Additive Manufacturing 21 (2018) 183–190

Contents lists available at ScienceDirect

Additive Manufacturing
journal homepage: www.elsevier.com/locate/addma

Full Length Article

Taguchi based process optimization for dimension and tolerance control for T
fused deposition modelling
⁎
Shahrain Mahmooda, , A.J. Qureshib, Didier Talamonac
a
Newcastle University, Singapore
b
University of Alberta, Canada
c
School of Engineering, Nazarbayev University, Kazakhstan & Conjoint, School of Engineering, The University of Newcastle, Australia

A R T I C LE I N FO A B S T R A C T

Keywords: This paper presents an experimental approach to investigate the effects of variation in the process parameter
Fused deposition modelling settings, found commonly in most fused deposition modelling printers, on the geometrical properties of the
Dimensional accuracy printed parts. A benchmark component was designed to include simple geometric features which allows for
Geometrical characteristics measurement for both dimensional accuracy and geometric characteristics. Taguchi’s design of experiment
Benchmark component
statistical approach was used to establish the relationship between varying process parameter settings on the
geometrical properties of the benchmark component. The critical process parameters affecting both the di-
mensional accuracy and geometric characteristics are identified and the theoretical optimum print settings were
found.

1. Introduction with a variety of mechanical properties are also available for use with
the FDM process. FDM machines are easy to use and such simple ma-
Additive manufacturing (AM) refers to a group of manufacturing chine operation appeals to end users with no prior 3D printing ex-
processes that create physical parts or components required for an as- perience. In terms of comparison with conventional manufacturing
sembly directly from a computer aided design (CAD) model without the processes, specifically for thermoplastic manufacturing, FDM process
need for any additional tooling. It is a layer based manufacturing pro- offer a significantly lower economic and technological entry barrier to
cess whereby 3-dimensional solid objects are formed through material manufacturing one-off or small lots as compared to conventional
deposition layer after layer, under the control of a computer numerical manufacturing such as injection molding or machining. The FDM pro-
control (CNC) software and hardware. These processes can create cess also offers more design flexibility and freedom as compared to
physical parts of complex geometries directly unlike the subtractive or conventional manufacturing processes.
transformative manufacturing processes. Some of the common AM In the FDM process, the material is fed through a heating element
processes include stereolithography (SLA), laminated object manu- and the molten material then goes through an extruder nozzle. The
facturing (LOM), selective laser sintering (SLS), selective laser melting extruded material is deposited selectively onto a build platform. Once a
(SLM) and fused deposition modelling (FDM). These processes are layer is completed, the build platform is lowered, and the next layer is
commonly used by engineers and developers in creating parts for visual deposited over the previous layer. The process is repeated until the solid
aids, presentation and prototyping models. The ability to manufacture part is created. The component is built through the bonding and soli-
components directly has significantly shortened the cycle time in the difying of the deposited layer with the previous layer. The movement of
product design stage. These processes are now being used in the man- the extruder nozzle is guided by the toolpath generated by the software.
ufacture of one-off customized components, end user products and have For a consumer grade desktop FDM 3D printer, the printing process is
replaced the more conventional manufacturing processes in small-scale determined by setting the numerous parameters available to the user.
manufacturing environments [1–6]. As with any other manufacturing processes, the FDM process has
The FDM process in particular has become highly accessible due to specific characteristics relating to attainable product geometric prop-
expiry of patents, lower cost of ownership unlike other laser based AM erties; achievable dimensional accuracy, surface finishing [9,10], pro-
systems [7,8], simple maintenance and an abundant supply of eco- cess repeatability, and tolerances [11] of the printed components. Such
nomically priced filaments for 3D printing. A wide range of materials properties may not be of concern for parts manufactured solely for

⁎
Corresponding author.
E-mail addresses: [email protected] (S. Mahmood), [email protected] (A.J. Qureshi), [email protected] (D. Talamona).

https://doi.org/10.1016/j.addma.2018.03.009
Received 18 November 2017; Received in revised form 19 January 2018; Accepted 7 March 2018
Available online 12 March 2018
2214-8604/ © 2018 Published by Elsevier B.V.
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

Fig. 1. Benchmark component model, all dimensions in mm.

