Use of Artificial Neural Network for Prediction of Mechanical Properties of α-Al 2 O 3 Particulate-reinforced Al–Si10Mg Alloy Composites Prepared by using Stir Casting Process

Abstract

In this article, the tensile strength, hardening behavior, and density properties of different α-Al₂O₃ particle size (μm)-reinforced metal matrix composites (MMCs), produced by using stir casting process, are predicted by designing a backpropagation (BP) neural network that used gradient-descent learning algorithm. Artificial neural network (ANN) is an intelligent technique that can solve nonlinear problems by learning from the samples. Therefore, some experimental samples are prepared at first to train the ANN to provide (to estimate) tensile strength, hardening behavior, and density properties of the MMCs produced for any given α-Al₂O₃ particle size (μm). The most important point is that after the ANN has been trained using some experimental samples, it gives approximately correct outputs for some of the experimental inputs that have not been used in the training. First, to prepare the training and test (checking) set of the network, some results are experimentally obtained and recorded in a file on a computer. In the experiments, α-Al₂O₃ particles are supplied commercially. α-Al₂O₃ ceramic powder of a varying particle size of 10 vol% is prepared, and then this ceramic powder with different α-Al₂O₃ particle sizes is added to Al–Si10Mg alloy in melt condition by stir casting process. The effect of reinforced particle size on the tensile strength, hardness resistance, and density properties of α-Al₂O₃-reinforced MMCs have been investigated. Mechanical tests reveal that tensile strength and hardness resistance of the α-Al₂O₃ ceramic powder composites decrease with increasing reinforced α-Al₂O₃ particle size. Then, neural network is trained using the prepared training set, also known as the learning set. In the preparation of the ANN training module, the aim of the use of the model is to predict the tensile strength, hardening behavior, and density properties for any given α-Al₂O₃ particle size by using some experimental results. Different α-Al₂O₃ particle sizes (μm) are used as the input, and tensile strength, hardening behavior, and density properties are used as outputs in the neural network training module. The tensile strength, hardening behavior, and density properties of the produced MMCs are estimated for different α-Al₂O₃ particle sizes using neural network efficiently instead of time-consuming experimental processes. At the end of the training process, the test data are used to check the system accuracy. Simulation results confirm the feasibility of this approach and show a good agreement with experimental results for a wide range of MMCs produced.