You are on page 1of 12

Error compensation for fused deposition

modeling (FDM) machine by correcting


slice files
Kun Tong, Sanjay Joshi and E. Amine Lehtihet
Department of Industrial and Manufacturing Engineering, The Pennsylvania State University, University Park, Pennsylvania, USA

Abstract
Purpose The purpose of this research is to extend the previous approach to software error compensation to fused deposition modeling (FDM)
machines and explores the approach to apply compensation by correcting slice files.
Design/methodology/approach In addition to applying the stereolithography (STL) file-based compensation method from earlier research; a new
approach using the slice file format to apply compensation is presented. Under this approach, the confounded effects of all errors in a FDM machine are
mapped into a virtual parametric machine error model. A 3D artifact is built on the FDM machine and differences between its actual and nominal
dimensions are used to estimate the coefficients of the error functions. A slice file compensation method is developed and tested on two types of parts
as a means for further improving the error compensation for feature form error improvement. STL file compensation is also applied to a specific FDM
3000 machine and the results are compared with those of a specific SLA 250 machine.
Findings The two compensation methods are compared. Although, the slice file compensation method theoretically allows higher compensation
resolution, the actual machine control resolution of the FDM machine can be a limitation which makes the difference between STL compensation and
slice file compensation indistinguishable. However, as the control resolution is increased, this method will make it possible to provide a higher degree of
compensation.
Originality/value Compensation method applied to slice file format is developed for FDM machines and its limitations are explored. Based on the
experimental study, dimensional accuracy of parts is considerably improved by the software error compensation approach.

Keywords Rapid prototypes, Software prototyping, Error analysis

Paper type Research paper

Introduction technique improvement, support structure generation, tool


path planning, process parameters tuning and build
Rapid prototyping (RP) machines are widely used in many orientation optimization. Good process planning can
industries to assist in the design, manufacture and improve machine accuracy to some extent, but with the
commercialization of a product (Dutta et al., 2001; Wohlers, current RP technology, even the best tuned system will still
1995). Dimensional accuracy of these processes is still an produce parts with considerable systematic errors. Error
obstacle preventing this technology from moving towards a compensation can be used to further reduce errors.
primary production process (Kruth et al., 1998). Very little work has been done on error compensation for
There are two general approaches which can be used to RP machines. Philpott and Green (1995) presented an
improve the accuracy of a process such as RP. The first iterative error compensation strategy for one RP process. This
approach attacks the problem through Error avoidance and approach compensates for cumulative error build-up during
seeks to eliminate the source of an error. The second replication without knowledge of the error creating
approach strives to cancel the effect of an error without mechanism. However, since no machine error models were
removing the error source and is known as Error built in this strategy, the iterative process needs to be repeated
compensation (Hocken, 1995; Sartori and Zhang, 1995; for every new part. Nee et al. (2001) constructed a correction
Zhang et al., 1985). Most past research on RP accuracy table with n n lattice points to improve the stereolithography
improvement falls within the error avoidance category and has (STL) process accuracy. This correction table is calculated
focused on different aspects of the RP process (Arni and according to the configuration of the Galvano-mirror system
Gupta, 1999; Cheng et al., 1995; Gargiulo and Belfiore, 1991; and is used mainly to compensate for the error in positioning
Kulkarni and Dutta, 1996; Kumar and Dutta, 1997; Lin et al., the laser beam on the platform.
2001; Liu et al., 1998a, b; Lynn and Rosen, 1999, 2000; Nee Tong et al. (2004) presented a more comprehensive method
et al., 2001; Onuh and Hon, 2001; Wah et al., 2001; for RP machines error evaluation and error compensation
Zhou et al., 2000). These include data file correction, slicing using virtual parametric errors, inspired by the technique
developed over the years for parametric evaluation of CMM
The current issue and full text archive of this journal is available at errors (Belforte et al., 1987; Elshennawy, 1987; Ferreira
www.emeraldinsight.com/1355-2546.htm and Liu, 1986; Hocken, 1995; Kunzman et al., 1990;
Mou and Liu, 1995; Ni, 1997; Sartori and Zhang, 1995;

Rapid Prototyping Journal


14/1 (2008) 414 Received: 1 November 2005
q Emerald Group Publishing Limited [ISSN 1355-2546] Revised: 1 September 2007
[DOI 10.1108/13552540810841517] Accepted: 1 October 2007

