Available online at www.sciencedirect.

com

ScienceDirect
Materials Today: Proceedings 5 (2018) 11268–11276 www.materialstoday.com/proceedings

ICMMM - 2017

Artificial Neural Network Prediction On Wear Of Al6061 Alloy

Metal Matrix Composites Reinforced With -Al2O3
G B Veeresh Kumara*, R Pramoda, C S P Raob, P S Shivakumar Goudac
a. Department of Mechanical Engineering, Amrita Vishwa Vidyapeetham, Amrita University, Bangalore Campus, Karnataka State,
India.
b. Department of Mechanical Engineering, National Institute of Technology, Warangal, Telangana State, India.
c. Department of Mechanical Engineering, Visvesvaraya Technological University, SDM College of Engineering and Technology,
Dharwad, Karnataka, India.

Abstract

The exceptional performance of composite materials in comparison with the monolithic materials have been
extensively studied by researchers. Among the metal matrix composites Aluminium matrix based composites have
displayed superior mechanical properties. The aluminium 6061 alloy has been used in aeronautical and automotive
components, but their resistance against the wear is poor. To enhance the wear properties, Aluminium Oxide (Al203)
particulates have been used as reinforcements. In the present investigation Back propagation (BP) technique has
been adopted for Artificial Neural Network (ANN) modelling. The wear experimentations were carried out on a pin-
on-disc wear monitoring apparatus. For conduction of wear tests ASTM G99 was adopted. Experimental design was
carried out using Taguchi L27 orthogonal array. The sliding distance, weight percentage of the reinforcement
material and applied load have a substantial influence on the height damage due to wear of the Al6061 and Al6061-
Al2O3 filled composites. The Al6061 with 6 wt% Al2O3 composite displayed an excellent wear resistance in
comparison with other composites investigated. A non-linear relationship between density, applied load, weight
percentage of reinforcement, sliding distance and height decrease due to wear has been established using an artificial
neural network. A good agreement has been observed between experimental and ANN model predicted results.

© 2017 Elsevier Ltd. All rights reserved.

Selection and/or Peer-review under responsibility of International Conference on Materials Manufacturing and Modelling (ICMMM - 2017).

Keywords: Metal Matrix Composites, Al6061, Al2O3, Sliding Wear, Artificial Neural Networks ;

* Corresponding author. Tel.: +91-944-832-3291; fax: +91-802-844-0092.

E-mail address: veeru232@gmail.com

2214-7853 © 2017 Elsevier Ltd. All rights reserved.

Selection and/or Peer-review under responsibility of International Conference on Materials Manufacturing and Modelling (ICMMM - 2017).
 G B Veeresh Kumar et al / Materials Today: Proceedings 5 (2018) 11268–11276 11269

