Composites Journal of Reinforced Plastics and - CiteSeerX

Journal of Reinforced Plastics and Composites http://jrp.sagepub.com

Application of Probablistic Neural Network for the Development of Wear Mechanism Map for Glass Fiber Reinforced Plastics V. Srinivasan, K.V. Maheshkumar, R. Karthikeyan and K. Palanikumar Journal of Reinforced Plastics and Composites 2007; 26; 1893 originally published online Oct 17, 2007; DOI: 10.1177/0731684407082632 The online version of this article can be found at: http://jrp.sagepub.com/cgi/content/abstract/26/18/1893

Published by: http://www.sagepublications.com

Additional services and information for Journal of Reinforced Plastics and Composites can be found at: Email Alerts: http://jrp.sagepub.com/cgi/alerts Subscriptions: http://jrp.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav

Downloaded from http://jrp.sagepub.com at PENNSYLVANIA STATE UNIV on February 7, 2008 © 2007 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

Application of Probablistic Neural Network for the Development of Wear Mechanism Map for Glass Fiber Reinforced Plastics V. SRINIVASAN, K. V. MAHESHKUMAR AND R. KARTHIKEYAN Department of Manufacturing Engineering, Annamalai University Annamalainagar – 608002, India K. PALANIKUMAR* Department of Mechanical & Production Engineering Sathyabama University, Chennai – 119, India ABSTRACT: Glass fiber reinforced plastic composite materials are finding increased applications due to their excellent properties. Fiber reinforced plastic (FRP) composite materials comprise soft matrix and fiber elements. This article presents the tribological aspects of glass fiber reinforced plastic composites. The wear test is carried out for FRP in a pin-on-roller wear tester. The wear rate obtained from different sliding speeds and normal pressure are plotted as a contour map. The scanning electron microscopy images are taken to study the deformations that occurred in the wear zone. The regions of different wear mechanisms are identified using scanning electron microscopy. The regions of wear mechanisms are classified using probabilistic neural networks and superimposed over wear rate contours. The wear mechanisms observed using scanning electron microscopy, along with wear rate data, are used for the construction of wear mechanism maps. KEY WORDS: fiber reinforced plastics, tribology, wear mechanisms, probabilistic neural network.

INTRODUCTION industries are searching for lighter weight, higher strength and safer material to meet the demands of structural designs and for economic benefit. In order to extend the application area of plastics, plastic composites are developed by adding reinforcement materials to the plastics matrix. Some of the reinforcements used in structural and industrial applications are carbon, aramid and glass fibers; the most commonly used is glass fiber. Owing to the strength of the fibers, the low strength of the plain plastic is compensated for. Plastic composite material has therefore become one of the new competitive materials in engineering [1]. Fiber reinforced plastic (FRP) is a relatively new class of composite material manufactured from fibers and resins, and has proven efficient and economical for the development and repair of new and deteriorating structures. The mechanical properties of FRPs make them ideal for

F

OLLOWING TECHNOLOGICAL DEVELOPMENTS,

*Author to whom correspondence should be addressed. E-mail: [email protected] Figures 1, 2, 4 and 8 appear in color online: http://jrp.sagepub.com

Journal of REINFORCED PLASTICS

AND

COMPOSITES, Vol. 26, No. 18/2007

0731-6844/07/18 1893–14 $10.00/0 DOI: 10.1177/0731684407082632 ß SAGE Publications 2007 Los Angeles, London, New Delhi and Singapore Downloaded from http://jrp.sagepub.com at PENNSYLVANIA STATE UNIV on February 7, 2008 © 2007 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.

1893

1894

V. SRINIVASAN

ET AL.

widespread applications in various industries worldwide. The enhancement of the mechanical and structural properties due to the addition of fibers makes FRPs ideal materials for aircraft parts, aerospace structures, railways, machine parts and other industrial applications [2,3]. The mechanical properties of most reinforcing fibers are considerably higher than those of un-reinforced resin systems. The mechanical properties of the fiber/resin composite are therefore dominated by the contribution of the fiber to the composite. The four main factors that govern the fiber’s contribution are: 1. 2. 3. 4.

