Professional Documents
Culture Documents
https://doi.org/10.1007/s00170-019-04568-3
ORIGINAL ARTICLE
Received: 24 July 2019 / Accepted: 4 October 2019 / Published online: 18 November 2019
# Springer-Verlag London Ltd., part of Springer Nature 2019
Abstract
Additive manufacturing (AM) technologies such as fused deposition modeling (FDM) have been widely used in various fields of
today’s manufacturing industries such as transportation, aerospace, and medical because of its ability to produce parts of complex
designs with less manufacturing time and cost. However, a proper selection of input process parameters is a vital aspect in order to
obtain the best quality of the printed part. This paper presents several approaches, namely response surface methodology, particle
swarm optimization, and symbiotic organism search, to find the optimal parameter settings for better surface quality, i.e., surface
roughness of the FDM printed part. Layer height, print speed, print temperature, and outer shell speed were considered as the
input parameters and surface roughness as the output response. The experimental design was carried out using response surface
methodology (RSM) method. Then, the relationship between the input parameters and the surface roughness was established
using regression model. Once the accuracy of the model had been validated, the model was then coupled with particle swarm
optimization (PSO) and symbiotic organism search (SOS) to optimize the input parameters that would lead to minimum surface
roughness. Experimental results showed that the surface roughness obtained using PSO and SOS have improved about 8.5% and
8.8%, respectively, compared with the conventional method, i.e., RSM. A good agreement between the predicted surface
roughness and the experimental values was also observed.
Keywords Fused deposition modeling (FDM) . Surface roughness . Particle swarm optimization . Symbiotic organism search .
Response surface methodology
at the same time adopting the parameters of the selected range. the layer thickness had more affect on the surface roughness
High product quality, low cost, and a fast production process whereby when the layer thickness was increased, the surface
are precedents of goals that should be accomplished. To meet roughness increased as well. In another study by Chung Wang
at least one target, the best set of parameters needs to be ob- et al. [17], the Taguchi method was used to optimize the fused
tained. There are several optimizations that have been done on deposition modeling (FDM) process in terms of the ultimate
3D printer such as dimensional accuracy [6], mechanical tensile strength, dimension accuracy, and surface roughness.
properties [7], building time [8], and also surface roughness The selected parameters for this study were layer thickness,
[9]. deposition orientation, deposition style, support style, and
Among the tangible problems that occur are the flexural build location. Based on the experimental result, it was ob-
strength of the product, the surface roughness, and dimension- served that the layer thickness had the greatest affect on the
al accuracy. However, the obvious problem that will under- surface roughness of the sample. Jiang et al. [18] examined the
mine perception of the first user is the surface quality of the effect of process parameters such as part orientation, layer
product. The rough surface of a product will affect the aesthet- thickness, and raster angle on surface roughness of FDM-
ic value of the product and indirectly reduce the selling rate of built parts. The experimental design was performed using re-
the product [10]. Logically, surface roughness can be classi- sponse surface methodology (RSM). Results showed that part
fied as a very important parameter in the 3D printing industry. orientation was the most prominent parameter influencing the
The quality of surface finish in the 3D printing industry be- surface roughness.
comes more important when multiple printed parts are used Based on the literature review for the surface roughness
for customized product applications. Good surface quality is optimization of 3D printer using conventional method, it can
important not only for functionality and appearance but also be clearly seen that the conventional method has already wide-
for cost-effectiveness and overall prototyping time reduction ly been used on 3D printer. Based on the previous studies, the
[11]. Therefore, in order to obtain the best surface quality, the most common methods used are Taguchi method, full factorial
optimum process parameters can be found through several method, and response surface method. For RSM, the method
optimization approaches such as conventional or nonconven- is not too popular and rarely used because it is time consuming
tional techniques. compared with the Taguchi method. However, Mohamed
For conventional optimization approach, various works et al. [5] stated that the RSM can be counted as an excellent
have been done to gain the optimal parameter settings of the method for optimization because it made very low standard
FDM 3D printer for the fabrication of 3D parts in order to errors on experimental validation. Furthermore, the RSM
achieve the best surface roughness. Peng and Yan [12] con- method can give a better solution for 3D printer optimization
ducted a study on the effect of process parameters to the mul- that has more than one response because of its ability to deal
tiple objective analysis which is surface roughness and energy with higher degrees of fitting models and multi-objectivity.
consumption for desktop FDM 3D printer using full factorial Nowadays, only a few nonconventional methods such as
design. Anitha et al. [13] experimentally obtained the optimal metaheuristic techniques have been substantially studied to
parameters (layer thickness, road width, and speed deposition) solve optimization problems in the FDM 3D printing process.
