Antonio Lanzotti

Fraunhofer JL IDEAS-CREAMI,
Department of Industrial Engineering,
University of Naples Federico II,
P.le Tecchio, 80,
Naples 80125, Italy
e-mail: antonio.lanzotti@unina.it
On the Geometric Accuracy
Domenico Maria Del Giudice
Fraunhofer JL IDEAS-CREAMI,
Department of Industrial Engineering,
of RepRap Open-Source Three-
University of Naples Federico II,
P.le Tecchio, 80,
Naples 80125, Italy
Dimensional Printer
e-mail: domenicomaria.delgiudice@unina.it In the field of additive manufacturing (AM) processes, there is a significant lack of scien-
tific data on the performance of open-source 3D printers in relation to process parameter
Antonio Lepore values. The purpose of this paper is to assess the impact of the main process parameters
Fraunhofer JL IDEAS-CREAMI, on the accuracy of a set of typical geometric features, as obtained with an open-source
Department of Industrial Engineering, 3D printer, the RepRap Prusa-Mendel I2. For this purpose, a benchmarking part was set
University of Naples Federico II, up, composed of elementary shapes, representing a series of different geometric features.
P.le Tecchio, 80, By means of a DoE approach, it was possible to assess the effects of two process
Naples 80125, Italy parameterslayer thickness (Lt) and flow rate (Fr)on five geometric features: cube,
e-mail: antonio.lepore@unina.it sphere, cylinder, cone, and angled surface. A high resolution Laser Scanner was used to
evaluate the variation between the acquired geometric feature and the corresponding 3D
Gabriele Staiano computer-aided design (CAD) nominal model. On the basis of experimental results, it
Fraunhofer JL IDEAS-CREAMI, was possible to analyze and discuss the main effects of the above-mentioned process pa-
Department of Industrial Engineering, rameters on each geometric feature. These results can help RepRap users in the correct
University of Naples Federico II, selection of process parameters with the aim of improving the quality of prototypes.
P.le Tecchio, 80, [DOI: 10.1115/1.4031298]
Naples 80125, Italy
e-mail: gabriele.staiano@unina.it Keywords: open-source 3D Printers, geometric features, process parameters, laser
scanner
Massimo Martorelli1
Fraunhofer JL IDEAS-CREAMI,
Department of Industrial Engineering,
University of Naples Federico II,
P.le Tecchio, 80,
Naples 80125, Italy
e-mail: massimo.martorelli@unina.it

1 Introduction However, even if this partnership represents a milestone for the

AM community, standard methods for the assessment of the accu-
The term AM is used to describe those technologies which
racy of AM systems have not been actually defined yet. In fact,
allow the production of physical objects, made up of various
there is a significant lack of scientific data on the accuracy and
materials, through an additive process. The AM manufacturing
repeatability between, for example, commercial AM systems
process produces objects by adding the material layer-upon-layer,
using proprietary methods to define the accuracy of their systems
directly from a 3D data model input.
and open-source AM systems.
In recent years, a new generation of AM techniques has rapidly
Over the years, various benchmarking parts for evaluating
become available to the public, due to the expiration of some AM
the accuracy and repeatability of AM processes have been proposed:
patents and to open-source movements, which allowed significant
Kruth [8], Lart [9], Ippolito et al. [10], Juster and Childs [11,12], Shel-
cost reductions.
labear [13], Mahesh et al. [14], Sercombe and Hopkinson [15].
The RepRap (replicating rapid prototyper) Project, which was
Unfortunately, none of these benchmarking parts comprehensively
developed by Adrian Bowyer at the University of Bath (UK) [1],
includes all the geometric features sufficient to establish the desired
is one of the best-known open-source projects.
accuracy/repeatability of all the AM process parameters. Recently,
The current increasing interest on the extraordinary potential of
we used the benchmarking part proposed by Fahad and Hopkinson
these systems is confirmed by the development of International
[16], with the aim of evaluating the impact of the main process pa-
Standards [27] by the Partner Standard Development Organiza-
rameters in a RepRap 3D printer [17] on the system accuracy.
tion (PSDO) cooperation agreement signed in 2011 between the
In a symmetrically repeatable sequence, the part includes ele-
ISO Technical Committee 261 on AM and the ASTM Interna-
mentary shapes representative of the main geometric features:
tional Committee F42.
cube, cylinder, sphere, cone, and angled surface. Geometric fea-
tures have been widely and successfully used in traditional manu-
1
Corresponding author. facturing processes [1821]. For example, the feature-based
Contributed by the Design for Manufacturing Committee of ASME for
publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received March 15,
approach allows the design evaluation and its modification or
2015; final manuscript received July 28, 2015; published online September 2, 2015. redesign into a functionally acceptable one, which is compatible
Assoc. Editor: Christopher Williams. with the selected manufacturing process.

