Abstract
Resistance spot welding (RSW) of aluminum is widely applied in lightweight vehicle body manufacturing; however, the irregular oxide layer on aluminum surfaces degrades weld quality and limits the predictability of joint performance. To address this issue, this study proposes a machine learning (ML)-based model incorporating contact voltage as a key variable to improve the accuracy of tensile strength prediction in aluminum RSW. Experimental data, including preheating current, welding current, and electrode force, were analyzed in a time-series framework, with particular emphasis on monitoring contact voltage between materials to enhance prediction performance. The results demonstrate that the ML model considering contact voltage reduces prediction error by approximately 30% compared with conventional models, thereby significantly improving the accuracy of tensile strength prediction. These findings highlight that contact voltage measurement is a critical factor for evaluating aluminum RSW quality and suggest its practical contribution to quality assurance in automotive body manufacturing.