A machine learning approach to mitigating hardness degradation in the heat-affected zone of AA7075-T6 alloy

Friction stir welding of precipitation-strengthened aluminum alloys presents a significant hardness difference between the heat-affected zone (HAZ) and the base metal (BM). A wide range of processing, materials, machine and tool parameters (such as tool rotational rate , tool traverse speed, and tool geometry, especially shoulder diameter), are known to affect the HAZ hardness, making it difficult to navigate the vast combinatorial parameter search space for optimum HAZ hardness. In this study, machine learning (ML) techniques were applied to identify the key processing parameters that affect the hardness difference between the HAZ and BM in AA7075-T6 alloy. Later, an adaptive design strategy was implemented to iteratively determine the optimal combination of processing parameters needed to obtain a targeted hardness difference in the HAZ. Initial data was collected from our random experiments as well as data within reported literature. The initial dataset was then used to train adaptive design tools, and the ML predicted processing parameters were validated via subsequent experiments. It was demonstrated that the ML-guided approach required the generation of only six experimental data points to attain a hardness difference of 41.00 HV 0.3 in the HAZ, compared to trial-and-error based conventional method in which 32 data points were needed to obtain a HAZ hardness difference of 42.00 HV 0.3, thereby significantly increasing the efficiency of process optimization for a targeted mechanical property. Finally, high fidelity mathematical equations were extracted from the data to predict hardness difference in the HAZ as a function of processing parameters using explainable artificial intelligence.