Journal of Mechanical Design Copyright V

C 2015 by ASME OCTOBER 2015, Vol. 137 / 101703-1

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 On the one hand, the feature-based approach would seem less Table 1 PLA properties
useful for the AM processes since even complex parts are
obtained layer by layer. On the other hand, in order to be accepted Physical property Nominal value
as a mainstream process, it is necessary for the AM that all geo-
metric features of the test parts are certified and that they consis- Specific gravity (23  C) 1.24 to 1.26 g/cm3
Melt mass-Fr
tently meet the critical geometric tolerance requirements. 210  C/2.16 kg 6.0 to 78.0 g/10 min
Therefore, the influence of the AM process parameter on the accu- 190  C/2.16 kg 1.5 to 36.0 g/10 min
racy of critical geometric features of the manufactured part Diameter tolerance 0.10 mm/0.15 mm
[2226] is very interesting. Ovalidity tolerance (max) 6 5%
The purpose of this paper is to study the effect of process
parameters on the accuracy of five typical geometric features Mechanical property Nominal value
cube, sphere, cylinder, cone, and angled surfaceusing an open-
Tensile modulus (23  C) 2020 to 3550 MPa
source 3D printer. Tensile strength yield (23  C) 15.5 to 72.0 MPa
Tensile strength break (23  C) 14 to 70 MPa
2 Materials and Methods Tensile elongation yield (23  C) 9.8 to 10.0%
Tensile elongation break (23  C) 0.50 to 9.20%
In this study, the open-source RepRap Prusa-Mendel I2 3D Flexural modulus (23  C) 2392 to 4930 MPa
printer was used (Fig. 1). The calibration phase of RepRap 3D Flexural strength (23  C) 48 to 110 MPa
printers is fundamental. It allows the motors to move the correct
distance each time they move the build platform or the extruder,
so as to obtain objects with the same dimensional characteristics,
even if they are fabricated by different RepRap of the same type.
In particular, the effect of the gravity, which is here due to the
Y-carriage, is not significant because the mass of the fabricated
part is very small compared to that of the build platform. The
printer was calibrated using a dial indicator with an accuracy of
610 lm through the Mitutoyo 2046-08 (Mitutoyo, Japan) mag-
netic base. Marlin and Cura open-source software were used to
get the final 3D objects printed in 2.85 mm gray polylactic acid
(PLA) biodegradable material. The values of the main physical
and mechanical properties of the material are reported in Table 1.

2.1 Geometric Features. The benchmarking part proposed

by Fahad and Hopkinson [16] (Fig. 2) was taken into account for
this study as in Ref. [17]. Five geometric featurescube, sphere,
cylinder, cone, and angled surface(Fig. 2) were selected. For
each geometric feature, the deviation di (i.e., the shortest distance Fig. 2 Five geometric features selected in the benchmarking
from the ith point of the cloud to the CAD nominal model) is part
recorded over the n point of the cloud. Then, the accuracy of the
manufactured part is defined as the root mean square error (RMSE)
2.2 Process Parameters. Several process parameters are
s involved in the experimentation. Starting from the previous results
1X n [17] obtained through the main slicing software (CURA, KIS-
RMSE d2 (1) SLICER, SLIC3R), the following process parameters were
n i1 i
considered:
over the n points. (1) Lt, the height of each slice deposited by the 3D printer,
which is the feed rate along the vertical axis before extrud-
ing a new layer (atop the previous one).
(2) Fr, a measure of the material quantity that comes out from
the extruder; it is expressed as a percentage of the revolu-
tions of the electrical motor for extruding 1 mm of filament.
(3) Deposition speed (Ds), the speed of the printing process.
(4) Bed temperature, the temperature of the printing bed. This
parameter depends on the material.
(5) Printing temperature, the extrusion temperature of the
material.
(6) Wall thickness, the thickness of the outside/inside shells in
the horizontal direction. In combination with the nozzle
size, it defines the number and the thickness of the perime-
ter lines.
(7) Bottom/top thickness, the amount of solid layer put down
the bottom and the top of the part.
(8) Fill density, the quantity of material inside the part.
The Ds is set to 30 mm/s. This value represents the better set-
ting found in Ref. [17], through a 33 full factorial design of experi-
ment (DoE) on the factors Lt, Fr (set to the same levels), and Ds.
The bed and the printing temperatures are set on the reference
Fig. 1 The open-source RepRap Prusa Mendel I2 used in this values suggested by the PLA producers since they are mainly
case study related to the material used.