for the best surface quality for an ABS prototype using Some of the popular metaheuristic methods for solving sur-
Taguchi method. The study showed that a minimum layer face roughness optimization in FDM 3D printer are genetic
thickness will improve surface finish. Vijeth et al. [14] studied algorithms (GAs), nondominated sorting genetic algorithm II
the parameters that affect the surface roughness of truncheon (NSGA II), and teaching-learning-based optimization
design test artifacts. From the Taguchi statistical analysis, it (TLBO). Among them, GA is the most common algorithm
was found that build inclination and layer thickness signifi- that has been applied for surface roughness optimization.
cantly influenced the surface quality of the part. Nancharaiah Thrimurthulu et al. [19] used GA to investigate the optimum
et al. [15] studied the influence of the process parameters of part deposition orientation for surface roughness. The sample
layer thickness, raster angle, air gap, and road width on the was created with different build orientations and layer thick-
dimensional accuracy and surface quality. The experimental nesses. The expectation of the created sample was verified,
study was implemented using Taguchi method where three and it was in great concurrence with the outcome distributed
levels for each factor had been set. It was observed that layer before. This study revealed that the created model can be
thickness and road width were the most influential parameters utilized to anticipate the optimum part orientation for any
for surface roughness compared to the raster angle and air gap. complex freeform surfaces. In another study, Asadollahi
Horvath et al. [16] conducted two sets of experiments using Yazdi et al. [20] employed NSGA II to search for the optimum
full factorial method to minimize the surface roughness of the layer thickness and orientation angle by minimizing the pro-
samples. The sample material used was ABS 400 polymer. duction time and material mass with the constraint of surface
The three parameters involved in the study were layer thick- roughness and mechanical strength of FDM products. Layer
ness, model temperature, and part fill style. It was found that thickness and part orientation were used as the input
Int J Adv Manuf Technol (2019) 105:5121–5137 5123
parameters. Experimental results showed that the surface minimize the surface roughness of the FDM 3D part. The
roughness could be improved by decreasing the layer thick- objective was to obtain the optimum parameters of FDM 3D
ness. Ghetiya and Patel [21] investigated the effect of layer printer for good surface quality. The experiments began with
thickness, flow rate, and feed rate on the surface roughness of the regression model development between the process factors
test specimen (implant replica). ABS plus plastic was used as and output response (surface roughness) using central com-
a material to match with the application of the test specimen of posite design (CCD) method of RSM. The regression model
this study which is for biomedical implants. TLBO method served as the prediction model for the surface roughness
was used to achieve good surface roughness simultaneously which in turn became the objective function while the lower
for all responses. Experimental validation had found that min- and upper limits became the constraints. Furthermore, the ob-
imum surface roughness can be achieved by lowering the feed jective function (i.e., surface roughness model) was evaluated
rate and increasing the flow rate of the FDM process. Pandey in PSO and SOS optimization methods to obtain the optimal
et al. [22] performed a study about the effect of the optimal parameters that lead to the minimum surface roughness.
part deposition on the surface roughness and build time. The Results obtained from different optimization techniques were
optimization method used was multi-criteria GA. It was found compared in order to look for the best minimum surface
that the orientation for part deposition significantly affected roughness. Finally, experimental confirmation were per-
the surface roughness and build time of the part produced. In formed to validate the efficiency of the propose methods and
another study by Byun and Lee [23], the optimal part orienta- the error between the predicted surface roughness and the
tion which was believed to be able to improve surface rough- experimental values were observed.
ness was investigated. GA method was used to search for the
optimal orientation for better surface roughness. From the
analysis, part orientation was the most significant factor that 2 Methodology
affected the surface roughness of the printed part.
Based on the literature review for the surface roughness Figure 1 indicates the workflow of the proposed methodology.
optimization of 3D printer using nonconventional method, it This study begins with 3D CAD modeling design of the spec-
can be seen that the meta-heuristic method is still at the early imen that is developed using CAD software (CATIAV5). The
stages and needs to be emphasized in the 3D printing indus- CAD drawing is converted into standard tessellation language
tries. Furthermore, particle swarm optimization (PSO) and (STL) file. This file is used to orientate the model for the
symbiosis organism search (SOS) methods have not yet prov- building process and mathematical slicing that will be printed
en its performance in the optimization of surface roughness in layer-by-layer by the FDM machine. Then, the conversion of
FDM 3D printer. Thus, the optimization of surface roughness the 3D model into printing instruction code is done using the
on 3D printer by using PSO and SOS method can be consid- Ultimater Cura 4.0 software. This software has several fea-
ered as new approaches whereby both methods offer great tures such as to modify the process parameters setting, infill
robustness with better and fast convergence towards the global density of each layer, calculate the tool paths, and to regulate
optimal solution. Both methods have shown their capability in the heated printer extruder head of the FDM 3D machine
other fields of practical optimization problems, i.e., engineer- through its USB port or SD card port for the specimen.
