DNN-Based Quality Monitoring of Al-Cu Dissimilar Dual-Beam Laser Welding Using Spectrometer and Photodiode Signals

As electric vehicles (EVs) become more widespread, the demand for reliable, high-speed joining of battery components such as copper (Cu) and aluminum (Al) is rapidly increasing. Laser welding is well-suited for automated EV production due to its precision, speed, and ease of integration. However, the significant mismatch in thermal and metallurgical properties between Cu and Al leads to unstable weld formation, which can degrade mechanical performance such as joint strength and penetration consistency. Ensuring weld reliability thus requires real-time monitoring of mechanical and geometric quality indicators during welding. To address this need, this study proposes an optical signal-based monitoring method for dual-beam laser welding of Al-Cu dissimilar metals. Optical emissions were captured using a spectrometer and photodiodes, then transformed into FFT-based features combined with welding parameters. A multi-output deep neural network (DNN) was trained to simultaneously predict tensile strength (regression), penetration mode (multi-class classification), and gap presence (binary classification). Experimental results showed a strong correlation between optical signals and weld quality, validating the feasibility of real-time quality prediction. This approach enables data-driven in-process monitoring for automated quality assurance in EV battery welding applications.