1. Introduction

The expanding interest for lessening in weight, saving energy and to reduce pollution in auto vehicles has prompt
to the improvement of novel, lightweight vehicle materials amid the most recent decade. Metal Matrix Composites
(MMCs) are a class of materials that try to consolidate the high solidness, stiffness with the sturdiness and damage
resistance gave by a metal matrix [1]. Lin Ye, et al [2] classified the uses of composite materials in military and
airplanes, clarified in insight about composite materials and structures in airframes. The earthenware particulate
filled composites display enhanced resistance to scratch [3] and they discover applications in car motor segments as
blocks of cylinders, pistons, cylinder embed rings, calipers [4], microwave channels, vibrator segment, contactors,
impellers and space structures. Among a different of aluminum compounds, Al6061 is highly investigated for its
fantastic properties. Al6061 combinations are exceedingly impervious to erosion, have magnificent extricability and
show modest strength. Al6061 are of much use in the fields of high way and building development, marine and car
applications [5]. U. Sanchez-Santana, et al [6] concentrated the impact of Laser Stun Handling surface treatment on
the wear of Al6061-T6 and demonstrated that wear rate decreased by around 68%, because of compressive lingering
stress field initiated. C. S. Ramesh et al, presumed that Al6061–TiO2 composite showed higher hardness, bring down
wear coefficient, when contrasted with matrix compound. From their work, it is apparent that Al6061–8 wt% TiO2
had the most reduced wear coefficient [7]. C. S. Ramesh et al, [8] considered wear conduct of hot expelled Al6061
with SiC, Al2O3 and cerium oxide filled composites and presumed Al6061-cerium oxide had the most reduced wear
rates under indistinguishable examination situations.
Wear an intricate marvel, where the genuine interaction territory among two exteriors contrasted and clear region
of contact is constantly little, in actuality restricting purposes of interaction amid exterior asperities. The essential
tribological factors that regulate the wear execution of filled aluminum composites are physical and mechanical
variables (outward to material experiencing surface collaboration) and material elements (characteristic for the
material experiencing surface connection) [9]. Wear is the most vital explanation behind damage and further
subsequent disappointment of machine components. A considerable measure should be directed to concentrate the
tribological conduct. This outcome in wastage of both labor and cash [10]. ANNs help in diminishing the price of
investigation when executed with attention and adequate information.
Late advance in information science and ability figuring gadgets offered a catalyst for the designing group to
reorganization its customary R&D benchmarks. Especially, Artificial Intelligence (AI), shows remarkable viability
in pleasing the profoundly requesting necessities of new era issues. AI fills in as an effective answer for complex
issues, for which customary clear consistent calculations are normally wasteful. A few variations beginning from
essential AI idea can be found in applications, in particular expert system (information based framework), genetic
algorithms, inductive learning, fuzzy logic and artificial neural network. ANNs were customized and prepared
utilizing a progression of run of the mill inputs and their comparing expected yields, to build up a certain multi-
dimensional and non-direct connection among them and prevent investigating the connection for a confounded
system. Enlivened by the natural sensory system, an ANN approach is an interesting numerical device, may be
utilized towards recreate a widespread assortment of multiple logical and engineering issues. Neural Network (NN)
systems may be prepared and altered via utilizing specific measure test information for very much planned neural
system. After the system has learned out on how to take care of the material issues, new information from the
analogous space may be anticipated by avoiding the conduction of long trials [11].
A few analysts have created ANN demonstrate utilizing MATLab [12, 13]. From the above ANN may be
effectively executed for forecasting of tribological properties of different materials. The goal of the current
assessment is to manufacture Al6061-Al2O3 composites, containing weight fractions of particulates from 2 to 6 wt%
in the means of 2. An ANN strategy is connected to foresee the tribological behaviour of Al6061- Al2O3 MMCs
utilizing neural system tool kit of MATLab and the trial and ANN outcomes be present thought about.
11270 G B Veeresh Kumar et al./ Materials Today: Proceedings 5 (2018) 11268–11276

2. Experimentation

2.1. Material Details

The alloy for the current reviews chose Fenfee Metallurgicals, Bangalore, for the supply of Al6061 as ingots.
Aluminum Oxide (Al2O3) filler material chose was of size 150µm from M/s. Snam Abrasives Private Limited,
Hosur, TN. The compositional and characteristic details about Al6061 and Al2O3 filler ingredients, composite
production techniques, physical, mechanical and wear property evaluation and the obtained results are published in
our before work [14-17].

2.2. Wear Tests

The test examples were as cylinders of 10 mm and height of 25 mm, the counter plate was steel EN31 of HRC 60
hardness. The height loss due to wear of the pin of cylindrical shape in microns was noted amid every wear test
utilizing a transducer of LVDT of precision 1.0 microns. Because of pin wear amid rubbing against counterpart,
cylindrical pin persistently moves to re-set up interaction through counter plate. The straight descending
development of the cylindrical pin indicates the amount of wear and documented through LVDT.
The exterior finish of the cylindrical pin plus counter plate were kept up by 0.1 µm Ra. Dry-sliding wear trials
were directed utilizing a PC supported cylinder on disc wear testing instrument on steady sliding speed (V=2.62
ms−1), load of 10-60 N in the means of 10 N, 6 km gliding separation of 6 km and track diameter 100 mm were
kept up. The tests were led as per ASTM G99 standard. Amid wear testing, tallness misfortune of pin noted in
microns.

