J Polym Res (2008) 15:147–160

DOI 10.1007/s10965-007-9154-2

A modeling approach for prediction of erosion behavior

of glass fiber–polyester composites
Amar Patnaik & Alok Satapathy & S. S. Mahapatra &
R. R. Dash

Received: 28 June 2007 / Accepted: 17 September 2007 / Published online: 20 October 2007
# Springer Science + Business Media B.V. 2007

Abstract In recent years, a fairly good number of articles as crack formation, fiber fragmentation and matrix body
dealing in characterization of solid particle erosion of glass deformation. Finally, popular evolutionary approach known
fiber reinforced composites are available but exhaustive genetic algorithm (GA) is used to generalize the method of
study on this vital aspect leading to understand erosion finding out optimal factor settings for minimum wear rate.
phenomenon is hardly found in the literature. Therefore, in
the present work, a theoretical model based on principle of Keywords Erosion modeling . Polyester composite;
conservation of particle kinetic energy is developed to de- Taguchi design . ANN . GA
termine wear rate of glass–polyester composites due to
multiple impact erosion. Room temperature erosion tests
are then carried out to study the effect of various control Introduction
factors in an interacting environment on the erosion be-
havior of these composites. For this purpose, design of Solid particle erosion is a general term used to describe
experiments approach utilizing Taguchi’s orthogonal arrays mechanical degradation (wear) of any material subjected
is adopted to test the specimens on air jet type erosion test to a stream of erodent particles impinging on its surface.
configuration. The results indicate that erodent size, fiber The effect of solid particle erosion has been recognized by
loading, impingement angle and impact velocity are the many researchers Wahl and Hartenstein [1] for a long time.
significant factors in the order of their influence on wear Damage caused by erosion has been reported in several
rate. Taguchi approach enables to determine optimal param- industries for a wide range of situations. Examples have
eter settings that lead to minimization of erosion rate. been sited for transportation of airborne solids through
Artificial neural network (ANN) approach is applied to the pipes by Bitter [2], boiler tubes exposed to fly ash by Raask
erosive wear data to reach at acceptable predictive models. [3] and gas turbine blades by Hibbert and Roy [4]. Solid
Scanning electron microscopy of the eroded surface of the particle erosion is the progressive loss of original material
composites is performed for observation of the features such from a solid surface due to mechanical interaction between
that surface and solid particles. Various applications of
A. Patnaik (*)
polymers and their composites in erosive wear situations
Mechanical Engineering, N.I.T. Hamirpur, are reported by Pool et al. [5], Kulkarni and Kishore [6] and
Hamirpur, Himachal Pradesh, 177005, India Ruff and Ives [7] in the literature. But solid particle erosion
e-mail: [email protected] of polymers and their composites have not been investigat-
A. Satapathy : S. S. Mahapatra
ed to the same extent as for metals or ceramics. However, a
Mechanical Engineering, N.I.T. Rourkela, number of researchers Barkoula and Karger-Kocsis [8],
Rourkela, Orissa, 769008, India Tewari et al. [9] have evaluated the resistance of various
types of polymers and their composites to solid particle
R. R. Dash
Mechanical Engineering,
erosion. It is widely recognized that polymer and their
Gandhi Institute of Engineering and Technology, composites have poor erosion resistance. Their erosion
Gunupur, Orissa, India rates (Er) are considerably higher than metals. Also, it is
148 A. Patnaik, et al.

well known that the erosion rate of polymer composites is P Force on the indenter (N)
even higher than that of neat polymers as reported by Häger Hv Hardness (N/m2)
et al. [10]. The solid particle erosion behavior of polymer m Mass of single erodent particle (kg)
composites as a function of fiber content has been studied to a M Mass flow rate of the erodent (kg/s)
limited extent by investigators like Miyazaki and Takeda [11]. N Number of impact per unit time (s−1)
Tilly and Sage [12] have investigated the influence of ρc Density of composite (kg/m3)
velocity, impact angle, particle size and weight of impacted ρ Density of erodent (kg/m3)
abrasives on nylon, carbon–fiber-reinforced nylon, epoxy ηnormal Erosion efficiency with normal impact
resin, polypropylene and glass–fiber-reinforced plastic. η Erosion efficiency
Lindsley and Marder [13] found impact velocity (v) to be Erth Erosion wear rate (kg/kg)
a critical test variable in erosion, and that it can easily over
Solid particle erosion is a wear process in which the
shadow changes in other variables, such as target material,
material is removed from a surface by the action of a high
impact angle etc. Sundararajan and Manish [14] suggested
velocity stream of erodent particles entrained in a high
that in addition to velocity, solid particle erosion is governed by
velocity fluid stream. The particles strike against the surface
the impact angle, particle size, particle shape and hardness. The
and promote material loss. During flight, a particle carries
impact of above parameters has been studied independently
momentum and kinetic energy which can be dissipated during
keeping all parameters at fixed levels. Therefore, visualization
the impact due to its interaction with a target surface. As far as
of impact of various factors in an interacting environment
erosion study of polymer matrix composites is concerned, no
really becomes difficult. To this end, an attempt has been made
specific model has been developed and thus the study of their
to analyze the impact of more than one parameter on solid
erosion behavior has been mostly experimental. However,
particle erosion of PMCs, because in actual practice the
Mishra [22] proposed a mathematical model for material
resultant erosion rate is the combined effect of impact of more
removal rate in abrasive jet machining process in which the
than one interacting variables. An inexpensive and easy-
material is removed from the work piece in a similar fashion.
to-operate experimental strategy based on Taguchi’s parameter
This model assumes that the volume of material removed is
design has been adopted to study effect of various parameters
same as the volume of indentation caused by the impact.
and their interactions. The experimental procedure has been
This has a serious limitation as in a real erosion process
successfully applied by Mahapatra and Patnaik [15, 16, 17, 18,
the volume of material removed is actually different from the
19, 20, 21] for parametric appraisal in wire electrical
indentation volume. Further, this model considers only the
discharge machining (WEDM) process, drilling of metal
normal impact i.e. α=90° whereas in actual practice,
matrix composites, and erosion behavior of metal matrix
particles may impinge on the surface at any angle (0°≤α≤
composites such as aluminium reinforced with red mud.
90°). The proposed model addresses these shortcomings in
The aim of the present study is, therefore, to investigate the
an effective manner. It considers the real situation in which
erosion behavior of polyester matrix composites based on
the volume of material removed by erosion is not same as
Taguchi method under various testing conditions. Further
the volume of material displaced and therefore, additional
more, the analysis of variance are employed to investigate the
term “erosion efficiency (η)” is incorporated in the erosion
most significant control factors and their interactions. Finally,
wear rate formulation. In the case of a stream of particles
evolutionary approach known as genetic algorithm has been
impacting a surface normally (i.e. at α=90°), erosion
applied for optimal factor settings to minimize the erosion rate.
efficiency (ηnormal) defined by Sundararajan and Manish
[14] is given as
Mathematical model 2Er Hv
hnormal ¼ ð1Þ
rV2
Nomenclature
But considering impact of erodent at any angle α to the
The following symbols are used in this paper: surface, the actual erosion efficiency can be obtained by
r Chord length of the indentation (m) modifying Eq. 1 as
d Erodent diameter (m)
δ Indentation depth (m) 2Er Hv
h ¼ ð2Þ
ev Volumetric wear loss per particle impact (m3) r V 2 sin2 a
EV Total volumetric erosion wear rate (m3/s)
α Angle of impingement (degree) The model is based on the assumption that the kinetic
V Impact velocity (m/s) energy of the impinging particles is utilized to cause micro-
A modeling approach for prediction of erosion behavior 149

