Validation of hardness and tensile strength of al 7075 based hybrid composites using artificial

Proceedings of the 2nd International Conference on Current Trends in Engineering and Management ICCTEM -2014
INTERNATIONAL JOURNAL OF MECHANICAL ENGINEERING
17 – 19, July 2014, Mysore, Karnataka, India
AND TECHNOLOGY (IJMET)
ISSN 0976 – 6340 (Print)
ISSN 0976 – 6359 (Online)
Volume 5, Issue 9, September (2014), pp. 280-285
© IAEME: www.iaeme.com/IJMET.asp
Journal Impact Factor (2014): 7.5377 (Calculated by GISI)
www.jifactor.com
280

IJMET
© I A E M E
VALIDATION OF HARDNESS AND TENSILE STRENGTH OF AL 7075
BASED HYBRID COMPOSITES USING ARTIFICIAL NEURAL NETWORKS
M. Sreenivasa Reddy1*, Dr. Soma. V. Chetty2, Dr. Sudheer Premkumar3, Reddappa H N4
1Research Scholar, JNTUH, Hyderabad and
Faculty, Department of Mechanical Engineering,
R.L. Jalappa Institute of Technology, Doddaballapur, Karnataka
2Principal, Kuppam Engineering College, Kuppam, (A. P.)
3Professor and Head, Department of Mechanical Engineering, JNTUH, Hyderabad, (A. P.)
4 Department of Mechanical Engineering, Bangalore Institute of Technology, Bangalore
ABSTRACT
Hybrid composites are relatively new class of materials characterized by excellent properties and dimensional
stability than those of usual composites. Based on the extensive literature review, it is concluded that majority of
investigations were carried out on Aluminium alloy based composite materials involving Silicon Carbide and Alumina as
reinforcements. The investigations using fly ash and E-glass have been carried out with matrix aluminium alloys other
than Al 7075 alloy. In the present investigation a new class of hybrid composite, Al 7075 alloy reinforced with fly ash
particulates, E-glass short fibers has been formed and experimental results have been validated by using Artificial Neural
Networks (ANNs). The ANN predictions were in very good agreement with experimental results, with correlation
coefficient 0.99918.
Key words: Al 7075, ANN, E-Glass, Fly Ash.
1. INTRODUCTION
A composite is defined as a structural material created artificially by combination of two or more materials
having dissimilar characteristics. The ingredients are unified at macroscopic level and are not soluble in each other. In the
combination, one constituent is known as matrix phase and the other as reinforcing phase which is embedded in the
matrix to give desirable characteristics [1]. The key feature of composites is that they generally exhibit the optimum
qualities of their constituents and often some attractive qualities that neither of the constituents possesses. Composite
materials are striking since they offer the possibility of attaining property combinations which are not possible to obtain
in monolithic materials and which can result in a number of vital service benefits [2].
1.1 Metal Matrix Composites
Metal Matrix Composites (MMCs) consists of either pure metal or an alloy as the matrix material, whereas the
reinforcement generally a ceramic material. The family of materials classified as metal-matrix composites (MMCs)
comprises a very wide range of advanced composites of immense importance to industrial (automobile), aerospace and
defense applications.
Metal matrix composites have usually of lighter metal such as (Al, Mg or Ti) or a super alloy (Ni based or Co
based super alloy). Aluminium is the most familiar matrix for the metal matrix composites. Metal Matrix Composites are
being progressively more used in aerospace and automobile industries due to their superior properties coupled with

considerable weight savings over unreinforced alloys [3]. The commonly used metallic matrices include Al, Mg, Ti, Cu
and their alloys. These alloys are chosen as matrix materials for the production of MMCs. The reinforcements being used
are fibers, whiskers and particulates [4].
281
1.2 Hybrid Composites

