.
on Figure 8 have all the same value for volume and tray ratio Basically, the statistical significance of presented average
and should be manufactured at the same average manufacturing speed model means that the variations of
manufacturing speed (on the same machine). In reality, response values due to the influence of both factors are much
there are some deviations due to carpet structure, support greater than the variations caused by different vertical heights
manufacture or possible actuators’ cleaning time. Therefore, of experiment repetition. This fact was proved for all tested
the purpose of this experiment’s repetition was to prove that machines as expected due to the manufacturing speed being
these deviations are statistically insignificant compared to measured in cm3 of fabricated products per hour.
changes in average manufacturing speed due to the influence The next step was the establishment of a regression model
of volume and tray ratio factors. that would describe the response of average manufacturing
The manufacturing time is needed, in order to calculate the speed regarding the values of volume and tray ratio factors:
average manufacturing speed of individual experiment run.
During this research, manufacturing times were acquired Average manufacturing speed ¼ b0 þ b1 · volume ratio
from the software packages of tested machines. It was þ b2 · tray ratio
presumed that the software-predicted manufacturing times þ b12 · volume ratio · tray ratio
are accurate enough for the purpose of this experiment
(Campbell et al., 2008). The first step was to prepare 12
Because the experiment introduced in this paper is based on
different build tray setups in the machine’s software and read
2k design, a potential concern is the assumption of linearity in
the estimated times manufacturing each setup (Figure 9).
the factor effects. For the purpose of this experiment the 2k
In the next step, the actual tray ratio values and average
design works quite well, even if the linearity assumption holds
manufacturing speeds of the test trays were calculated. Table I
only very approximately. Firstly, this experiment is not
presents the results of EDEN330 speed evaluation.
intended for the establishment of a response function that will
The result analysis is based on the analysis of variance
be able to predict the manufacturing speeds of individual
(ANOVA) method. In this research, statistical software was
parts exactly. Instead, the main goal is to establish the range
used in order to speed up the process. ANOVA of EDEN330 and evaluate the limits of average manufacturing speed that
average manufacturing speed evaluation results are shown in are achievable by the tested machine. Second, by adding the
Table II. All tested machines were evaluated using the same interaction term to the main effects, this model is capable of
method, but EDEN330 results are used as an example, to representing some curvature in the response function. This
explain the statistical method and significance of the curvature results from the twisting of the plane induced by
assessment. interaction terms (Montgomery, 2001).

5.2 Application of a test part for manufacturing


accuracy evaluation
Figure 8 Test part’s assembly for experiment repetition
A test part consists of three pairs of partial spherical segments
Test 3 that are used for manufacturing accuracy evaluation.
Segments have to be measured with a 3D measuring device,
either a coordinate measuring machine or a high-end optical
scanner being obvious choices (Hodolic et al., 2005). The
Test 2
measurement results are the coordinates of those sphere
centres defined by the segments. Accuracy evaluation is then
based on comparing the distances between the paired spheres’
Test 1 centres of a finished part and its CAD model. This method
for accuracy evaluation was established with the intent of
reducing the influence of surface roughness as much as
possible.
Regardless of which measuring method is used, the results
are always analyzed as follows. Three vectors are defined,
based on coordinates of sphere centres. Vectors are used for to
following reasons. First, this method enables the calculation
of angles between vectors as additional evaluation of
manufacturing accuracy and, secondly, by using this method
the position and orientation of the measurement coordinate
system (Acko, 2007) is unimportant (Figure 10).

