Heat-resistant aluminum alloy design using explainable machine learning

The high-temperature strength of aluminum alloys must be enhanced for improving their applicability across industries. This study proposes a machine learning approach for developing heat-resistant aluminum alloys. Using a combination of correlation-based screening and genetic algorithms, feature selection was performed on descriptors derived from the atomic compositions of alloys. Then, alloy compositions and descriptors were used as input variables of the model to improve its robustness and applicability due to the richness of information. Four distinct alloys were discovered by employing Bayesian optimization within the framework of a quaternary alloy system. The best alloy demonstrated an exceptional high-temperature strength of 175 MPa at 300 °C in the absence of heat treatment. Microstructural analyses of these alloys indicated the critical role of vanadium-rich intermetallics in enhancing the high-temperature strength of aluminum alloys. Furthermore, the output of the model was explained using the SHapley Additive exPlanations method. The findings emphasize the critical importance of titanium and vanadium in enhancing the high-temperature strength of aluminum alloys tailored for environments with high thermal stress.