visual aids and presentation models. However, they are important for benchmark component manufactured using a consumer grade FDM
prototyping models and components manufactured as part of an as- desktop printer.
sembly, which are aimed at appraising the fit and function of the dif-
ferent components. Engineers and designers would require information 2. Methods
relating to the assembly performance in terms of the achievable di-
mensional accuracy and tolerances. Currently, most of the FDM printer The FDM printer used in this work was the Makerbot Replicator 2X,
manufacturers, report their machine specifications in the form of axis from Makerbot Industries LLC, and 1.75 mm diameter ABS filament.
resolutions which relate to the positioning accuracy in the X- and Y- This FDM printer has a build volume measuring 246 × 152 × 155 mm
axes, and layer height in the Z-axis. However, translating these re- with positioning precision stated as 0.011 mm in the XY direction
solutions into the final geometric form and dimensional capability is (horizontal plane) and 0.025 mm in the Z direction (vertical plane).
not possible solely by using this information. Components were printed in the XYZ orientation at the center of the
A number of research studies have focused on the geometric cap- build platform [17].
ability of the FDM process. It has been found that, of the AM processes For this investigation, a benchmark component was designed to
currently available, the FDM was the least suitable process to manu- include basic geometric features of varying sizes. In order to get the
facture fine to medium sized features [12]. It was also reported that for geometric features free of additional artefacts, it was decided that the
feature dimensions smaller than 2.0 mm, the deviation from nominal features on the benchmark component should not require any addi-
dimension was larger and significant deviation was observed for cir- tional build supports (temporary structures which are used when
cular features [13]. Furthermore, it was found that deviations vary for printing structures with overhangs or unsupported features which can
internal features (holes) and external (bosses) under the same printing be removed after the printing process). Build supports may affect the
conditions [14]. Based on a U-shaped test part with a hole, it was found measurement for dimensional accuracy if these supports are not re-
that dimensions in the X-Y plane were undersized while oversized in the moved properly. The component was designed so that measurements
Z-direction [15]. In another investigation, it was found that deviations for dimensional accuracy and geometrical characteristics; form, or-
in the x-direction were oversized, they were larger for feature sizes of ientation and location [18], can be performed on the benchmark
more than 50 mm, while in the y-direction, the deviations were nega- component. The parameters to be considered in this investigation were
tive values for feature sizes of more than 30 mm. In the Z-direction, identified, based on the user parameters available in the printer’s user
dimensions vary greatly across all nominal height ranging from 3 to interface, and a suitable design of experiment method was adopted. The
100 mm [16]. benchmark components were printed according to the experimental
For most of the earlier research, the effects on the geometric prop- plan, measurement results were analyzed and critical parameters af-
erties were conducted by varying only a few parameters available such fecting both the dimensional accuracy and geometric characteristics
as layer thickness, contour width and nozzle speed. It has been shown were identified. The printing conditions which optimizes both dimen-
that deviations between nominal and measured dimensions on a printed sional accuracy and geometric characteristics of the printed component
component can be affected by varying any of the available printing were also found based on the results obtained.
parameters. However, to fully understand the effects of parameter
setting variations on the geometrical properties, all parameters avail-
3. Benchmark component
able in FDM 3D printer hardware and software should be considered.
This paper focuses on the FDM process in context of the final geo-
Benchmark components used in earlier research varied in sizes
metric shape characterization and optimization. This article in-
while incorporating elements ranging from simple geometric features to
vestigates the effects of varying parameter settings on the geometrical
highly complex structures which for some, required the use of build
properties; dimensional accuracy and geometrical characteristics of a
supports in the printing process. The benchmark component proposed

