A KNN-SVR Data Mending Method for Insufficient Data of Magnetic Flux Leakage Detection

Abstract

In magnetic flux leakage (MFL) detection, transient fault appears unavoidably on individual sensor when we collect magnetic flux leakage signals, which makes MFL data insufficient. Data mending for insufficient data concerns the accuracy of the defects inversion. A precise data mending method based on K Nearest Neighbor-Support Vector Regression (KNN-SVR) is introduced, which effectively reduces the training cost of SVR and greatly improves the accuracy of the algorithm. The method is tested by experiment data obtained. The results demonstrate that the proposed method can improve the accuracy rate of data mending of insufficient data with an acceptable time cost.