Research On Simulation And Identification Of Pipeline Weld Defects Based On Magnetic Flux Leakage Principle

Due to the limitation of welding technology level and lax on-site construction quality control,some domestic pipelines have serious weld defects such as incomplete fusion,incomplete penetration and insufficient filling after they are put into operation,which will do great harm to the pipeline operation.Therefore,the detection and identification of pipeline weld defects is very important.Aiming at the weld defects in the pipeline,this paper studies the recognition method of magnetic flux leakage signal of pipeline weld defects based on deep learning.The main work includes the following aspects:(1)Various weld defects existing in the pipeline are summarized,classified and quantified.The finite element simulation model of magnetic flux leakage of weld defects is established.Through the analysis of the simulation results,in addition to the defect size,there are many factors affecting the distribution of magnetic flux leakage,such as lift off value and weld reinforcement.(2)How to deduce the size of weld defects from the distribution of leakage magnetic field and realize the identification of weld defects is the focus of this paper.Firstly,the research on single weld defect recognition is started,and the recognition and prediction models based on convolution neural network and cyclic neural network are established.Through comparison,the accuracy of cyclic neural network is 2.8% higher and the speed is 9.7% faster than convolution neural network.Cyclic neural network has more advantages in identifying and predicting single weld defects.(3)For the identification of multiple weld defects,based on the target detection algorithm of Yolo v3,this paper realizes the identification and prediction of multiple weld defects and achieves a good effect of weld defect detection.The evaluation indexes of identification and prediction: the precision rate is 60.37% and the recall rate is 71.62%,which has a certain practical application value.In order to make the prediction model more suitable for the prediction of weld defects,the optimization of clustering optimization a priori frame and adaptive adjustment learning rate are carried out.Through image expansion,the two evaluation indexes are improved by 3.8% and4.5% respectively.