indentation in the composite material and the material loss is a

measure of the indentation. The erosion is the result of
cumulative damage of such non-interacting, single particle
impacts. The model further assumes the erodent particles to be
rigid, spherical bodies of diameter equal to the average grit
size. It considers the ductile mode of erosion and assumes
the volume of material lost in a single impact is less than
the volume of indentation. The model is developed with the
simplified approach of energy conservation which equals
the erodent kinetic energy with the work done in creating the
indentation.
The model for ductile mode erosion proceeds as follows.
Fig. 2 Resolution of impact velocity in normal and parallel directions
From the geometry of Fig. (1), r2 ¼ d
%

 Now applying conservation of energy to the single impact

d % erosion process, kinetic energy associated with the normal
The volume of indentation ¼:% 2

2 3 velocity component of a single erodent particle is equal to the
work done in the indentation of composite. The energy of
So, the volumetric wear loss per particle impact is given by impact introduces a force P on the indenter to cause the
indentation in the composite. Thus,
ev ¼ Volume of indentation
η

 1 2 2 1
2 d % mv sin a ¼
P
d ð4Þ
¼η
: %  and neglecting % 3 terms 2 2
2 3
:
d
%2 So,
¼
η
2  
1 p d3 1 2 
r v2 sin2 a ¼ p r HV d
Considering N number of particle impacts per unit time, 2 6 2
the volumetric erosion wear loss will be
On solving;
p
d
d2
EV ¼ N
h ð3Þ r
V 2 d 2 sin2 a
2 d2 ¼ ð5Þ
6HV

The impact velocity will have two components; one The number of erodent particle impacting the target is
normal to the composite surface and one parallel to it. At estimated from the known value of erodent mass flow rate,
zero impact angles, it is assumed that there is negligible wear M as
because eroding particles do not practically impact the target M
N ¼ ð6Þ
surface [8]. Consequently, there will be no erosion due to the : d3 ρ
parallel component and the indentation is assumed to be 6
caused entirely by the component normal to the composite Substituting the value of % in Eq. 3
surface as shown in Fig. (2).
:
d
d 2
V 2
sin2 !
ρ M
6
EV ¼

η
2
6HV :
d3 ρ
V 2
sin2 !
EV ¼
η
2HV
Erosion rate (Er) defined as the ratio of mass lost due to
erosion to the mass of erodent is now expressed as.
rc
h
V 2
sin2 a
Er ¼ ð7Þ
2HV

Material removal by impact erosion wear involves complex

Fig. 1 Scheme of material removal mechanism in ductile mode mechanisms. A simplified theoretical model for such a
150 A. Patnaik, et al.