4
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Trapet and Waldele, 1991; Zhang et al., 1985, 1988). A typical research. First, the error compensation method used in
CMM has three linear carriages, designed to move the SLA case (Tong et al., 2004) is duplicated on this FDM
independently along the X, Y, or Z axis. However, each machine and used for comparison; then, a compensation
carriage has six degrees of freedom and hardware method based on correcting slice files is developed and
construction usually cannot completely suppress the tested. Machine error model and software error compensation
undesired translational or rotational movements. As a of the SLA machine were fully described in a previous
result, each axis has three translational errors and three publication (Tong et al., 2004) and will not be repeated here.
rotational errors. A three-axis machine has a total of 18 The rest of this paper is organized as follows: the next section
parametric errors. Assuming rigid body kinematics, the presents the machine error model of a FDM 3000 system and
volumetric error of the machine can be written as a function its parametric error functions derived from the 3D artifact.
of the 18 parametric errors. In some studies (Elshennawy, Followed by the section that compares the compensation
1987; Hocken, 1995; Sartori and Zhang, 1995; Zhang et al., results between a specific FDM 3000 machine and a specific
1985), non-orthogonal coordinate systems are assumed and SLA 250 machine from the previous research and presents the
squareness errors between each pair of axes are treated slice file compensation method. The last section draws
separately, which makes a total of 21 parametric errors. The conclusions from the results of this research.
differences between these two methods and their equivalence
in modeling the machine errors were discussed by Kruth et al.
(1994). In this research, squareness errors are included in the Mathematical error model of FDM machine
straightness errors of the axes and thus 18 parametric errors The FDM system builds a prototype by depositing melted
are used to build the error model. These 18 errors and the plastic. A plastic filament is unwound from a spool and is
notation used are summarized in Table I. delivered to an extrusion head (Figure 1). The extrusion head
In the RP processes, the error budget is quite large and contains the nozzle for delivery of the melted plastic, heating
includes error sources other than axes motion errors (Fadel elements, and incorporates mechanisms which allow the flow
and Kirschman, 1996; Jurrens, 1999; Lee et al., 1995). In this of the melted plastic to be turned on and off. The nozzle is
research, the approach is to map all these confounded errors mounted in the head which can travel in one horizontal XY
into 18 virtual parametric errors without explicit plane. As the nozzle is moved over the table driven by the
consideration of each specific error source. Unlike the case required geometry, it deposits a thin bead of extruded plastic
of CMMs and machine tools, the 18 virtual parametric to form each layer. The plastic hardens immediately after
errors of a RP machine cannot be measured directly on the being extruded from the nozzle and bonds to the layer below.
machine, and the artifact method is thus used as an indirect The entire system is contained within a chamber that is held
measurement of these errors. That is, the RP machine is used at a certain temperature.
to manufacture a specially designed artifact, and a CMM is
used as a master scale to measure artifact characteristics.
FDM machine error model
Measurements are then used to infer the parametric errors of
In general, three axes machines can be classified according to
the RP machine, and an error compensation routine can then
the motion of the tool tip and work piece as XYFZ, XYZF,
be applied to the build files of any part prior to processing on
XFYZ or FYXZ. In this representation, F designates the fixed
the machine. Tong et al. (2004) applied compensation to a
machine foundation while letters appearing to the right of F
SLA 250 machine and demonstrated considerable
indicate probe motion and letters appearing to the left of F
improvement in dimensional accuracy of parts produced
indicate work piece motion (Elshennawy, 1987). In the FDM
using this compensation approach.
machine, the table can move up and down along the Z axis,
Since, the slice file format used for SLA machines is
while the tool tip, or the extrusion nozzle, can move in both Y
proprietary, compensation was implemented by correcting
and X directions. In addition, the X axis is carried by the Y
STL files. The main purpose of this paper is to extend the
previous research and investigate a compensation method carriage. According to these features, the FDM machine
which corrects slice files. Fused deposition modeling (FDM) Figure 1 The FDM machine
machines use the SSL file format for its slice files. This format
is readily accessible and was used to investigate software
Plastic
compensation on a FDM 3000 machine available for the Filament
Table I 18 Parametric errors of a three axes system Support
Axis X Y Z Filament
Heated Head
Scale error d x (x ) dy(y) dz(z) Moving in X, Y
Straightness error dy(x), dz(x) dx(y), dz(y) dx(z), dy(z)
Rotational error
Roll 1 x (x ) 1y(y) 1z(z) Nozzle
Pitch 1y(x) 1z(y) 1 x (z )
Yaw 1z(x) 1 x (y ) 1y(z)
Notes: dj (i ) is the undesired translational error of axis i in the j direction. Support Part
1j (i ) is the rotational error of axis i arising from the undesired rotational Structure
movement of the carriage around axis j. Both dj (i ) and 1j (i ) are functions of
the position of the carriage along axis i
Table, Moving Vertically