ing structures, power systems (Flexible AC Transmission Once the printing process is completed, the surface roughness
System (FACTS)), human resources management, transporta- of the specimen will be measured using contact-type surface
tion, etc. [24–27]. In manufacturing areas such as machining roughness machine.
and rapid prototyping, PSO has been widely applied for pro- Next, the experimental plan is conducted using response
cess optimization and proved to be efficient and effective. In surface methodology. With this method, a sequence of exper-
contrast, no studies have been conducted using SOS in the imental runs will be performed to obtain the optimum re-
field of machining and rapid prototyping. Since the SOS al- sponse. At this stage, the experiments are carried out accord-
gorithm is notably new, the capability of SOS optimization ing to the number of runs suggested from the experimental
approach in solving the surface roughness of FDM 3D part design method, i.e., DOE. The surface roughness for all the
is very compelling to be further explored and investigated. samples is measured using surface roughness testing machine
The novelty of this study lies in the newly approach of (SV-C4500 Mitutoyo Formtracer. The surface roughness (Ra)
FDM optimization based on the recently developed SOS and corresponding to the each parameter setting is recorded.
PSO algorithms specifically in the surface quality of the FDM Design Expert software is used to analyze the relationship
part. Both methods are more robust, simple, fast convergence between the surface roughness and the input parameters (i.e.,
towards the global optimal solution and ability to escape from layer height, print speed, print temperature, and outer shell
local optima compared to the conventional optimization speed) and to obtain the regression model for the output re-
methods. In this work, three different optimization methods, sponse. The accuracy of the predicted model is determined
i.e., RSM, PSO, and SOS, were carried out in order to based on the statistical analysis using analysis of variance
5124 Int J Adv Manuf Technol (2019) 105:5121–5137
(ANOVA) method where the significance of the input param- 3 Experimental work
eters, lack of fit, and the coefficient of determination (R2) can
be observed. The coefficient of determination (R2) is impor- In the experimental study, all the samples were printed using
tant to be known in order to check the fitness of the regression 3D printer machine (Tarantula Prusa I3). The overall specifi-
model to the experimental data. cation for the FDM 3D printer is shown in Table 1. The print-
After a good model of the process is obtained, the optimi- ing material used for the model was polylactic acid (PLA)
zation process will be conducted to get the optimal value of which is a biodegradable thermoplastic polymerized from nat-
FDM 3D printer parameters that can produce the minimum ural sources. It can be prototyped fast, is cost-effective, and
surface roughness. The optimization process is performed capable of printing complex structures. Table 2 shows the
using conventional method (RSM), and two popular nature- material properties for PLA. The microstructures of the spec-
inspired algorithms: PSO and SOS. The regression model of imens and fracture surfaces were determined using a digital
the surface roughness will become the objective/fitness func- microscope (Micro Topper), and the surface roughness of the
tion for PSO and SOS. Simulation results obtained from all the specimen was measured using a SV-C4500 Mitutoyo
optimization techniques is analyzed and compared. Finally, Formtracer. In order to minimize the error, the three measure-
the experimental confirmation test is performed to validate ment points were averaged to represent the surface roughness
the efficiency of the proposed algorithms. The optimization of each specimen.
techniques that can produce the minimum surface roughness In this study, the ASTM D790 and ISO178 were applied
will be considered as the best optimization approach. for testing flexural samples. As shown in Fig. 2, the sample
Int J Adv Manuf Technol (2019) 105:5121–5137 5125
size for ASTM is 3.2 mm × 12.7 mm × 125 mm. The 3D much different from each other with a range between 2.3 and
printed sample geometry was modeled using CATIA software 3.9 μm.
exported as a STL file and imported into the 3D printing
software.
