You are on page 1of 7

Int J Adv Manuf Technol (2015) 79:823829

DOI 10.1007/s00170-015-6879-7

ORIGINAL ARTICLE

Evaluation of analytical modeling for improvement of surface


roughness of FDM test part using measurement results
Sadegh Rahmati & Ebrahim Vahabli

Received: 6 October 2014 / Accepted: 4 February 2015 / Published online: 19 February 2015
# Springer-Verlag London 2015

Abstract In additive manufacturing (AM) processes, the tessellation of CAD model and the slicing procedure are the
significant factors resulting unsatisfactory surface quality,
where the topics related to surface roughness have been a
key issue in this regard. In this paper, analytical models which
have been presented to express surface roughness distribution
in fused deposition modeling (FDM) are assessed according to
the variations in surface build angle by considering the main
factors which crucially affect surface quality. Analytical
models are verified by implementation and comparison with
empirical data derived from the comprehensive FDM fabricated test part. Finally, the most accurate model for estimation of
surface roughness in the process planning stage for optimization of effective parameters have been introduced upon calculating the mean absolute percentage error (MAPE) as performance criteria of each model in various equal ranges.
Keywords Additive Manufacturing . Surface Roughness .
Analytical Model

1 Introduction
Emphasize on reduction of product manufacturing time is
followed by essential changes in manufacturing processes
and resulted in the birth of a new race of manufacturing techS. Rahmati (*)
Department of Mechanical Engineering, Majlesi Branch, Islamic
Azad University, Isfahan, Iran
e-mail: rahmati@rapidtoolpart.com
E. Vahabli
Department of Mechanical Engineering, Science and Research
Branch, Islamic Azad University, Tehran, Iran
e-mail: Ebrahim.vahabli@chmail.ir

nologies. This technology is able to produce the part directly


from CAD model on a layer by layer deposition principle
without any need to tools, dies, jigs and fixtures, and human
intervention. This method therein part manufacturing is additive, introduced as additive manufacturing (AM) technology
[1]. AM has also enabled industry to benefit from different
rapid tooling (RT) techniques [2]. Due to the nature of layered
manufacturing in AM processes, quality of manufacturing
part is usually lower than parts manufactured by NC machines. Furthermore, most of layered manufacturing technologies use support structure for prevention of deviation during
stacking of layers that after support separation, plenty of separated support burrs residue on the part surface and upraise the
surface roughness [3].
Among AM processes, fused deposition modeling (FDM)
uses heated thermoplastic filament which is extruded in a
semi-molten state from the tip-off nozzle in a prescribed manner in a temperature-controlled environment for building the
part through a layer by layer deposition method until completion of part manufacturing. The part is fabricated by deposition of layers contoured in a plane (x-y) two dimensionally.
The third dimension (z) results from single layers being
stacked up on top of each other but not as a continuous zcoordinate. Due to the layered production of the part, the surface finish of the build part is unavoidably excessively rough.
This problem is raised in wide applications of FDM process in
different industries. Due to the importance of surface roughness subject, various studies and efforts have been done to
solve this issue [4, 5].
Poor surface finish in AM processes is often affected by
tessellation of the original CAD model and the slicing procedure during fabrication process. In slicing of tessellated CAD
model, containment problem results in distortion of the original CAD model of the designed form. In addition to containment problem, deposition of sliced layers results in another
problem called staircase effect or stair-stepping effect [6]. This

824

problem may be varied based on the surface build angle [7]. In


this paper, upon designing and building the specific test part,
well-known analytical models are evaluated. The mean absolute percentage error (MAPE) of each model within specified
ranges is calculated in order to introduce the best analytical
model. The general model parameters that are used in all analytical models are shown in Fig. 1 [6], where, denotes build
orientation (or surface build angle), t is layer thickness (slice
thickness), and N is the angle between vertical axis and normal to surface axis.

2 Analytical modeling
The concept of cusp height tolerance was introduced by
Dolence et al. [8] and used for interpretation of surface roughness. Offering a theoretic formula, they calculated cusp height
in a 2-dimensional and 3-dimensional part. Campbell et al. [9]
introduced a mathematical model for estimation of arithmetic
mean surface roughness in various AM processes given by
Eq. (1).