process may appear inadequate unless its assessment against Table 1 Parameters of the setting
experimental results is made. So for the validation of the Control factors Symbols Fixed parameters
proposed model erosion tests on the composites are conducted
at various operating conditions. Velocity of Factor A Erodent Silica sand
impact
Fiber loading Factor B Erodent feed 10.0±1.0
rate (g/min)
Experimental program
Stand-off distance Factor C Test temperature RT
Impingement Factor D Nozzle diameter (mm) 3
Materials angle
Erodent size Factor E Length of nozzle 80
Cross plied E-glass fibers (360 roving taken from Saint Govion) (mm)
are reinforced in unsaturated isophthalic polyester resin
(supplied by Ciba-Giegy Ltd. India) to prepare the composites.
The composite slabs are made by conventional hand-lay-up and an adjustable sample clip. The velocity of the eroding
technique. Two percent cobalt nephthalate (as accelerator) is particles is determined using double disc method [7]. The
mixed thoroughly in isophthalic polyester resin and then 2% samples were cleaned in acetone, dried and weighed to an
methyl–ethyl–ketone–peroxide (MEKP) as hardener is mixed accuracy of ±0.1 mg accuracy using a precision electronic
in the resin prior to reinforcement. E-glass fiber and polyester balance. These are then eroded in the test rig for 10 min and
resin have modulus of 72.5 and 3.25 GPa respectively and weighed again to determine the weight loss. The procedure
possess density of 2.59 and 1.35 gm/cc respectively. Three is repeated till the erosion rate attains a constant value
composites of different glass fiber weight fractions (30, 40 and called steady state erosion rate. The ratio of this weight loss
50 wt.%) are fabricated. The castings are put under load for to the weight of the eroding particles causing the loss (i.e.
about 24 h for proper curing at room temperature. Specimens of testing time × particle feed rate) is then computed as the
suitable dimension are cut using a diamond cutter for normalized erosion rate.
mechanical characterization and erosion test.
Mechanical characterization
Test apparatus
Micro-hardness measurement is done using a Leitz micro-
The room temperature erosion test facility used in the hardness tester equipped with a square based pyramidal
present investigation is illustrated schematically in Fig. 3. (angle 136° between opposite faces) diamond indenter by
The set up is capable of creating reproducible erosive applying a load of 24.54 N. The tensile test is performed on
situation for assessing erosion wear resistance of the flat dog-bone shaped composite specimens as per ASTM D
prepared composite samples. The conditions (confirming 3039-76 test standards an universal testing machine Instron
to ASTM G 76 test standards) under which erosion tests are 1195. Three point bend test is conducted in the same machine
carried out are listed in Table 1. Dry silica sand (density at across head speed of 10 mm/min to evaluate the flexural
2.5 gm/cc) is used as the erodent. The particles fed at strength of the composites. Finally, the eroded surfaces of
constant rate are made to flow with compressed air jet to some selected samples are examined by scanning electron
impact the specimen, which can be held at various angles microscope JEOL JSM-6480LV.
with respect to the flow direction of erodent using a swivel
Experimental design
Erodent feeder
Design of experiment is a powerful analysis tool for model-
ing and analyzing the influence of control factors on per-
formance output. The most important stage in the design of
Erodent experiment lies in the selection of the control factors.
Air jet Therefore, a large number of factors are included so that
α Specimen
Compressor non-significant variables can be identified at earliest
Swivel
opportunity. The operating conditions under which erosion
Test section
tests were carried out are given in Table 2. The tests were
conducted as per experimental design given in Table 3
under room temperature.
Five parameters viz., impact velocity, fiber loading, stand-
Fig. 3 A schematic diagram of the erosion rig off distance, impingement angle and erodent size, each at
A modeling approach for prediction of erosion behavior 151

Table 2 Levels for various control factors better characteristic, which can be calculated as logarithmic
Control factor Level transformation of the loss function as shown below.

I II III Units
Smaller is the better characteristic : ð8Þ
S 1 X 2 
A: Velocity of impact 32 45 58 m/s ¼  10 log y (8)
B: Fiber loading 30 40 50 %
N n
C: Stand-off distance 120 180 240 Mm where n the number of observations, and y the observed
D: Impingement angle 30 60 90 degree data. “Lower is better” (LB) characteristic, with the above
E: Erodent size 300 500 800 μm S/N ratio transformation, is suitable for minimizations of
erosion rate. The standard linear graph by Glen [23] and
Madhav [24], as shown in Fig. 4, is used to assign the
three levels, are considered in this study in accordance with factors and interactions to various columns of the orthog-
L27 (313) orthogonal array design. In Table 3, each column onal array. Solid particle erosion is characterized by a large
represents a test parameter and a row gives a test condition number of factors such as impact velocity, fiber loading,
which is nothing but combination of parameter levels. Five stand off-distance, impingement angle, and erodent size.
parameters each at three levels would require 35 =243 runs in Out of all these factors, velocity predominantly governs the
a full factorial experiment. Whereas, Taguchi’s factorial rate of erosion.
experiment approach reduces it to 27 runs only offering a The plan of the experiments is as follows: the first
great advantage. column was assigned to impact velocity (A), the second
The experimental observations are transformed into a column to fiber loading (B), the fifth column to stand-off
signal-to-noise (S/N) ratio. There are several S/N ratios distance (C), the ninth column to impingement angle (D)
available depending on the type of characteristics. The S/N and tenth column to erodent size (E), the third and fourth
ratio for minimum erosion rate coming under smaller is column are assigned to (A×B)1 and (A×B)2, respectively to

Table 3 Orthogonal array for L27 (313) Taguchi design

L27(313) 1A 2B 3(A×B)1 4(AxB)2 5C 6(B×C)1 7(B×C)2 8(A×C)1 9D 10 E 11(A×C)2 12 13

1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 2 2 2 2 2 2 2 2 2
3 1 1 1 1 3 3 3 3 3 3 3 3 3
4 1 2 2 2 1 1 1 2 2 2 3 3 3
5 1 2 2 2 2 2 2 3 3 3 1 1 1
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
8 1 3 3 3 2 2 2 1 1 1 3 3 3
9 1 3 3 3 3 3 3 2 2 2 1 1 1
10 2 1 2 3 1 2 3 1 2 3 1 2 3
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
13 2 2 3 1 1 2 3 2 3 1 3 1 2
14 2 2 3 1 2 3 1 3 1 2 1 2 3
15 2 2 3 1 3 1 2 1 2 3 2 3 1
16 2 3 1 2 1 2 3 3 1 2 2 3 1
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
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
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
24 3 2 1 3 3 2 1 1 3 2 2 1 3
25 3 3 2 1 1 3 2 3 2 1 2 1 3
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
152 A. Patnaik, et al.