5 RSM
Xi þ X j
main advantages, i.e., able to find the best global optima re- Mutual Vector ¼ ð6Þ
gardless of the initial setting of its parameter values, uses 2
fewer control parameters, and fast convergence rate to reach BF is the benefit factor of either 1 or 2.
for the final value. Other important features of SOS algorithm 2. Commensalism phase: The stage in which an organism
are simple and easily adaptable for integer and discrete opti- is developing a relationship that benefits itself while
mization, useful for optimizing multi-modal search spaces, not affecting others. Shark and remora fish is one ex-
and quite affective in nonlinear constraint optimization includ- ample of commensalism. The commensal relationship
ing penalty functions [32]. The SOS algorithm is a population is modeled in following equation:
based on symbiotic organisms by using mutualism, commen-
salism, and parasitism. The details of the operation of each
X inew ¼ X i þ rand ð−1; 1Þ* X best −X j ð7Þ
phase are as follows:
3. Parasitism phase: Is the phase where an organism is
1. Mutualism phase: In this phase an organism creates a
building a relationship that benefits itself but harms
relationship that brings benefit to each other. For example,
others. An example of parasitism is a plasmodium par-
the interaction between bees and flowers shows that bees
asite, which uses its association with the anopheles
obtain benefit from the nectar and distribute the pollen,
mosquitoes to pass parasite into the human hosts.
which facilitates the pollination process. The following
There are no equations involve in this phase. The pro-
equations show the mutualism relationship:
cess is performed only by duplicating the existing or-
X inew ¼ X i þ rand ð0; 1Þ*ðX best −Mutual Vector:BF 1 Þ ð4Þ ganism to become the parasite organism. Parasite or-
ganism with better fitness value will kill other organ-
X jnew ¼ X j þ rand ð0; 1Þ*ðX best −Mutual Vector:BF 2 Þ ð5Þ isms from the ecosystem.
Int J Adv Manuf Technol (2019) 105:5121–5137 5129
7.1 Adaptation of SOS technique to 3D printer value (i.e., lowest Ra) will continue to the next phase. In the
optimization problem commensalism phase, again a new organism, Xinew, is calcu-
lated according to commensal symbiosis as from Eq. (7)
In this study, the target of the optimization process is to deter- where Xj is randomly selected from the ecosystem to com-
mine the optimal set values of process parameters that con- mensal with Xi. The fitness value is evaluated for Xinew and
tribute to the minimum surface roughness, Ra, value. To opti- compared with the previous Xi. If Xinew fitness is better than
mize the surface roughness, it is important to determine the its pre-interaction fitness, then the former will be chosen as the
problems of process parameters which consist of layer height solution for the next iteration. For the third phase, parasitism
(A), print speed (B), print temperature (C), and outer shell organism is created from organism, Xi (i.e., duplicating organ-
speed (D). Figure 4 shows the SOS optimization process. ism Xi and modifying its value). The fitness of the parasite
The SOS optimization begins with the initialization process. organism is compared with the organism Xj that is randomly
In the initialization process, there are four important parame- selected from the ecosystem. If the fitness of the parasite or-
ters that need to be set: number of iteration, number of organ- ganism is better than the organism Xj, organism Xj will be
isms in the ecosystem, organism size, and range of organisms. eliminated from the ecosystem and replaced with the parasite
The number of organism reflects to the number of input pa- organism. The fitter organism will survive for the next itera-
rameters of the FDM 3D printer and range of organisms refer tion. For each iteration, the minimum fitness value (i.e., sur-
to the range of its input parameters. After initialization is com- face roughness) will be stored iteratively until it converges to
pleted, the mutualism phase is carried out. At this phase, two the global optimum value. This value is the optimum input
organisms need to be mutual. Organism Xi, which is the ith parameter setting for the FDM 3D printer to produce the best
member of the ecosystem, is mutual with Xj that has been minimum value of surface roughness.
selected randomly from the ecosystem. Both will go through
the mutualistic relationship from Eqs. (4) and (5) and become
new organisms (Xinew and Xjnew). Both new organisms will 8 Results and analysis
be evaluated for its fitness value and compared with the pre-
vious organisms (Xi and Xj). Fitter organisms are selected as a 8.1 Variance analysis
solution for the next iteration. The fitter organism is referred to
as the best fitness function evaluated from the surface rough- ANOVA has been done as shown in Table 6 to observe the
ness regression model. In this case, organism with the fittest influence of the process parameters which are layer thickness
(A), print speed (B), print temperatures (C), and outer shell to the response which are layer thickness and print speed (AB),
speed (D) on output response which is surface roughness. layer thickness and outer shell speed (AD), and print temper-
The cubic model was chosen in this analysis because cubic ature and outer shell speed (CD).
model are more accurate based on R squared (R2) compared to Figure 5 a shows the interaction plot between surface
the quadratic model. The significance of each parameter to the roughness and layer thickness with respect to the print speed.
