关键词: 焊接缺陷, 旋转磁场, 磁光检测, 有限元仿真, 主成分分析, BP-AdaBoost强分类器

Abstract: Accurate detection and classification of weld defects plays an important role in ensuring the quality of welding products. A magneto-optical imaging nondestructive detection method of weld defects under rotating magnetic field excitation is studied. The mechanism of magneto-optical imaging of weld defects based on Faraday magneto-optical effect is analyzed, the magnetic field distribution of weld defects is analyzed by using finite element simulation, and a best lift-off degree is found to collect magneto-optical images. The mechanism of rotating magnetic field is analyzed, a pair of cross yokes is used to generate a rotating magnetic to excite the weldment. Also, a magneto-optical sensor is used to obtain the images of weld defects. Using PCA to extract and reduce the dimension of gray-scale features of column pixels in magneto-optical images, the AdaBoost algorithm combined with BP neural network is used to establish a BP-AdaBoost weld defects classification model. Experimental results show that the proposed method can effectively improve the recognition accuracy of No.45 steel weld joint defects (curve crack, linear crack, sag, less penetration), the overall recognition rate of BP-AdaBoost weld defects classification model reaches 98.88%, and the detection and classification of weld defects can be realized effectively.

Key words: weld defects, rotating magnetic field, magneto-optical detection, finite element simulation, PCA, BP-AdaBoost strong classifier