Research On Intelligent Recognition Technology Of Metal Loss Defect Shape Based On Magnetic Flux Leakage Principle

At the moment when China is vigorously developing its economy,it is extremely important for the safety and inspection standards of storage tanks and long-distance pipelines.Magnetic flux leakage testing is one of the more mature testing methods at present.Magnetic flux leakage testing technology is used to timely detect the size and location of defects,ensuring the safety of storage and transportation of storage tanks and pipelines.Therefore,this paper focuses on how to combine magnetic flux leakage testing and artificial intelligence to efficiently detect defect types.In this paper,firstly,a three-dimensional finite element model of magnetic flux leakage(MFL)is established for three types of defects:rectangular groove,smooth pit and perforation by using ANSYS Maxwell software,and a large number of sample data are simulated;By analyzing and comparing the changes of the axial and radial curves of the leakage magnetic field of three kinds of defects under different parameters,four different characteristic values are extracted in batches by using python programming,and the four different characteristic values are reduced from high-dimensional space to two-dimensional space by T-SNE algorithm;Finally,it introduces three different multi classification machine learning algorithms that are most commonly used at present.The three algorithms are used to train models and predict defect types through the trained models.The T-SNE algorithm is used to reduce the dimension of defect feature points and visualize them.It can be intuitively distinguished that three different defect types have different distributions in two-dimensional space because of different defect types.The same type of defect feature points are close to each other,and different types of defect points are scattered far away;Different algorithm models are trained through machine learning simulation,and data sets are imported into three different algorithm models for training.On the premise of small number of samples,the accuracy of SVM algorithm is low,only 89.28%;The accuracy of random forest and GBDT is similar,more than 94%,but the accuracy of GBDT for a small amount of data is higher than that of random forest;In this paper,three different classification and recognition algorithms are used to judge different types of defects,which provides a reference for future intelligent identification of defect types.