101703-2 / Vol. 137, OCTOBER 2015 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Table 2 List of control and held factors The range of variation of the control factors (Table 3) was
selected by prior technological information on the expected opti-
Factor Type Value mal values, even though the full ranges of printer parameters
could be more extended. In particular, the better accuracy is
Lt (mm) Control Table 3 expected at low values of Lt, which is bounded below the diame-
Fr (%) Control Table 3
Ds (mm/s) Held-constant 30
ter of the nozzle (0.35 mm). However, our previous experience
Wall thickness (mm) Held-constant 0.7 [17] showed a poor quality of the manufactured parts when Lt was
Bottom/top thickness (mm) Held-constant 0.6 set at 0.05 mm. Therefore, the minimum level of Lt should be set
Fill density (%) Held-constant 20 above such a value.
Bed temperature ( C) Held-constant 80 The Fr, based on the opinion of RepRap experts, should be set
Printing temperature ( C) Held-constant 200 to 100%, 105%, and 110%.
Table 4 shows the 32 full factorial design adopted in this paper.
For each treatment, three replicates are planned; 27 PLA bench-
Table 3 Three levels chosen for the control factors marking parts (viz., 9 treatments  3 replicates) were fabricated
and 135 geometric features (27 benchmarking parts  5 typical
Control factor Label Low (1) Intermediate (0) High (1) Unit geometric features) were examined. In order to reduce the effects
of variability, treatments were performed in a random order. In
Lt Lt 0.10 0.15 0.20 mm this way, the accuracy of the geometric test parts, described in
Fr Fr 100 105 110 % Sec. 2.1, can be evaluated.

Table 4 32 Full factorial design 2.3 Laser Scanner Acquisition. A high-resolution Laser
Scanner (D700 Scanner3Shape, Denmark) was used to get
Treatment Lt Fr the point clouds of each geometric feature (Fig. 3). The accu-
racy of this noncontact reverse engineering (RE) system is
1 1 1 620 lm.
2 1 0 On the basis of a strict protocol, all printed parts were acquired
3 1 1 by the same operator.
4 0 1
5 0 0
Using the iterative closest point algorithm [27,28] of Geo-
R

6 0 1 magicV software of 3D Systems, the point cloud of each geomet-

7 1 1 ric feature was aligned with the CAD nominal model. In such a
8 1 0 way, the analysis of the deviations di s was carried out.
9 1 1 All the areas at the edge between the base of the benchmarking
part and the different geometric features (gray colored in Fig. 4)
must be excluded from the analysis because they are too critical
The wall and the bottom/top thicknesses as well as the fill den- for the RE System. Therefore, the common subset of scan data
sity are set to the values reported in Table 2 based on the previous was extracted by considering all the points above a common suita-
laboratory experiences, since they mainly impact on the mechani- ble offset plane parallel to the base of the benchmarking part.
cal properties of the manufactured part. Apart from this, the number of points n of the cloud, automatically
Therefore, only the Lt and Fr parameters are considered as con- determined by the acquisition procedure, is as high as needed not
trol factors. to introduce additional bias due to the RE System. In fact, it is

Fig. 3 Benchmarking part RE acquisition by D700 Laser Scanner Fig. 4 Tensile test
specimen

