Feature Extraction and Classification Recognition of Molten Pool in Multi-layer and Multi-pass Welding of Medium and Thick Plates

Multi-Layer and Multi-Pass Welding (MLMPW) for medium thick plates is widely used in fields such as marine engineering, nuclear welding, and high-speed rail welding. The implementation of feature extraction and classification of different layer weld pools in multi-layer and multi pass welding of medium and thick plates can accurately distinguish the types of weld pools in different layers during the welding process, whether the current weld pool meets normal weld pool parameters, and is of great significance for path planning, real-time correction, and penetration prediction control of multi-layer and multi pass welding. This article is based on the welding experiment of multi-layer and multi-pass welding of medium and thick plates, and carries out image processing and feature extraction of the weld pool in multi-layer and multi-pass welding of medium and thick plates based on visual information and current and voltage information. The classification of the weld pool in four layers and seven passes is also studied. This article uses a U-net network-based fusion pool image segmentation method to segment and process the fusion pool image, and provides relevant evaluation indicators to objectively evaluate the segmentation effect of the U-net fusion pool image. The edge of the segmented fusion pool image is extracted, and the definition of the maximum width of the fusion pool and the curvature radius of the three parts is given, and relevant parameters are calculated, and classify the four layers and seven weld seams according to the calculated pool parameters and provide the corresponding classification parameter range.