surface roughness was evaluated by the significance effect The minimum point of surface roughness from the graph was
(p values). This value indicates the importance of the param- achieved at the minimum value of layer thickness and print
eters that influence the output response if the value is less than speed of 0.18 mm and 36 mm/s, respectively. However, the
0.05. maximum point of surface roughness increased linearly with
Table 6 shows the layer thickness and print speed having a the increasing of layer thickness and print speed. Thus, it
p value < 0.0001 followed by outer shell speed and print shows that the decreasing of layer thickness and print speed
temperature. This indicates that both of these parameters have would give better surface roughness. In a previous study,
considerable affect on the surface roughness since the p values Anitha et al. [24] also found that the decreasing of print speed
of these are less than 0.05. The lower the layer thickness and would improve the surface quality. Therefore, the print speed
print speed, the higher surface quality can be achieved. is one of the important factors that needs to be considered
However, it will lead to longer build time as the print speed while performing the FDM process.
has been reduced. For the interaction between parameter, layer Figure 5 b shows the interaction effect of the layer thick-
thickness and print speed (AB); layer thickness and outer shell ness and outer shell speed on surface roughness. Minimum
speed (AD); print temperature and outer shell speed (CD); surface roughness decreased linearly with the decreasing of
layer thickness, print speed, and print temperature (ABC); lay- layer thickness and increasing outer shell speed. However, the
er thickness, print speed, and outer shell speed (ABD); and outer shell speed had only a little bit of effect on the surface
layer thickness, print temperature, and outer shell speed roughness as the value was reduced. The minimum surface
(ACD) significantly influenced the response. roughness can be achieved at lower layer thickness and higher
An interaction effect is the simultaneous effect of two or outer shell speed which was consistent with previous study by
more input parameters on at least one output response in Nancharaiah et al. [15]. Thus, it is revealed that both layer
which their joint affect is significantly greater (or significantly thickness and outer shell speed have an influence on the sur-
less) than the sum of the parts. The presence of interaction face roughness.
effects in this study is important because it tells how the pro- Figure 5 c shows the interaction effect of print temperature
cess parameters of 3D printer work together to impact the and outer shell speed on surface roughness. Based on the plot,
surface roughness of the printed sample. From the ANOVA, the surface roughness decreased proportionally when higher
there are three interaction effects that have significant impact outer shell speed reacted with the lower print temperature.
Int J Adv Manuf Technol (2019) 105:5121–5137 5131
Fig. 5 Interaction effect analysis for the surface roughness. a Layer thickness and print speed. b Layer thickness and outer shell speed. c Print speed and
outer shell speed
However, the surface roughness also remained low at lower One of the analyses is to examine a normal probability
shell speed and higher print temperature. This situation occurs plot of the residuals where the data of surface roughness
because, as the outer shell speed was increased, the material are plotted against a theoretical normal distribution in or-
was released faster from the extruder. At that time, a higher der to form an approximate straight line. The residuals
print temperature was required in order to release the material present the normality and response transformation of the
quickly. The best surface roughness was obtained at the 40 model. In this study, the normal probability plot of resid-
mm/s outer shell speed and 205 °C print temperature. This uals is depicted in Fig. 7. It is observed that most of the
interaction was the most significant interaction as the p value data points are located along the line which indicates that
was < 0.0001. the error of the model are normally distributed and ade-
Comparisons of the regression predicted results and the quately fit a straight line with no apparent problems in
experimental results of the surface roughness were also normality and response transformation.
performed. The experimental and predicted surface rough-
ness values were compared as shown in Fig. 6. For a good 8.2 Regression model for surface roughness
fit, the points are located in the vicinity of the fitted line,
with narrow confidence bands. Points on the left or right The regression equation and coefficient of the cubic model
of the plot, furthest from the mean, have the most lever- was obtained by carrying out a rigorous analysis with the
age and effectively try to pull the fitted line towards the experimental data using Design Expert software. The signifi-
point. Points that are vertically distant from the line rep- cance of the model was examined using ANOVA. The best
resent possible outliers. Figure 6 shows that the points model was achieved by observing three factors which are p
that have been plot are mostly close to the fitted line so value of the model, p value of the lack of fit, and R2 value.
the cubic model that had been generated can be consid- First of all, the model was evaluated by the p value of model
ered as a good prediction in estimating the predicted sur- and lack of fit to make sure whether the model was significant
face roughness values. or not.
