Advanced ensemble machine learning and response surface methodology for optimizing and predicting tribological performance of CMT-WAAM fabricated Al5356 alloy

This study investigates the tribological behavior of Cold Metal Transfer-Wire Arc Additive Manufactured (CMT-WAAM) Al 5356 alloy, optimizing and predicting Specific Wear Rate (SWR) and Coefficient of Friction (COF) using a hybrid Response Surface Methodology (RSM) and Ensemble Machine Learning (EML) approach. Wear tests were conducted using a pin-on-disc tribometer, with loads of 30-50 N, sliding velocities of 1-2 m/s, and 600-1800 m distances. RSM was employed to develop predictive models, and an ensemble model incorporating Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), and Random Forest Regressors was built to enhance prediction accuracy. SEM and EDS analyses of worn surfaces revealed key wear mechanisms, including abrasive, adhesive, and oxidative wear. Oxidation layers were observed, particularly under higher loads and velocities, contributing to wear debris formation. The optimal wear parameters identified through the desirability approach were Load = 33 N, Velocity = 1.8 m/s, and Sliding Distance = 1485 m, which were experimentally validated. The regression models developed using RSM were compared with predictions from EML models (ANN, LSTM, RF), demonstrating higher accuracy for EML with lower error percentages (SWR: 1.15%, COF: 1.15%) compared to RSM (SWR: 2.74%, COF: 1.79%). The results demonstrate that both RSM and Ensemble Machine Learning effectively predict wear behavior, providing valuable insights into optimizing wear resistance in CMT-WAAM-fabricated Al5356 components. This work contributes to understanding the tribological performance of aluminum alloys in additive manufacturing, offering practical guidance for improving the durability and performance of WAAM-fabricated components.