5
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

corresponds to a ZFYX type machine. The kinematic axes the model. Each parametric error function in equation (3)
chain vector diagram in Figure 2 is used to build the error can be represented by a 3rd order polynomial function. For
model for the FDM machine, where: example:
2 3 2 3 2 3    
x dx x dx y dx z 3 2 1 3
dx x xtx1 x xtx2 x xtx3 5x 2 3x
6 7 6 7 6 7 2 2
6 7 6 7 6 7     4
X 6 dy x 7 Y 6 y dy y 7 Z 6 dy z 7 3 2 1
4 5 4 5 4 5 1z x xrz1 x xrz2 x xrz3 5x3 2 3x
dz x dz y z dz z 2 2
2 3 2 3 2 3 1
xt 0 XP In general, itjs are used to represent coefficients of dj(i) and irjs
6 7 6 7 6 7 are used to represent coefficients of 1j(i).To fully define the error
6 7 6 7 6 7
T 6 yt 7 6 0 7 W 6 Y P 7 model of a FDM machine, 42 such coefficients need to be
4 5 4 5 4 5
zt 0 ZP determined.

By equating the two vector chains from origin O to the 3D artifact and parametric error functions of the FDM
extrusion nozzle and taking into account the rotational errors 3000 machine
of the X, Y and Z axes, the following equation is obtained: The same 3D artifact used in Tongs previous research (Tong
et al., 2004) is used for this research and is shown in Figure 3.
~ R21 ZW
Z ~ Y
~ R21 Y X
~ R21 Y R21 XT
~ 2 It comprises 169 cylinders at 13 height levels and located at
the intersections of 13 x levels and 13 y levels. The position of
where: the top center point of each cylinder is measured and can be
2 3 written as a function of its nominal position and the error
1 1z u 21y u coefficients. The heights of the cylinders are arranged in a way
6 7 such that there are as many x, y and z combinations as
6 1x u 7
Ru 6 21z u 1 7 possible to form independent equations and all the top
4 5
1y u 21x u 1 surfaces of the cylinders are easily reachable by the CMM
2 3 probe during measurement. This artifact covers a working
1 21z u 1y u envelope of 200 mm (x) 200 mm (y) 100 mm (z).
6 7 The artifact is built with a layer thickness of 0.010 in.
6 21x u 7
and R21 u 6 1z u 1 7 (0.254 mm). A reference coordinate system is defined with the
4 5
21y u 1x u 1 left front corner of the platform as the origin. The 3D artifact
is then measured on a Carl Zeiss ECLIPSE 550 CMM.
A part coordinate system is defined with the top surface of the
represent the rotational error of axis u.
base as the x-y plane and the left front corner of the 3D
Substituting X, Y, Z, T, W and rotational matrixes into
artifact as the x, y datum. The coordinate of the top central
equation (2) and rearranging the terms, the expressions for
point of each cylinder is measured for estimation of the error
the coordinates of the depositing nozzle in the work piece
function coefficients. The x and y coordinates of the center
coordinates system Xp, Yp and Zp are obtained: point are found by measuring around the periphery of each
X p x dx x dx y 2 dx z y 1z z cylinder and computing a best fit center.
A non-linear programming problem is defined to solve for
z 1y z tx the coefficients, with the objective function to minimize the
Y p y dy x dy y 2 dy z x1z y 2 1z z sum of square of all the residual non-repeatable errors t,
3 where:
2 z 1x z ty
Z p 2z dz x dz y 2 dz z x21y y 1y z Figure 3 3D artifact
2 y 1x z tz

tx, ty, and tz are the non-repeatable error terms which represent
the errors due to non-repeatable components. Since, the FDM
machine has a zero tool tip offset xt yt zt 0, the effect of
rotational errors 1x x; 1y x; 1z x and 1x y do not appear in
Figure 2 FDM machine axes chain vector diagram