Hybrid composite is a composite material in which two or more reinforcing materials are combined into a metal
matrix [5]. Hybrid composites are formed to improve the properties. These are relatively new class of materials
characterized by excellent properties and dimensional stability at elevated temperatures than those of usual composites.
Hybrid composites find their applications in aerospace sector, automobile engine parts such as drive shafts, cylinders,
pistons and brake rotors.
2. ARTIFICIAL INTELLIGENCE
Recent developments in informatics and high capability computing devices have offered a new spring board for
the research community to rationalize its traditional Research and Development approach. Predominantly, Artificial
Intelligence (AI) an information processing technique exhibits outstanding effectiveness in facilitating the highly
demanding requirements of new generation problems. AI serves as an influential solution to complex problems for which
conventional methods are usually incompetent. Several variants originating from fundamental AI concept can be found in
application namely expert system, fuzzy logic, inductive learning, genetic algorithms and Artificial Neural Networks
ANN [6].
2.1 Prediction of Properties of Composites by Software Techniques
Recent trends in informatics and high capability computing devices has offered a new brand platform for the
research community to reshuffle its traditional RD criteria. In particular Artificial Intelligence (AI), an information
processing technique, exhibits remarkable effectiveness in accommodating the highly demanding requirements of current
generation problems. AI serves as a powerful solution to complex problems, for which traditional methods are
inefficient. One such variant originated from fundamental AI concept is Artificial Neural Network (ANN).
ANN can be tailored and trained using a series of distinctive inputs and their corresponding predictable outputs
to establish an inherent non-linear and multi-dimensional correlation between them, establishing the constitutive relation
for a complicated system. Inherently capable with talents in adaptability, robustness and parallelism, the ANN technique
has found extensive applications in pattern recognition, classification and functional approximation signal processing and
system identification [7]. Inspiring by the biological nervous system, ANN approach is an enthralling mathematical tool,
which can be used to simulate a wide variety of complex scientific and engineering problems. ANN can be tailored and
trained by using a certain amount of experimental data to a well-designed ANN. After the network has learnt to solve
material problems, new data from the similar domain can then be predicted without performing too many long
experiments [8].
Back Propagation Neural Network (BPNN) is a Multi-Layer Perceptron (MLP) network consisting of an input
layer, hidden layer and an output layer. There may be one or more hidden layers and each hidden layer may consist of a
number of neurons (nodes). Hence, when BPNN is made use of, there is a wide scope to look at the influence of network
architecture (number of hidden layers and neurons) on the estimation performance of the network [9].
In recent times, ANNs have established a great deal of attention as a prediction and modeling tool in many research
areas. ANNs can be defined as massively parallel distributed processors, which have capability to store experimental
knowledge and make it available to use [10]. The main feature of the network is that the network establishing the
relationship is trained directly by examples without any complex formulae about the nature of the problem. The ANN
method is appropriate when huge database is available, it is complicated to find an exact solution for a problem by
mathematical methods and when the data set is partial, noisy and complex.
2.2 Artificial Neural Networks
A neural network is an artificial illustration of human brain that tries to simulate its learning process. An
Artificial Neural Network (ANN) is often called as neural network. Artificial Neural Networks are considered as
artificial intelligence modelling techniques. They have a hugely interconnected structure analogous to brain cells of
human neural networks and consist of huge number of simple processing elements called neurons, which are arranged in
different layers in the network: an input layer, an output layer and one or more hidden layers. ANN is an interconnected
group of artificial neurons that uses a mathematical model or computational models for information processing based on
a connected approach to computation. The artificial neural networks are made of interconnecting neurons which may
share some properties of biological neurons.