The The The The

basic mechanical properties of the fiber itself. surface interaction of fiber and resin (the ’interface’). amount of fiber in the composite (fiber volume fraction). orientation of the fibers in the composite.

Tribology is crucial to modern industry. Many tribological components such as brakes, clutches, driving wheels, gears, cams, bearings, and seals are applied in machinery. The friction and wear appearing in these products leads to high energy losses. Therefore, research in tribology leads to substantial economical savings and better performance of machines. Since FRP is used for many such applications, study of wear behavior has become essential. When material is lost from a loaded surface, entirely through some form of mechanical interaction, the concentration, size and shape of the debris particles carry important information about the state of surfaces from which they were generated. By constructing a wear mechanism map, a single dominant mechanism, which influences the wear rate of a specimen, can be identified. It can be constructed by summarizing the data for wear; showing how the mechanisms interface; and allowing the dominant mechanism, for any given set of conditions, to be identified. Williams [4] has presented the wear mechanism maps for various materials. He has concluded that when material is lost from a loaded surface through some form of mechanical interaction, the concentration, size and shape of the debris particle carry important information about the state of the surface. The full exploitation of this information, and the ability to be able to predict quantitatively future performance or life, requires an understanding of the sources and mechanisms of generation of the extracted and sampled particulate debris. In many cases, it is instructive to display the running conditions of a given contact on some form of operational or wear map. Briscoe et al. [5] have constructed wear mechanism maps for polymers in dry sliding conditions to illustrate the behavior of polymeric surfaces when subjected to scratching, and their dependence on contact conditions. Cenna et al. [6] have analyzed the abrasive wear behavior of polymer matrix composites and concluded that the abrasion resistance of reinforced composite materials is a consequence of the micro-mechanics that occur during abrasive wear, which in turn are strongly dependent upon the hardness of the wear media. Thus, the selection of an appropriate reinforcing material for enhancing the abrasion resistance of a composite surface is strongly governed by the material properties of both the abrasive particle and the reinforcement. Lim et al. [7] have constructed a wear mode map and wear transition map for steel, to predict the type and severity of wear and explain the changes in wear rate for a wide range of operating conditions. From the results of Ho et al. [8] it has been asserted that the wear volume loss in the sliding direction perpendicular to the melt flow direction is mostly lower than the sliding direction parallel with the melt flow direction. Xin et al. [9] have studied the friction and wear properties of the polytetrafloroethylene (PTFE) composites reinforced