68
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Figure 9 Test trays’ setup

Test tray 1 Test tray 2 Test tray 3

Test part: Type 1 Test part: Type 1 Test part: Type 1


Volume ratio: Low Volume ratio: High Volume ratio: Low
Tray ratio: Low Tray ratio: Low Tray ratio: High
Test tray 4 Test tray 5 Test tray 6

Test part: Type 1 Test part: Type 2 Test part: Type 2


Volume ratio: High Volume ratio: Low Volume ratio: High
Tray ratio: High Tray ratio: Low Tray ratio: Low
Test tray 7 Test tray 8 Test tray 9

Test part: Type 2 Test part: Type 2 Test part: Type 3


Volume ratio: Low Volume ratio: High Volume ratio: Low
Tray ratio: High Tray ratio: High Tray ratio: Low
Test tray 10 Test tray 11 Test tray 12

Test part: Type 3 Test part: Type 3 Test part: Type 3


Volume ratio: High Volume ratio: Low Volume ratio: High
Tray ratio: Low Tray ratio: High Tray ratio: High

Each vector points from a sphere centre in an individual pair qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi


that is closer to the test part’s inner corner to the sphere y¼ y2x þ y2y þ y2z
centre that is further away. The vector components are qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
calculated from centres’ coordinates by the following z¼ z2x þ z2y þ z2z
equations:
x ¼ ðxx ; xy ; xz Þ ¼ ððXX2 2 XX1 Þ; ðYX2 2 YX1 Þ; ðZX2 2 ZX1 ÞÞ As defined by the test part’s setup, nominal (CAD) distances
  between paired spheres’ centres are equal to one third of the
y ¼ yx ; yy ; yz ¼ ð ðXY2 2 XY1 Þ; ðYY2 2 YY1 Þ; ðZY2 2 ZY1 Þ Þ test part’s edge or sphere’s diameter. Deviations between
 
z ¼ zx ; zy ; zz ¼ ð ðXZ2 2 XZ1 Þ; ðYZ2 2 YZ1 Þ; ðZZ2 2 ZZ1 Þ Þ nominal and measured distances are used for dimensional
accuracy evaluation:
The next step is to calculate each vector’s length:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Dimensional deviation x ¼ x 2 D
x ¼ x2x þ x2y þ x2z Dimensional deviation y ¼ y 2 D

69
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Table I Results of EDEN330 average manufacturing speed evaluation


Tray Part type Number of parts Build time (h) Volume ratio Tray ratio Average manufacturing speed (cm3/h)
1 1 11 6.40 0.25 0.10 11.70
2 1 11 6.40 0.69 0.10 31.80
3 1 81 17.55 0.25 0.74 31.43
4 1 81 17.55 0.69 0.74 85.38
5 2 11 12.38 0.25 0.10 12.10
6 2 11 12.38 0.69 0.10 32.87
7 2 81 34.28 0.25 0.74 32.18
8 2 81 34.28 0.69 0.74 87.42
9 3 11 18.35 0.25 0.10 12.25
10 3 11 18.35 0.69 0.10 33.27
11 3 81 51.00 0.25 0.74 32.45
12 3 81 51.00 0.69 0.74 88.15

Table II ANOVA of EDEN330 test results


Response 1 Average fabrication speed
ANOVA for selected factorial model
Analysis of variance table [partial sum of squares – type III]
Source Sum of squares Df Mean square F-value p-value Prob > F
Model 9,314.67 3 3,104.89 4,141.24 ,0.0001 Significant
A-volume ratio 4,285.76 1 4,285.76 5,716.26 ,0.0001
B-tray ratio 4,144.83 1 4,144.33 5,528.28 ,0.0001
AB 884.08 1 884.08 1,179.17 ,0.0001
Pure error 6.00 8 0.75
Cor total 9,320.67 11
Notes: The model F-value of 4,141.24 implies the model is significant; there is only a 0.01 percent chance that a “Model F-Value” this large could occur due to
noise; values of “Prob . F” , 0.0500 indicate model terms are significant; in this case A, B and AB are significant model terms; values .0.1000 indicate the
model terms are not significant

0 1
Figure 10 Definition of vectors based on virtual sphere centres,
defined by segments B xx · yx þ xy · yy þ xz · yz C
xy ¼ arcoss@qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA
x2x þ x2y þ x2z · y2x þ y2y þ y2z
0 1
B xx · zx þ xy · zy þ xz · zz C
xz ¼ arcoss@qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA
xx þ x2y þ x2z · z2x þ z2y þ z2z
2

