Processing method for failure data in pipeline MFL inline inspection

Abstract

Since the accuracy of defect detection is strongly influenced by the data validity of pipeline magnetic flux leakage(MFL) inline detection, processing approaches for MFL failure data have been designed, based on the MFL failure signal characteristic analysis, to avoid, as far as possible, the influence of failure data. The approaches, such as threshold over-limit method, neighborhood difference threshold method, signal area, local zero-crossing rate, etc., were used for five types of failure data including over-limit failure, peak failure, continuous oversmoothing, single-channel data drift and sensor jitter, with the purpose of detection and rejection of failure data. With K-Nearest Neighbor(KNN) algorithm, the redundancy of the training set can be reduced by means of Support Vector Regression(SVR) algorithm, and the missing data interpolation method based on KNN and SVR was thus designed for missing data interpolation.The results demonstrate that, the data processing method for pipeline MFL failure is capable of accurately interpolating the missing MFL data in terms of different degrees of data missing, and considered to be of great reference to the failure data processing that frequently occurs in practical engineering.

Keywords

• pipeline
• mfl inspection
• failure data
• missing data interpolation
• data cleaning