Fig. 4 Common subset of scan data extracted for each geometric feature

Journal of Mechanical Design OCTOBER 2015, Vol. 137 / 101703-3

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 5 Three-dimensional colored maps of deviations for the first replicate

worth noting that by decreasing n (even to the 25% of the initial population without making any assumption on the underlying sta-
points) further investigations on the variability of the RMSE have tistical distribution.
not shown a significant decrease for the RMSE nor a computa- Figure 6 shows the box-plots of the RMSE grouped by geomet-
tional time saving. ric feature in order to compare their (empirical) accuracy distribu-
tions over the different treatments (Table 5).
3 Results The RMSE distributions of the cylinder and the cone show very
different mean and dispersion values with respect to the others. The
The accuracy of the manufactured part defined in Eq. (1) is RMSE distributions of the sphere, the angled surface, and the cube
assumed as response. As an example for the first replicate, Fig. 5 appear similar, although the sphere obtains the lowest values.
shows the 3D colored map of the measured deviations di s. The These graphical results are also analytically investigated by sta-
measure variability (shown in Fig. 5) also takes into account the tistical T-Test (one-tail), having checked the normality assump-
expected roughness for a fused deposition modeling process that tion of those distributions. In particular, the following crucial
depends on the surface orientation [29]. comparisons were considered:
Moreover, for each geometric feature, experimental data and
response are collected in Table 5. (1) sphere versus angled surface
(2) angled surface versus cube
3.1 Box-Plots and T-Test. Box-plots are a convenient graph- (3) cube versus cylinder
ical way to display variation in samples having a statistical (4) cylinder versus cone

101703-4 / Vol. 137, OCTOBER 2015 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Table 5 Full factorial design for geometric features RMSE the sphere results significantly more accurate than the angled
responses (the dotted line divides two replicates) surface
the angled surface is not significantly more accurate than the
Factor Response: RMSE (mm) cube
the cube is significantly more accurate than the cylinder
Angled
the cylinder is significantly more accurate than the cone
Treatment Replicate Lt Fr Sphere surface Cylinder Cube Cone

1 1 1 1 0.03 0.04 0.09 0.04 0.09

2 1 1 0 0.03 0.05 0.08 0.05 0.09 3.2 Analysis of Variance (ANOVA): Main Effect and
3 1 1 1 0.03 0.04 0.05 0.04 0.09 Interaction Plots. The ANOVA was used in order to analyze the
4 1 0 1 0.04 0.04 0.06 0.04 0.08 differences between group means. The ANOVA table (Table 6)
5 1 0 0 0.03 0.04 0.08 0.04 0.08 for RMSE shows that, at a significance level a 0:05, the mean
6 1 0 1 0.03 0.04 0.07 0.04 0.11 effect of the factor Lt is significant for the sphere, the angled sur-
7 1 1 1 0.04 0.06 0.08 0.07 0.10 face, the cube, and the cone; the Fr factor is significant for the cyl-
8 1 1 0 0.04 0.06 0.08 0.07 0.10
inder; the (Lt  Fr) interaction is significant only for the cone.
9 1 1 1 0.05 0.06 0.06 0.05 0.10
Figure 7 shows the main effect plots of the five geometric fea-
1 2 1 1 0.03 0.04 0.07 0.05 0.09 tures. The main effect plot is utilized to select the levels of the sig-
2 2 1 0 0.03 0.04 0.05 0.04 0.08 nificant factors that are expected to maximise the accuracy of
3 2 1 1 0.03 0.04 0.06 0.04 0.09 each geometric feature (viz., minimize the RMSE):
4 2 0 1 0.05 0.04 0.12 0.04 0.09
5 2 0 0 0.04 0.04 0.06 0.05 0.08 the level 1 of the Lt factor (0.10 mm) for the sphere and the
6 2 0 1 0.03 0.04 0.05 0.05 0.10 cube
7 2 1 1 0.05 0.06 0.09 0.06 0.10 the level 0 of the Lt factor (0.15 mm) for the angled surface
8 2 1 0 0.05 0.05 0.06 0.05 0.11 the levels 1 of the Lt factor (0.15 mm) and 0 of the Fr factor
9 2 1 1 0.04 0.04 0.06 0.05 0.12
(105%) for the cone
1 3 1 1 0.03 0.05 0.06 0.04 0.09 the level 1 of the Fr factor (105%) for the cylinder
2 3 1 0 0.04 0.04 0.07 0.04 0.09
3 3 1 1 0.03 0.04 0.06 0.03 0.08 The RMSE increase due to the main effect of the Lt factor
4 3 0 1 0.04 0.04 0.07 0.05 0.09 (Fig. 8(a) and top-right corner Fig. 7) is smaller than its decrease
5 3 0 0 0.04 0.04 0.08 0.04 0.07 due to the interaction effect (Lt  Fr, Fig. 8(b)) when the Lt level
6 3 0 1 0.04 0.04 0.06 0.04 0.10 change from 1 to 0. Therefore, the level 0 (0.15 mm) of the Lt
7 3 1 1 0.06 0.06 0.08 0.06 0.12 factor and the level 0 (105%) of the Fr factor is the optimal
8 3 1 0 0.05 0.06 0.11 0.05 0.12 setting.
9 3 1 1 0.04 0.05 0.06 0.06 0.12
Strictly speaking, from Fig. 7 the optimal level of the Lt fac-
tor for the angled surface is 0. However, as in the cone and the
cube cases, the main effect is fairly constant between levels 1
and 0.
Therefore, in conclusion, a common recommendation to set Lt
and Fr at levels 1 and 1, respectively, can be given to maximize
the accuracy of all considered geometric features. This is true
especially when conic surfaces are not predominant, otherwise the
interaction effect analysis suggests the different setting of the Lt
and Fr factors discussed above (Fig. 8).