184
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

in this article is a more complete design as investigated by Mahmood Table 2

et al. [19]. The proposed design (Fig. 1) consists of a square base which L27 Orthogonal array, level control for each experiment.
allows for straightness measurements of length, for features along the
Parameters from Table 1
X-axis, and width, for features along the Y-axis with the edges of the
base used as datum for perpendicularity and parallelism measurements Expt A B C D E F G H J K L M N
and the surface of the base was used to measure flatness.
1 1 1 1 1 1 1 1 1 1 1 1 1 1
The geometric features on the square base are designed with sizes of
2 1 1 1 1 2 2 2 2 2 2 2 2 2
5, 10 and 20 mm with protruding (bosses) and depressed (holes) 3 1 1 1 1 3 3 3 3 3 3 3 3 3
structures from the surface. The effect on the geometrical properties of 4 1 2 2 2 1 1 1 2 2 2 3 3 3
the printed components due to the variation in feature sizes, within the 5 1 2 2 2 2 2 2 3 3 3 1 1 1
same component, can be observed by the inclusion of features with 6 1 2 2 2 3 3 3 1 1 1 2 2 2
7 1 3 3 3 1 1 1 3 3 3 2 2 2
different dimensions. Geometric features are duplicated at two different
8 1 3 3 3 2 2 2 1 1 1 3 3 3
locations on a same plane on the benchmark component to allow for 9 1 3 3 3 3 3 3 2 2 2 1 1 1
repeatability measurements. The circular and square bosses allow for 10 2 1 2 3 1 2 3 1 2 3 1 2 3
the measurements of diameter (Ø), length (l), width (w) and thickness 11 2 1 2 3 2 3 1 2 3 1 2 3 1
12 2 1 2 3 3 1 2 3 1 2 3 1 2
(h). The circular bosses also allow for the measurement of roundness
13 2 2 3 1 1 2 3 2 3 1 3 1 2
while concentricity can be measured using the circular stepped features 14 2 2 3 1 2 3 1 3 1 2 1 2 3
[20]. The square holes and bosses allow measurements of perpendicu- 15 2 2 3 1 3 1 2 1 2 3 2 3 1
larity and parallelism with respect to the straight edges of the square 16 2 3 1 2 1 2 3 3 1 2 2 3 1
base while the central diagonal feature allows for angularity measure- 17 2 3 1 2 2 3 1 1 2 3 3 1 2
18 2 3 1 2 3 1 2 2 3 1 1 2 3
ment.
19 3 1 3 2 1 3 2 1 3 2 1 3 2
20 3 1 3 2 2 1 3 2 1 3 2 1 3
21 3 1 3 2 3 2 1 3 2 1 3 2 1
4. Design of experiment
22 3 2 1 3 1 3 2 2 1 3 3 2 1
23 3 2 1 3 2 1 3 3 2 1 1 3 2
To understand the effects of varying parameter settings on the 24 3 2 1 3 3 2 1 1 3 2 2 1 3
geometrical properties of a printed component, an evaluation of the 25 3 3 2 1 1 3 2 3 2 1 2 1 3
control parameters that may influence the dimensional accuracy and 26 3 3 2 1 2 1 3 1 3 2 3 2 1
27 3 3 2 1 3 2 1 2 1 3 1 3 2
geometrical characteristics of the benchmark component was per-
formed. For the 3D printer used in this research, a total of 13 para-
meters are available and are considered for this investigation. The Taguchi’s method involves the use of orthogonal array (OA), which
parameter settings for this printer are easily changed via the printer chooses the control level combination of the parameter for each ex-
user interface. Each of the parameters considered was assigned three periment. Based on Taguchi’s experimental design, a L27 OA was re-
levels of control as shown in Table 1. quired to investigate the effects of 13 parameters, each with three levels
Given the numerous parameters involved, and the experimental of control. The L27 OA consists of 27 rows, representing 27 experiments,
nature of the work, a methodology that results in identification of the each a unique combination of the control levels for the different para-
significance of the parameter while minimizing the number of experi- meters (Table 2).
ments needed was required. Several statistical and heuristic approaches All 27 benchmark components were printed based on the experi-
exist for parameter ranking, modelling, and optimization. Taguchi’s mental design as shown in Table 2. On average, each benchmark
parametric design of experiment (DOE) method is a statistical method component required about one hour and a half to print. The build
for identification and optimization of parameter, significance and sen- platform was allowed to cool sufficiently before the component was
sitivity. Taguchi’s DOE methods have been widely used in various removed.
process optimization and product design studies [21–23]. Other The features on the benchmark components were measured using a
methods include semi parametric methods such as Automated Neural coordinate measuring machine (CMM). The CMM was programmed to
Networks (ANS) [24,25], geometric programming (GP) [26] and system perform the measurements automatically to avoid errors during the
identification (SI) [27]. Given the extensive experimental nature and measurement process which may occur when measured manually. Jigs
physical measurements required for assessing the dimensional and and fixtures were used to secure the benchmark components in place to
geometric variation, the Taguchi’s DOE method was selected to reduce allow for repeatability and ease of measurement (Fig. 2).
the number of experiments required as well as to obtain the significance For the 27 components, all dimensions were measured for geometric
and sensitivity of the parameters [28]. features and then the dimensional deviation was calculated, i.e. the
difference between the nominal and the averaged values. The dimen-
Table 1 sional deviations together with the measurement results for geometric
Parameters and the assigned control levels.
characteristics were then used to calculate the Signal-to-Noise Ratio
Parameters Level 1 Level 2 Level 3 (SNR) and Means in order to evaluate the performance of the FDM
printer. The “Smaller-Is-Better” SNR response characteristics was used
A: Chamber temperature (°C) 35 45 55 to analyze the influence of the parameters on both dimensional accu-
B: Layer thickness (mm) 0.16 0.2 0.24
racy and geometric characteristic. The parameters were then ranked
C: Extruder temperature (°C) 207 218 230
D: Platform temperature (°C) 110 121 132 according to the delta values (delta value = maximum response –
E: Number of shells 1 2 3 minimum response) across all three levels of control, with the largest
F: Infill shell spacing multiplier 0.56 0.7 0.84 delta in the response to SN ratio representing the most sensitive para-
G: Inset distance multiplier 0.8 1.0 1.2 meter.
H: Floor/roof thickness (mm) 0.64 0.8 0.94
J: Infill pattern hexagonal linear diagonal
K: Infill density (%) 25 50 75 5. Results & analysis
L: Infill speed (mm/s) 72 90 108
M: Outline speed (mm/s) 24 32 40
N: Inset speed (mm/s) 54 72 90 The experimental results for dimensional accuracy and geometric
characteristics were recorded and analyzed.

185
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

Fig. 2. CMM measurement of benchmark component.

