Intelligent Metal Welding Defect Detection Model on Improved FAST-PNN

Abstract

In order to solve the problem of accurate and efficient detection of welding defects in the process of batch welding of metal parts, an improved Probabilistic Neural Network (PNN) algorithm was proposed to build an automatic identification model of welding defects. Combined with the characteristics of the PNN model, the structure and algorithm flow of the FAST-PNN algorithm model are proposed. Extraction of welding defect image texture features of metal welded parts by a Gray Level Co-occurrence Matrix (GLCM) screens out the characteristic indicators that can effectively characterize welding defects. Weld defect texture features are used as input to build a defect classification model with FAST-PNN, for accurate and efficient classification of welding defects. The results show that the improved FAST-PNN model can effectively identify the types of welding defects such as burn-through, pores and cracks, etc. The classification recognition accuracy and recognition efficiency have been significantly improved. The proposed defect welding identification method can accurately and effectively identify the damage types of welding defects based on a small number of defect sample images. Welding defects can be quickly identified and classified by simply collecting weld images, which helps to solve the problem of intelligent, high-precision, fast real-time online detection of welding defects in modern metal structures; it provides corresponding evidence for formulating response strategies, with a certain theoretical basis and numerical reference.