Machine Learning-Assisted Ultrasonic Rolling Optimization for Gradient Nanomechanical Enhancement of TA1-Welded Joints

This study explores the potential of ultrasonic surface rolling process (USRP) technology in enhancing the nanoscale mechanical properties of TA1 laser-welded joints. The nanoscale effects of the USRP process on the joint surface were explored by nanoindentation, three machine learning models (GRP, BPNN, SVR) combined with the TOPSIS algorithm were used to predict and screen the optimal process combination for the nanoscale mechanical properties (USRP-5, [Formula: see text] 250[Formula: see text]N, [Formula: see text] 2500[Formula: see text]mm/min, [Formula: see text] 0.05[Formula: see text]mm, [Formula: see text] 35). Finally, the specimen gradient was explored for the nanomechanical properties to validate it. The results show the surface nanohardness of USRPed TA1 laser-welded joints increased by 79.38% and the elastic modulus by 4.39% compared with the original specimen. Among the three machine-learning models, the SVR was the most effective, and the optimal parameter was the USRP-5. The nanohardness and elastic modulus of the USRP-5 sample gradually decreased to original values (at 800[Formula: see text][Formula: see text][Formula: see text]m) with the increase of the distance from the surface. This work introduces an innovative approach combining nanoindentation and machine learning to investigate the micromechanical properties of USRPed TA1 laser-welded joints at the nanoscale.