Prediction of mechanical properties of friction stir processed 6061-O aluminium alloy using machine learning

This research focuses on the application of machine learning models for predicting friction stir-processed (FSPed) AA6061-O under varying processing conditions. In this research, a square pin profile was employed, and FSP process parameters such as number of passes (1, 2, and 3), tool rotational speed(TRS: 700 rpm, 1000 rpm, and 1300 rpm), and travel speed (TS: 25, 45, and 65 mm/min). The mechanical properties, such as microhardness and tensile strength, are measured and used as outputs for the machine learning models. According to the Taguchi L27 orthogonal array, experiments are designed, and the employed machine learning models (Training: 70 % of data, Testing: remaining) are J48, random forests, and regression algorithms. The machine learning analysis is exploratory due to the limited dataset size (27 experiments), and the reported predictive performance should therefore be interpreted as a proof of concept rather than a fully generalizable model. Measured mechanical properties suggest the dynamic recrystallization-induced grain refinement with increasing number of passes. Experimental results confirmed that, with increasing passes, mechanical properties show an increasing trend and align with prior findings. The random forest algorithm shows high internal consistency, and the random committee algorithm also shows a higher degree of predictability (R 2 > 0.95) than other machine learning models. This research supports industries by enabling trend analysis and the selection of process parameters to improve the mechanical properties of FSPed AA6061-O.