B(2) 360

350

Tensile strength (MPa)

(3,4) (6,7)
340

D(9) E(10) (12) (13) 330

A(1) (8,11) C(5) 320

Fig. 4 Standard linear graphs for L27 array
310
Fiber loading
estimate interaction between impact velocity (A) and fiber 300
loading (B), the sixth and seventh column are assigned to 30 40 50
(B×C)1 and (B×C)2 respectively, to estimate interaction Fiber loading (%)
between the fiber loading (B) and stand-off distance (C), the Fig. 6 Effect of fiber loading on tensile strength of glass fiber
eighth and eleventh column are assigned to (A×C)1 and polyester composites
(A×C)2 respectively, to estimate interaction between the
impact velocity (A) and stand-off distance (C). The remain- recommending any composite as a candidate for structural
ing columns are assigned to error columns respectively. applications.

Steady state erosion

Results and discussion
Erosion wear behavior of materials can be grouped as
Mechanical properties ductile and brittle categories although this grouping is not
definitive. Thermoplastic matrix composites usually show
Figure 5 shows the micro-hardness values for different com- ductile behavior and have the peak erosion rate at around
positions. It is seen that with the increase in fiber content in 30° impact angle because cutting mechanism is dominant in
the composite, its hardness value improves although the erosion. While the thermosetting ones erode in a brittle
increment is marginal. Figures 6 and 7 show the variation of manner with the peak erosion occurring at normal impact.
tensile and flexural strengths of the composites with the fiber However, there is a dispute about this failure classification
content. A gradual increase in tensile strength as well as as the erosive wear behavior depends strongly on the
flexural strength with the weight fraction of fiber is noticed. experimental conditions and the composition of the target
It clearly indicates that inclusion of glass fiber improves the material [8]. Figure 8 shows the impact angle dependence
load bearing capacity and the ability to withstand bending of of the erosion rate of polyester composites with different
the composites. Similar observations have been reported by fiber content. The curves are plotted with the results of
Harsha et al. [25] for other fiber reinforced thermoplastics erosion tests conducted for different impingement angle
such as polyaryletherketone composites. It may be mentioned keeping all other parameters constant (impact velocity=
here that both tensile and flexural strengths are important for 32 m/s, stand-off distance=120 mm and erodent size=

400
40
Flexural strength (MPa)

350
Vicker's microhardness (Hv)

35
300

30 250

25 200

30 wt%, 40wt% and 50wt% 150

20 30 35 40 45 50
30 40 50 Fiber loading (%)
Fiber loading (%) Fig. 7 Effect of fiber loading on flexural strength of glass fiber
Fig. 5 Variation of microhardness vs. fiber loading polyester composites
A modeling approach for prediction of erosion behavior 153

400
Micro-ploughing on ductile matrix
350
Erosion rate (mg/kg)

300

250

200 Crack Propagation

30 wt%
40 wt% (longitudinal and transverse)
50 wt%
150
15 30 45 60 75 90
Impingement angle (Degree)
Fig. 8 Erosion rate vs. Angle of impingement for different fiber Fig. 10 SEM micrograph (× 1,000) of GF Polymer composite eroded
loading surface (impact velocity 58 m/s, fiber loading 50%, S.O.D 120 mm,
impingement angle 60° and erodent size 300 μm)
300 μm). It can be seen that the peaks of erosion rates are
located at an angle of 60° for all the samples irrespective of is formation of transverse cracks that break these fibers.
fiber content. This shows semi-ductile erosion behavior of Figure 10 presents the microstructure of the composite
the composite. It is further noted (Fig. 8) that the erosion eroded at high impact velocity (58 m/s), at lower stand-off
rate increases with increase in fiber content. Sundararajan distance (120 mm) and at an impingement angle of 60°.
et al. [26] concluded that this behavior is attributed to the Here the propagation of crack along transverse as well as
fact that the harder the material, larger is the fraction of the longitudinal direction is well visualized. On comparing this
crater volume that is removed. In this investigation higher micro-structure with that of the same composite eroded at a
hardness values have been noted for composites with higher lower impact velocity (45 m/s), higher stand-off distance
fiber loading and this is therefore the reason why the (240 mm) and higher impingement angle (90°), it can be
composites exhibit declining erosion resistance with the seen that in the second case the breaking of glass fibers is
increase in fiber content. more prominent (Fig. 11). It appears that cracks have grown
To identify the mode of material removal, the morphol- on the fibers giving rise to breaking of the fibers into small
ogies of eroded surfaces are observed under scanning fragments. Further the cracks have been annihilated at the
electron microscope. Figure 9 shows the local removal of fiber matrix interface and seem not to have penetrated
resin material from the impacted surface resulting in through the matrix. Change in impact angle from oblique to
exposure of the fibers to the erodent flux. This micrograph normal changes the topography of the damaged surface
also reveals that due to sand particle impact on fibers there very significantly. Figure 11 shows the dominance of

Transverse crack Fiber Fragmentation

Fig. 9 SEM micrograph (× 250) of GF Polymer composite eroded Fig. 11 SEM micrograph (× 1,000) of GF Polymer composite eroded
surface (impact velocity 58 m/s, fiber loading 50%, S.O.D 120 mm, surface (impact velocity 45 m/s, fiber loading 50%, S.O.D 240 mm,
impingement angle 60° and erodent size 300 μm) impingement angle 90° and erodent size 800 μm)
154 A. Patnaik, et al.