0 1
B yx · zx þ yy · zy þ yz · zz C
Vector z yz ¼ arcoss@qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA
Vector y 2 2 2
yx þ yy þ yz · zx þ zy þ zz 2 2 2

Angular deviation xy ð+ Þ ¼ xyð+ Þ 2 90+


z
Angular deviation xzð+ Þ ¼ xzð+ Þ 2 90+
y Angular deviation yzð+ Þ ¼ yzð+ Þ 2 90+
x
Vector x Coordinate directions

6. Research results
Four different machines currently installed in Slovenia were
tested with the presented method. These machines also
Dimensional deviation z ¼ z 2 D represent four different AM technologies (Wohlers, 2008;
Table III).
Additionally, angular manufacturing accuracy is evaluated by Testing was carried out on two test parts with minimal
calculating the angles between vectors, and comparing them and maximal possible volume ratios, as defined by their
to angles defined by the test part’s setup: setup. Manufacturing time data was acquired from the

70
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Table III Tested AM machines


Machine AM technology Manufacturer Installed User
EDEN330 PolyJet Objet Geometries 2003 UNI Maribor
SLA 3500 Stereolitografy 3D Systems 1998 Gorenje
EOSINT P385 Laser sint ering EOS 2006 RTCZ Hrastnik
Prodigy Plus Fused deposition modelling Stratasys 2005 Status

machines’ software. Accuracy evaluation was based on ten manufacturing speed. There is also a significant positive
manufactured test parts (five for each volume ratio) on each interaction of tray and volume ratio terms resulting in a
machine. The test parts’ tray layout and build parameters used twisting of the response surface. This fact is clearly presented by
for manufacturing were always optimized for best accuracy the curvatures of borders between interval areas (Figure 12). It is
results, as defined in an individual machine’s software. shown that volume ratio (part’s density) and tray ratio affect the
The test parts were measured using a Zeiss UMC 850 building speed and that it especially depends on the effective use
coordinate measuring machine (Figure 11). of the whole building tray of the machine. This is due to the
printing head technology, which is very productive as compared
6.1 EDEN 330 results to “scanning” (laser based systems) or even “plotting” fused
Figure 12 presents the average manufacturing speed response deposition modelling (FDM systems) technologies.
surface as a result of EDEN330 machine testing. The significant Accuracy evaluation was based on measuring ten finished
influence of tray and volume ratio on response can be seen, test parts. Five parts were designed with 0.25 volume ratio
resulting in a range from 10 to 120 cm3/h of achievable average and five parts with 0.69. All parts were designed with 30 mm
basic edges resulting in 10 mm nominal sphere centre
Figure 11 Test part’s coordinate measuring distances. The same part setup was used throughout the
research presented in this paper (Figures 13-16).
Test results show much smaller deviations in the X than the Y
and Z axes. There is no notable difference between the
dimensional deviations for higher and lower volume ratio parts.
However, there are slightly larger angular deviations for higher
volume ratio parts that can be attributed to deformations caused
by material shrinkage (Brajlih et al., 2006).

Figure 13 Dimensional deviations (EDEN330 0.25 volume ratio)


Polyjet part 0.25
10.10
10.08
10.06
10.04
10.02 X actual
(mm)

10.00 Y actual
9.98 Z actual
9.96 Nominal
9.94
Figure 12 EDEN330 Average manufacturing speed response surface 9.92
9.90
1 2 3 4 5
Average manufacturing speed (EDEN 330)

(cm3/h)
140 120-140 Figure 14 Dimensional deviations (EDEN330 0.69 volume ratio)
120 100-120
Polyjet part 0.69
100 80-100 10.10
60-80 10.08
80 10.06
40-60
60 10.04
20-40
10.02 X actual
40
(mm)