90
Ra 1000tsin
tan90
1
4

where Ra is the arithmetic-mean-surface roughness (micron), t is the layer thickness (mm), and is the build angle
(degree). This model was tested for a few angles that none of
them contains downward facing. In their report, this model
predicts the surface roughness appropriately in a number of
angles. This estimation was reported in the range of 45 to
180 for the FDM process. Mason et al. [10] presented a
model for estimation of arithmetic mean surface roughness.

Int J Adv Manuf Technol (2015) 79:823829

In this model, surface roughness is indeed equal to half of the


cusp height presented in the Dolende model which is given by
Eq. (2).
Ra

1000t
cos90
2

where Ra is the arithmetic-mean-surface roughness


(micron), t is the layer thickness (mm), and N is the
normal build angle in degree which is equals to (90).
Pandey et al. [6, 7, 11, 12] realized that layer thickness
and build orientation are assumed as the most important
process variables which affect the surface finish.
According to the analytical observations, the edge profile of the part built by the FDM process is parabola.
According to these observations, a semi-empirical model
was proposed. However, the formula used in this paper
is presented in Eq. (3), where, Ra is the arithmeticmean-surface roughness (micron), t is the layer thickness (mm), is the build angle (degree), Ra90 is the
equivalent for the roughness of the build angle (90),
and w is a dimensionless adjustment parameter for supported facets (chosen to be 0.2 in all FDM systems). It
is notable that for Ra in the build angles between 0 to
70 , the average value of the sum is utilized.

8
t
>
;
0 70

>
> 69:28e72:36
>
cos
.
<
20 90Ra70 70Ra90 Ra90 Ra70 ; 70 < < 90
Ra 1
>
>
>
117:6  t;
90
>
:
90 < 180
Ra90 1 w;

8
0;
0; =2;
>
<
 2
 






2
2
2
2
R1 R2 1 sin90
Ra 1000t
R1 R2 1 4
4
>
cos90
tan90 sin90 ; O = W

:
4
1000t
1000t 3



1000t  cos90 
Ra

2 
cos

Fig. 1 Surface roughness estimation parameters model [6]

Byun et al. [13] proposed a new roughness profile characterizing the surfaces made by layered manufacturing processes. For the FDM process, R1 and R2 were set to 0.045 and
0.01 mm respectively and maximum overhang angle was set
to 30 . Likewise, the profile roughness Ra as a function of f(t,
,R1,R2) is calculated by Eq. (4), where, Ra is arithmetic-mean

Int J Adv Manuf Technol (2015) 79:823829

825

Fig. 4 Final test part after washing and assembly operation

Fig. 2 Pyramid-shaped test part for measurement of surface roughness


[9]

surface roughness (micron), t is the constant layer thickness


for a facet (mm), is the build angle (degree), R1 is the radius
of fillet (mm), and R2 is the radius of the corner (mm). Ahn
et al. [14] presented a theoretical model for calculation of
arithmetic mean surface roughness; therein, the edge profile
was assumed as ellipse which is given by Eq. (5); in which t
denotes the layer thickness in millimeters, the build angle in
degrees, and is the surface profile angle in degrees.

3 Test part design


To calculate the surface roughness in different build angles, a
suitable and comprehensive model is required. In the applied
studies, researches and papers written for evaluation of surface
roughness, a pyramid-shaped test part with varied surface
build angles was selected by some researchers as a prototype
for measuring the surface roughness. The angle between upward vector tangents to surface with the vertical axis is called
the surface build angle. Pyramid test part is shown in Fig. 2;
therein, surface build angles are equal to 10, 15, 30, and 45 ,
respectively.
Pandey et al. [6] reported the surface roughness measurement for the pyramid test part with side angles of 10, 15, 30,
and 45 and layer thickness of 0.254 mm. In their research, by
offering a formula, they introduced a solution for estimation of
surface roughness using layer thickness and build orientation
parameters. Mason et al. [10] developed a customized FDM
machine with the nozzle thickness of 2 mm for building the
test part therein resulted in very rough surfaces with high

surface roughness indexes. Other researchers such as


Campbell et al. [9] and Ahn et al. [14] presented a more comprehensive test part for assessment of their analytical models.
This test part, therein, was firstly introduced by Reeves and
Cobb [15] as depicted in Fig. 3, called Truncheon, has a suitable geometry which enables the measurement of surface
roughness for a wide range of surface build angles.
In the present study, with the choice of a step angle of 5 ,
the truncheon test part in total consists of 36 cuboids to span
surface build angles between 0 to 180 , where the test part
was designed in CATIAv5. The dimension of each cuboid is
assumed 103030 mm3 for accurate measurement by surface scanner and considering machine platform dimension of
254254305 mm3, where the overall dimension of the truncheon is 2203030 mm3.