micro-chipping and micro-cracking phenomena. It can be impact, where erosion rates are highest for brittle materials.
seen that multiple cracks originate from the point of impact, In the present study also, the peak erosion rate shifts to
intersect one another and form wear debris due to brittle impingement angle (60°) and it is clearly due to the brittle
fracture in the fiber body. After repetitive impacts, the nature of glass fibers. So although polyester is a ductile
debris in platelet form are removed and account for the material, the presence of fibers makes the composite
measured wear loss. The occurrence of peak erosion rate at relatively more sensitive to impact energy which increases
60° impact is understandable. In this case, both abrasion when the impact mode pattern changes from tangential (α=
and erosion processes play important roles. The sand 0°) to normal (α=90°). This explains the semi-ductile nature
particles after impacting, slide on the surface and abrade of the glass–polyester composites with respect to solid
while dropping down. The wear and subsequently the particle erosion.
damage are therefore more than that in the case of normal From Table 4, the overall mean for the S/N ratio of the
impact. Marks of micro-ploughing on the ductile polyester erosion rate is found to be −48.97 dB. Figure 12 shows
matrix region seen in Fig. 10 support this argument. graphically the effect of the six control factors on erosion
Polyester is a thermoplastic polymer and it is known that rate. The analysis was made using the popular software
it shows a ductile erosion response. So a possible reason for specifically used for design of experiment applications
the semi-ductile erosion behavior exhibited by the polyester known as MINITAB 14. Before any attempt is made to use
based composites in the present investigation is that the this simple model as a predictor for the measures of
glass fibers used as reinforcements are a typical brittle performance, the possible interactions between the control
material. Their erosion is caused mostly by damage factors must be considered. Thus factorial design incorpo-
mechanism such as micro-cracking. Such damage is rates a simple means of testing for the presence of the
supposed to increase with the increase of kinetic energy interaction effects.
loss of the impinging sand particles. According to Hutchings Analysis of the result leads to the conclusion that factor
et al. [27] the kinetic energy loss is a maximum at normal combination of A1, B2, C1, D1 and E2 gives minimum

Table 4 Experimental design using L27 orthogonal array

Experiment. Impact velocity Fiber loading Stand-off distance Impingement angle Erodent size Erosion rate S/N Ratio
no. (A) (m/s) (B) (%) (C) (mm) (D) (degree) (E) (μm) (Er) mg/kg (dB)

1 32 30 120 30 300 309.83 −49.8225

2 32 30 180 60 500 315.25 −49.9731
3 32 30 240 90 800 305.19 −49.6914
4 32 40 120 60 500 186.07 −45.3936
5 32 40 180 90 800 272.79 −48.7166
6 32 40 240 30 300 230.96 −47.2707
7 32 50 120 90 800 287.69 −49.1785
8 32 50 180 30 300 279.85 −48.9385
9 32 50 240 60 500 255.25 −48.1393
10 45 30 120 60 800 288.86 −49.2137
11 45 30 180 90 300 249.80 −47.9518
12 45 30 240 30 500 255.25 −48.1393
13 45 40 120 90 300 239.76 −47.5955
14 45 40 180 30 500 249.18 −47.9304
15 45 40 240 60 800 298.23 −49.4910
16 45 50 120 30 500 261.17 −48.3385
17 45 50 180 60 800 364.31 −51.2294
18 45 50 240 90 300 389.94 −51.8201
19 58 30 120 90 500 315.10 −49.9690
20 58 30 180 30 800 245.19 −47.7901
21 58 30 240 60 300 219.89 −46.8441
22 58 40 120 30 800 261.27 −48.3418
23 58 40 180 60 300 239.76 −47.5955
24 58 40 240 90 500 210.66 −46.4716
25 58 50 120 60 300 369.47 −51.3516
26 58 50 180 90 500 452.81 −53.1183
27 58 50 240 30 800 391.45 −51.8535
A modeling approach for prediction of erosion behavior 155

Main Effects Plot (data means) for SN ratios

Interaction Plot (data means) for SN ratios
A B C
-48.0 C
-48.0 12
18
-48.6 24
-49.2 -48.5
Mean of SN ratios

-49.8

SN ratios
-50.4
32 45 58 30 40 50 120 180 240 -49.0
D E

-48.0

-48.6 -49.5
-49.2

-49.8

-50.4 -50.0
30 60 90 300 500 800 32 45 58
A
Signal-to-noise: Smaller is better
Signal-to-noise: Smaller is better
Fig. 12 Effect of control factors on erosion rate Fig. 14 Interaction graph between A×C for erosion rate

erosion rate. The interaction graphs are shown in Figs. 13, efficiency can be made only on the basis of experimental
14 and 15. As for as minimization of erosion rate is data. Hence, the values of erosion efficiencies of these
concerned, factor B and E have significant effect whereas composites calculated using Eq. 2 are summarized in
factor C has least effect. It is observed from Fig. 13 that the Table 5 along with their hardness values and operating
interaction between A×B shows most significant effect on conditions. It clearly shows that erosion efficiency is not
erosion rate. But the factor C individually has less exclusively a material property; but also depends on other
contribution on output performance, and their combination operational variables such as impingement angle and
of interaction with factor A and B (A×C and B×C) is shown impact velocity. The erosion efficiencies of these compo-
in Figs. 14 and 15 can be neglected for further study. sites under normal impact (ηnormal) vary from 3 to 6%, 6–
9% and 9–12% for impact velocities 58, 45 and 32 m/s
Erosion efficiency respectively. The value of η for a particular impact velocity
under oblique impact can be obtained simply by multiply-
The hardness alone is unable to provide sufficient correla- ing a factor 1/sin2α with ηnormal. Similar observation on
tion with erosion rate, largely because it determines only velocity dependence of erosion efficiency has previously
the volume displaced by each impact and not really the been reported by few investigators Roy et al. [28], and
volume eroded. Thus a parameter which will reflect the Arjula and Harsha [29].
efficiency with which the volume that is displaced is The theoretical erosion wear rate (Erth) of the polyester-
removed should be combined with hardness to obtain a GF composites are calculated using Eq. 7. These values are
better correlation. The erosion efficiency is obviously one compared with those obtained from experiments (Erexpt)
such parameter. This thought has already been reflected in conducted under similar operating conditions. Seventy five
the theoretical model but the evaluation of erosion percent of data collected from erosion test is used for

Interaction Plot (data means) for SN ratios Interaction Plot (data means) for SN ratios
-47 B -47 C
30 12
40 18
50 24
-48
-48

-49
SN ratios
SN ratios

-49
-50

-50
-51

-52 -51

32 45 58 30 40 50
A B
Signal-to-noise: Smaller is better
Signal-to-noise: Smaller is better

Fig. 13 Interaction graph between A×B for erosion rate Fig. 15 Interaction graph between B×C for erosion rate
156 A. Patnaik, et al.