0-20 10.00 Y actual


20 0.7 9.98 Z actual
io

0 0.4 9.96 Nominal


rad

0.1 0.2 9.94


y

0.3 0.4 0.5


Tra

0.6 0.7 0.1 9.92


0.8 0.9 9.90
Volume radio 1 2 3 4 5

71
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Figure 15 Angular deviations (EDEN330 0.25 volume ratio) performance with relatively complex parts, filling the whole
working space of the machine.
Polyjet part 0.25
92.00 Accuracy evaluation results show significantly larger
deviations of test parts in X-axis then the other two. This
91.50
can be attributed to the fact that the tested machine has the
91.00 longest running period (approximately ten years) from among
90.50 the machines evaluated during the research and that the
Angle (°)

XY actual
90.00 XZ actual machine’s scanning system has been worn out. It can be
89.50 YZ actual presumed, that the testing of a new SLA machine would yield
Nominal similar deviations to the presented Y and Z results in all
89.00
machine’s axes (Figures 18-21).
88.50
88.00 6.3 EOSINT P385 results
1 2 3 4 5
Test results show much higher average manufacturing speed
values achievable by the EOSINT P385 machine than by the
other tested machines. Interaction of terms has a similar positive
Figure 16 Angular deviations (EDEN330 0.69 volume ratio) Figure 18 Dimensional deviations (SLA 3500 0.25 volume ratio)
Polyjet part 0.69
92.00 SLA part 0.25
10.10
91.50 10.08
91.00 10.06
90.50 10.04
Angle (°)

XY actual
90.00 XZ actual (mm) 10.02 X actual
YZ actual 10.00 Y actual
89.50
Nominal 9.98 Z actual
89.00
9.96 Nominal
88.50
9.94
88.00 9.92
1 2 3 4 5
9.90
1 2 3 4 5

6.2 SLA 3500 results


SLA 3500 results show a much narrower range of achievable Figure 19 Dimensional deviations (SLA 3500 0.69 volume ratio)
average manufacturing speeds than with EDEN 330. An
SLA part 0.69
interesting fact is the negative interaction term of volume and 10.10
tray ratio. This is presented by different curvatures of borders 10.08
(compared to EDEN 330 results) on a response surface 10.06
(Figure 17). Tray ratio for the SLA 3500 has no or little 10.04
10.02 X actual
influence on achievable manufacturing speed, especially at
(mm)

10.00 Y actual
very dense parts, i.e. parts with high volume ratio. On the
9.98 Z actual
other hand volume ratio of parts increases the manufacturing 9.96 Nominal
speed but the effect is much smaller as compared to EDEN 9.94
330. The results show that the SLA 3500 achieves the best 9.92
9.90
1 2 3 4 5
Figure 17 SLA 3500 average manufacturing speed response surface
Average manufacturing speed (SLA 3500) Figure 20 Angular deviations (SLA 3500 0.25 volume ratio)
(cm3/h) SLA part 0.29
50 45-50 91.00
45
40-45 90.50
40
35 35-40
90.00
30 30-35
25 25-30 89.50
XY actual
Angle (°)

20 20-25 89.00 XZ actual


15
15-20 88.50 YZ actual
10 0.7
5 10-15 Nominal
88.00
5-10
io

0 0.4
rat

0.1 0.2 0-5 87.50


ay

0.3 0.4 0.5


Tr

0.6 0.7 0.1


0.8 0.9 87.00
Volume ratio 1 2 3 4 5

72
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Figure 21 Angular deviations (SLA 3500 0.69 volume ratio) Figure 24 Dimensional deviations (EOSINT P385 0.69 volume ratio)
SLA part 0.69 LS part 0.69
91.00 10.10
10.08
90.50
10.06
90.00
10.04
89.50 10.02 X actual
XY actual

(mm)
Angle (°)

