Wear Behavior Analysis and Gated Recurrent Unit Neural Network Prediction of Coefficient of Friction in Al10Cu-B4C Composites

Aluminum-based metal matrix composites reinforced with B4C are advanced materials recognized for their exceptional combination of lightweight properties, high hardness, and superior wear resistance. These characteristics make them perfectly suited for applications demanding exceptional performance in extreme mechanical and tribological environments. This study investigates the wear behavior, microstructural characteristics, and predictive modeling of Al10Cu-B4C composites fabricated via powder metallurgy with varying B4C contents (0, 2.5, 5, and 7.5 wt.%). The addition of B4C microparticles to Al10Cu composites significantly influenced their tribological properties with 2.5 wt.% B4C achieving a 21.74% reduction in the coefficient of friction (COF) and 7.5 wt.% B4C providing a remarkable 65.00% improvement in wear resistance. Microstructural analysis using SEM and EDS was conducted on the unreinforced materials and the reinforced composites both before and after the wear tests. To further analyze and predict the tribological performance, a Gated Recurrent Unit neural network was developed to predict COF values. The need for this model arises from its potential to cost-effectively facilitate the prediction of COF, diminishing the need for extensive experimental testing while being noted for its simplicity and ease of implementation in practical applications. The model achieved excellent accuracy with an R2 of 0.9965 for the test set and 0.9917 for the validation set. Additionally, feature importance analysis using Random Forest models identified reinforcement-related features as the dominant predictors for both COF and mass wear. These findings demonstrate the potential of Al10Cu-B4C composites for emerging industrial applications, where enhanced wear resistance and controlled friction are critical for improving efficiency and durability under rigorous operating conditions. Furthermore, this study highlights the efficacy of neural network models in accurately predicting COF, providing a powerful tool for optimizing the performance of advanced composite materials.