Penetration Prediction During Laser Welding Of Variable Thickness Plate Based On Multifeature Fusion Of Plasma

Tailor rolled blanks(TRB)is one of the main materials to realize lightweight of automobile.At the same time,laser welding is widely used in automobile industry as the connection technology of automobile materials.In the process of laser welding,the penetration of TRB is required to change adaptively with the change of material thickness.However,the change of penetration can not be measured directly.The penetration state can only be obtained by indirectly monitoring the change of light,sound and electrical signals.In this paper,the penetration state is predicted by indirectly monitoring the plasma profile and spectral signal.Firstly,the monitoring platform of "plasma profile spectrum" is built.It includes HWF50 laser 3D manufacturing system,electric displacement platform and control system,shape acquisition system and spectrum acquisition system.It can realize the real-time collection of plasma profile and spectral characteristics.The swing height,swing angle and the area of plasma are extracted by image processing.The spectral intensity,electron temperature and electron density are calculated by spectral analysis.Secondly,the relationship between plasma characteristic information and penetration state is established.In view of the problem of realizing real-time monitoring of penetration state in laser welding process,this paper divides the penetration state into "over-penetration","penetration" and "none-penetration" by observing the penetration depth and weld morphology,and establishes the relationship between the plasma shape characteristics and spectral characteristics.It is found that there is a correlation between the multi-feature information of plasma and the state of penetration.The feasibility of using plasma to indirectly monitor the state of penetration is verified.Finally,the fusion state prediction based on multi-feature information fusion of plasma is realized.BP neural network is used to classify the penetration state,and DS evidence theory fusion algorithm is used to fuse the plasma shape feature and spectral feature.The results of DS evidence theory fusion and BP neural network training are compared,which show that the accuracy of BP neural network with 6×3 structure model is higher than that of DS evidence theory fusion method,and the accuracy can reach 97%.In this paper,the penetration state prediction is realized by studying and monitoring the plasma characteristic information,and the correlation between the plasma characteristic information and the penetration state is analyzed.The fusion method of BP neural network and DS evidence theory is proposed to predict the penetration state,which provides guidance for the multi-sensor monitoring and information fusion of the penetration state,so as to realize the intelligent monitoring of laser welding.