89.00 XZ actual 10.00 Y actual


YZ actual 9.98 Z actual
88.50
Nominal 9.96 Nominal
88.00
9.94
87.50 9.92
87.00 9.90
1 2 3 4 5 1 2 3 4 5

effect on response as with the EDEN330 results. The tray ratio


Figure 25 Angular deviations (EOSINT P385 0.25 volume ratio)
significantly influences the manufacturing speed, whereas the
volume ratio is not so important. The fact makes this machine a LS part 0.25
perfect choice for service providers or manufacturers of end 92.00
parts that would always fill the machine to the top. The machine 91.50
is not suitable for development departments requiring low 91.00
volume or individual production of parts (Figures 22-26).
In terms of accuracy, EOSINT P385 results show somewhat 90.50
Angle (°)
XY actual
larger dimensional deviations in comparison to EDEN330, but 90.00 XZ actual
still well in within currently-accepted levels for an AM system 89.50 YZ actual
based on laser sintering technology (Dolinsek, 2004). Nominal
89.00

Figure 22 EOSINT P385 average manufacturing speed response 88.50

surface 88.00
1 2 3 4 5
Average manufacturing speed (EOSINT P385)

Figure 26 Angular deviations (EOSINT P385 0.69 volume ratio)


500 (cm3/h)
450 450-500 LS part 0.69
92.00
400 400-450
350 350-400 91.50
300 300-350 91.00
250 250-300
90.50
Angle (°)

200 200-250 XY actual


150 90.00 XZ actual
150-200
100 0.9 YZ actual
100-150 89.50
50 0.7 Nominal
0.5 50-100 89.00
io

0
rat

0.1 0.2 0.3 0-50


ay

0.3 0.4 88.50


Tr

0.5 0.6 0.1


0.7 0.8 88.00
0.9 1 2 3 4 5
Volume ratio

Figure 23 Dimensional deviations (EOSINT P385 0.25 volume ratio) 6.4 Prodigy plus results
By observing the response surface, it can be noted that prodigy
LS part 0.25
10.10 plus achieves the lowest average manufacturing speed compared
10.08 to the other tested machines (Figures 27-31). There is also no
10.06 interaction term effect (either positive or negative), resulting in
10.04 straight borders between average speed interval borders on a
10.02 X actual
response surface. It can be seen that tray ratio makes little or no
(mm)

10.00 Y actual
influence on the manufacturing speed. Volume ratio makes a
9.98 Z actual
difference but in a very low extent. This machine is predominantly
9.96 Nominal meant for developmental tasks, for manufacturing of prototypes
9.94 or concept models, where production of several parts at once
9.92 means no advantage to the process.
9.90
Prodigy plus accuracy evaluation shows somewhat larger
1 2 3 4 5 deviations than for the rest of the tested machines. Especially

73
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

Figure 27 Prodigy plus average manufacturing speed response surface Figure 31 Angular deviations (prodigy plus 0.69 volume ratio)
Average manufacturing speed (prodigy plus) FDM part 0.69
92.00
91.50
10 (cm3/h) 91.00
8 8-10
90.50
6 6-8
90.00

Angle (°)
0.9 XY actual
4 4-6
0.7 89.50 XZ actual
2 2-4
0.5 89.00 YZ actual

io
0

rat
0.3 0-2 88.50 Nominal
0.1 0.2

y
0.3 0.4

Tra
0.5 0.6 0.1 88.00
0.7 0.8
Volume ratio 0.9 87.50
87.00
1 2 3 4 5

Figure 28 Dimensional deviations (prodigy plus 0.25 volume ratio) interesting is a change in dimensional deviation behaviour
between parts with lower and higher volume ratios. This can
FDM part 0.25 be attributed to the shrinkage condition change with a higher
10.10
10.08
mass of melted material deposited.
10.06
10.04 7. Summary
10.02 X actual The following diagrams compare the average manufacturing
(mm)

10.00 Y actual speeds for the tested machines. The results are separated into
9.98 Z actual low and high volume ratio levels (Figures 32 and 33).
9.96 Nominal The EOSINT P385 machine achieves the highest average
9.94 manufacturing speeds, regardless of both ratio values. The
9.92 comparison between EDEN330 and SLA 3500 test results is
9.90
1 2 3 4 5 Figure 32 Average manufacturing speed comparison (0.25 volume
ratio)

