Machine Vision Welding Defect Detection Based on FPGA

Abstract

Two defect detection algorithms are proposed in this paper because of the various types of defects in pole welding and the lack of obvious characteristics. Among them, the Halcon based defect detection algorithm can quickly and accurately detect most defect types; however, it cannot accurately distinguish between the bursting point and the welding penetration defects. Therefore, a defect detection algorithm based on Ray-Shooting is proposed. The Ray-Shooting model is a round core designed to extract pixel intensity characteristics around seed points. Firstly, based on the characteristics of the welding image, the OTSU algorithm and morphological operation were used to extract the welding area. Then, distance transformation is used to obtain the center point of the welding area and generate Ray-Shooting model at this point to extract features. Finally, the types of welding defects are classified based on the extracted features based on FPGA. Experiments show that this algorithm has high robustness. It can detect and classify the defects of pole pole pad, such as false welding, missed welding, burst point, welding penetration, etc. The accuracy rate of all test pictures is 100%.