282
3. NEURAL NETWORK MODELING

Neural networks have learning and generalization capabilities. They are competent to learn the correlation
connecting input values and the predictable outcome, more significantly generalize the relationship. Learning is the
process in which random-valued parameters (weight and bias) of a neural network are adopted through a continuous
process of simulation by the environment, with which the network is embedded, called training phase. There are several
of training methods to train neural networks but the back propagation model emerged as the most accepted training
method. Subsequent to the training phase, ANN parameters are set and the system is deployed to solve the problem at
hand (the testing phase). Due to this ability, neural networks are found to be as a useful tool in pattern recognition and
classification.
3.1 Layers of a Neural Network
The capability of a single neuron is confined; complex functions can be realized through connecting several
such neurons to form layers of neuron network as shown in Fig. 1. A layer is defined as group of parallel neurons. The
common type of ANN consists of 3 layers:
i. Input layer ii. Hidden layer iii. Output layer
A layer of input units is linked to a layer of hidden units which is coupled to a layer of output units. Patterns are
presented to the networks through input layers, which communicates to one or more hidden layers where the real
processing is done through a system of weighed connections. The hidden layers then link to an output layer.
X1
X2
X3
Y1
Y2
Figure 1: A simple neural network
In order to train a neural network to perform a specific task, the weight of each unit is adjusted in such a way
that the error between the preferred output and the actual output is abridged and Regression Co-efficient (R value)
between preferred output and the actual output is nearer to one. This process requires the neural network to compute an
error derivative of the weight. The back-propagation algorithm is most extensively used technique for obtaining the value
of MSE and R. During training, the predicted output has been compared with the desired output, and the R value and
MSE is calculated. If the R value and MSE value is more than a prescribed limiting value, it is back propagated from
output to input, and weight are further modified to obtain the R and MSE value within a prescribed limit. The artificial
neural network back propagation algorithm is implemented in MATLAB language.
3.2 Training Using Artificial Neural Network
There are two ways of training the data namely incremental and batch training. In incremental training the
weights and biases of the network are updated every time an input is fed to the network. In batch training the weights and
biases are merely updated subsequent to all the inputs are fed. Data will be loaded individually in the command window.
The following function can be considered as an example.
Input = [0 10 20 30 40 50 60 70 80 90 100]
In this, input is the variable name and 0, 10, 20…..100 are the inputs for analysis. The values are then fed automatically
onto the workspace which has a set of variables built up during a session, using the MATLAB software and stored in
memory. Variables are added to the workspace by using functions, running M-files and loading stored workspaces. The
obtained values from the data cursor mode are then loaded onto the command window using a dissimilar variable name
with same function for the purpose of regression analysis which is to be carried out using the neural network toolbox.

283

By invoking ‘nftool’ syntax in the command window, neural network fitting tool Graphical User Interface
(GUI) is opened. The ‘nftool’ leads by means of solving a data fitting problem, solving it with a two-layer feed-forward
network trained with Levenberg-Marquardt.
Using the inputs and target options in the select data window, the desired output values are obtained. Fig. 2
shows a Neural Network of the model.
Figure 2: Neural Network Model
Generally one epoch of training is stated as a single presentation of all input vectors to the network. The
network is then updated in accordance with the results of all those inputs. Training will continue till a maximum number
of epochs occur, the performance goal is achieved. Training automatically stops when generalization stops improving, as
indicated by an increase in the mean square error of the validation samples. Training multiple times will generate
different results due to varied initial conditions and sampling.
Under the option plots, by using Regression, plot regression syntax is invoked. The plot regression (targets,
inputs)’ plots the linear regression of targets relative to inputs. These plots are regression plots for the output with respect
to training, validation and test data.
3.3 Mean Square Error
The most important parameter for the neural network performance evaluation is Mean Square Error (MSE) and
Regression co-efficient (R value). Considering the Regression co-efficient, MSE for the network training to evaluate the
performance of network, the optimized network structure is identified by comparing the R value and MSE value of each
structure. This R value and MSE value calculation is used to determine the amount of variance between the expected and
actual output.
Figure 3: Training progress
Input data is fed into the neural network, corresponding weights and bias are extracted. Then weights and bias
are integrated in the program which is used to calculate and predict the values. When the training has been completed, it
is necessary to make sure the network performance and decide if any variations are desired to the training procedure,
network architecture or the data sets. Fig. 3 shows the performance plot for four neurons at 70% training set. The

iteration at which the validation performance has approached a minimum was eight. The training was continued for five
more iterations and stopped.
284
3.4 Regression Analysis