3.1 Fabrication and measurement of the test part


The prototype is fabricated by FDM-Dimension sst 1200es.
The test part is sliced in CatalystEx and is made horizontally
with layer thickness of 0.254 mm. Model interior is solid and
support sparse method used for fabrication. The material used
for fabrication is ABSPLUS. Inside, the test part is filled and
fabricated with support sparse method. At the final stage, the
test part is put into the chemical solution for 1 day and support
material dissolved therein thoroughly. Due to the estimation of
surface roughness, no surface polishing operation is implemented as depicted in Fig. 4. The surface roughness measurement is done by surface scanner MahrSurf MFW250 and Ra
values are derived accordingly (Table 1).

Table 1
No.

Parameter

Value

1
2

Sampling length (lr)/cut off length (lc)


Evaluation length (ln)

0.8 mm
4 mm

3
4

Traversing length (lt)


Profile resolution
(measuring range)
Traversing speed (vt)
Points
Direction

5.6 mm
250 m/7 nm

5
6
7
Fig. 3 Schema of the truncheon test part [14]

Scanning parameters

0.5 mm/s
11200
Perpendicular to the
deposition direction

826

Int J Adv Manuf Technol (2015) 79:823829

Fig. 5 Measured Ra for the truncheon test part with layer thickness of
0.254 mm

Fig. 6 Estimation of surface roughness by Mason model for the


truncheon test part with layer thickness of 0.254 mm

3.1.1 The surface roughness

study (according to DIN EN ISO 4287 standard) are presented


in Table 1.

The surface roughness values measured by the scanner for the


test part with layer thickness of 0.254 mm are presented in
Fig. 5. Two separate roughness measurements are performed
for the test part on the same surface for each build angle as
presented in Table 2, while the scanning parameters set for this

Table 2 Empirical surface roughness (Ra) values for each build angle
with layer thickness of 0.254 mm
Build
angle
(degree)

Build
Ra-report no. Raavea
(micron) angle
(degree)
Ra-A Ra-B

Ra-report no. Raave


(micron)
Ra-A Ra-B

0
5
10
15
20
25
30
35
40

17.56
17.72
20.37
21.94
20.06
26.98
24.48
27.43
28.56

17.82
19.1
19.33
20.56
22.82
20.32
25.32
27.63
29.52

17.69
18.41
19.85
21.25
21.44
23.65
24.9
27.53
29.04

90
95
100
105
110
115
120
125
130

1.4
18.55
18.2
18.68
19.14
17.14
21.1
22.55
24.63

2.72
17.77
17.56
18.4
20.02
23.54
21.22
22.63
23.37

2.06
18.16
17.88
18.54
19.58
20.34
21.16
22.59
24

45
50
55
60
65
70
75
80
85

32.52
31.92
34.52
39.26
42.01
38.42
31.3
23.08
18.27

28.74
32.54
34.4
40.1
39.55
42.5
31.4
20.38
17.59

30.63
32.23
34.46
39.68
40.78
40.46
31.35
21.73
17.93

135
140
145
150
155
160
165
170
180

24.66
26.63
21.86
27.83
29.61
32.5
33.21
20.22
18.78

26.92
25.45
26.22
30.11
30.45
30.82
35.63
20.74
17.92

25.79
26.04
24.04
28.97
30.03
31.66
34.42
20.48
18.35

a
It is notable that Raave is denoting the average value between the Ra-A
and Ra-B values which is considered as Ra value and used in evaluation
procedure

4 Accuracy assessment of analytical models


In the present study, MATLAB software is used for modeling,
whereas Figs. 6 to 10 exhibit the performance of the Mason,
Campbell, Pandey, Byun, and Ahn models for estimation of
surface roughness in layer thickness of 0.254 mm for varied
surface build angles in comparison with the measured empirical data for the truncheon test part. As it is observed in the
Mason model, a weak estimation of surface roughness is obtained therein for prototype with layer thickness of 0.254 mm.
This model provides an appropriate estimation in angles of 0
to 15 and 165 to 180 .
In Fig. 7, the Campbell model shows a better estimation of
surface roughness in surface build angles of 45 to 135 and