Development of Wear Mechanism Map for Plastics

1895

with various amounts of potassium titanium whiskers under dry sliding conditions. They inferred that the wear rate of PTW/PTFE composite decreases dramatically when PTW content increases from 1 to 20 wt%, but only decreases slightly when PTW content increases from 20 to 40 wt%. The PTFE composite reinforced with 20 wt% PTW has the best wear resistance, which is over 1000 times higher than that of pure PTFE. The friction coefficient decreases with increasing PTW content. Iglesias et al. [10] have shown that the friction and wear of the polymer-matrix composites reinforced with discontinuous aluminum chips increase with increasing sliding distance or speed. The polymer-matrix composite based on phenolic resin reinforced with discontinuous nanostructured Al 6061 particles shows lower friction and wear. From their studies, Neogi et al. [11] have stated that the addition of PET in plastic improves the wear resistance of the plastic by reducing the wear losses. Su et al. [12] have concluded that the character of the transfer film and the difference in composite structures, coupled with the difference in the properties of the particles themselves, account for the difference in the wear resistance and friction reduction of Nomex fabric composites filled with PFW and nano-SiO2. Unal et al. [13] concluded that the sliding speed has stronger affect on the specific wear rate of Polyamide 66(PA 66), POM (poly oxy methylene), ultra high molecular weight polyethylene (UHMWPE), polyphenylene-sulfide (PPS) 30% glass fiber reinforcement and aliphatic polyketone (APK) polymers than the applied pressure. PA 66 and POM showed micro-cutting and plastic deformation morphology surfaces, and PPS showed a rough surface with glass fiber particles exposed in the matrix, while APK polymer showed a wavy morphology surface with peaks and valleys. Chang et al. [14] stated that the addition of nano-TiO2 further reduces the frictional coefficient and the contact temperature of the nanocomposite, which may be due to a rolling effect of nanoparticles. The probabilistic neural network (PNN) is a neural network classifier introduced by Specht in the 1990s. The theoretical framework for the probabilistic neural network is an old but very powerful theory of how to build classifiers, based on Bayes, and termed ‘Bayesian Classification’. This theory explains how one can build an optimal classifier provided that the distributions from which the data emanate are known. The main purposes of the present study are: (1) to determine the wear rate of FRP as a function of force and velocity; (2) to develop a wear mechanism map for FRP using the developed model; and (3) to study wear mechanism using scanning electron microscopy (SEM). EXPERIMENTAL PROCEDURE The experiments are conducted in a pin-on-roller wear tester. The material selected for the wear testing is epoxy resin fiber reinforced plastic composites supplied by Hydro S & S Industries, Pondicherry, India. In the present study the epoxy resins are used as a matrix for its high thermal and mechanical properties. It has temperature resistance up to 1408C wet/2208C dry and also has high water resistance. It has tensile strength of 8 Mpa. E-Glass fibers, the most common type of fibers used in many applications, have a maximum tensile strength of 2400 MPa, tensile modulus of 69 GPa, density 2.5 g/cc and specific modulus of 27. The GFRP can be manufactured by filament winding, resin transfer molding, or by pultrusion. For this study the GFRP is manufactured by the pultrusion technique with a fiber volume fraction of 35%. The material, which has the form of a cylindrical


1896

V. SRINIVASAN

ET AL.

rod size 10 12 mm, is used for the experimentation. The diameter of the glass fiber is 24 mm. The wear behavior of the composite against the hardened steel roller has been evaluated. The diameter and width of the hardened steel roller used in the wear tester is 60 mm and 12 mm, respectively. The tests are carried out in dry conditions, for five minutes for each specimen. The wear mass loss of the samples is determined by an electronic balance with an accuracy of 0.001 mg. The wear rate of the specimen has been obtained by measuring the weight loss of the specimen. Table 1 shows the experimental conditions used and Table 2 shows the specification of the machine used for experimentation. Figure 1 shows the experimental setup used for experimentation and Figure 2 shows the specimens used for wear tests.

Table 1. Experimental parameters for material wear.

Speed, rpm Load, Kg Temp, 8c

1

2

3

200 1 50

250 1.5 75

300 1.7 100

Table 2. Machine specification. Minimum load Maximum load Speed range Roller diameter Flat

– 42 N – 750 N – 30 – 800 rpm – 60 mm – 12 mm thick

Figure 1. Friction and wear testing machine used for the experiments.



1897

The wear rates as well as wear mechanism maps are drawn using normalized wear rate, normalized force and normalized velocity. The normalizations of variables can help in overcoming the complexity of using data from different sources, using specimens of different shapes and sizes with respect to wear rate, force and sliding velocity. The normalized velocity, force and wear rate are obtained using the following relations [15]: Normalized force, F¼

F An Ho

F ¼ force, Ho ¼ hardness:

ð1Þ

Normalized velocity, V¼

V ro a

a ¼ thermal diffusivity

V ¼ velocity; ro is the radius of a pinð5 103 mÞ ð2Þ

Normalized wear rate, W¼

W An

W ¼ wear rate; An ¼ nominal contact area:

ð3Þ

The wear rate has been calculated by using the relation: Wear rate ¼

Weight lost in time mm3 =min Density

ð4Þ

and the weight loss has been calculated by using: Weight loss ¼

WeightBefore WeightAfter g=min Time

Figure 2. FRP reinforced with epoxy resin specimens.


