Intelligent Detection to Laser Welding Seam Defect Using Adaptive Joint Wavelet Dictionary Learning of PEC Signal

Abstract

The pulsed eddy current (PEC) is an effective method for the online detection of laser welding seam defects. The joint wavelet dictionary learning method is proposed for solving the separation problem of the broad frequency harmonic and local non-smooth distortion of the PEC signal. The Haar and Gabor wavelet is adopted to be the basic function, which is extended to be the over complete wavelet dictionary library by cyclic migration. The sparse representation of the defect PEC signal is obtained by combining the joint wavelet dictionary with the orthogonal matching pursuit algorithm. The feature parameters of the PEC signal are calculated and inputted into the support vector machine to detect the laser welding seam defect intelligently. The validity of the proposed method is further verified by the experimental results, demonstrating the effectiveness of the classification identification and quantitative assessment of the pore and crack.