Before accepting any model, the adequacy of the model Table 7 shows the p value for the model is less than 0.0001
must be checked by an appropriate statistical analysis. which implies that the selected model is significant. The term
5132 Int J Adv Manuf Technol (2019) 105:5121–5137
Fig. 6 Comparison of
experimental and predicted
surface roughness values
significant means that the model fits enough data from the from the actual response and 0 shows the model fits worse
experiment. The lack of fit is another important feature needed than the mean of the response. The value of R2 shows the
to assess whether the model fits the data well or not. It is accuracy of the model in terms of the percentage value.
shows that the lack of fit p value of 0.2721 implies that the From Table 7, the R2 value is 91.11% and the adjusted R2
lack of fit is not significant relative to the pure error. There is a is 86.67 % that can be considered as an adequate model in
27.21% chance that a lack of fit this large could occur due to representing the surface roughness of the FDM process. The
noise. The lack of fit should be insignificant before it will adjusted R2 is an altered version of R2 that was adjusted
further be used as the predicted model for the response. based on the number of predictors in the model. The adjust-
The regression model was developed together with the ed R2 increases only if the new term improves the model
coefficient of determination (R2). R2 is the proportion of more than would be expected by chance. However, it is
variability in the response described by the model. R2 ap- always lower than the R2. The regression model for the
proaches 1 when the model fits the data perfectly predicted surface roughness is shown in the Eq. (8).
studied wisely for affective optimization. Table 10 shows the process was selected to present the target of the optimization
parameter settings for PSO algorithm. The main settings that solution as follows:
need to be focused included the population number (no. of
0:10≤ A ≤0:33 ð10aÞ
particles), no. of iteration (particle steps), and PSO momen-
tum (w). The population number was set to 60 which is suffi- 36:00≤ B ≤ 60:00 ð10bÞ
cient to obtain the best surface roughness. Apart from that, the
number of iteration was set to 100 because from the prelimi- 185:00≤ C ≤ 205:00 ð10cÞ
nary test, it was observed that most of the runs that have been
29:00≤ D ≤ 40:00 ð10dÞ
done converged between 40 and 80 iterations. If the number of
iterations was increased, there would be no difference in the In this optimization, the process parameters were represent-
PSO results. The PSO momentum (w) controls the speed of ed by the position and velocity of particles while the surface
the graph to convergence. From the w value, there are two roughness is represented by the fitness function. By using the
categories in searching the optimum value that can be used combination of Eqs. (10a) to (10d) and the setting value from
which are exploration and exploitation. Exploration means the Table 10, the optimization was performed. The system ran
searching process can go further but not the deeper in the area based on the settings that have been set to achieve the opti-
and vice versa for the exploitation. Commonly, for the explo- mum setting of parameters in order to minimize the surface
ration setting, the value of w is greater than 1 or equal and roughness. Table 11 shows that the optimum surface rough-
exploitation is below than 1. In this study, w was set to 1. ness given by PSO is 2.176 μm. The set of process parameters
C1 is the acceleration constant for the cognitive parameter that was predicted by the system are layer thickness of 0.15
and C2 is the acceleration constant for the social parameter mm, print speed of 37.77 mm/s, print temperature of 202.09
which controls the position of the particles. The values of C1 °C, and outer shell speed of 39.69 mm/s.
which controls the position of pbest (local best) must be the Figure 9 shows the parameter profile for the PSO optimi-
same with the value of C2 which controls the position of gbest zation. In this optimization, all the process parameters started
(global best) to make sure the position of the particle is bal- to converge when it reached the optimum value of between 55
anced. Therefore, both values have been set to 1.5 which is the to 60 iterations. The convergence profile in Fig. 10 indicates
common value used in PSO optimization. In order to achieve
the optimal process parameter, the range based on DOE
Table 10 Optimal solution of PSO setting
Table 9 Overall optimization result by using RSM
Parameter Setting value
Process parameter Optimum value
Population number (no. of particles) 60
Layer thickness (mm) 0.18 Iteration number (particle steps) 100
Printing speed (mm/s) 0.36 Dimension (number of process parameter) 4
Printing temperature (°C) 205.00 C1 (acceleration constant for the cognitive parameter) 1.5
Outer shell speed (mm/s) 40.00 C2 (acceleration constant for the social parameter) 1.5
Predicted surface roughness (μm) 2.295 W (PSO momentum) 1.0
Int J Adv Manuf Technol (2019) 105:5121–5137 5135
and experimental value. From Table 14, it can be observed Fig. 12 SOS convergence Profile
that the predicted surface roughness for PSO and SOS are less
than that of the RSM optimization method. After going 9 Conclusions
through the experiment validation, the error for each method
was computed. The experimental validation of surface rough- In this work, the optimization of FDM parameters to improve
ness for RSM was 2.444 μm compared with the predicted surface roughness has been implemented using RSM, PSO,
value, 2.295 μm. There was about 6.10% of error between and SOS. The regression model for the surface roughness was
the experiment and predicted value. This is most probably developed using ANOVA statistical analysis. The major infer-
due to the inaccuracy of the polynomial estimation that lead ences of the research are presented below:
to poor representation of the optimal FDM process parame-
ters. In contrast, the experimental validations of surface rough- 1. ANOVA analysis and surface interaction plot have re-
ness for PSO and SOS improved by about 2.68% and 2.21% vealed that layer thickness and print speed have signifi-
of error, respectively. The results for PSO and SOS also cant effect on surface roughness followed by outer shell
showed that the surface roughness reduced by 8.5% and speed and print temperature. Better surface finish can be
8.8%, respectively, compared with the conventional method, obtained with a proper combination of lower layer thick-
i.e., RSM. It can be concluded that the use of metaheuristic ness, low print speed, high print temperature, and high
methods such as PSO and SOS can improve the surface outer shell speed.