ð5Þ

1898

V. SRINIVASAN

ET AL.

WEAR MECHANISM MAPS The wear mechanism map can be developed using a probabilistic neural network. The wear mechanism map is plotted using the normalized load, normalized speed and normalized wear rate. The map is plotted by taking normalized velocity on the X-axis and normalized force on the Y-axis. The steps involved for the development of wear mechanism maps are shown below [16]. . Normalized force and velocity is given as an input vector. . A sample of data points covering the entire mechanism is collected from the numerical wear map. . New PNN is a toolbox used to design a probabilistic neural network. It creates a twolayer network. The first layer has a radial basis transfer function neuron; it calculates the layer’s output from its net input with the help of a Euclidean distance weight function, by applying weights to an input to get weighed input. The net input is calculated using a net input function, by combining its weighed input and biases. The second layer has competitive transfer function neurons, which also calculate the layer’s output from its net input; with the help of a dot product weight function by adding weighted input. The net input is calculated using a net input function. . Spread is used to specify the typical distance between input vectors. . Simulation is carried out by adding the output from the two layers, corresponding to the class to which it belongs. The output layer neuron produces a binary output value corresponding to highest probability density function, and makes the classification decision.

The probabilistic neural network used for fiber reinforced plastic composites is shown in Figure 3. The network paradigm also uses Parzen estimators, which were developed to construct the probability density function required by Bayes theory [17]. This theory explains how one can build an optimal classifier provided that the distributions from which the data emanate are known. Consider a pattern vector X with m dimensions that belongs to one of two categories K1 and K2. Let F1 (X) and F2 (X) be the probability density functions (pdf) for the classification categories K1 and K2, respectively. From Bayes’ decision rule, X belongs to K1 if: F1 ðXÞ L1 P2 > : F2 ðXÞ L2 P2

ð6Þ

F1 ðXÞ L1 P2 > F2 ðXÞ L2 P2

ð7Þ

Conversely, X belongs to K2 if:

where L1 is the loss or cost function associated with misclassifying the vector as belonging to category K1 while it belongs to category K2, L2 is the loss function associated with misclassifying the vector as belonging to category K2 while it belongs to category K1, P1 is the prior probability of occurrence of category K1, and P2 is the prior probability of occurrence of category K2. In many situations, the loss functions and the prior probabilities can be considered equal. Hence the key to using the decision rules given by Equations (6) and (7) is to estimate the probability density functions from the training patterns.


1899

Development of Wear Mechanism Map for Plastics X1

X2

X3

Input layer

Pattern layer

Summation layer

Output layer

X1-Normalized force X2-Normalized velocity X3-Temperature

Figure 3. Architecture of PNN.

In the PNN, a non-parametric estimation technique known as Parzen windows is used to construct the class-dependent probability density functions for each classification category required by Bayes’ theory. This allows determination of the chance that a given vector pattern lies within a given category. Combining this with the relative frequency of each category, the PNN selects the most likely category for the given pattern vector. Both Bayes’ theory and Parzen windows are theoretically well established, have been in use for decades in many engineering applications, and are treated at length in a variety of statistical textbooks. If the jth training pattern for category K1 is Xj, then the Parzen estimate of the pdf for category K1 is: " T # n x xj x xj 1 1 X exp 2 F1 ðxÞ ¼ m 2 ð2Þ 2 m nm j¼1

ð8Þ

where n is the number of training patterns, m is the input space dimension, j is the pattern number, and is the smoothing parameter. Figure 3 shows the basic architecture of the PNN. The first layer is the input layer which represents the m input variables (X1, X2, . . . , Xm). The input neurons merely distribute all of the variables X to all neurons in the second layer. The pattern layer is fully connected to the input layer, with one neuron for each pattern in the training set [16]. The weight values of the neurons in this layer are set equal to the different training patterns. The summation of the exponential term in Equation 8 is carried out by the summation layer neurons. DISCUSSION The wear rate for glass fiber reinforced plastics has been calculated by experimental evaluation. The mean effects of wear rate with respect to sliding speed, load and temperature have been plotted in Figure 4.