W T

Z X

O Y

6
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

compensation for the FDM machine is not as great as that


X
169
for the SLA machine. The reasons behind this are discussed
t tx;i 2 ty;i2 tz;i 2 5
i1
in the next section.
Comparison of FDM and SLA compensation results
The coefficients are listed in Table II for each polynomial
Recall that the layer thickness parameter in the FDM
error function.
machine is about 0.010 in. (0.254 mm), that is 2.5 times
Using these coefficients, the parametric error functions of
thicker than that used in the SLA 250 machine (0.004 in. or
all the three axes are determined and shown in Figures 4
0.1016 mm). This implies larger Z value quantization for the
and 5. The following observations can be made:
FDM machine and thus less control of dimensions in the Z
.
The parametric errors are not always linear functions and
direction. To test whether this is the reason for lower
not always symmetric about the x 0 or y 0 axis. This
volumetric error reduction by compensation for the FDM
allows for more accurate compensation to the model than
machine, the volumetric errors are decomposed into errors in
what is possible by applying simple homogenous shrinkage
the XY plane and errors in the Z direction; the reduction
factor compensation.
ratios by compensation are then computed for XY plane
. In terms of magnitude, scale errors of the three axes are
errors and Z direction errors, respectively. Figure 8 shows
the largest translational error.
these decomposed errors before and after compensation for
.
The scale error of X axis and scale error of Y axis in most
SLA 250 and FDM 3000 machines.
of its range are positive, which implies an oversized
It can be seen from Figure 8 that:
prototype. .
For both the FDM machine and the SLA machine, errors
in the XY plane are greatly reduced by compensation,
Compensation methods and results with an average reduction ratio of 0.383 and 0.371,
respectively. Also the average XY plane error values after
STL file compensation applied to a FDM 3000 machine
compensation are about the same for both machines,
The STL file compensation method, discussed in detail by
0.0014 in. (0.0356 mm) for FDM and 0.0018 in.
Tong et al. (2004), was applied to a FDM 3000 machine using
(0.0457 mm) for SLA. No significant difference is found
the coefficients determined in the previous section. The same between FDM and SLA.
test part used in the SLA machine study is used in this .
The Z direction error of the FDM machine is hardly
research (Figure 6). This part has 49 cylinders of the same improved by compensation at all. The average reduction
diameter as those in the 3D artifact but positioned at different ratio is 1.041, much worse than the reduction ratio of
locations and with random selected heights. Two copies are 0.395 in the SLA case. However, the average Z direction
made by the same FDM machine, one with compensation error values after compensation are again about the
and one without compensation. The volumetric error, i.e. the same for both machines, 0.0032 in. (0.0813 mm) for
distance between the actual point and the nominal point, of FDM and 0.0027 in. (0.0686 mm) for SLA. This implies
the top surface center point of each cylinder is calculated. that, with the proposed software error compensation
The ratio of the volumetric error after compensation and the method, a dimensional accuracy of around 0.003 in.
volumetric error before compensation is calculated for each (0.0762 mm) in the Z direction can be reached for both
point and the histogram of the 49 ratios is shown in machines. The difference in the reduction ratios are due
Figure 7(a). On average, it was found that volumetric error is to the different Z direction errors in the uncompensated
reduced from 0.0063 in. (0.1600 mm) to 0.0037 in. parts.
(0.0940 mm), which represents around 68.7 percent of its .
Owing to layer quantization, Z direction dimensions are
original value. This is considerable improvement. However, less controllable. This explains why after compensation
due to non-repeatable factors in the RP process and Z value the residual Z direction error is larger than the XY plane
quantization in layered manufacturing (Tong et al., 2004), the error.
ratios follow a statistical distribution and a number of data
points are greater than 1.0, which implies a larger volumetric Slice file compensation for form error reduction
error after compensation. For comparison, the same test As shown in Figure 9(a), for a cylinder, all vertices in the STL
result of the SLA machine used by Tong et al. (2004) is shown model lie on the circumference of the two bounding circles.
in Figure 7(b). Clearly, the improvement through Compensation applied to these vertices will compensate for

Table II Coefficients of polynomial error functions


Translational error coeff_1 (in.) coeff_2 (in.) coeff_3 (in.) Rotational error coeff_1 (rad) coeff_2 (rad) coeff_3 (rad)
dx(x) 0.0024437 2 0.0001088 0.0000024 1y(y) 0.0000054 20.0000123 20.0000005
dy(x) 0.0009650 2 0.0000994 0.0000033 1z(y) 0.0000898 20.0000076 20.0000002
dz(x) 20.0017483 0.0001725 20.0000003 1x(z) 2 0.0008350 20.0002165 20.0000183
dx(y) 20.0000396 2 0.0000888 0.0000053 1y(z) 0.0019100 0.0007025 0.0000721
dy(y) 20.0020189 0.0005293 20.0000244 1z(z) 2 0.0006774 20.0002798 20.0000304
dz(y) 20.0013518 0.0000971 0.0000012
dx(z) 20.0059439 2 0.0023896 20.0002756
dy(z) 0.0022604 0.0012610 0.0001175
dz(z) 0.0113696 0.0050943 0.0004876

7
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Figure 4 Parametric error functions of FDM 3000 machine (1-8)