Fig. 7 Estimation of surface roughness by Campbell model for the


truncheon test part with layer thickness of 0.254 mm

Int J Adv Manuf Technol (2015) 79:823829

827

Fig. 8 Estimation of surface roughness by Pandey model for the


truncheon test part with layer thickness of 0.254 mm

Fig. 10 Estimation of surface roughness by Ahn model for the truncheon


test part with layer thickness of 0.254 mm

also it overestimates the surface roughness for the build angles


between 0 to 30 and also 155 to 180 .
Moreover, according to Eq. (1), this model cannot predict
the surface roughness for angles between 0 to 180 . As it is
observed in Fig. 8, the Pandey model provides a relatively
appropriate estimation of surface roughness more than other
models; this model has an appropriate estimation performance
for surface build angle range within 0 to 65 and 95 to 135,
while this model has a weak performance for 80 to 95 and
also 155 to 180 .
Figure 9 shows the Byun model performance for surface
roughness estimation which demonstrates overestimation of
surface roughness within 95 to 135 , while estimation performance within 0 to 45 and 135 to 180 is more fitted.
According to Eq. (2) and Eq. (5), the Ahn model has a phase
shift of (degree) relative to the Mason model; thus as presented in Fig. 10, the Ahn model provides a similar estimation
like the Mason model with phase shift () of 5 .

4.1 Performance of surface roughness estimation for different


analytical models
The estimation error values in different models are calculated
for four angle ranges including 0 to 45, 45 to 90, 90 to 135,
and 135 to 180 in layer thickness of 0.254 mm. Analytical
models are compared by using mean absolute percentage error
(MAPE) criteria. MAPE gives a normalized error, permitting
efficiency comparison of different models coming from different data sets, as presented in Eq. (6).
MAPE% 1



X  X act X est 
  100


 X
N
act
i

where Xact is the actual value of Ra, Xest is denoting models


estimation value of Ra, and N is the number of patterns. As it is
observed in Table 3, within 0 to 45 angles, minimum error is
obtained for the Pandey model (13.56 %) and later the Byun
model (39.88 %), and other models have higher errors. Within
45 to 90 angles, minimum error is obtained for the Campbell
model (67.88 %). Within 90 to 135 , the Pandey model
(16.54 %) has the minimum error. Within 135 to 180 , the
Byun model (45.57 %) shows lower error than the other
Table 3 Mean absolute percentage error values (%) of surface
roughness of 0.254-mm layer thickness for different analytical models

Fig. 9 Estimation of surface roughness by Byun model for the truncheon


test part with layer thickness of 0.254 mm

Byun
Ahn
Campbell
Mason
Pandey

0 45

45< 90

90< 135

135< 180

39.88
138.33
1279.15
112.91
13.56

96.13
950.35
67.88
932.79
183.03

175.93
434.21
59.28
451.24
16.54

45.57
76.30
1299.83
89.29
99.49

828

Int J Adv Manuf Technol (2015) 79:823829


1400

1279.15

1299.83

1200
950.35

1000

932.79

Byun model
Ahn model
Campbell model
Mason model
Pandey model

800
600

200

451.24

434.21

400
138.33
39.88

112.91
96.13
13.56

67.88

183.03 175.93
59.28

0
0 45

45< 90

16.54 45.57 76.3

90< 135

89.29 99.49

135< 180

Fig. 11 Mean absolute percentage error values (%) for different analytical models

models. Considering the calculated values as presented in


Table 3, the Pandey and Byun models may be used due to
lower errors for surface roughness prediction.
In the designed parts with surface build angles of 0 to 45
and 90 to 135 with the specified layer thickness, the Pandey
model provides an appropriate estimation for optimum design
and also for the surface build angles of 135 to 180 ; the Byun
model is a proper choice for desirable and optimum design.
But, in surface build angles within 45 to 90 , all above models
have higher errors for prediction of roughness. In Fig. 11,
different models are compared and their estimation performance can be seen. Therefore, based on these assessments,
the Pandey and Byun models exhibit lower prediction error
than other models and the best estimation performance belongs to the Pandey model as presented in Fig. 11.
The Ahn and Campbell models despite benefitting from the
similar test part in their model evaluation, nevertheless do not
lead to desirable results. It may be due to the changes made in
the test part design and higher accuracy of surface measurement device utilized in this research. Although, in ref [14], an
empirical model is used for perusing the surface roughness
distribution in accordance with the measured data that provides more precise prediction, but due to the assessment of
the theoretical estimation models, models introduced in this
paper are utilized merely for comparison of the derived theoretical data with measured data.