1900

V. SRINIVASAN

ET AL.

Wear rate (mm3/min)

Mean effects plot of sliding speed in m/s VS wear rate mm3/min 0.3 0.2 0.1 0 0.6

0.7

0.8

0.9

Wear rate (mm3/min)

Sliding speed (m/s)

1 Wear rate mm3/min

Mean effects plot of load in N VS wear rate mm3/min

0.3 0.2 0.1 0 8

10

12

14 Load (N)

16

18

Wear rate mm3/min

Wear rate (mm3/min)

Mean effects plot of temperature in °C VS wear rate mm3/min 0.4 0.3 0.2 0.1 0 40

60

80 Temperature (°C)

100

120 Wear rate mm3/min

Figure 4. Mean effects of wear rate, mm3/min against sliding speed, load and temperature (for epoxy resin).

Figure 4 clearly shows that when the sliding speed increases, the wear rate increases linearly. The graphical plot between applied load and wear rate shows that when the applied load increases, the wear rate increases. The wear rate, when plotted at low temperature, decreases; it increases when the temperature rises because melt lubrication reduces the wear rate considerably by reducing the friction. Wear Rate Maps From the experimental values, wear rate maps were drawn using MATLAB software by taking normalized velocity on the x axis, normalized force on the y axis and normalized wear rate on the z axis, using cubic interpolation as shown in Figure 5.


1901

Development of Wear Mechanism Map for Plastics 0

−4

Normalized force (log scale)

−1 −4 −3.5

−2

0 −0 5 −1 −1.5

−3

−3

−2.5

−3.5

−4 −5

−6 −7

−6 −7 −9 −8

−2

0

−8 −8.5

−8.5

−5

−3.5 −4 −5.5−4.5 −6.5 −7.5

−3 −2.5

0.5

1

1.5 2 2.5 3 Normalized velocity (log scale)

3.5

4

Figure 5. Field of dominance of different wear mechanisms.

Nominal area, An ¼

d2 ¼ 102 ¼ 78:53 mm2 : 4 4

F¼

F 9:81 ¼ 0:5678: ¼ An Ho 78:53 0:22

Normalized force,

Normalized velocity, V¼

V ro 0:628 5 103 ¼ ¼ 10161:81: a 3:09 107

Normalized wear rate, W¼

W 0:111 ¼ 1:413 103 : ¼ An 78:53

The plot between sliding speed and wear rate shows that when the sliding speed increases, the wear rate increases linearly for glass fiber reinforced plastic. An increase in sliding speed increases the temperature in the contact surfaces, thereby increasing the wear rate. Due to low coefficient of thermal expansion of FRP, the wear rate decreases considerably at higher sliding speeds. From the plot of applied load against wear rate, it is found that wear rate of glass fiber reinforced plastic composites increases rapidly with load, due to friction in contacting surfaces; when the load reaches around 15 N, frictional force


1902

V. SRINIVASAN

ET AL.