roughness of the 3D printed part due to its ability to search 2. Results from optimization algorithms have shown that
for the best parameter setting for the optimal response. SOS and PSO gave better prediction of surface roughness
compared to RSM. The predicted surface roughness from
the simulation results have shown that both methods
(PSO and SOS) the ability to improve the surface quality
of FDM 3D parts. It is about 5% reduction of surface
roughness compared to the RSM method.
3. From the experimental confirmation test, it is observed
that SOS gave better surface quality compared to RSM
and PSO. SOS produced the lowest surface roughness of
2.229 μm followed by PSO, 2.236 μm, and RSM, 2.444
μm, respectively. The results for PSO and SOS have im- 10. Boér J, Blaga P (2018) Reducing production costs by monitoring
the roughness of raw product surfaces. Procedia Manuf 22:202–
proved by 8.5% and 8.8%, respectively, compared to
208. https://doi.org/10.1016/j.promfg.2018.03.031
RSM. These findings reveal that the nonconventional 11. Alsoufi MS, Elsayed AE (2018) Surface roughness quality and
method gives better optimization results compared to the dimensional accuracy-a comprehensive analysis of 100% infill
conventional method. printed parts fabricated by a personal/desktop cost-effective FDM
3D printer. Mater Sci Appl 9:11–40
12. Peng T, Yan F (2018) Dual-objective analysis for desktop FDM
Since SOS optimization method is relatively new, the ca- printers: energy consumption and surface roughness. Procedia
pability of this method in solving other types of responses CIRP 69:106–111
such as mechanical strength, build time, dimensional accura- 13. Anitha R, Arunachalam S, Radhakrishnan P (2001) Critical param-
eters influencing the quality of prototypes in fused deposition
cy, etc. would be very interesting to be further explored.
modelling. J Mater Process Technol 118(1):385–388. https://doi.
Furthermore, this work could be extended to other addictive org/10.1016/S0924-0136(01)00980-3
manufacturing process such as selective laser melting, selec- 14. Reddy V, Flys O, Chaparala A, Berrimi CE, Amogh V, Rosén BG
tive laser sintering, stereo-lithography, electron beam melting, (2018) Study on surface texture of Fused Deposition Modeling.
Procedia Manuf 25:389–396
etc. Each of the process has their own parameters settings
15. Nancharaiah T, Raju DR, Raju VR (2010) An experimental inves-
which is generally difficult to obtain. Therefore, a reliable tigation on surface quality and dimensional accuracy of FDM com-
and promising approach such as metaheuristic optimization ponents. Int J Emerg Technol 1(2):106–111
should be chosen in order to achieve the best output response. 16. Horvath D, Noorani R, Mendelson M Improvement of surface
roughness on ABS 400 polymer using design of experiments
(DOE). In: Materials Science Forum, 2007. Trans Tech Publ, pp
Acknowledgments This research study was supported by the researchers
2389-2392
from University Malaysia Perlis. The authors would like to express their
17. Chung Wang C, Lin T-W, Hu S-S (2007) Optimizing the rapid
gratitude to University Malaysia Perlis for their guidance in order to
prototyping process by integrating the Taguchi method with the
complete this research study.