5.1. Dimensional accuracy mainly under-sized feature sizes. The same cannot be said for the boss
features, the interaction between the different parameter settings have
The calculated dimensional deviation represents the dimensional resulted in measurements which ranged from being undersized to
accuracy achievable via the FDM process; the smaller the deviation, the oversized.
better the machine’s performance. The average dimension, measure- Table 4 shows the parameter rankings, based on the delta values, of
ment range and deviation for each feature is shown in Table 3. This each feature on the benchmark component. The ranking for each
table represent the measurement results taken across all 27 compo- parameter are then added together, resulting in a cumulative sum of all
nents. the rankings. The parameters are then ranked again based on the cu-
The effects of parameter variation on the dimensional accuracy of mulative sum of all the individual rankings, with the least sum ranked
the printed components is summarized in Table 3. It can be seen that, the most significant while the most sum ranked the least significant.
overall, the deviation from nominal value increases with increasing It can be seen from Table 4 that number of shells (E), was the most
feature size and that the deviations are generally smaller for bosses as significant parameter affecting the accuracy of the printed benchmark
compared to holes for the same feature size. It was observed that, for component. This was followed by the inset distance multiplier (G); the
features of the same size, the deviation from nominal values was gen- space between adjacent shells, chamber temperature (A), infill shell
erally larger and significant for circular features [13]. The results from spacing multiplier (F); the amount of overlap between the innermost
this investigation are in agreement with the findings of [15] where it shell and the adjacent infill extrusion, and the infill density (K). These
was stated that dimensions for holes and features in the X-Y plane were five parameters were identified as the most sensitive parameters in
undersized. The overall averaged measurements for all the features determining the printing accuracy of the benchmark component.
were also smaller than the nominal values. This could be possibly due to Similarly, for the response to Means, the ranking of each parameter
the shrinkage of the ABS when it is heated [29]. Shrinkage has been for all geometric features are added together and the parameters were
shown to be more prevalent in the X-Y plane [30]. The results also show ranked according to the cumulative sum. The parameter ranking for the
the effects of varying parameter settings on the dimensional accuracy of response to Means is shown in Table 5.
the printed parts. For circular features, it was also found that the de- It can be seen from Table 5 that number of shells (E) was the most
viations for holes and bosses vary under the same printing conditions critical parameter affecting Means followed by inset distance multiplier
[14]. For hole features, varying the parameter settings resulted in (G), chamber temperature (A), infill shell spacing multiplier (F) and

Table 3
Measurement results from benchmark components (dimensions in mm).

Feature Nominal Holes Bosses

Average Range Deviation Average Range Deviation

diameter, Ø 5 4.77 4.41–4.97 0.23 4.85 4.48–5.16 0.15

10 9.63 9.23–9.84 0.37 9.86 9.68–10.15 0.14
20 19.53 19.23–19.74 0.47 19.77 19.47–20.14 0.23
length, l 5 4.83 4.49–4.96 0.17 5.02 4.87–5.42 0.02
10 9.71 9.37–9.89 0.29 9.92 9.66–10.16 0.08
20 19.69 19.33–19.88 0.31 19.79 19.54–20.09 0.21
80 – – – 78.89 78.42–79.28 1.11
width, w 5 4.76 4.47–4.96 0.24 4.97 4.82–5.41 0.03
10 9.74 9.42–9.90 0.26 9.92 9.78–10.19 0.08
20 19.72 19.38–19.99 0.28 19.79 19.61–20.07 0.21
80 – – – 79.05 78.63–79.56 0.95
thickness, h 2 – – – 1.98 1.60–2.30 0.02
4 – – – 4.04 3.68–4.38 0.04

186
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

Table 4
Parameter ranking for SNR (dimensional accuracy).

Feature Parameters from Table 1

A B C D E F G H J K L M N

Diameter CH.5 2 9 6 12 1 8 4 5 11 13 10 7 3
CH.10 6 3 7 11 1 4 2 10 12 5 8 13 9
CH.20 3 9 4 6 2 10 1 13 12 5 11 7 8
CB.5 7 12 9 8 3 6 1 11 2 13 10 4 5
CB.10 5 3 13 6 2 12 1 10 11 9 4 8 7
CB.20 6 10 5 3 2 7 1 13 9 4 11 12 8
Length SH.5 2 8 7 9 3 4 1 5 12 11 6 13 10
SH.10 5 3 4 6 2 7 1 11 13 9 8 12 10
SH.20 2 12 9 13 3 6 1 8 7 5 11 4 10
SB.5 3 13 8 5 1 4 6 9 10 11 7 2 12
SB.10 12 9 3 6 1 10 8 11 4 2 7 5 13
SB.20 5 8 3 4 2 12 1 11 10 6 9 13 7
SB.80 8 3 12 5 2 7 1 9 10 4 11 13 6
Width SH.5 2 9 10 6 3 5 1 12 8 11 13 4 7
SH.10 8 5 4 9 2 7 1 13 11 3 12 10 6
SH.20 3 10 11 5 2 7 1 6 12 4 8 13 9
SB.5 4 6 8 5 2 3 13 11 10 12 7 1 9
SB.10 2 9 7 8 4 3 1 12 11 5 13 6 10
SB.20 6 7 4 9 2 5 1 8 13 3 12 10 11
SB.80 8 4 10 5 1 6 12 11 3 7 13 9 2
Height Z.2 8 4 10 5 1 6 12 11 3 7 13 9 2
Z.4 2 1 3 13 10 5 6 12 11 8 9 7 4
Sum 109 157 157 159 52 144 77 222 205 157 213 182 168
Overall Ranking 3 7 6 8 1 4 2 13 11 5 12 10 9

