Online monitoring and process knowledge database-based feedback control of penetration depth during laser welding using coherent optical sensing

Assessment of penetration depth can reflect internal forming situations of laser beam welding, and controlling it by regulating process parameters has especially become a key issue in ensuring welding quality. This paper adopts coherent optical sensing to monitor penetration depth online, and implements its feedback control based on a process knowledge database. After utilizing empirical mode decomposition to process the original keyhole depth signal measured via coherent light, the reconstructed signal is found to exhibit a strong correlation with the penetration depth curve, and a deep neural network is built to accurately predict the penetration depth from the reconstructed keyhole depth. In the process knowledge database designed and created to store standard information on different expected penetration depth data under diverse combinations of process parameters, through the welding speed's alteration to produce variations in the scale of “penetration depth-heat input”, an artificial neural network is constructed to characterize the accurate mapping between penetration depth levels and heat input values. Closed-loop experiments demonstrate that when the monitored penetration depth curve varies abnormally during welding, the real-time control subsystem can refresh the required welding speed in accordance with the heat input gap predicted by the penetration depth gap and subsequently feed the penetration depth back to the desired level via executing the new welding speed. In all verification results, the global and local monitoring errors do not exceed 0.2127 and 1.0802 respectively, and the total monitoring-control latency of the entire system is between 0.2 s and 0.4 s. © 2026 Elsevier Ltd