Prediction of Tribological Behaviour of AA5083/CSA-ZnO Hybrid Composites Using Machine Learning and Artificial Intelligence Techniques

Aluminium Alloys AA5083 dispersed with varying fractions of reinforcement was fabricated through the stir casting method. In varying weight percentage combinations, zinc oxide (ZnO) and coconut shell ash (CSA) particles were combined to create hybrid reinforcement particles. Using a pin-on-disc tribometer, the wear characteristics of the developed AA5083 hybrid composites were estimated. The volumetric proportion of hybrid reinforcement particles CSA (3, 6, 9 and 3 ZnO wt%), load (20, 30, 40 N), sliding velocity (2, 3, and 4 m/s), Cumulative Time (4.16, 5.55, and 8.33 min), and sliding distance are some of the experimental parameters (1000 m). Wear analysis revealed effective bonding and homogeneous dispersion of hybrid reinforcement particles onto the AA5083. Analysis of Specific Wear Rate (SWR) results showed that Specific Wear Rate rose with load, sliding velocity, and sliding duration while decreasing with hybrid particle dispersion. This research proposes the use of several intelligent classification techniques using Machine Learning (ML) and Artificial Neural Network (ANN) to predict the wear rate of an AA 5083 hybrid composite. For estimating wear quantities, the algorithms Random Forest (RF), Neural Network (NN), and k-nearest neighbours (kNN) are utilized. Six inputs are utilized to train and evaluate the Machine Learning (ML) algorithms: the Applied Load (N), Sliding Velocity, Sliding Speed, Cumulative Time, Percentage of Reinforcements, and Sliding Distance. The output is the Specific Wear Rate (SWR). The RF, NN, and KNN algorithms all produced success rates of correlation between experimental to anticipated of 0.90, 0.84, and 0.90, respectively. The same model data was utilised to train and evaluate Artificial Neural Networks (ANN), with the Multilayer Perceptron (MLP) network having the lowest Mean Square Error (MSE) to improve machine learning prediction accuracy. Maximum estimate error range of 0.1%, training and cross-validation of 0.00000496 and 0.0261, respectively, with linear correlation coefficient in testing of 0.9999 or 99.9% better prediction accuracy rate. The AA 5083 composites were designed and implemented using this machine learning and artificial neural network model for forecasting specific wear rate.