Scale Error of X Axis Straightness Error of X Axis


0.35 0.07
0.3 0.06
0.25 0.05
xtx (mm)

xty (mm)
0.2 0.04
0.15 0.03
0.1 0.02
0.05 0.01
0 0.00
0 50 100 150 200 250 0 50 100 150 200 250
X coordinates (mm) X coordinates (mm)

Straightness Error of X Axis Straightness Error of Y Axis


0.08 0
0.06 0 50 100 150 200 250
0.01
0.04
0.02
0.02
xtz (mm)

ytx (mm)
0 0.03
0.02 0 50 100 150 200 250
0.04
0.04
0.05
0.06
0.08 0.06
0.1 0.07
X coordinates (mm) Y coordinates (mm)

Scale Error of Y Axis Straightness Error of Y Axis


0.12 0
0.1 0 50 100 150 200 250
0.01
0.08 0.02
0.06
0.03
yty (mm)

ytz (mm)

0.04
0.04
0.02
0.05
0
0.02 0 50 100 150 200 250 0.06
0.04 0.07
0.06 0.08
Y coordinates (mm) Y coordinates (mm)

Straightness Error of Z Axis Straightness Error of Z Axis


0.14 0.10
0.12 0.08
0.1 0.06
ztx (mm)

zty (inch)

0.08
0.04
0.06
0.02
0.04
0.00
0.02 100 80 60 40 20 0
0.02
0
100 80 60 40 20 0 0.04
Z coordinates (mm) Z coordinates (inch)

the dimensional error in the size and positioning error of the displacement of the top and bottom bounding vertices.
object. However, when the slices are made to create the layers This means that feature form error along the Z direction
that drive the part building process, the body of the cylinder cannot be compensated. A form error states how far an actual
will only inherit compensation which propagates from the surface of feature varies from the desired form implied by

8
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Figure 5 Parametric error functions of FDM 3000 machine (9-14)

Scale Error of Z Axis Roll of Y Axis


0.15 0
50 0 50 100 150 200 250
0.1
100

yry (Arc. Sec)


0.05
150
ztz (mm)

0 200
100 80 60 40 20 0 250
0.05
300
0.1
350
0.15 400
Z coordinates (mm) Y coordinates (mm)

Pitch of Y Axis Pitch of Z Axis


40 180
30 160
20 140
120
yrz (Arc. Sec)

10

zrx (Arc. Sec)


0 100
10 0 50 100 150 200 250 80
60
20
40
30
20
40
0
50 100 80 60 40 20 0
Y coordinates (mm)
Z coordinates (mm)

Yaw of Z Axis Roll of Z Axis


0 80
100 80 60 40 20 0 70
50 60
zrz (Arc. Sec)
zry (Arc. Sec)

50
100
40
150 30
20
200 10
0
250 100 80 60 40 20 0
Z coordinates (mm) Z coordinates (mm)

Figure 6 49-Cylinders test part


the drawing. Typical examples of form errors include flatness,
straightness, roundness, cylindricity, parallelism, and
perpendicularity. A more refined compensation would be to
apply the compensation to each contour of each layer,
generated after the STL file is sliced, as shown in Figure 9.
Each contour is essentially made up of linear, sequentially
connected segments which define the boundary of a layer and
drive the build tool during execution of that layer. Applying
compensation to each slice of the slice file provides resolution
of compensation equal to that of the machine layer resolution.
The slice file format of the SLA machine is proprietary, thus
only STL file compensation was developed for the SLA study.
The slice file format of the FDM machine, SSL file,
is readable and has the following format text in italic is
comment:

9
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Figure 7 Histograms of volumetric error reduction ratios


Histogram of Volumetric Error Reduction Ratios Histogram of Volumetric Error Reduction Ratios
22 22
20 20
18 18
16 16
14 14
Frequency

Frequency
12 12
10 10
8 8
6 6
4 4
2 2
0 0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 More 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 More
Volumetric Error Ratio Volumetric Error Ratio
Average Ratio = 0.687 Average Ratio = 0.355
Volumetric Error: 0.0063 inch 0.0037 inch Volumetric Error: 0.0097 inch 0.0035 inch
(0.1600 mm 0.0940 mm) (0.2464 mm 0.0889 mm)
(a) Histogram for FDM Machine (b) Histogram for SLA Machine

Figure 8 Histograms of XY plane and Z direction error reduction ratios

Histogram of XY Plane Error Reduction Ratios Histogram of XY Plane Error Reduction Ratios
22 22
20 20
18 18
16 16
Frequency