Table 5 Erosion efficiency of

GF-reinforced polyester resin Experiment. Impact velocity Density of Hardness of Erosion rate Erosion
no. (V) m/s eroding material eroding material (Er) mg/kg efficiency (η)
(ρ) kg/m3 (Hv) MPa

1 32 1738 32 309.83 43.70689

2 32 1738 32 315.25 12.83002
3 32 1738 32 305.19 10.76308
4 32 1874 34 186.07 7.462042
5 32 1874 34 272.79 9.479905
6 32 1874 34 230.96 32.10497
7 32 1932 39 287.69 11.12368
8 32 1932 39 279.85 43.28217
9 32 1932 39 255.25 11.38925
10 45 1738 32 288.86 5.944763
11 45 1738 32 249.80 4.454857
12 45 1738 32 255.25 18.2082
13 45 1874 34 239.76 4.213347
14 45 1874 34 249.18 17.51554
15 45 1874 34 298.23 6.047942
16 45 1932 39 261.17 20.42593
17 45 1932 39 364.31 8.22007
18 45 1932 39 389.94 7.624237
19 58 1738 32 315.10 3.382663
20 58 1738 32 245.19 10.52866
21 58 1738 32 219.89 2.724091
22 58 1874 34 261.27 11.05526
23 58 1874 34 239.76 2.926861
24 58 1874 34 210.66 2.228443
25 58 1932 39 369.47 5.018254
26 58 1932 39 452.81 5.329466
27 58 1932 39 391.45 18.42909

training whereas 25% data is used for testing. The by repeated impact, as is usually the case in the case of
parameters of three layer architecture of ANN model are ductile materials, the magnitude of η will be very low, i.e.
set as input nodes=5, output node=1, hidden nodes=12, η≤100%. In the case of brittle materials, erosion occurs
learning rate=0.01, momentum parameter=0.03, number of usually by spalling and removal of large chunks of
epochs=200,000 and a set of predicted output (ErANN) is materials resulting from the interlinking of lateral or radial
obtained. Table 6 presents a comparison among the cracks and thus η can be expected to be even greater than
theoretical, experimental and the ANN predicted results. 100% [29]. According to the categorization made by Roy
The errors calculated with respect to the theoretical results et al. [28], the erosion efficiencies of the composites under
are also given. It is observed that maximum error between the present study indicate that at low impact speed the
theoretical and experimental wear rate is 0–10%, whereas erosion response is semi-ductile (η=10–100%). On the
same between ANN prediction and experimental wear rate other hand at relatively higher impact velocity the compo-
is 0–14%. The error in case of ANN model can further be sites exhibit ductile (η<10%) erosion behavior.
reduced if number of test patterns is increased. However,
present study demonstrates application of ANN for predic- ANOVA and the effects of factors
tion of wear rate in a complex process of solid particle
erosion of polymer composites. In order to understand a concrete visualization of impact of
The magnitude of η can be used to characterize the various factors and their interactions, it is desirable to develop
nature and mechanism of erosion. For example, ideal analysis of variance (ANOVA) table to find out the order of
microploughing involving just the displacement of the significant factors as well as interactions. Table 7 shows the
material from the crater without any fracture (and hence results of the ANOVA with the erosion rate. This analysis
no erosion) will results in η=0. In contrast, if the material was undertaken for a level of confidence of significance of
removal is by ideal micro-cutting, η=1.0 or 100%. If 5%. The last column of the table indicates that the main
erosion occurs by lip or platelet formation and their fracture effects are highly significant (all have very small p values).
A modeling approach for prediction of erosion behavior 157

Table 6 Comparison of
theoretical, experimental and Experiment Erth Erexpt. ErANN Error (%) Error (%)
ANN results no. (mg/kg) (mg/kg) (mg/kg) (Erth −Erexpt) (Erth −ErANN)

1 314.84 309.83 340.33 1.5912 8.0945

2 321.69 315.25 308.95 2.0019 3.9601
3 304.17 305.19 291.93 0.3353 4.0226
4 201.98 186.07 176.31 7.8765 12.711
5 271.78 272.79 295.34 0.3716 8.6687
6 238.45 230.96 269.54 3.1411 13.038
7 286.69 287.69 251.33 0.3488 12.332
8 296.75 279.85 258.51 5.6950 12.886
9 276.81 255.25 292.07 7.7887 5.5125
10 309.29 288.86 289.53 6.6054 6.3882
11 248.76 249.80 267.10 0.4180 7.3727
12 266.73 255.25 238.76 4.3039 10.485
13 238.76 239.76 233.65 0.4188 2.1385
14 257.34 249.18 251.49 3.1693 2.2722
15 319.46 298.23 321.04 6.6455 0.4947
16 289.38 261.17 293.27 9.7484 1.3438
17 378.38 364.31 327.33 3.7184 13.491
18 387.95 389.94 351.06 0.5155 9.5067
19 314.14 315.10 272.29 0.3056 13.321
20 259.21 245.19 248.82 5.4087 4.0069
21 227.76 219.89 206.00 3.4553 9.5529
22 281.37 261.27 318.62 7.1436 13.239
23 248.75 239.76 277.37 3.6140 12.312
24 209.66 210.66 229.48 0.4769 9.4519
25 387.34 369.47 349.40 4.6135 9.7947
26 451.81 452.81 391.23 0.2213 13.408
27 405.27 391.45 363.25 3.4101 10.369