starts to melt the specimen, but melt lubrication considerably reduces the wear rate. The compressive strength of FRP is sufficient to withstand a load of above 15 N; therefore the wear rate decreases gradually. It is found that mild wear is predominant at lower and intermediate speeds, i.e., from normalized velocity of 0 to 3 and at normalized load of 8 to 3.5. The maximum wear rate is found to be at higher magnitudes of normalized velocity and normalized force; whereas at lower magnitudes the wear rate climbs down to 9. The severe wear is predominant at the higher values of normalized force from 3 to 0, corresponding to lower velocities ranging from 0 to 2.5 (normalized condition); traces of severe wear can also be found at higher speeds, say from 3 to 4, corresponding to lower values of normalized force from 9 to 4. The wear at higher loads is due to increase in frictional force, which squeezes the FRP over the steel. As a result, asperity heating leads to melting of FRP at a rapid rate. Here the wear rate ranges from 3.5 to 4, i.e, the wear rate gradients are steeper than the mild wear. In the second category, at higher speeds the wear rate ranges from 3.5 to 2.5, which is comparatively higher than the previous one. Here, as the sliding speed is more, there is not sufficient time for the specimen to dissipate the heat, i.e., before dissipating the heat, the specimen will make contact for the next revolution of sliding. The ultra severe wear is predominant at higher magnitudes of normalized velocity and normalized force. The normalized velocity ranges from 2 to 4, normalized force ranges from 3 to 0, and the normalized wear rate ranges from 3 to 0. At higher magnitudes, melt lubrication and hydrodynamic loading takes place [15]. When the force reaches 3, frictional force increases, thereby increasing the temperature. As a result of this, melting of the specimen takes place; the melt will act as a lubricant, which will reduce the friction. There should therefore be considerable fall in wear rate of the specimen. However, the wear rate increases irrespective of decrease in frictional force, because hydrodynamic loading of melt will further increase the wear rate of the specimen i.e., the effect of melt lubrication is neutralized by the effect of hydrodynamic loading. Wear Transition Map One of the important problems in the construction of a wear map is the search for boundaries of regions of predominant wear, and transitions where the competing processes may change the wear rate. The wear modes can be estimated by two approaches. The mechanical approach is based on the ratio of contact pressure to hardness value [7]. There is a boundary between mild wear, severe wear and ultra severe wear in the mechanics of sliding wear. The structural approach is based on the definition of dominant wear mechanism and transition [7]. The wear rate of the dominant mechanism can be estimated with the help of structure and a chemical state of the surface layer. Lim and Ashby [15] assumed that the definite wear mechanism predominates in all regions with the exception of transitions, i.e., only one dominant wear mechanism affects the wear rate of a specimen. The map represents mild, severe and ultra severe wear and transition between each mechanism. In comparison with other maps, this map defines the wear behavior of FRP over a wide range of sliding velocities and loads, taking into account the hardness of the rubbing pairs. The transitions are classified as load dependant, velocity dependant and sliding distance dependant. In the load dependant transition, the graph will be linear, whereas for the velocity dependant it is a smooth curve.


1903

Development of Wear Mechanism Map for Plastics 0

−1


−1 −4

ULTRA SEVERE WEAR

−2 SEVERE WEAR

−3

−3

−4 −5 −6

MILD WEAR

−7 −9 −8

0

−4

0.5

1

1.5

2

2.5

4

3.5

4

Normalized velocity (log scale) Figure 6. Wear transition map.

The collected wear rate data are plotted with velocity in x-axis, load in y-axis, and wear rate in z-axis. The MATLAB software is used to construct the map. On the map it was observed that the contours of wear rate are difficult to identify the boundaries between the mechanisms. The boundaries for this transition are interpolated from wear rates with increasing loads. The volumetric wear rates of epoxy resin FRP are plotted against the applied loads for the tests conducted (Figure 6). At all sliding speeds, the wear rates increased with increase of the applied load. As the sliding conditions were changed, the slope of wear rate increased abruptly from mild (9 to 4) to severe (4 to 2.5) and from severe to ultra severe (3 to 0) (Figure 6). When two surfaces slide against each other, the work done against friction is turned to heat. This temperature rise may modify the mechanical properties of the sliding surfaces and it may make them melt, which all lead to an increase in wear rate. The heat is really generated at asperities (tiny contact areas), which make up the true area of contact at the sliding interface. The instantaneous temperature of these contact points is obviously higher than the average temperature of the surface [15]. Wear Mechanism Map After completion of the wear transition map, it is essential to construct a wear mechanism map. The wear mechanism map may be constructed in two ways; one is by the empirical method, the other is by the model based method. Lim and Ashby [15] have constructed a wear mechanism map for steels with different compositions. Here the empirical method was adopted, where the map is constructed purely based on the experimental values. The wear rate of a sliding surface is conventionally defined as volume lost per unit distance slide. Its dimensions are m2; it is a derivative with respect to distance not to time. For each mechanism, wear rate is a function of normal force acting across the sliding surfaces, their relative velocity, their initial temperature and the thermal, mechanical and


1904

V. SRINIVASAN

(a)

ET AL.