extruder temperature (C). The three most critical parameters identified Table 7 summarizes the print condition to achieve optimum di-
in the cumulative response to Means were similar to the cumulative mensional accuracy for the benchmark component used in this in-
response for SNR ranking. vestigation.
From the results for the response to Means (Table 5), the printing
conditions to produce a component with the least deviation can be
5.2. Geometric characteristics
found. Table 6 shows the control level combination optimized for
smallest deviation from nominal dimension of each geometrical feature.
The similar analysis approach for dimensional accuracy was used to
For each parameter, the frequency of the control levels were
evaluate the machine performance for the geometrical characteristics.
counted and the overall optimum combination for printing was found
The ranking for all the parameters across all the geometrical char-
by identifying the most common control level for each parameter across
acteristics was calculated and the overall ranking was found based on
all the geometric features.
the cumulative ranking.

Table 5
Parameter ranking for Means (dimensional accuracy).

Feature Parameters from Table 1

A B C D E F G H J K L M N

Diameter CH.5 2 13 10 9 1 4 3 6 7 12 11 8 5
CH.10 5 3 7 11 1 4 2 9 13 6 8 12 10
CH.20 3 6 4 7 2 5 1 13 12 9 11 10 8
CB.5 9 10 11 6 5 4 1 13 2 8 12 7 3
CB.10 3 4 12 7 2 9 1 13 11 5 6 10 8
CB.20 5 11 6 3 2 8 1 12 9 4 10 13 7
Length SH.5 2 10 13 8 3 4 1 5 12 11 6 7 9
SH.10 5 6 4 7 2 3 1 11 12 10 9 13 8
SH.20 3 11 6 13 2 5 1 7 10 8 12 4 9
SB.5 4 11 9 6 1 3 2 7 12 10 8 5 13
SB.10 9 10 3 11 1 8 5 12 4 2 7 6 13
SB.20 6 10 5 4 2 11 1 13 8 3 9 12 7
SB.80 9 4 11 3 2 6 1 8 10 5 13 12 7
Width SH.5 2 7 13 5 3 4 1 9 12 10 11 6 8
SH.10 5 8 3 11 2 4 1 13 9 6 12 10 7
SH.20 3 9 10 7 2 4 1 6 13 5 11 12 8
SB.5 4 7 13 5 1 2 11 6 12 10 8 3 9
SB.10 4 13 6 10 2 5 1 8 12 3 9 7 11
SB.20 5 7 3 10 4 6 1 9 11 2 8 13 12
SB.80 11 10 4 7 2 9 6 5 3 12 8 13 1
Height Z.2 11 10 4 7 2 9 6 5 3 12 8 13 1
Z.4 2 1 3 13 10 5 6 12 11 8 9 7 4
Sum 112 181 160 170 54 122 55 202 208 161 206 203 168
Overall Ranking 3 9 5 8 1 4 2 10 13 6 12 11 7

187
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

Table 6
Optimum control level combination for dimensional accuracy.

Feature Parameters from Table 1

A B C D E F G H J K L M N

Diameter CH.5 1 1 2 1 1 1 2 3 1 3 1 2 3
CH.10 1 1 2 3 1 3 3 2 1 2 3 1 2
CH.20 1 1 2 2 1 1 3 2 3 2 3 2 2
CB.5 3 3 3 3 3 1 1 2 3 3 3 1 1
CB.10 1 2 3 2 3 1 1 1 3 3 2 1 3
CB.20 2 2 3 1 2 2 1 2 3 3 2 1 3
Length SH.5 1 2 1 1 2 1 3 3 1 2 1 2 1
SH.10 1 1 2 1 1 3 1 2 1 2 1 1 2
SH.20 1 1 2 1 1 1 3 2 2 2 1 3 1
SB.5 1 1 1 2 2 3 3 3 3 1 3 1 1
SB.10 1 1 1 1 2 1 1 3 3 3 2 2 3
SB.20 3 2 3 1 2 2 1 3 3 3 2 1 3
SB.80 2 3 3 2 1 2 1 1 1 2 1 1 2
Width SH.5 1 2 1 1 2 3 3 2 3 1 1 2 3
SH.10 1 1 2 1 1 1 3 2 1 2 1 1 2
SH.20 1 1 2 3 1 1 3 2 3 2 1 3 2
SB.5 1 1 1 2 2 1 3 3 3 1 3 1 1
SB.10 2 3 1 2 2 1 1 3 3 3 2 2 3
SB.20 2 3 3 2 3 2 1 3 3 3 2 1 3
SB.80 3 3 3 1 3 2 1 1 1 3 2 1 2
Height Z.2 3 2 2 2 2 1 1 3 1 3 3 3 2
Z.4 1 1 3 1 2 1 3 2 2 3 2 2 1
Sum Level 1 14 11 6 11 8 13 11 3 8 3 8 12 6
Level 2 4 6 7 8 10 5 0 10 2 8 8 6 8
Level 3 4 5 8 2 4 3 9 9 12 11 5 3 7
Optimum Combination 1 1 3 1 2 1 1 2 3 3 1 1 2