Figure 29 Dimensional deviations (prodigy plus 0.69 volume ratio) Average manufacturing speed
(0.25 volume ratio)
160
FDM part 0.69
10.10 140
Average manufacturing

10.08
120
10.06
speed (cm3/h)

100
10.04 EDEN 330
80
10.02 X actual SLA 3500
(mm)

10.00 60
Y actual
Prodigy plus
9.98 Z actual 40
EOSINT P385
9.96 Nominal 20
9.94 0
0 0.2 0.4 0.6 0.8 1
9.92
Tray ratio
9.90
1 2 3 4 5

Figure 33 Average manufacturing speed comparison (0.69 volume


ratio)
Figure 30 Angular deviations (prodigy plus 0.25 volume ratio)
Average manufacturing speed
Polyjet part 0.25 (0.69 volume ratio)
92.00 350
91.50 300
Average manufacturing

91.00
250
90.50
speed (cm3/h)

90.00 200
Angle (°)

XY actual EDEN 330


89.50 XZ actual 150 SLA 3500
89.00 YZ actual Prodigy plus
100
88.50 Nominal EOSINT P385
88.00 50
87.50 0
0 0.2 0.4 0.6 0.8 1
87.00
1 2 3 4 5 Tray ratio

74
Speed and accuracy evaluation of AM machines Rapid Prototyping Journal
Tomaz Brajlih, Bogdan Valentan, Joze Balic and Igor Drstvensek Volume 17 · Number 1 · 2011 · 64 –75