14 14
Frequency

12 12
10 10
8 8
6 6
4 4
2 2
0 0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 More 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 More
XY Plane Error Ratio XY Plane Error Ratio
Average Ratio = 0.383 Average Ratio = 0.371
XY Plane Error: 0.0047 inch 0.0014 inch XY Plane Error: 0.0050 inch 0.0018 inch
(0.1194 mm 0.0356 mm) (0.1270 mm 0.0457 mm)
(a) Histogram of XY Plane Error (b) Histogram of XY Plane Error
Reduction Ratios in FDM Machine Reduction Ratios in SLA Machine

Histogram of Z Direction Error Reduction Ratios Histogram of Z Direction Error Reduction Ratios
22 22
20 20
18 18
16 16
Frequency

14 14
Frequency

12 12
10 10
8 8
6 6
4 4
2 2
0 0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 More 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 More
Z Direction Error Ratio Z Direction Error Ratio
Average Ratio = 1.041 Average Ratio = 0.395
Z Direction Error: 0.0039 inch 0.0032 inch Z Direction Error: 0.0076 inch 0.0027 inch
(0.0991 mm 0.0813 mm) (0.1930 mm 0.0686 mm)
(c) Histogram of Z Direction Error (d) Histogram of Z Direction Error
Reduction Ratios in FDM Machine Reduction Ratios in SLA Machine

10
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

The SSL file lists all the contours in each layer after slicing the vertically positioned features, such as vertical cylinders and
STL model. For each contour, the number of vertices and the vertical planes.
(X, Y) coordinates of each vertex are listed, followed by a To check the extra benefit in vertical feature form error
number of summary parameters including min X, max X, min reduction brought by SSL compensation, two test parts are
Y, max Y, perimeter, and area. The vertices are listed in a designed, as shown in Figures 10 and 11. Both parts have a
sequence such that when traveling along these vertices, the planar base, which is used to assist positioning the part
material is always to the left. relative to the origin of the platform. Figure 10 has a cylinder
When compensating using slice files, the compensation
of 1.8 in. (45.72 mm) diameter and Figure 11 has horizontal,
will be a two-step procedure. First, use the STL
compensation method discussed before to compensate only vertical and slanted planar surfaces. Two copies of each part
for the Z direction error for all vertices in the STL model. are built, one with STL compensation, and the other with
Then, the compensated STL model is sliced to generate the SSL compensation. The cylinder and the planes are measured
slice file. Compensation for X and Y errors will be applied and their form error data is listed in Table III. Compensation
to each vertex in the SSL file. By doing this two-step for the overall part size and feature position should stay
procedure, we can compensated for Z direction error and the same and will not be tested against STL file
take advantage of the higher resolution of slice file compensation.
compensation. Contrary to expectation, the results did not show any
Form error reduction by slice file compensation significant difference between the two compensation
The SSL file compensation approach compensates error at approaches. In order to explain this result, the machine
each individual layer and thus provides a better compensation control resolution is examined to see if the FDM machine has
resolution along the Z direction. A direct benefit over STL file high enough resolution to distinguish between STL
compensation would be improvement in the form error of compensation and SSL compensation.

11
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Figure 9 Compensation applied to the contours of all slices Table III Form error of cylinder and planes with STL compensation and
SSL compensation
Form error SSL compensation STL compensation
Cylindricity 0.00401 0.00455
Flatness
Plane 1 0.00293 0.00271
Plane 2 0.00190 0.00191
Plane 3 0.00134 0.00193
Plane 4 0.00249 0.00318
Plane 5 0.00220 0.00157
Plane 6 0.00169 0.00161

(a) STL Model


without compensation are known. Using the machine error
model, compute the (X, Y) coordinates of those points with
STL compensation and SSL compensation, respectively.
Slice 1 Apply Figures 12 and 13 show X and Y coordinates of those points
Compensation
after STL compensation and SSL compensation, respectively.
It is seen that coordinate differences between the STL line
Slice i and the SSL line are always less than 0.001 in. (0.0254 mm).
This means that if the machine control resolution in X and Y
axis is equal to or larger than 0.001 in. (0.0254 mm), then
Slice N these two types of compensated methods will cause no
difference in the machine tool tip motion.
Unfortunately, the FDM 3000 machine has a resolution of
(b) Slices
just around 0.001 in. (0.0254 mm) in both X and Y direction.
Thus, by STL file compensation, the machine control
Figure 10 Cylindrical test part resolution was already reached, and SSL compensation will
not introduce any further improvement. It thus remains to be
seen if the higher resolution of slice file compensation can still
be exploited by a machine control system with higher
resolution than that of an FDM 3000 machine.