Table 7 Similarly, from the response table for Means (Table 9), the most
Optimum print condition for dimensional accuracy. significant parameter affecting the geometric characteristics were layer
thickness (B), infill speed (L), infill shell spacing multiplier (F), number
Parameter Control Level Parameter Setting
of shells (E) and extruder temperature (C). It can be seen that from both
A: Chamber temperature (°C) 1 35 response tables for SNR (Table 8) and Means (Table 9), four parameters
B: Layer thickness (mm) 1 0.16 are common; layer thickness, extruder temperature, number of shells
C: Extruder temperature (°C) 3 230 and infill speed, in terms of the critical effects on the geometric char-
D: Platform temperature (°C) 1 110
E: Number of shells 2 2
acteristics of the printed components.
F: Infill shell spacing multiplier 1 0.56 The optimum control level combination for each individual geo-
G: Inset distance multiplier 1 0.8 metrical characteristic is shown in Table 10. The overall control level
H: Floor/roof thickness (mm) 2 0.8 combination optimized for geometrical characteristics is found by
J: Infill pattern 3 diagonal
identifying the most common control level for each individual para-
K: Infill density (%) 3 75
L: Infill speed (mm/s) 1 72 meter across all characteristics. The control level and its parameter
M: Outline speed (mm/s) 1 24 settings optimized for geometric characteristics is shown in Table 11.
N: Inset speed (mm/s) 2 72

6. Conclusion
Table 8 shows the SNR response for geometric characteristics, it can
be seen that layer thickness (B) was the most significant parameter, This paper presents and demonstrates a systematic methodology of
followed by the number of shells (E), extruder temperature (C), infill evaluating the achievable dimensional accuracy and geometric char-
speed (L) and floor/roof thickness (H). acteristics of a FDM machine. A comprehensive list of 13 parameters
were considered for this investigation. An experimental plan was

Table 8
Parameter ranking for SNR (geometric characteristics).

Parameters from Table 1

Characteristic A B C D E F G H J K L M N

Angularity 9 7 1 5 4 6 11 3 13 8 2 12 10
Flatness 11 7 3 5 8 4 13 12 6 1 2 9 10
Perpendicularity – X 3 1 10 8 5 6 12 7 9 11 2 13 4
Perpendicularity – Y 9 6 1 4 5 2 13 11 12 10 7 8 3
Parallelism – X 4 3 6 11 2 9 1 5 13 7 10 12 8
Parallelism – Y 2 4 11 10 1 7 3 5 12 9 13 6 8
Roundness 11 5 2 9 8 7 10 1 6 13 4 3 12
Concentricity 8 1 12 6 11 13 3 9 5 10 7 2 4
Sum 57 34 46 58 44 54 66 53 76 69 47 65 59
Overall Ranking 7 1 3 8 2 6 11 5 3 12 4 10 9

188
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

Table 9
Parameter ranking for Means (geometric characteristics).

Parameters from Table 1

Characteristic A B C D E F G H J K L M N

Angularity 6 3 2 8 5 4 12 9 11 7 1 13 10
Flatness 12 8 4 6 5 3 13 10 9 1 2 7 11
Perpendicularity – X 2 1 7 5 9 4 12 6 11 13 3 8 10
Perpendicularity – Y 7 3 1 9 5 2 13 4 11 12 6 10 8
Parallelism – X 4 7 6 13 2 5 1 3 11 9 8 12 10
Parallelism – Y 1 4 13 11 3 9 2 5 10 8 12 6 7
Roundness 11 6 2 5 10 9 7 1 8 13 4 3 12
Concentricity 7 4 13 5 8 10 2 12 3 11 9 1 6
Sum 50 36 48 62 47 46 62 50 74 74 45 60 74
Overall Ranking 7 1 5 10 4 3 9 6 13 12 2 8 11

