Decoding the wear-friction nexus in FSW dissimilar alloys via hybrid machine learning-evolutionary optimisation

Optimising the dry sliding tribology of friction stir welding for dissimilar aluminium alloys requires the resolution of the inherent conflict between the minimisation of wear and friction. In this scientific research, the wear-friction nexus of dissimilar AA5052-AA2014 joints was analyzed by introducing a hybrid Machine Learning and Multi-Objective Evolutionary Algorithm (ML-MOEA) framework to elucidate and optimise the wear-friction nexus. A Gaussian Process Regression (GPR) model and a Neural Network (NN) were built for high-fidelity simulation of wear rate and coefficient of friction (COF), respectively, based on the training of data from an investigation of a systematic pin-on-drum test. Non-dominated Sorting Genetic Algorithm II (NSGA-II) was then applied to true multi-objective optimisation and produced an explicit Pareto front yielding a measure of the performance trade-off. The optimisation identified two specialised regimes of optimal performance, one for near-zero wear (using a cylindrical pin and moderate-to-high load/velocity) and one for low friction (using a triangular pin and low load/velocity), thus showing that selection of pin geometry is intrinsically objective-specific and unattainable using conventional methods. Experimental validation was performed to validate the predictive accuracy of the framework, and SEM analysis was used to relate the optimal regimes to different wear mechanisms.