(b)

Figure 7. (a) SEM micrograph of as-received GFRP specimen (500x); (b) SEM micrograph of as-received GFRP specimen (1000x).

chemical properties. If the mechanisms do not interact, then the dominant mechanism is the one which for a given force, velocity, temperature, etc has the greatest wear rate. Sliding may heat the surfaces, but the local temperature does not appear among the independent variables because it too directly depends on them [15]. The advantage of normalized variables can help in overcoming the complexity in using data from different sources; using specimens of different shapes and sizes can be correlated by using a normalized wear rate, force and sliding velocity. The worn out specimen has been studied using scanning electron microscopy (SEM). Wear mechanisms were identified through this study. The specimens in the form of received condition are presented in Figure 7(a) and (b). The friction and wear behavior of polymide and polymide based composites sliding against stainless steel has been analyzed by Jia et al. [18]. In analysis of wear, there are four different wear mechanisms, ironing, ploughing, brittle fracture and micro-cutting. The field boundaries of different wear regime are presented in Figure 8. When the worn out surface is smooth and shiny, then the region can be called ironing, which occurred at low speed and load. For the test this mechanism was attained at the sliding speed of 0.628 ms1 and at an applied load of 9.81 N, and is shown in Figure 8(a). Plastic ploughing is the region in which only a small proportion of the displaced material is actually detached from the surface, although the surface topography is modified. The deep longitudinal cracks observed on the sample are consistent with a repeated ploughing mechanism causing surface fatigue. For the test, this mechanism was attained at a sliding speed of 0.785 ms1 and an applied load of 14.715 N, as shown in Figure 8(b). Material is plastically deformed under load through ploughing; and with subsequent applications of loads, micro-cracks are created along the deformed materials, which are eventually removed by brittle fracture. For the test, this mechanism was attained at sliding speed of 0.785 ms1, and an applied load of 16.667 N is shown in Figure 8(c). When a much higher proportion of the plastically deformed material is lost as wear debris, then it is micro-cutting. The scratch-like grooves along the wear direction, present on the surface, are formed by micro-cutting by the abrasive particles. For the test, this mechanism was attained at a sliding speed of 0.942 ms1 and an applied load of 16.667 N, and is shown in Figure 8(d).


1905


(a) SEM micrograph for the condition of 0.785 ms−1 and an applied load of 16.667 N

(b) SEM micrograph for the condition of 0.942 ms−1 and an applied load of 16.667 N

0


−1 Micro-cutting

−2 −3 −4

Plastic ploughing

−5 −6

Brittle fracture

−7 −8

Ironing Elastic

0

0.5

1 1.5 2 2.5 3 Normalized velocity (log scale)

(c) SEM micrograph of epoxy resin FRP. For the condition of 0.628 m−1 and an applied load of 9.81 N

3.5

4

(d) SEM micrograph of epoxy resin FRP. For the condition of 0.785 ms−1 and an applied load of 14.715 N.

Figure 8. Field boundaries of different wear regime.


1906

V. SRINIVASAN

ET AL.

CONCLUSIONS . The wear rate for epoxy resin FRP has been determined as a function of force and velocity. . The wear mechanism maps showing the mechanisms of wear have been developed. . From the study of fracture phenomena, it is found that there are mainly four mechanisms of wear observed: ironing, plastic ploughing, brittle fracture and microcutting. . Probabilistic neural network has been used for the development of a wear mechanism map. The wear mechanism map can be used for the selection of optimum working conditions.

