Line Laser Vision-Guided Weld Seam Tracking System Research

Traditional welding robot processing relies heavily on demonstration teaching or offline programming, resulting in low flexibility and poor adaptability in the production process. Under the influence of complex and changing welding environment, there are still difficulties in weld seam recognition; when the position of the weldment is shifted or deformed by the high temperature of the welding torch, it will cause the error between the actual welding trajectory and the preset trajectory. This paper mainly focuses on the optimization and innovation of weld seam image recognition, after completing the calibration of each part of the welding system, using convolutional neural network to detect and recognize the weld seam, and carrying out a series of processing for the collected images, on this basis, it is proposed to establish an improved Faster Region-based Convolutional Neural Networks model, adjust and optimize the model structure, and at the same time, use the Feature Pyramid Networks algorithm for target detection. Through the comparison of the detection performance of different networks, it is proved that the improved Faster-RCNN network is more accurate in weld seam detection.