5 Conclusion
In layered production processes, the tessellation of the CAD
model and the slicing procedure are assumed as the major
factors affecting creation of rough surfaces. In slicing a tessellated CAD model, containment problem results in distortion
of the original CAD model from the designed form. In addition to the containment problem, deposition of the sliced
layers leads to another problem called staircase effects. To
improve the FDM part surface roughness, modeling of the
surface roughness distribution for optimizing the effective

parameters before the fabrication process is used for more


precise planning of AM process.
In the present paper, the most well-known models used for
improvement of surface roughness are utilized based on analytical modeling. Later, the specific test part called truncheon
is designed to evaluate the above models. Analytical models
are evaluated by the comparison of the performance criteria
(MAPE) in different ranges. According to the provided assessments, best model for prediction of each range is introduced,
where altogether, Pandeys model exhibits the best prediction
for surface roughness. It is notable that although different test
parts are used for model evaluation in their research (pyramid
shape), results demonstrates high predictability of this model
for varied parts.
The main advantage of this research is the use of a comprehensive test part which enables the assessment of surface roughness in all surface build angles, including all factors affecting the
real rough surface creation such as roughness arising out of
staircase effects, support burrs, material properties, and other
factors which is not possible to achieve them in other ways.

References
1. Tolio T, Ceglarek D, ElMaraghy HA, Fischer A, Hu SJ, Laperrire L,
Newman ST, Vncza J (2010) SPECIES-Co-evolution of products,
processes and production systems. CIRP Ann-Manuf Technol 59:
672693
2. Rahmati S. (2014) Direct Rapid Tooling, Comprehensive Materials
Processing, Masood S, ed.; ELSEVIER Ltd.; Vol. 10, 303-344. doi:
10.1016/B978-0-08-096532-1.01013-X
3. Ahn D, Kweon J, Kwon S, Song J, Lee S (2009) Representation of
surface roughness in fused deposition modeling. Elsevier, Journal of
Materials Processing Technology, pp 55935600
4. Onuh SO, Yusuf YY (1999) Rapid prototyping technology: applications and benefits for rapid product development. J Intell Manuf 10:
301311
5. Upcraft S, Fletcher R (2003) The rapid prototyping technologies.
Assem Autom 23(4):318330
6. Pandey PM, Reddy NV, Dhande SG (2003) Slicing procedures in
layered manufacturing: a review. Rapid Prototyp J 9(5):274288

Int J Adv Manuf Technol (2015) 79:823829


7. Nourghassemi A (2011) Surface Roughness Estimation for FDM
Systems, MSc Thesis, Ryerson University, Canada
8. Dolenc A, Makalea I (1994) Slicing procedures for layered
manufacturing techniques. Comput Aided Des 26(2):119126
9. Campbell RI, Martorelli M, Lee HS (2002) Surface roughness visualization for rapid prototyping models. Comput Aided Des 34(10):717725
10. Mason A (2006) Multi-axis hybrid rapid prototyping using fusion deposition Modeling, MSc Thesis, Ryerson University,
Canada
11. Thrimurthulu K, Pandey PM, Reddy N (2004) Optimum part deposition orientation in fused deposition modeling. Int J Mach Tools
Manuf 44:585594

829
12. Pandey PM, Thrimurthulu K, Reddy NV (2004) Optimal part deposition orientation in FDM by using a multicriteria genetic algorithm,
Int J Prod Res. 42:4069-4089.
13. Byun HS, Lee KH (2006) Determination of optimal build direction in
rapid prototyping with variable slicing. Int J Adv Manuf Technol 28:
307313
14. Ahn DK, Kwon SM, Lee SH (2008) Expression for surface roughness distribution of FDM processed parts. Int. Conf. on Smart
Manufacturing Application, Korea, pp 490493
15. Reeves PE, Cobb RC (1997) Reducing the surface deviation of
stereolithography using in-process techniques. Rapid Prototyp J 3(1):
2031