Ensemble machine learning for predicting and enhancing tribological performance of Al5083 alloy with HEA reinforcement

Abstract

This study investigates the tribological behavior of Al5083 alloy reinforced with AlCoCrFeNiSi high-entropy alloy (HEA) particles using friction stir processing (FSP). Wear characteristics were analyzed using pin-on-disc experiments across varying HEA volume percentages, disc speeds, and test durations, revealing significant improvements in wear resistance with increasing HEA content. Machine learning techniques, including artificial neural networks (ANN) and long short-term memory (LSTM) networks, were employed to predict specific wear rate (SWR) and coefficient of friction (COF) with high accuracy. The ensemble model combining ANN and LSTM architectures achieved R-squared values of 0.9653 for SWR and 0.9718 for COF, with a root mean square error (RMSE) of 0.024 and 0.017 for SWR and COF respectively, indicating robust predictive capabilities. Cross-validation further validated the model's effectiveness, achieving an average prediction error of 2.13% for SWR and 1.89% for COF. Response surface methodology (RSM) optimization refined process parameter relationships, identifying conditions that minimize SWR to 3.57 × 10^–6 mm³/Nm and COF to 0.237. Scanning electron microscopy (SEM) analysis of worn surfaces confirmed the effectiveness of HEA reinforcement in mitigating wear mechanisms, enhancing the material's durability by 45% compared to the unreinforced alloy. This comprehensive approach advances the understanding of HEA-reinforced composites. It provides practical insights for optimizing material performance in industrial applications, contributing to developing high-performance materials tailored for durability and wear resistance.

Graphical Abstract

This is a visual representation of the abstract.