3.3 Robust Choice of the Process Parameters. A typical

issue on the choice of process parameters arises when the variabil-
ity on the optimal value of the response (i.e., the accuracy of the
manufactured part) has to be minimized. This is required in order
to set process parameters which are also (as little as possible)
insensitive to noise factors (i.e., different geometric features) that
may be controlled in an experimental environment. According to
the Taguchis methodology for robust parameter design, data
Fig. 6 Box-plots of the RMSE grouped by geometric features
(Table 7) are summarized with the mean RMSE (over the treat-
ment replicates) of the point clouds averaged across all the geo-
The T-tests 1, 3, and 4 result significant at a significance level metric features and the corresponding signal-to-noise ratios (S/N).
a 0:05, while the T-test 2 is not significant at the same level. In The higher the S/N, the more robust the setting. In other words,
conclusion: the setting of the control factors chosen by maximizing the S/N
Table 6 ANOVA for the RMSE of the manufactured parts for each geometric feature (p-values lower than a 5 0.05 are highlighted
in bold and underlined)

Sphere Angled surface Cylinder Cube Cone

Source Degrees of Sequential sum of

freedom (DF) squares (Seq SS) p-value Seq SS p-value Seq SS p-value Seq SS p-value Seq SS p-value

Lt 2 0.00110 0.00 0.00127 0.00 0.00047 0.44 0.00147 0.00 0.00282 0.00
Fr 2 0.00014 0.15 0.00010 0.14 0.00216 0.04 0.00014 0.23 0.00047 0.04
Lt  Fr 4 0.00010 0.55 0.00008 0.48 0.00018 0.95 0.00010 0.68 0.00071 0.05
Error 18 0.00060 0.00040 0.00487 0.00080 0.00107
Total 26 0.00194 0.00185 0.00767 0.00252 0.00507

Journal of Mechanical Design OCTOBER 2015, Vol. 137 / 101703-5

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 7 Main effect plots obtained for each geometric feature for RMSE

Fig. 8 Main (Lt) and interaction (Lt 3 Fr) effects for the cone

Table 7 Average and S/N across all the geometric features of the mean RMSE

Factor Response: RMSE (mm)