more complicated. When fabricating thin-walled parts (low Brajlih, T., Drstvensek, I., Kovacic, M. and Balic, J. (2006),
volume ratio) and a lesser portion of tray surface covered (low tray “Optimizing scale factors of the polyJet rapid prototyping
ratio), SLA 3500 is able to achieve higher average manufacturing procedure by genetic programming”, Journal of
speeds than EDEN330. When considering manufacturing of Achievements in Materials and Manufacturing Engineering,
parts with high volume ratio values, EDEN330 achieves higher Vol. 16 Nos 1/2, pp. 101-6.
manufacturing speeds regardless of the part’s tray ratio value. The Campbell, I., Combrinck, J., de Beer, D. and Barnard, L. (2008),
prodigy plus machine achieves the lowest average manufacturing “Stereolithography build time estimation based on volumetric
speeds as expected, considering that it is within a lower price range calculations”, Rapid Prototyping Journal, Vol. 14 No. 5,
than the rest of the tested machines. pp. 271-9.
Due to the relatively small sample sizes, it is more difficult Chuk, R.N. and Thomson, J.V. (1998), “Rapid prototyping
to objectively compare machines’ manufacturing accuracy. techniques for wind tunnel model fabrication”, Rapid
Test parts’ deviations are rather similar on all machines, with Prototyping Journal, Vol. 4 No. 4, pp. 185-96.
the exception of slightly bigger deviations for the prodigy plus Dimitrov, D., van Wijck, W., Schreve, K. and de Beer, N.
machine. It can be assumed, that the larger X-axis deviations (2006), “Investigating the achievable accuracy of three
for the SLA 3500 machine are due to wear, and that the dimensional printing”, Rapid Prototyping Journal, Vol. 12
testing of a new SLA machine would yield much better No. 1, pp. 42-52.
manufacturing accuracy results. Dolinsek, S. (2004), “Investigation of direct metal laser
The presented results show that different machines are able to sintering process”, Journal of Mechanical Engineering, Vol. 50
achieve different values and ranges of average manufacturing No. 4, pp. 229-38.
speeds. It confirms that both ratios used to describe workspace Drstvensek, I. (2004), Layered Technologies, Faculty of
yield significantly influence AM speed. The defined test part Mechanical Engineering, Maribor, Slovenia.
and the presented method of research show a capability of Ficko, M., Drstvensek, I., Brezocnik, M. and Balic, J. (2005),
recognizing the differences in speeds and accuracies of various “Prediction of total manufacturing costs for stamping tool
AM machines. This method presents a universal, relatively on the basis of CAD-model of finished product”, Journal of
simple and highly adaptive approach to the problem of Materials Processing Technology, Vol. 164, pp. 1327-35.
objectively evaluating speed and accuracy for AM machines. Han, A.W.L. and Tseng, Y.C. (2002), “A robust process
As such, it enables an objective comparison among all AM optimization for a powder type rapid prototyper”, Rapid
machines currently present on the market, regardless on which Prototyping Journal, Vol. 8 No. 3, pp. 180-9.
AM technology they are based on. Previous research mostly Han, W., Jafari, A.M. and Seyed, K. (2003), “Process
dealt with the accuracy evaluation of specific machine types, speeding up via deposition planning in fused deposition-
where machines’ speeds were not taken into account. The based layered manufacturing processes”, Rapid Prototyping
presented method provides a solution to fill the current research Journal, Vol. 9 No. 4, pp. 212-8.
gap and can also be used for decision making and evaluation in Hodolic, J., Budak, I. and Stević, M. (2005), “3D-digitization
everyday practice. For example, previous research has already in production engineering”, Proceedings – 9th International
established that in general LS based machines build parts faster Conference Mechanical Engineering, Bratislava, Slovakia.
than FDM based machines (Lan et al., 2005). This fact was also Lan, H., Ding, Y. and Hong, J. (2005), “Decision support
confirmed by research presented in this paper. However, the system for rapid prototyping process selection through
specific decision making, regarding specific machines and integration of fuzzy synthetic evaluation and an expert
predicted workspace yield with which the machine will operate, system”, International Journal of Production Research, Vol. 43
can only be made after the machines are evaluated by the No. 1, pp. 169-94.
method presented in this paper (compare EDEN330 v. Mahesh, M., Wong, Y.S., Fuh, J.Y.H. and Loh, H.T. (2004),
SLA3500 results). Finally, the method is presumed to be “Benchmarking for comparative evaluation of RP systems and
general enough to enable the evaluation of the AM machines processes”, Rapid Prototyping Journal, Vol. 10 No. 2, pp. 132-5.
that will be launched onto the market in future and their direct Montgomery, D.C. (2001), Design and Analysis of Experiments,
comparison to already existing ones. 5th ed., Wiley, New York, NY.
Future research should be focused on evaluating as many Pahole, I., Drstvensek, I., Ficko, M. and Balic, J. (2005),
different AM machines as possible by the presented method. This “Rapid prototyping processes give new possibilities to
will enable their placement among already evaluated machines numerical copying techniques”, Journal of Materials
regarding their achievable speed and accuracy. Also, some Processing Technology, Vol. 164, pp. 1416-22.
additional criteria (parts’ mechanical properties, surface quality, Stopp, S., Wolff, T., Irlinger, F. and Lueth, T. (2008), “A new
etc.) or even some economical aspects (machine costs, available method for printer calibration and contour accuracy
material costs, etc.) could be added to evaluation. This additional manufacturing with 3D-print technology”, Rapid
data could prove very useful for potential users but present an Prototyping Journal, Vol. 14 No. 3, pp. 167-72.
entirely different subject beyond the scope of this paper. Wohlers, T. (2008), State of the Industry, Additive Layered
Manufacturing, Faculty of Mechanical Engineering,
References Maribor, Slovenia.

Acko, B. (2007), “Calibration of measuring instruments on a


Corresponding author
coordinate measuring machine”, Advances in Production
Engineering and Management, Vol. 2 No. 4, pp. 127-34. Igor Drstvensek can be contacted at: drsti@uni-mb.si

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com


Or visit our web site for further details: www.emeraldinsight.com/reprints

75

You might also like