From Table 7, one can observe that the fiber loading (p= Confirmation experiment
0.004), erodent size (p=0.145), impingement angle (p=
0.252) and impact velocity (p=0.265) have great influence The confirmation experiment is the final test in the design
on erosion rate. The interaction of impact velocity × fiber of experiment process. The purpose of the confirmation
loading (p=0.029) shows significance of contribution on experiment is to validate the conclusions drawn during the
the erosion rate and the factor stand-off distance (p=0.493) analysis phase. The confirmation experiment is performed
and impact velocity × stand-off distance (p=0.150), fiber by conducting a new set of factor settings A2B3D2E3 to
loading × stand-off distance (p=0.162) present less signif- predict the erosion rate. The estimated S/N ratio for erosion
icance of contribution on erosion rate. rate can be calculated with the help of following prediction
equation:
   
b
h1 ¼ T þ A 2  T þ B 3  T
Table 7 ANOVA table for erosion rate
     

þ A2 B3  T  A2  T  B3  T
Source df Seq SS Adj SS Adj MS F P
   
þ D2  T þ E 3  T ð9Þ
A 2 2.3056 2.3056 1.1528 1.88 0.265
B 2 35.4646 35.4646 17.7323 28.95 0.004
C 2 1.0737 1.0737 0.5369 0.88 0.483
D 2 2.4297 2.4297 1.2149 1.98 0.252
h1 Predicted average
E 2 3.9765 3.9765 1.9882 3.25 0.145 T Overall experimental average
A×B 4 21.5781 21.5781 5.3945 8.81 0.029 A2 ; B3 ; D2 and E 3 Mean response for factors and
A×C 4 7.5740 7.5740 1.8935 3.09 0.150 interactions at designated levels.
B×C 4 7.1630 7.1630 1.7908 2.92 0.162
Error 4 2.4498 2.4498 0.6125 By combining like terms, the equation reduces to
Total 26 84.0150
h1 ¼ A2 B3 þ D2 þ E 3  2T ð10Þ
158 A. Patnaik, et al.

A new combination of factor levels A2, B3, D2 and E3 is suitable to use for further analysis. Here, the resultant
used to predict deposition rate through prediction equation objective function to be maximized is given as:
and it is found to be h1 ¼ 50:8283 dB. For each
Maximize Z ¼ 1=f ð13Þ
performance measure, an experiment was conducted for a
different factors combination and compared with the
result obtained from the predictive equation as shown in
f Normalized function for erosion rate
Table 8.
The resulting model seems to be capable of predicting Subjected to constraints:
erosion rate to a reasonable accuracy. An error of 2.48% for
Amin  A  Amax ð14Þ
the S/N ratio of erosion rate is observed. However, the error
can be further reduced if the number of measurements is
increased. This validates the development of the mathe- Bmin  B  Bmax ð15Þ
matical model for predicting the measures of performance
based on knowledge of the input parameters. Dmin  D  Dmax ð16Þ

Factor settings for minimum erosion rate

Emin  E  Emax ð17Þ
In this study, an attempt is made to derive optimal settings The min and max in Eqs. (14) to (17) shows the lowest
of the control factors for minimization of erosion rate. The and highest control factors settings (control factors) used in
single-objective optimization requires quantitative determi- this study (Table 2).
nation of the relationship between erosion rates with Genetic algorithm (GA) is used to obtain the optimum
combination of control factors. In order to express, erosion value for single-objective outputs to optimize the single-
rate in terms of mathematical model in the following form objective function. The computational algorithm is imple-
is suggested. mented in Turbo C++ and run on an IBM Pentium IV
machine. Genetic algorithms (GAs) are mathematical
Er ¼ K0 þ K1
A þ K2
B þ K3
D þ K4
E optimization techniques that simulate a natural evolution
process. They are based on the Darwinian Theory, in which
þ K5
A
B ð11Þ the fittest species survives and propagate while the less
successful tend to disappear. Genetic algorithm mainly
depends on three types of operators viz., reproduction,
Here, Er is the performance output terms and Ki (i=0, crossover and mutation. Reproduction is accomplished by
1………..5) are the model constants. The constant are copying the best individuals from one generation to the
calculated using non-linear regression analysis with the next, what is often called an elitist strategy. The best
help of SYSTAT 7 software and the following relations are solution is monotonically improving from one generation to
obtained. the next. The selected parents are submitted to the
crossover operator to produce one or two children. The
Er ¼ 1:521  1:633A  1:387B þ 0:088D ð12Þ crossover is carried out with an assigned probability, which
is generally rather high. If a number randomly sampled is
þ 0:078E þ 1:221AB inferior to the probability, the crossover is performed. The
r2 ¼ 0:98 genetic mutation introduces diversity in the population by
an occasional random replacement of the individuals. The
mutation is performed based on an assigned probability. A
The correctness of the calculated constants is confirmed
as high correlation coefficients (r2) in the tune of 0.96 are
obtained for Eq. 1 and therefore, the models are quite
Table 9 Optimum conditions for performance output