REFERENCES 1. Ning, X. (2005). Analysis of the Tribological Behaviour in Transversely Isotropic Materials Utilizing Analytical and Finite Element Methods, Materials Letters, 59: 175–179. 2. Wu, J. and Cheng, X. H. (2006). The Tribological Properties of Kevlar Pulp Reinforced Epoxy Composites under Dry Sliding and Water Lubricated Condition, Wear, 261: 1293–1297. 3. Lua, X., Wong, K. C., Wong, P. C., Mitchell, K. A. R., Cotter, J. and Eadie, D. T. (2006). Surface Characterization of Polytetrafluoroethylene (PTFE) Transfer Films During Rolling–Sliding Tribology Tests using X-ray Photoelectron Spectroscopy, Wear, 261: 1155–1162. 4. Williams, J. A. (2005). Wear and Wear Particles–Some Fundamentals, Journal of Tribology International, 38: 863–870. 5. Briscoe, B. J., Pelillo, E., Sinha, S. K. and Evans, P. D. (1996). Scratching Maps for Polymers, Wear, 200: 137–147. 6. Cenna, A. A., Doyle, J., Page, N. W., Beehag, A. and Dastoor, P. (2000). Wear Mechanisms in Polymer Matrix Composites Abraded by Bulk Solids, Wear, 240: 207–214. 7. Lim, S. C., Ashby, M. F. and Brunton, J. H. (1987). Wear-rate Transitions and their Relationship to Wear Mechansims, Acta Metall, 35(63): 342–348. 8. Ho, K. C. and Jeng, M. C. (1997). Tribological Characteristics of Short Glass Fiber Reinforced Polycarbonate Composites, Wear, 206: 60–68. 9. Xin, F., Xiaosong, D., Yijun, S., Huaiyuan, W., Shenghua, S. and Xiaohua, L. (2006). A Study on the Friction and Wear Behavior of Polytetrafluoroethylene, Filled with Potassium Titanate Whiskers, Wear, 261: 1208–1212. 10. Iglesias, P., Bermu´dez, M. D., Carrio´n, F. J., Chandrasekar, S. and Compton, W. D. (2007). A Study of the Wear Behavior of Polymer–Matrix Composites Containing Discontinuous Nanocrystalline Alloy Reinforcements, Tribology International 40: 479–489. 11. Neogi, S., Hashmi, S. A. R. and Chand, N. (2003). Role of PET in Improving Wear Properties of PP in Dry Sliding Condition, Bull Material Science, 26: 579–583. 12. Su, F.-H., Zhang, Z.-Z. and Liu W.-M. (2007). Tribological and Mechanical Properties of Nomex Fabric Composites Filled with Polyfluo 150 Wax and Nano-SiO2, Composites Science and Technology, 67: 102–110. 13. Unal, H., Sen, U. and Mimaroglu, A. (2004). Dry Sliding Wear Characteristics of Some Industrial Polymers against Steel Counter Face, Tribology International 37: 727–732. 14. Chang, L., Zhang, Z., Zhang, H. and Friedrich, K. (2005). Effect of Nanoparticles on the Tribological Behavior of Short Carbon Fiber Reinforced Poly(etherimide) Composites, Tribology International, 38(I): 966–973. 15. Lim, S. C. and Ashby, M. F. (1987). Wear Mechanism Maps, Acta Metall, 35 (4): 1–24. 16. Berthold, M. R. and Diamond, J. (1998). Constructive Training of Probabilistic Neural Networks, Neurocomputing, 19(1–3): 167–183. 17. Bolat, B. and Yildirim, T. (2003). Performance Increasing Methods for Probabilistic Neural Networks, Pakistan Journal of Information and Technology, 2(3): 250–255. 18. Jia, J. H., Zhou, H. D., Gao, S. Q. and Chen, J. M. (2003). A Comparative Investigation of the Friction and Wear Behavior of Polymide Composites under Dry Sliding and Water-lubricated Condition, Material Science and Engineering A, 356: 48–53.