Treatment Lt Fr Sphere Angled surface Cylinder Cube Cone Average S/N

1 1 1 0.03 0.04 0.07 0.05 0.09 0.06 24.33

2 1 0 0.03 0.04 0.07 0.04 0.09 0.06 24.10
3 1 1 0.03 0.04 0.06 0.05 0.09 0.05 24.78
4 0 1 0.04 0.04 0.08 0.05 0.08 0.06 24.06
5 0 0 0.04 0.04 0.07 0.04 0.08 0.06 24.53
6 0 1 0.03 0.04 0.06 0.04 0.10 0.06 23.52
7 1 1 0.05 0.06 0.08 0.07 0.10 0.07 22.60
8 1 0 0.04 0.06 0.09 0.06 0.11 0.07 22.49
9 1 1 0.04 0.05 0.06 0.05 0.11 0.07 23.02

minimizes the variability transmitted from the different geometric Lt and its interaction with Fr results significant at a level a smaller
features. The treatment 3 is therefore the more robust combination than 0.04. Therefore, it can be stated that the variation of the
maximizing the accuracy of the manufactured test part. The treat- RMSE average across the five geometric features is most influ-
ment 5 is the second choice (in terms of S/N). enced by the Lt factor.
Table 8 shows the results of the ANOVA test for the two Lt and The main effects plot for the RMSE average across all the geo-
Fr factors with the assumption of the Lt  Fr additive interaction. metric features (Fig. 9) and the interaction plot (Fig. 10) provide

101703-6 / Vol. 137, OCTOBER 2015 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Table 8 ANOVA for RMSE average across all the geometric fea- The cylinder and the cone are significantly affected by
tures (a 5 0.05; p-values lower than a 5 0.05 are highlighted in changes in the Fr. In particular, the level 1 (110%) maximizes
bold and underlined) the accuracy of cylinders, while the level 0 (105%) is the opti-
mal choice for cones. Therefore, the suggestion to increase the
RMSE average Fr over 100% is correct. The effects on the accuracy are accept-
able, since improvements are also confirmed in the range of
Source DF Seq SS p-value
105110%. The level 1 for the Fr is a robust choice even for the
Lt 2 0.00103 0.00 sphere, the angled surface, and the cube for which the Fr is not
Fr 2 0.00003 0.53 significant (see the maximum S/N value in Table 7).
Lt  Fr 4 0.00028 0.04 For all the considered process parameters, the cone and the
Error 18 0.00040 sphere always show the lowest and the highest accuracy,
Total 26 0.00174 respectively. All these results are in accordance with experts
expectations.
Moreover, in order to improve the global accuracy of all geo-
metric features (e.g., in the case they have to be manufactured all
together as in the benchmarking part), some practical suggestions
can be given as follows:
The Lt should be set at 0.10 mm in order to maximize the
mean accuracy. This choice is also the most robust, even if it
involves a (small) loss in accuracy for the cone.
The Fr should be set at 110% in order to account also for
interaction effects.