3. Results and Discussions

3.1. ANN Model

Figure 1 demonstrates ANN design, which usually comprises various layers. To input layer the information
examples were connected, the output layer where the yield is gotten, and the layers amid the layers are the hidden
layers. Unseen (hidden) layers were named on the grounds that their yields are not straightforwardly observable. The
hidden layers expansion empowers the system to extricate complex-arrange insights which were especially
significant after the measurement of information layer stays substantial [18]. Neurons in every layer were organized
to going before plus ensuing layer neurons with every inter-connection have related association weight (strength).
These systems are ordinarily alluded to as multilayer perceptrons (MLP) [19]. The info flag spreads from side to
side by the network in a frontward heading, in a level by level premise. The unseen layers fill accompanying needs:
(a) improve nonlinearity to framework, (b) report associations amid information factors. There were many covered
layers, with numerous levels of nonlinearity and cooperations. The choice on the number of unseen layers stays very
critical variable to be noticed while tackling an issue [20]. The ANN demonstrate development and computational
points of interest are very much recorded in the writing [21]. Hence, the nitty gritty data won't be displayed here.
Every Input (info) information layer gets input flag Xi then communicates this Signal (flag) to other units in the
unseen layer. Each unseen unit Hi entireties its weighted info flag and applies its actuation function to figure yield
flag
Hj   i 1
Wij X i 
 
Ok    i1Wik H j 
 
Wij is the weight from info unit Xi to shrouded unit Hj. Yield flag of the unseen unit Hj directed other units in the
yield layer. Every yield unit Ok totals its weighted info flag and put on its enactment function toward process its
yield flag
 G B Veeresh Kumar et al / Materials Today: Proceedings 5 (2018) 11268–11276 11271

Figure 1. An artificial neural network with input, output and hidden layers.

Wjk is the weight from the unseen unit Hj toward yield unit Ok. The BP teaching algorithm depends on limiting
the sum square error (SSE), computed amid the NN preparing stage as
2
1 P
E   d P OP 
2 i 1
Where, dP is the obtained yield and OP is the computed yield from the ANN for the Pth pattern. Amid preparing,
an ANN is given the information of a large number of times, which is alluded to as cycle. After every cycle, the
error between the ANN yield (anticipated) and desired qualities are spread in reverse to modify the weight in a way
numerically ensured to scope [22]. The prepared neural system must be tried by providing testing information. In the
event that the testing error is considerably more contrasted with preparing mistake (training error), the system is said
to over-fit the information. An appropriately fitted system will give about equivalent preparing and testing mistakes.
In the current examination BP algorithm was utilized for ANN modelling. An ANN with 4 – [5 – 7]2 – 1 setup
was chosen similar to the network shown in figure 1 (with four info, one yield and two unseen layers with 12 unseen
hubs, 5 and 7 neurons in the neurons first and second covered layers respectively), with material and testing (two
parameters each) as information and wear tallness misfortune as yield.
ANN model were demonstrated utilizing NN (Neural Network) tool kit of MATLAB 7.7. NN device is an
acronym for neural network apparatus with realistic UI (GUI) module. The got trial wear information were
supplementary partitioned into preparing and testing set. The information was gathered from leading 24 tests. An
aggregate of 960 information were gathered out of which 816 were chosen as preparing information and the rest was
utilized as verifying information for check. A tan sigmoid transference function helped among the layers (aside from
yield layer), and direct linear transference function were utilized amongst last unseen layer and the yield layer to
abstain from constraining the yield an incentive to little range. This ANN model has been utilized to anticipate wear-
tallness misfortune. Subsequently best design of ANN and algorithms for learning were obscure ahead of time,
experimentation strategy was utilized amid preparing procedure to discover arrange attributes (number of shrouded
layers, number of handling units i.e., neurons and estimations of association weights between neurons, learning
algorithms) for coordinating a specific information/yield work. The feedforward backpropagation (BP) system,
utilizing LMBP algorithms with Mean Square Error (MSE) as the execution computing constraint were prepared.
The streamlining procedure were gotten by utilizing a TRAINLM work which was accessible inside the neural
network tool compartment. TRAINLM is a system preparing capacity that overhauls weightiness and inclination
standards in a BP algorithm as per Levenberg Marquardt improvement. Levenberg Marquardt algorithm is
profoundly proficient strategy aimed at taking care of non-straight maximization issues [23].
11272 G B Veeresh Kumar et al./ Materials Today: Proceedings 5 (2018) 11268–11276