Gray relational analysis. Rapid Prototyp J 13(5):304–315
18. Jiang R, Kleer R, Piller FT (2017) Predicting the future of additive
manufacturing: A Delphi study on economic and societal implica-
References tions of 3D printing for 2030. Technol Forecast Soc 117:84–97
19. Thrimurthulu K, Pandey PM, Venkata Reddy N (2004) Optimum
part deposition orientation in fused deposition modeling. Int J Mach
1. Yan Q, Dong H, Su J, Han J, Song B, Wei Q, Shi Y (2018) A review
Tool Manuf 44(6):585–594. https://doi.org/10.1016/j.ijmachtools.
of 3D printing technology for medical applications. Engineering
2003.12.004
4(5):729–742. https://doi.org/10.1016/j.eng.2018.07.021
20. Asadollahi-Yazdi E, Gardan J, Lafon P (2018) Multi-objective op-
2. Singamneni S, Yifan L, Hewitt A, Chalk R, Thomas W (2019)
timization of additive manufacturing process. IFAC-PapersOnLine
Additive manufacturing for the aircraft industry: a review. J
51(11):152–157. https://doi.org/10.1016/j.ifacol.2018.08.250
Aeronaut Aerospace Eng 8(214):2
21. Rao RV, Rai DP (2016) Optimization of fused deposition modeling
3. Deswal S, Narang R, Chhabra D (2019) Modeling and parametric
process using teaching-learning-based optimization algorithm. Eng
optimization of FDM 3D printing process using hybrid techniques
Sci Technol Int J 19(1):587–603
for enhancing dimensional preciseness. Int J Interact Des Manuf
22. Pandey PM, Thrimurthulu K, Reddy* NV (2004) Optimal part
(IJIDeM):1-18
deposition orientation in FDM by using a multicriteria genetic al-
4. Lyu J, Manoochehri S (2019) Multi-objective optimization based gorithm. Int J Prod Res 42(19):4069–4089
on machine reliability and process-dependent product quality for 23. Byun HS, Lee* KH (2005) Determination of the optimal part ori-
FDM system. Int J Adv Manuf Technol:1-10 entation in layered manufacturing using a genetic algorithm. Int J
5. Griffiths C, Howarth J, De Almeida-Rowbotham G, Rees A, Prod Res 43(13):2709–2724. https://doi.org/10.1080/
Kerton R (2016) A design of experiments approach for the optimi- 00207540500031857
sation of energy and waste during the production of parts 24. Khatir S, Wahab MA, Benaissa B, Köppen M Crack identification
manufactured by 3D printing. J Clean Prod 139:74–85 using eXtended IsoGeometric analysis and particle swarm optimi-
6. Boursier J-F, Fournet A, Bassanino J, Manassero M, Bedu A-S, zation. In: Fracture, fatigue and wear, 2018. Springer, pp 210-222
Leperlier D (2018) Reproducibility, accuracy and effect of auto- 25. Prasad D, Mukherjee V (2016) A novel symbiotic organisms search
clave sterilization on a thermoplastic three-dimensional model algorithm for optimal power flow of power system with FACTS
printed by a desktop fused deposition modelling three- devices. Eng Sci Technol Int J 19(1):79–89
dimensional printer. Vet Comp Orthopaed Traumatol 31(06):422– 26. Tran D-H, Cheng M-Y, Prayogo D (2016) A novel Multiple
430 Objective Symbiotic Organisms Search (MOSOS) for time–cost–
7. Leite M, Varanda A, Ribeiro AR, Silva A, Vaz MF (2018) labor utilization tradeoff problem. Knowl-Based Syst 94:132–145
Mechanical properties and water absorption of surface modified 27. Yamada T, Febri Z (2015) Freight transport network design using
ABS 3D printed by fused deposition modelling. Rapid Prototyp J particle swarm optimisation in supply chain–transport
24(1):195–203 supernetwork equilibrium. Transport Res E Log 75:164–187
8. Srivastava M, Rathee S, Maheshwari S, Kundra T (2019) 28. Noordin MY, Venkatesh VC, Sharif S, Elting S, Abdullah A (2004)
Estimating percentage contribution of process parameters towards Application of response surface methodology in describing the per-
build time of FDM process for components displaying spatial sym- formance of coated carbide tools when turning AISI 1045 steel. J
metry: a case study. Int J Mater Prod Technol 58(2-3):201–224 Mater Process Technol 145(1):46–58
9. Mohamed OA, Masood SH, Bhowmik JL (2015) Optimization of 29. Nee CY, Saad MS, Mohd Nor A, Zakaria MZ, Baharudin ME
fused deposition modeling process parameters: a review of current (2018) Optimal process parameters for minimizing the surface
research and future prospects. Adv Manuf 3(1):42–53 roughness in CNC lathe machining of Co28Cr6Mo medical alloy