5 Concluding Remarks
The ISO and the ASTM Technical Committees are working on
the development of International Standards for AM to be adopted
worldwide. However, standard methods for the assessment of the
accuracy of the AM systems have not been defined yet. The
Fig. 9 Main effects plot for RMSE average across all the geo- benchmarking part used in this paper can be suitably fabricated by
metric features
different AM processes, to assess the accuracy and repeatability
of the system.
In this paper, the benchmarking part was used to analyze the
impact of the Lt and the Fr process parameters on the accuracy of
the production of five typical geometric features in an open-
source 3D printer.
Even if some limits of the present investigation are being dis-
cussed here, the results obtained clearly show that the Lt is a sig-
nificant parameter in four out of five of the considered geometric
features with respect to the accuracy of the manufactured parts.
The Fr results a significant parameter in two out of five of them.
The best results (except for the cone) for all the considered geo-
metric features are achieved at the lower (0.10 mm) and the higher
(110%) levels of the Lt and the Fr, respectively. Therefore, the
common recommendation to set Lt and Fr at 0.10 mm and 110%,
respectively, can be given when conic surfaces are not predomi-
nant. Otherwise, the interaction effect analysis suggests to set Lt
and Fr at 0.15 mm and 105%.
Fig. 10 Interaction plot of the RMSE average across all the Conversely, the higher level of the Lt (0.20 mm) and the lower
geometric features level of the Fr (100%) produce significant loss in accuracy for all
the expected optimal (more robust) combination by setting Lt at the considered geometric features (except for the cone). However,
level 1 (0.10 mm) and Fr at level 1 (110%). if the interactions of different geometric features are considered,
some potential nonlinear effects of process parameters cannot be
excluded (e.g., the interaction effect is significant for the cone).
4 Discussion The results discussed in this paper provide the AM community
The present study deals with the effects of two process parame- with additional scientific data useful to investigate the process
ters on the accuracy of geometric features manufactured using an parameter impact on the quality of parts obtained through a Rep-
open-source 3D printer. Starting from the results collected in the Rap 3D printer. Being quality a critical factor for a successful
experimental phase, it can be stated that: industrial application of the AM processes, future work should be
carried out to understand the effects of process parameter on mul-
Sphere, angled surface, cube, and cone are significantly tifeature parts which can meet any geometric tolerance specifica-
affected by changes in Lt. The level 0 (0.15 mm) maximizes the tions and to validate predictive models useful to anticipate the
accuracy of the angled surface, whereas the level 1 (0.10 mm) expected accuracy of parts which may be obtained through new
is the best choice for the other ones. This results in accordance generations of 3D printer.
with the rule of thumb which empirically suggests the opti-
mal set of the Lt at one-fourth of the nozzle (in this test the di-
ameter is 0.35 mm). The choice of level 1 for the angled Acknowledgment
surface is acceptable being the accuracy decrease negligible This work was supported by the POR FSE 2007/13 Project enti-
when compared to the optimal level 0. tled WISCH, Work Into Shaping Campanias Home.