3.2. Implementation

This section exhibits the points of interest of the ANN display training, testing and validation. The figures 2 (a) to
(d) consequences of regression graphs of training, testing, validation and normal of all. The graphs demonstrate a
normal regression of R=0.99944 estimation, i.e. a regression esteem near one implies the anticipated qualities were
in cozy association with yield as the information in charts lie on trend line. Similarly acknowledged for training,
testing and validation independently.

Figures 2 a-d. The regression graphs of training, testing, validation and complete of all
Figure 3, execution of the NN system, was measured utilizing MSE with the ages (epochs) i.e. quantity of cycles
did from the system so as to accomplish conjecture. An ANN accomplished steady-state after 108 cycles of
preparing. The error arc of preparing were demonstrated in Figure 3, and demonstrates the decrease of blunder over
the ages by the NN framework amid training, testing and approval. Speculation was ceased at the 114th age as there
are no changes. The finest execution were acquired at 108th age, which the MSE in the midst of get ready and
endorsement was seen to be 1.9825.

3.3. Prediction by ANN

Upon neural system effectively prepared and utilized toward anticipate newer outcomes in a similar information
space. New information sets are developed by the exploratory outcomes which were given to the NN system to the
forecast of wear attributes. The whole area learning got from the current examples is put away as weights connected
with every association between neurons in computerized shape.
G B Veeresh Kumar et al / Materials Today: Proceedings 5 (2018) 11268–11276 11273

Figure 3. ANN Mean squared error (MSE) performance.

11274 G B Veeresh Kumar et al./ Materials Today: Proceedings 5 (2018) 11268–11276

Figures 4 (a) to (f). Forecast for the association of tallness loss as a function of gliding distance and Al2O3 wt% filler at 10 to 60N load.

Figures 5 (a) to (d). Al6061 alloy and Al2O3 filled composites forecast for the association of height loss due to wear as a function of gliding
distance and applied load.
 G B Veeresh Kumar et al / Materials Today: Proceedings 5 (2018) 11268–11276 11275

Figures 4 (a) to (f), were sketched from constructing the utilization of put away area information of the NN
system and demonstrate the relationship amid the distinctive tribological properties (%’age weight of filler, load,
sliding distance and wear tallness misfortune (error)).
The expectation of wear tallness loss of the Al6061-Al2O3 filled MMCs as an element of gliding distance and
wt%’age of filler was appeared in figures 4 (a) to (f), at various loads. At steady load of 10N, fig 4 (a), demonstrate
the wear tallness misfortune diminishes as Al2O3 wt% filler increments from 0 to 6% and the wear tallness
misfortune increments bit by bit as the sliding distance increments. A comparative pattern was noticed in figures 4
(b) to (f), and may be reasoned that as the gliding distance expands the wear tallness misfortune increments and
declines with the expansion in the Al2O3 filler in the Al6061. The bigger gliding distance, the increment of
temperature of the gliding exteriors is inevitable. This causes softening of the alloy and composite stick exteriors
prompting overwhelming distortion at higher gliding distances. Promote the Al6061-6 wt% Al2O3 composite shows
bring down wear tallness misfortune. This change in wear resistance credited to the hardness Al2O3 particulates.
Figures 5 (a) to (d), demonstrate Al6061 wear tallness loss and Al2O3 filled composites as an element of load and
sliding distance. It was noticed from figures, that as the load increments 10 to 60 N, the wear tallness misfortune
increments quickly. Further at higher gliding distances for all composites examined, at all distances, the wear
tallness misfortune increments. With respect above the wt% of filler material, load and sliding distances have the
noteworthy impact on wear tallness loss of the Al6061 alloy and Al2O3 filled composites. Comparative outcomes
were acquired by different scientists [24]. Assist, the Al6061-6 wt% Al2O3 composite indicated amazing wear
resistance than alternate composites considered.
The forecast of wear tallness loss of the stick (pin) (Al6061-base and its Al2O3 filled composites) by way of a
wt% of Al2O3 filler material, load and sliding distance are appeared in the figures 4 and 5. The non-direct
association ANN system for forecast of height misfortune due to wear has been studied. The experimentation
outcomes demonstrates very much trained NN system can accurately foresee the height misfortune due to wear. In
light of the outcomes got from the prepared ANN show, anticipated outcomes were observed to be in close
concurrence with the test comes about. The created ANN display demonstrated great execution at an apparent level
inside a error scope of ±5%. From the outcomes, it can be watched that the expectation precision is sensible.
Besides, if the trial information base for preparing the system could be extended, the system can possibly yield
better outcomes. The outcomes uncovered that the all-around prepared ANN system may be utilized to foresee new
information from a similar learning area in this manner keeping away from reiteration of long haul tests, wastage of
labor and money [25].

