A method for detecting welds and weld defects based on deep learning

Abstract

The invention discloses a method for detecting weld seams and weld seam defects based on deep learning. The YOLOV3 network is used to realize the detection of weld seams and / or weld seam defects; the training steps of the network: the workpiece image is frame-selected by the positioning frame for the weld seam , marking, as a training data set; use the positioning frame to frame the weld defect and mark the defect type as the training data set I; obtain the coordinate x of the positioning frame p 、y p , and the width and height dimensions w p 、h p ; Initialize the network; randomly call the input tensor a j Carry out training calculations and output the detection results; use the detection results to calculate the error function loss of the prediction results; combine the gradient descent method to adjust the weight W and the bias value b, and cycle like this to obtain a trained network; this method, for multiple welds, Simultaneous detection of various defect types, one-time measurement can realize welding seam identification and positioning and defect detection, effectively improving measurement efficiency and accuracy.

Description

technical field [0001] The invention relates to the field of defect detection, in particular to a deep learning-based weld seam and a method for detecting weld seam defects. Background technique [0002] With the development of automation technology, industrial welding robots have been widely used in the field of processing and manufacturing, and have become the main automation equipment. The new industrial welding robot uses remote laser welding technology to overcome the limitations of traditional welding (such as arc welding robot posture The limitation of the welding torch is limited by the size of the workpiece), which has the advantages of fast welding speed of the workpiece and small thermal deformation. [0003] In contrast, an efficient weld quality detection method is needed to match the processing cycle requirements. Conventional structured light sensors have high measurement accuracy and can measure three-dimensional parameters. The efficiency of quality inspect...

AI Technical Summary

Problems solved by technology

[0004] During the welding seam detection process, due to problems such as image deformation, image shooting quality, and angle, the existing two-dimensional image detection methods such as template matching cannot effectively detect the specific defect types and locations in the weld seam; in order to solve the above problems, The present invention proposes an intelligent weld defect detection method based on the principle of deep learning, which detects multiple welds and various defect types synchronously, realizes weld identification and positioning and defect detection in one measurement, and effectively improves measurement efficiency

Images