Journal of Mechanical Design OCTOBER 2015, Vol. 137 / 101703-7

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 References [17] Lanzotti, A., Martorelli, M., and Staiano, G., 2015, Understanding Process
Parameter Effects of RepRap Open-Source Three-Dimensional Printers
[1] Jones, R., Haufe, P., Sells, E., Iravani, P., Olliver, V., Palmer, C., and Bowyer, Through a Design of Experiments Approach, ASME J. Manuf. Sci. Eng.,
A., 2011, RepRap-The Replicating Rapid Prototyper, Robotica, 29(1), pp. 137(1), p. 011017.
177191. [18] Shah, J., and Mantyla, M., 1995, Parametric and Feature-Based CAD/CAM:
[2] ISO/ASTM 52915, 2013, Standard Specification for Additive Manufacturing Concepts, Techniques, and Applications, Wiley-Inter-Science, New York.
File Format (AMF) Version 1.1. [19] Chen, Y. M., Wen, C.-C., and Ho, C., 2003, Extraction of Geometric Charac-
[3] ISO/ASTM 52921, 2013, Standard Terminology for Additive Manufacturing- teristics for Manufacturability Assessment, Rob. Comput. Integr. Manuf.,
Coordinate Systems and Test Methodologies. 19(4), pp. 371385.
[4] ISO 17296-1, 2014, Additive ManufacturingGeneralPart 1: Terminology. [20] Gayretli, A., and Abdalla, H. S., 1999, A Feature-Based Prototype System for
[5] ISO 17296-4, 2014, Additive ManufacturingGeneral PrinciplesPart 4: the Evaluation and Optimisation of Manufacturing Processes, Comput. Ind.
Overview of Data Processing Technologies, ASTM Fact Sheet. Eng., 37(12), pp. 481484.
[6] ISO 17296-3, 2014, Additive ManufacturingGeneral PrinciplesPart 3: [21] Ip, C. Y., and Regli, W. C., 2006, A 3D Object Classifier for Discriminating
Main Characteristics and Corresponding Test Methods. Manufacturing Processes, Comput. Graph., 30(6), pp. 903916.
[7] ISO 17296-2, 2015, Additive ManufacturingGeneral PrinciplesPart 2: [22] Zhang, Y., and Bernard, A., 2014, Using AM Feature and Multi-Attribute De-
Overview of Process Categories and Feedstock. cision Making to Orientate Part in Additive Manufacturing, High Value Manu-
[8] Kruth, J. P., 1991, Material Incress Manufacturing by Rapid Prototyping facturing: Advanced Research in Virtual and Rapid Prototyping, P. J. da Silva
Techniques, CIRP Ann., 40(2), pp. 16031615. Bartolo, A. C. Soares de Lemos, A. M. H. Pereira, A. J. dos Santos Mateus, C.
[9] Lart, G., 1992, Comparison of Rapid Prototyping Systems, First European Ramos, C. dos Santos, D. Oliveira, E. Pinto, F. Craveiro, H. M. Coelho da
Conference on Rapid Prototyping, University of Nottingham, Nottingham, UK, Rocha Terreiro Galha Bartolo, H. de Amorim Almeia, I. Sousa, J. M. Matias, L.
pp. 243254. Dur~ao, M. Gaspar, N. M. F. Alves, P. Carreira, T. Ferreira, and T. Marques,
[10] Ippolito, N. R., Iuliano, L., and de Filippi, A., 1994, A New User Part for Per- eds., Taylor & Francis Group, London.
formance Evaluation of Rapid Prototyping Systems, Third European Confer- [23] Rosen, D. W., 2007, Computer-Aided Design for Additive Manufacturing of
ence on Rapid Prototyping and Manufacturing, University of Nottingham, Cellular Structures, CAD Appl., 4(5), pp. 585594.
Nottingham, UK, pp. 327339. [24] Qian, X., and Dutta, D., 2001, Feature Based Fabrication in Layered Man-
[11] Juster, N. P., and Childs, T. H. C., 1994, Linear and Geometric Accuracies ufacturing, ASME J. Mech. Des., 123(3), pp. 337345.
From Layer Manufacturing, CIRP Ann., 43(1), pp. 163166. [25] Moroni, G., Syam, W. P., and Petr o, S., 2014, Towards Early Estimation of
[12] Juster, N. P., and Childs, T. H. C., 1994, A Comparison of Rapid Prototyping Part Accuracy in Additive Manufacturing, Procedia CIRP, 24th CIRP Design
Processes, Third European Conference on Rapid Prototyping and Manufactur- Conference, Vol. 21, pp. 300305.
ing, University of Nottingham, Nottingham, UK, pp. 3552. [26] Moroni, G., Syam, W. P., and Petr o, S., 2015, Functionality-Based Part Orien-
[13] Shellabear, M., 1999, Benchmarking Study of Accuracy and Surface Quality tation for Additive Manufacturing, Procedia CIRP, 25th CIRP Design Confer-
in RP Models, RAPTEC, Task 4.2, Report No. 2. ence, pp. 17.
[14] Mahesh, M., Wong, Y. S., Fuh, Y. H., and Loh, H. T., 2004, Benchmarking for [27] Besl, P. J., and McKay, N. D., 1992, A Method for Registration of 3-D
Comparative Evaluation of RP Systems and Processes, Rapid Prototyping J., Shapes, IEEE Trans. Pattern Anal. Mach. Intell., 14(2), pp. 239256.
10(2), pp. 123135. [28] Pottmann, H., Leopoldseder, S., and Hofer, M., 2002, Simultaneous Registra-
[15] Sercombe, T. B., and Hopkinson, N., 2006, Process Shrinkage and Accuracy Dur- tion of Multiple Views of a 3D Object, PCV02, Archives of the Photogram-
ing Indirect Laser Sintering of Aluminum, Adv. Eng. Mater., 8(4), pp. 260264. metry, Remote Sensing and Spatial Information Sciences, Vol. XXXIV, Part
[16] Fahad, M., and Hopkinson, N., 2012, A New Benchmarking Part for Evaluat- 3A, Commission III, pp. 265270.
ing the Accuracy and Repeatability of Additive Manufacturing (AM) [29] Campbell, R. I., Martorelli, M., and Lee, H. S., 2002, Surface Roughness Visu-
Processes, 2nd International Conference on Mechanical, Production and Auto- alisation for Rapid Prototyping Models, Comput. Aided Des., 34(10),
mobile Engineering (ICMPAE 2012), Singapore, April 2829, pp. 234238. pp. 717725.

101703-8 / Vol. 137, OCTOBER 2015 Transactions of the ASME

Downloaded From: http://mechanicaldesign.asmedigitalcollection.asme.org/ on 02/02/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use