Conclusions
This research extends the approach of software error
compensation to improve the accuracy of a FDM machine
by considering correction of both STL and SSL file formats.
The STL file compensation method was applied to the FDM
3000 machine and reduced the average volumetric error by
around 30 percent. In comparison with the SLA 250
Figure 11 Planar test part machine, this reduction ratio is smaller, but the error values
after compensation are about the same for both machines.
These values can be thought of as the residual errors after
compensation and define the ability of the proposed error
4 2 compensation method. Owing to lager layer quantization, the
1
residual error in the Z direction is larger than the residual
3
5 error in the XY plane.
Slice file compensation method was developed as a means
to improve the compensation to account for form errors and
6 tested using two test parts. Although it theoretically allows
higher compensation resolution, the actual machine control
resolution of the FDM 3000 machine is not high enough
and thus the difference between STL compensation and slice
Effect of machine control resolution file compensation is not distinguishable. Only when
Consider one vertical line on the cylinder surface located at the machine control resolution is considerably higher than
(X 2.700, Y 2.700) and then consider the points the difference between STL and SSL compensation, SSL
obtained after slicing. The coordinates of those points compensation should be used in stead of STL compensation.

12
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Figure 12 X coordinates of the line

Compare SSL compensation and STL compensation


(Line Nominal X Coordinates = 2.700 inch)
3.5
Line with SSL Compensation
3
Line with STL Compensation

Z coordinates (inch)
2.5

2
Nominal Line
1.5

0.5

0
2.700 2.701 2.702 2.703 2.704 2.705
X Coordinates (inch)

Figure 13 Y coordinates of the line

Compare SSL compensation and STL compensation


(Line Nominal Y Coordinates = 2.700 inch)
3.5
Line with SSL Compensation
3
Line with STL Compensation
Z coordinates (inch)

2.5

1.5

0.5
Nominal Line
0
2.700 2.701 2.702 2.703 2.704 2.705
Y Coordinates (inch)

References Fadel, G.M. and Kirschman, C. (1996), Accuracy issue in


CAD to RP translations, Rapid Prototyping Journal, Vol. 2
Arni, R.K. and Gupta, S.K. (1999), Manufacturability
No. 2, pp. 4-17.
analysis for solid freeform fabrication, Proceedings of
Ferreira, P.M. and Liu, C.R. (1986), A contribution to the
DETC, ASME Design Engineering Technical Conference,
analysis and compensation of the geometric error of a
pp. 147-58.
machining center, Annals of the CIRP, Vol. 35 No. 1,
Belforte, G., Bona, B., Canuto, E., Donati, F., Ferraris, F.,
Gorini, I., Morei, S., Peisino, M. and Sartori, S. (1987), pp. 259-62.
Coordinate measuring machines and machine tools self- Gargiulo, E.P. and Belfiore, D.A. (1991), Photopolymer
calibration and error correction, Annals of the CIRP, Vol. 36 solid imaging process accuracy, Intelligent Design and
No. 1, pp. 359-64. Manufacturing for Prototyping, Vol. 50, pp. 81-95.
Cheng, W., Fuh, J.Y.H., Nee, A.Y.C., Wong, Y.S., Loh, H.T. Hocken, R. (1995), Software correction of precision
and Miyazawa, T. (1995), Multi-objective optimization of machines, A report from Precision Engineering
part-building orientation in stereolithgraphy, Rapid Laboratory of UNC Charlotte to National Institute of
Prototyping Journal, Vol. 1 No. 4, pp. 12-23. Standards and Technology in fulfillment of Contract No.
Dutta, D., Prinz, F.B., Rosen, D. and Weiss, L. (2001), NIST-60NANB2D1214.
Layered manufacturing: current status and future trends, Jurrens, K.K. (1999), Standards for the rapid prototyping
Journal of Computing and Information Science in Engineering industry, Rapid Prototyping Journal, Vol. 5 No. 4,
Transactions of ASME, Vol. 1, pp. 60-71. pp. 169-78.
Elshennawy, A.K.M. (1987), Performance evaluation of Kruth, J.P., Leu, M.C. and Nakagawa, T. (1998), Progress
coordinate measuring machines, PhD thesis, The in additive manufacturing and rapid prototyping, Annals of
Pennsylvania State University, University Park, PA. the CIRP, Vol. 47 No. 2, pp. 525-40.