Table 10 the system. This paper identifies the critical parameters affecting both
Optimum control levels for geometric characteristics. the dimensional accuracy and geometric characteristics of the printed
components and the print condition to achieve optimum dimensional
Parameters from Table 1
accuracy and geometrical characteristics.
Characteristic A B C D E F G H J K L M N The research work presented provides the optimized results for both
the dimensional accuracy and geometric characteristics. However, it is
Angularity 2 2 1 2 1 1 2 2 3 1 2 1 1 limited to only linear and circular features. This work can further ex-
Flatness 2 3 3 2 3 2 3 2 3 3 3 2 2
plore the effects on more complex features such as overhangs, gradients
Perpendicularity – X 2 2 1 3 2 3 2 1 2 1 1 1 3
Perpendicularity – Y 3 2 1 1 2 1 2 3 3 1 3 1 3 and curvature. Based on the results of this experiment, future studies
Parallelism – X 1 1 1 2 2 1 2 1 3 2 2 2 3 can incorporate complex features such as overhangs, gradients and
Parallelism – Y 2 2 1 2 1 1 1 3 1 1 2 1 2 curvatures. The benchmark component has been designed to provide
Roundness 3 1 1 1 2 3 1 2 1 1 3 3 1 repeatability measurements across X-Y plane only. This results into bi-
Concentricity 1 1 2 2 1 3 2 2 1 3 3 2 3
Sum Level 1 2 3 6 2 3 4 2 2 3 5 1 4 2
planar repeatability. However, in order to achieve tri-planar repeat-
Level 2 4 4 1 5 4 1 5 4 1 1 3 3 2 ability, further tests with the test part rotated at 90° to the X-Y plane
Level 3 2 1 1 1 1 3 1 2 4 2 4 1 4 should be conducted. Furthermore, the applicability of the current work
Optimum Combination 2 2 1 2 2 1 2 2 3 1 3 1 2 is limited to the parameter envelope of Table 1 and any changes to
values beyond these envelopes should be investigated via separate DOE.
Table 11
Optimum print condition for geometric characteristics.
Acknowledgement

Parameter Control Level Parameter Setting The authors would like Mr Liang Hao Jie at Nanyang Polytechnic for
manufacturing the benchmark components and performing the mea-
A: Chamber temperature (°C) 2 45
surements.
B: Layer thickness (mm) 2 0.2
C: Extruder temperature (°C) 1 207
D: Platform temperature (°C) 2 121 References
E: Number of shells 2 2
F: Infill shell spacing multiplier 1 0.56
[1] W.A. Kalender, R. Petzold, H.-F. Zeilhofer, Rapid prototyping technology in medi-
G: Inset distance multiplier 2 0.8 cine—basics and applications, Comput. Med. Imaging Graph. 23 (1999) 277–284,
H: Floor/roof thickness (mm) 2 0.8 http://dx.doi.org/10.1016/S0895-6111(99)00025-7.
J: Infill pattern 3 diagonal [2] B. Dabrowski, W. Swieszkowski, D. Godlinski, K.J. Kurzydlowski, Highly porous
K: Infill density (%) 1 25 titanium scaffolds for orthopaedic applications, J. Biomed. Mater. Res. – Part B
L: Infill speed (mm/s) 3 108 Appl. Biomater. 95 (2010) 53–61, http://dx.doi.org/10.1002/jbm.b.31682.
M: Outline speed (mm/s) 1 24 [3] D. Thomas, G. Sabin, Additive manufacturing for the aerospace industry, American
N: Inset speed (mm/s) 2 72 Helicopter Society 68th Annual Forum, Fort Worth, Texas, 1–3.05.2012, 2012, pp.
1876–1888.
[4] T. Ferreira, P.J. Bártolo, I. Campbell, Additive manufacturing in jewellery design,
ASME 2012 11th Biennial Conference on Engineering Sysems Design and Analysis,
developed using Taguchi’s design of experiment method and a custo- Nantes, France, 2–4.06.2012, 2012, pp. 1–8.
mized benchmark component was designed. Twenty-seven benchmark [5] S. Negi, S. Dhiman, R.K. Sharma, Basics and applications of rapid prototyping
components were printed based on the experimental plan and mea- medical models, Rapid Prototyp. J. 20 (2014) 256–267.
[6] A.K. Misra, J.E. Grady, R. Carter, Additive manufacturing of aerospace propulsion
surements were performed on the components. The significance and components, Additive Manufacturing for Small Manufacturers Conference,
ranking of these parameters was established. Pittsburgh, PA, 01.10.2015, 2015.
The measurement results have shown that, for dimensional accu- [7] F. Abdullah, Fused deposition modeling (FDM) mechanism, Int. J. Sci. Eng. Res. 7
(2016) 41–43.
racy, the deviation from nominal values increases with increasing fea- [8] J. Mireles, D. Espalin, D. Roberson, B. Zinniel, F. Medina, R. Wicker, Fused de-
ture sizes for all geometric features and the deviation from nominal position modeling of metals, International SFF Symposium (2012) 836–845.
dimensions are larger for recessed features (holes) as compared to ex- [9] E. Vahabli, S. Rahmati, Improvement of FDM parts’ surface quality using optimized
neural networks – medical case studies, Rapid Prototyp. J. 23 (2017) 825–842,
truded features (bosses) in the X-Y plane. Furthermore, measurement
http://dx.doi.org/10.1108/RPJ-06-2015-0075.
results for recessed features were all undersized whereas for extruded [10] P.K. Garg, R. Singh, I. Ahuja, Multi-objective optimization of dimensional accuracy,
features the results ranged from undersize to oversize. surface roughness and hardness of hybrid investment cast components, Rapid
Prototyp. J. 23 (2017) 845–857, http://dx.doi.org/10.1108/RPJ-10-2015-0149.
This investigation addresses both, the dimensional variation as well
[11] A. Armillotta, S. Bianchi, M. Cavallaro, S. Minnella, Edge quality in fused deposition
as the geometric variation of the features on a benchmark component. modelling: II. Experimental verification, Rapid Prototyp. J. 23 (2017) 686–695,
Geometric form and tolerances have a significant impact on the as- http://dx.doi.org/10.1108/RPJ-02-2016-0021.
sembly constraints and define the proper assembly and functioning of [12] M. Mahesh, Y.S. Wong, J.Y.H. Fuh, H.T. Loh, Benchmarking for comparative

