Taguchi and Artificial Neural Network Approaches for Optimization of Wear Behaviour of AA7178/MOS2/TiB2 Hybrid Composites

Introduction: Aluminium matrix composites (AMCs) are a type of innovative material, where the main matrix material is aluminium reinforced with particles, fibres, etc. Their mechanical and thermal properties can be improved by combining the reinforcements with the lightweight, corrosion- resistant characteristics of aluminium. The current study investigates the wear behaviour of stir casting AA7178 /MOS2/ TiB2 hybrid metal matrix composites. Method: The Pin-on-Disc (POD) apparatus is utilized to perform wear tests in compliance with the ASTM G99 standard, ensuring the accuracy and reliability of the results. A total of 16 wear tests are conducted by systematically varying critical wear parameters: applied load (AL), sliding speed (SL), sliding distance (SD), and the weight percentage of MoS2/TiB2 reinforcements (RT). The experimental design employs the Taguchi method to optimize wear behaviour parameters, specifically focusing on wear rate (WRT) and coefficient of friction (COF). Moreover, the Minitab statistical software is used for signal-to-noise ratio (SN) analysis and ANOVA testing. Result: ANOVA result indicates that the applied load is the most significant factor affecting WRT (73.04%), followed by reinforcement percentage (17.86%), sliding distance (7.33%), and sliding speed (1.77%). For COF, the applied load also has the highest contribution (36.3%), followed by reinforcement percentage (22.2%), sliding distance (13.5%), and sliding speed (7.7%). Conclusion: An ANN model is designed and trained using the LM algorithm to improve the accuracy of the WRT and COF predictions. The ANN is setup with the (4–10–2) architecture with four neurons in the input layer to represent the wear parameters. ANN has an R² coefficient as high as 0.98437, which shows outstanding fit between the predicted and experimentally determined values. In addition, the model has an extremely low MSE of 0.0036536, hence a very high accuracy in prediction. It is effectively trained, as it reaches the optimal performance by epoch 4, demonstrating the rapid convergence of the LM algorithm during ANN training.