Multi- source data diagnostics and weld penetration monitoring in oscillating laser-arc hybrid welding

This study proposes a multi-sensor data diagnostics and weld penetration monitoring approach to address the challenge of real-time penetration measurement in oscillating laser–arc hybrid welding of aluminum alloys. A multi-source off-axis monitoring platform was established to collect molten pool images and spectral signals. The geometric features of the keyhole and arc were extracted through image processing. Multi-source data diagnostics were then performed by combining principal component analysis (PCA), the sliding window algorithm, and empirical mode decomposition (EMD). An online weld penetration monitoring model based on the Long Short-Term Memory (LSTM) network was developed to effectively learn the temporal dependencies among the multi-source features. The model hyperparameters were optimized using a random search algorithm, achieving online weld penetration monitoring for oscillating laser–arc hybrid welding with an R2of 0.9897, an RMSE of 0.1964, and an MAE of 0.1521.