4. Conclusions

The accompanying inferences are drawn from the reviews on tribological conduct of Al6061 and Al2O3 filled
composites utilizing ANN systems.
The wear tallness loss of the Al6061 and Al2O3 composites expanded as load and gliding distance expanded, yet
the same stood found to diminish as Al2O3 content expanded in the base material. The Al6061-6 wt% Al2O3
composite indicated brilliant resistance to wear than alternate materials contemplated. The non-direct association
among applied load, wt% of filler, gliding distance, density, and wear tallness misfortune were worked by utilizing a
ANN system. The outcomes got after the trained and tested ANN system were observed in-close concurrence with
the exploratory outcomes. An all-around prepared neural system can unequivocally anticipate the wear tallness loss
of Al6061 and Al2O3 composites.

Acknowledgements

The writers precise appreciation to the administration of School of Engineering, Amrita Vishwa Vidyapeetham
University, Bangalore, for their consolation and support amid the investigation.
 11276 G B Veeresh Kumar et al./ Materials Today: Proceedings 5 (2018) 11268–11276

References

[1]. Q.D. Qin, Y.G. Zhao, W. Zhou, “Dry sliding wear behavior of Mg2Si/Al composites against automobile friction material”, Wear 264 (2007)
654–661.
[2]. Schwartz MM. Composite materials. Upper Saddle River, NJ: Prentice-Hall PTR; 1997.
[3]. Kassim S. Al-Rubaie, Humberto N. Yoshimura, Jose Daniel Biasoli de Mello, “Two-body abrasive wear of Al–SiC composites”, Wear 233–
235 (1999) 444–454.
[4]. L. Ceschini, C. Bosi, A. Casagrande, G.L. Garagnani, “Effect of thermal treatment and recycling on the tribological behaviour of an
AlSiMg–SiCp composite”, Wear 251 (2001) 1377–1385.
[5]. B. Neitzel, M. Barth, M. Matic, “Weight Reduction of Discs Brake Systems with the Utilization of New Aluminum Material”, SAE
Technical Paper Series, 940335, Warrendale, PA, USA.
[6]. U. Sanchez-Santana, C. Rubio-Gonzalez, G. Gomez-Rosas, J.L. Ocana, C. Molpeceres, J. Porro, M. Morales, “Wear and friction of 6061-T6
aluminum alloy treated by laser shock processing” Wear 260 (2006) 847–854.
[7]. C.S. Ramesh, A.R. Anwar Khan, N. Ravikumar, P. Savanprabhu, “Prediction of wear coefficient of Al6061–TiO2 composites”, Wear 259
(2005) 602–608.
[8]. C.S. Ramesh, Mir Safiulla, “Short communication: Wear behavior of hot extruded Al6061 based composites”, Wear 263 (2007) 629–635.
[9]. R.K. Uyyuru., M.K. Surappa, S. Brusethaug, “Effect of reinforcement volume fraction and size distribution on the tribological behavior of
Al-composite/brake pad tribo-couple”, Wear 260, pp 1248–1255, 2006.
[10]. Hulya Kaçar Durmuş, Erdogan Ozkaya and Cevdet Meric, “The use of neural networks for the prediction of wear loss and surface roughness
of AA 6351 Al alloy”, Materials & Design, Vol 27, Issue 2, Pgs 156-159, 2006.
[11]. Z. Zhang, K. Friedrich, “Artificial neural networks applied to polymer composites: a review”, Composites Science and Technology 63