13
Fused deposition modeling Rapid Prototyping Journal
Kun Tong, Sanjay Joshi and E. Amine Lehtihet Volume 14 Number 1 2008 4 14

Kruth, J.P., Vanherck, P. and De Jonge, L. (1994), Self- Ni, J. (1997), CNC machine accuracy enhancement through
calibration method and software error correction for three- real-time error compensation, Journal of Manufacturing
dimensional coordinate measuring machines using artifact Science and Engineering Transactions of the ASME, Vol. 119
measurements, Measurement, Vol. 14, pp. 157-67. No. 4B, pp. 717-25.
Kulkarni, P. and Dutta, D. (1996), An accurate slicing Onuh, S.O. and Hon, K.K.B. (2001), Improving
procedure for layered manufacturing, Computer-aided stereolithgraphy part accuracy for industrial applications,
Design, Vol. 28 No. 9, pp. 683-97. International Journal of Advanced Manufacturing Technology,
Kumar, V. and Dutta, D. (1997), An assessment of data Vol. 17, pp. 61-8.
formats for layered manufacturing, Advances in Engineering Philpott, M.L. and Green, P.A. (1995), An error
Software, Vol. 28 No. 3, pp. 151-64. compensation strategy for replication by rapid
Kunzman, H., Trapet, E. and Waldele, F. (1990), Concept prototyping, Journal of Engineering for Industry, Vol. 117,
for calibration, acceptance test, and periodic inspection of pp. 423-9.
coordinate measuring machines using reference objects, Sartori, S. and Zhang, G.X. (1995), Geometric error
Annals of the CIRP, Vol. 39 No. 1, pp. 561-4. measurement and compensation of machines, Annals of the
Lee, S.J., Sachs, E. and Cima, M. (1995), Layer position CIRP, Vol. 44 No. 2, pp. 599-609.
accuracy in powder-based rapid prototyping, Rapid Tong, K., Lehtihet, E.A. and Joshi, S. (2004), Software error
Prototyping Journal, Vol. 1 No. 4, pp. 24-37. compensation for rapid prototyping, Precision Engineering,
Lin, F., Sun, W. and Yan, Y. (2001), Optimization with Vol. 28 No. 3, pp. 280-92.
minimum process error for layered manufacturing Trapet, E. and Waldele, F. (1991), A reference object based
fabrication, Rapid Prototyping Journal, Vol. 7 No. 2, method to determine the parametric error components of
pp. 73-81. coordinate measuring machines and machine tools,
Liu, W., Li, L. and Kochhar, K. (1998a), A method for Measurement, Vol. 9 No. 1, pp. 17-22.
assessing geometrical errors in layered manufacturing. Part Wah, P.K., Murty, K.G., Joneja, A. and Chiu, L.C. (2001),
1: error interaction and transfer mechanisms, International Tool path optimization in layered manufacturing,
Journal of Advanced Manufacturing Technology, Vol. 14, IIE Transaction, Vol. 34 No. 4, pp. 335-47.
pp. 637-43. Wohlers, T. (1995), Future potential of rapid prototyping
Liu, W., Li, L. and Kochhar, K. (1998b), A method and manufacturing around the world, Rapid Prototyping
for assessing geometrical errors in layered manufacturing.
Journal, Vol. 1 No. 1, pp. 4-10.
Part 2: mathematical modeling and numerical evaluation,
Zhang, G.X., Veale, R., Charlton, T., Borchardt, B. and
International Journal of Advanced Manufacturing Technology,
Hocken, R. (1985), Error compensation of coordinate
Vol. 14, pp. 644-50.
measuring machines, Annals of the CIRP, Vol. 34 No. 1,
Lynn, C.M. and Rosen, D.W. (1999), SLA-250 parts vs
pp. 445-8.
geometric tolerances: quantitative results, paper presented
Zhang, G.X., Ouyang, R., Lu, B., Hocken, R., Veale, R. and
at North American Stereolithography User Group
Donmez, A. (1988), A displacement method for machine
Conference, pp. 1-15.
geometry calibration, Annals of the CIRP, Vol. 37 No. 1,
Lynn, C.M. and Rosen, D.W. (2000), Usage of accuracy
pp. 515-8.
models in Stereolithography process planning, Rapid
Zhou, J.G., Herscovici, D. and Chen, C.C. (2000),
Prototyping Journal, Vol. 6 No. 2, pp. 77-87.
Mou, J. and Liu, R.C. (1995), A methodology for machine Parametric process optimization to improve the accuracy
tools error correction using reference parts, International of rapid prototyped stereolithography parts, International
Journal of Computer Integrated Manufacturing, Vol. 8 No. 1, Journal of Machine Tools and Manufacturing, Vol. 40,
pp. 64-77. pp. 363-79.
Nee, A.Y.C., Fuh, J.Y.H. and Miyazawa, T. (2001), On the
improvement of the stereolithography process,
Corresponding author
International Journal of Material Processing Technology,
Vol. 113, pp. 262-8. Sanjay Joshi can be contacted at: sjoshi@psu.edu

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


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

14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.