Control factors and performance characteristics Optimum

Table 8 Results of the confirmation experiments for erosion rate conditions

Optimal control parameters A: Impact velocity (m/s) 33.15

prediction experimental B: Fiber loading (%) 41.02
D: Impingement angle (degree) 59.45
Level A2 B3D2E3 A2 B3D2E3 E: Erodent size (μm) 500.0
S/N ratio for Erosion rate (dB) −50.8283 −49.5677 Erosion rate (mg/kg) 364.72
A modeling approach for prediction of erosion behavior 159

random number is used to determine if a new individual 5. The erosion efficiency (η) values obtained experimen-
will be produced to substitute the one generated by tally also suggest that the glass fiber reinforced
crossover. The mutation procedure consists of replacing polyester composites exhibit semi-ductile erosion re-
one of the decision variable values of an individual while sponse (η=10–60%) for low impact velocities. How-
keeping the remaining variables unchanged. The replaced ever, for relatively high impact velocity, they present a
variable is randomly chosen and its new value is calculated ductile erosion response (η<10%).
by randomly sampling within its specific range. In genetic 6. Two predictive models based on ANN approach is
optimization, population size, probability of crossover and proposed. It is demonstrated that that these models well
mutation are set at 50, 75, and 5% respectively for all the reflect the effects of various factors on the erosion loss
cases. Number of generation is varied till the output is and their predictive results are consistent with experi-
converted. Table 9 shows the optimum conditions of the mental observations.
control factors with optimum performance out put gives a 7. The rationale behind the use of genetic algorithm lies
better combination of set of input control factors. in the fact that genetic algorithm has the capability
to find the global optimal parameter settings whereas
the traditional optimization techniques are normally
stuck up at the local optimum values. The optimum
Conclusions
settings are found to be impact velocity=33.15 m/s,
fiber loading=41.02%, impingement angle=59.45°,
This analytical and experimental investigation into the
erodent size=500.0 μm, and resulting erosion rate=
erosion behavior of glass fiber reinforced polyester compo-
364.72 mg/kg as far as present experimental conditions
sites leads to the following conclusions:
are concerned.
1. Conservation of energy principle is applied to the multiple 8. This work leaves wide scope for future investigators to
impact erosion process and consequently a mathematical study the erosion behavior of such composites with
model based on ductile mode erosion is developed. To short fiber reinforcement and with particulate filling.
overcome the shortcomings of existing theoretical models
‘erosion efficiency’ term has been introduced. It has been
demonstrated that if supported by an appropriate magni-
tude of erosion efficiency, the model performs well for
References
polyester matrix composites for normal as well as oblique
impacts.
1. Wahl H, Hartenstein F (1946) Strahlverschleiss, Frankh’sche
2. Solid particle erosion characteristics of these compo- Verlagshandlung, Stuttgart
sites can be successfully analyzed using Taguchi 2. Bitter JGA (1963) Wear 6:169–190
experimental design scheme. Taguchi method provides 3. Raask E (1968) Wear 13:303–313
a simple, systematic and efficient methodology for the 4. Hibbert WA, Roy J (1965) Aero Soc 69:769–776
5. Pool KV, Dharan CKH, Finnie I (1986) Wear 107:1–12
optimization of the control factors. This approach not 6. Kulkarni SM, Kishore (2001) Polym Polym Compos 9:25–30
only needs engineering judgment but also requires a 7. Ruff AW, Ives LK (1975) Wear 35:195–199
rigorous mathematical model to obtain optimal process 8. Barkoula NM, Karger-Kocsis J (2002) Wear 252:80–87
settings. 9. Tewari US, Harsha AP, Hager AM, Friedrich K (2002) Wear
252:992–1000
3. The results indicate that erodent size, fiber loading, 10. Häger A, Friedrich K, Dzenis YA, Paipetis SA (1995) Proceedings
impingement angle and impact velocity are the significant of the ICCM-10, Canada Wood head Publishing Ltd., Cambridge,
factors in a declining sequence affecting the erosion wear pp 155–162
rate. Although the effect of impact velocity is less 11. Miyazaki N, Takeda T (1993) J Compos Mater 27:21–31
12. Tilly GP, Sage W (1970) Wear 16:447–465
compared to other factors, it cannot be ignored because 13. Lindsley BA, Marder AR (1999) Wear 225:510–516
it shows significant interaction with fiber loading. An 14. Sundararajan G, Manish R (1997) Tribol Int 30:339–359
optimal parameter combination is determined, which 15. Mahapatra SS, Patnaik A (2006) Solid Waste Technol Manag 32
leads to minimization of material loss due to erosion. (1):28–35
16. Mahapatra SS, Patnaik A (2006) Indian J Eng Mater Sci 13:493–
4. The composites exhibit semi-ductile erosion character- 502
istics with the peak erosion wear occurring at 60° 17. Mahapatra SS, Patnaik A (2006) Int J Adv Manuf Technol DOI
impingement angle. This nature has been explained by 10.1007/s00170-006-0672-6
analyzing the possible damage mechanism with the 18. Mahapatra SS, Patnaik A (2007) J Braz Soc Mech Sci 28(4):423–
430
help of SEM micrographs. It is concluded that the 19. Mahapatra SS, Patnaik A (2006) Inst Eng 87:16–24
inclusion of brittle fibers in ductile polyester matrix is 20. Mahapatra SS, Patnaik A (2006) The International Journal for
responsible for this semi-ductility. Manufacturing Science and Technology 8(1):5–12
160 A. Patnaik, et al.

21. Mahapatra SS, Patnaik A (2007) The International Journal for 26. Sundararajan G, Roy M, Venkataraman B (1990) Wear 140:369–381
Manufacturing Science and Technology 9(2):129–144 27. Hutchings IM, Winter RE, Field JE (1976) Proc Roy Soc Lond
22. Mishra PK (1997) Narosa, New Delhi Ser A 348:379–392
23. Glen SP (1993) Addison-Wesley, New York 28. Roy M, Vishwanathan B, Sundararajan G (1994) Wear 171:149–
24. Madhav SP (1989) Prentice-Hall, Eaglewood Cliffs, NJ. 161
25. Harsha AP, Tewari US, Venkatraman B (2003) Wear 254:693–712 29. Arjula S, Harsha AP (2006) Polym test 25:188–196