(2003) 2029–2044.
[12]. G. Partheepan, D.K. Sehgal, R.K. Pandey, “Fracture toughness evaluation using miniature specimen test and neural network”, Computational
Materials Science, 44, pp 523–530, 2008.
[13]. Arjula Suresh, A.P. Harsha, M.K. Ghosh, “Solid particle erosion studies on polyphenylene sulfide composites and prediction on erosion data
using artificial neural networks”, Wear 266, pp 184–193, 2009.
[14]. Veeresh Kumar.G.B, C.S.P.Rao, Selvaraj.N, Bhagyashekar.M.S., “Studies on Al6061-SiC and Al7075-Al2O3 Metal Matrix Composites”,
Journal of Minerals & Materials Characterization & Engineering, (JMMCE), USA, Volume 9, no.1, pp.47-59, 2009.
[15]. G.B. Veeresh Kumar, C.S.P. Rao, N Selvaraj, “Studies on mechanical and dry sliding wear of A16061-SiC Composites”, Journal of
Composites: Part-B Engineering (Elsevier Publications), volume 43, Issue 3, April 2012, Pages 1185-1191.
[16]. G.B. Veeresh Kumar, C.S.P. Rao, N Selvaraj, “Mechanical and dry sliding wear behavior of Al7075 alloy-reinforced with SiC particles”,
Journal of Composite Materials (SAGE Publications), Volume 46 Issue 10, May 2012, Pages 1201-1209.
[17]. G.B. Veeresh Kumar, A.R.K. Swamy, A. Ramesha, “Studies on Properties of As-Cast Al6061-WC-Gr Hybrid MMCs”, Journal of
Composite Materials (SAGE Publications), Volume 46 Issue 17, August 2012, Pages 2111-2122.
[18]. Schalkoff RJ. “Artificial neural networks”, McGraw-Hill; 1997.
[19]. S. Rajasekaran and G.A.Vijayalakshmi Pai, “Neural networks, fuzzy logic and genetic algorithms synthesis and application”, Prentice-Hall
of India Pvt., Ltd, New Delhi, 2004.
[20]. D. Aleksendric and C. Duboka, “Prediction of automotive friction material characteristics using artificial neural networks-cold performance”,
Wear, Volume 261, Issues 3-4, pages 269-282, 30 August 2006.
[21]. Munakata T., “Fundamentals of the new artificial intelligence beyond traditional paradigms”, New York: Springer-Verlag; 1998.
[22]. K.Genel, S.C Kurnaz, M. Durman “Modeling of tribological properties of alumina fiber reinforced zinc-Al composites using Artificial
Neural Network”, Material Science and engineering A, Volume 363, pages 203-210, December 2003.
[23]. H.R. Hafizpour, M. Sanjari, A. Simchi, “Analysis of the effect of reinforcement particles on the compressibility of Al–SiC composite
powders using a neural network model”, Materials and Design 30 (2009) 1518–1523.
[24]. S. Das, D.P. Mondal, S. Sawla, N. Ramakrishnan, “Synergic effect of reinforcement and heat treatment on the two body abrasive wear of an
Al–Si alloy under varying loads and abrasive sizes”, Wear 264, pp 47–59, 2008.
[25]. Veeresh Kumar G.B, C.S.P.Rao, Selvaraj.N, Bhagyashekar.M.S., “Prediction of Wear of Al6061-SiC Metal Matrix Composites using
Artificial Neural Networks”, International Journal of Computer Applications in Engineering, Technology and Sciences (IJ-CA-ETS) Volume
2, Issue 1, pp. 498-505, October 2009 – March 2010.