Research On X-ray Pipe Weld Porosity Detection Technology Based On Deep Learning

The safety performance of pipeline is closely related to its weld quality.From domestic gas pipeline to strategic energy pipeline,some weld defects may lead to pipeline explosion,causing serious economic losses and even casualties.Therefore,the weld defects must be strictly inspected.X-ray testing is one of the most widely used weld nondestructive testing methods because of its accurate,intuitive and nondestructive characteristics.However,X-ray testing results mainly rely on manual film evaluation.Manual film evaluation has the disadvantages of cumbersome operation,high work intensity,inconsistent testing results and so on.However,the existing automatic defect detection algorithms generally have problems such as low precision and slow speed,and a large number of threshold parameters depend on manual selection,so the degree of intelligence is relatively low.To solve the above problems,in order to improve the quality and automation level of welding porosity defect detection,an X-ray welding porosity defect detection algorithm based on deep learning is proposed.The main work is as follows:(1)A weld area location method based on LSTM network is proposed.According to the characteristics of weld imaging,the intensive curve is constructed to reduce the two-dimensional problem to a one-dimensional sequence problem.The weld boundary features are learned from the sequence through LSTM network to realize the weld area location,which greatly reduces the detection range and improves the calculation speed of233 %.(2)An improved defect target detection algorithm based on Yolo is proposed.Based on the residual learning strategy,the front-level feature extraction network is redesigned to ensure the real-time detection speed? The loss function used to guide the network parameter training is optimized,which reduces the computational complexity and improves the defect detection accuracy.(3)The prototype system of automatic detection of weld defects is designed and implemented,so that the staff can quickly obtain and visualize the detection results of weld defects through cloud computing through the client graphical interface and the powerful computing power of the server,which improves the evaluation efficiency of X-ray photos.In order to verify the detection effect,an experimental data set containing 1200 Xray detection images is constructed.Through experiments,the defect recall rate of the algorithm proposed in this paper is 99.3 %,and the average detection time of a single image is 32 milliseconds,which can meet the detection requirements of accuracy and efficiency.