Influence of laser welding defocus and penetration monitoring based on advanced optical sensors

Laser welding is being considered for material jointing to improve the production efficiency and reduce the residual stresses of the joints for its advantages. In this study, six advanced sensors were used to monitor metal vapor, spatters, keyhole and molten pool during high power laser welding. Image process technology and statistically theory were used to analyze the features which were measured from six advanced optical sensors, including the reflective light intensity, visible light intensity, metal vapor area, spatters number, molten pool area. Then, comparative analysis of support vector machine and artificial neural network models for welding molten depth prediction were compared. Experimental results show that the welding molten depth could be predicted more accurately by the nonlinear autoregressive network with external input (NARX-NN) model than the support vector machine (SVM) model.