189
S. Mahmood et al. Additive Manufacturing 21 (2018) 183–190

evaluation of RP systems and processes, Rapid Prototyp. J. 10 (2004) 123–135, production of flexible ABS object, J. Mater. Process. Technol. 169 (2005) 54–61,
http://dx.doi.org/10.1108/13552540410526999. http://dx.doi.org/10.1016/j.jmatprotec.2005.02.259.
[13] N.S.A. Bakar, M.R. Alkahari, H. Boejang, Analysis on fused deposition modelling [22] H.-G. Beyer, B. Sendhoff, Robust optimization – a comprehensive survey, Comput.
performance, J. Zhejiang Univ. 11 (2010) 972–977, http://dx.doi.org/10.1631/ Methods Appl. Mech. Eng. 196 (2007) 3190–3218, http://dx.doi.org/10.1016/j.
jzus.A1001365. cma.2007.03.003.
[14] F. Etesami, T. Griffin, Characterizing the accuracy of FDM rapid prototyping ma- [23] G. Taguchi, S. Chowdhurry, Y. Wu, Taguchi’s Quality Engineering Handbook, Wiley
chines for machine design applications, Syst. Des. 12 (2013) 12, http://dx.doi.org/ & Sons, 2005.
10.1115/IMECE2013-64972. [24] A. Garg, V. Vijayaraghavan, J. Zhang, J.S.L. Lam, Robust model design for eva-
[15] M.N. Islam, B. Boswell, A. Pramanik, An investigation of dimensional accuracy of luation of power characteristics of the cleaner energy system, Renew. Energy 112
parts produced by three-dimensional printing, Proceedings of the World Congress (2017) 302–331, http://dx.doi.org/10.1016/j.renene.2017.05.041.
on Engineering (WCE2013), I (2013) 5–8. [25] A. Garg, V. Vijayaraghavan, J. Zhang, S. Li, X. Liang, Design of robust battery ca-
[16] F. Knoop, V. Schoeppner, Analysis and optimization of the dimensional accuracy for pacity model for electric vehicle by incorporation of uncertainties, Int. J. Energy
FDM parts manufactured with ABS-M30, ASPE 2105 Spring Topical Meeting 60 Res. 41 (2017) 1436–1451, http://dx.doi.org/10.1002/er.3723.
(2015). [26] A. Garg, J. Li, J. Hou, C. Berretta, A. Garg, A new computational approach for
[17] ISO/ASTM 52921, Standard Terminology for Additive Manufacturing – Coordinate estimation of wilting point for green infrastructure, Measurement 111 (2017)
Systems and Test Methodologies, (2013) Geneva, Switzerland. 351–358, http://dx.doi.org/10.1016/j.measurement.2017.07.026.
[18] ISO 1101:2005, Geometrical Product Specifications (GPS)—Geometrical [27] Y. Huang, L. Gao, Z. Yi, K. Tai, P. Kalita, P. Prapainainar, A. Garg, An application of
tolerancing—Tolerances of Form, Orientation, Location and Run-out, (2005) evolutionary system identification algorithm in modelling of energy production
Geneva, Switzerland. system, Measurement 114 (2018) 122–131, http://dx.doi.org/10.1016/j.
[19] S. Mahmood, D. Talamona, K.L. Goh, A.J. Qureshi, Fast deviation simulation for measurement.2017.09.009.
fused deposition modelling process, 14th CIRP CAT 2016 – CIRP Conference [28] Raghu N. Kacker, Eric S. Lagergren, J.J. Filliben, Taguchi’s orthogonal arrays are
Computer Aided Tolerancing, Procedia CIRP, Gotherburg, Sweden, 2016, pp. classical designs of experiments, J. Res. Nat. Inst. Stand. Technol. 96 (1991) 577,
327–332, , http://dx.doi.org/10.1016/j.procir.2016.02.004. http://dx.doi.org/10.6028/jres.096.034.
[20] S. Moylan, J. Slotwinski, A. Cooke, K. Jurrens, M.A. Donmez, Proposal for a stan- [29] A. Gregorian, B. Elliot, R. Navarro, Accuracy improvement in rapid prototyping
dardized test artifact for additive manufacturing machines and processes, Solid machine (FDM-1650), Solid Freeform Fabrication Proceesings (2001) 77–84.
Freeform Fabrication Symposium Proceedings (2012) 902–920. [30] A.K. Sood, R.K. Ohdar, S.S. Mahapatra, Grey Taguchi method for improving di-
[21] B.H. Lee, J. Abdullah, Z.A. Khan, Optimization of rapid prototyping parameters for mensional accuracy of FDM process, Mater. Des. 30 (2009) 4243–4252.

190