Identification of Apparent Welding Defects Using Computer Vision Based On Deep Learning

One of the welding controls in health monitoring of structures is to visually control the appearance of welding defects (cracks, Spatter, Overlap, Lack of Fusion). Currently, according to regulations, the appearance quality of welding is controlled by an inspector visually. The accuracy of work in this method depends on the skill level of the inspector. Non using of equipment and technology leads to a high error in identifying visual defects. In this research, a method is proposed to be able to more accurately identify the appearance of welding defects with the help of imaging using machine vision based on deep learning. Convolutional network is used for deep learning to extract features from the image. The results show that the proposed method can identify welding defects with an acceptable accuracy (over 85%). Also, the results show that by using the proposed method, welding defects are evaluated more quickly compared to the traditional method.