Regression is the process of fitting models to the data. For validation, a regression plot is obtained which shows
the relationship between the outputs of network and the targets. If the training correlates, the network outputs and the
targets would be accurately equal but the relationship rarely correlates in general practice. Fig. 4 shows the regression
plots representing the training, validation and testing data. The dotted line indicates the perfect result - an output is equal
to targets. The thick line represents the best fit linear regression line between outputs and targets. The value of R is a
mark of relationship between the outputs and the targets. The value of R = 1, signifies that there is an exact linear
relationship between outputs and targets. When R is very close to zero, it implies that there is no linear relationship
between outputs and the targets.
Figure 4: Plots of Regression Analysis
The data consists of 40 experimental samples, 70% of the experimental data has been used for training the NN
model and 30% for validation and testing. In training session, validation is stopped after 13 iterations at Mean Square
Error (MSE) very close to zero with correlation coefficient 0.99918, which indicates excellent matching between the
experimental data and prediction of neural network model. Table 1 shows percentage error between experimental and
ANN predicted values of hardness and tensile strength.
Table 1: % Error Between Experimental and ANN Results
Hardness Tensile Strength
Sample Expt.
Value
ANN
value
%
error
Expt.
Value
ANN
value
%
error
Plain 75 75.215 0.28 130 129.893 0.08
C1 77 81.121 5.35 132 133.151 0.87
C2 86 86.097 0.11 133 132.863 0.10
C3 98 98.107 0.11 135 135.547 0.40
C4 107 98.608 7.84 136 138.250 1.65
C5 117 116.085 0.78 136 136.470 0.34
C6 120 120.020 0.02 139 139.643 0.46
C7 135 128.465 4.84 145 148.276 2.26
C8 138 137.092 0.65 148 148.691 0.46
C9 140 140.741 0.53 150 150.957 0.64

285
4. CONCLUSIONS

Al 7075 based hybrid composites were successfully formed by stir casting method with different weight
percentage of fly ash and E-glass reinforcements with Al 7075 as matrix material. Within purview of the study, following
conclusions have been drawn.
i. By making use of stir casting method, fly ash particulates and E-glass fibers can be successfully introduced in
the Al 7075 matrix alloy material to fabricate hybrid composite materials.
ii. Artificial Neural Network modeling has been carried out to link the input variables to the performance
measures. A neural network has been designed to validate the experimental results obtained. The neural network
was successfully trained and tested using the experimental data.
iii. Artificial Neural Networks can be used competently as prediction techniques in the area of material
characterization.
iv. The ANN predictions have revealed a good accordance with experimental results obtained, establishing a
correlation coefficient of 0.99918 and thus the experimental results have been validated.
5. REFERENCES
[1] V.S.R. Murthy, A.K. Jena, K.P. Gupta, G.S. Murty, Structure and Properties of Engineering Materials (Tata
Mc Graw-Hill, 2008).
[2] Robert M. Jones, Mechanics of composite materials (Scripta book company, Washington D.C 1975).
[3] T. Miyajima, Y Iwai, Effect of reinforcements on sliding wear behavior of aluminium matrix composites, Wear
255(2003), pp 606-616.
[4] M.D. Bermudez, G.Martinez-Nicolas, F.J. Carrion, Dry and lubricated wear resistance of mechanically-alloyed
aluminium-base sintered composites, Wear 248 (2001), pp 178-186.
[5] Khanna.O.P, Material Science and Metallurgy (Dhanpat Rai Publications,2007).
[6] G.B. Veeresh Kumar, C.S.P Rao, N. Selvaraj, Mechanical and tribological behaviour of particulate reinforced
Alumnium Metal Matrix Composites, a review, Journal of minerals and Material characterisation
Engineering, 10(2011), pp 59-91.
[7] Lin Ye, Zhongqing Su, Guang Meng, Functionalized composite structures for new generation airframes: a
review, Composites Science and Technology 65(2005), pp 1436-1446.
[8] Z. Zhang, K. Friedrich, Artificial neural networks applied to polymer composites: a review, Composites Science
and Technology 63(2003), pp 2029-2044.
[9] G. Krishna Mohana Rao, G. Ranga Janardhana D. Hanumantha Rao and M. Srinivasa Rao, Development of
hybrid model and optimization of surface roughness in electric discharge machining using artificial neural
networks and genetic algorithm, Journal of materials processing technology, 2 (9), (2009), pp 1512–1520.
[10] Rasit Koker, Necat Altinkok, Adem Demir, Neural network based prediction of mechanical properties of
particulate reinforced metal matrix composites using various training algorithms, Materials Design, 28,
(2007), pp 616-627.

Validation of hardness and tensile strength of al 7075 based hybrid composites using artificial

More Related Content

What's hot (19)

Viewers also liked (9)

Similar to Validation of hardness and tensile strength of al 7075 based hybrid composites using artificial (20)

More from IAEME Publication (20)

Recently uploaded (20)

Validation of hardness and tensile strength of al 7075 based hybrid composites using artificial