Irregular Defect Size Estimation for the Magnetic Flux Leakage Detection System via Expertise-Informed Collaborative Network

Defect size estimation plays a vital role in pipeline risk level assessment for the magnetic flux leakage (MFL) detection system. However, the estimation of irregular defect sizes always suffers from low accuracy and poor stability, mainly because they usually exhibit deformed signals and intricate MFL characteristics. To solve this issue, in this article, expertise is integrated to aid in capturing critical features, and a collaborative mechanism is presented to deeply exploit the MFL characteristics from multiview. Thus, a novel irregular defect size estimation method based on expertise-informed collaborative network (EIC-Net) is proposed. First, an expertise-informed feature mining network is designed to mine macrofeatures embedded with physical properties and microfeatures consistent with engineering experience so that coarse to fine MFL signal features can be captured. Second, a multiview collaborative decision network is proposed for the first time, where two parallel probabilistic decision makers are designed to model uncertainty caused by subjective cognitive differences from a new perspective of the probability theory and they collaborate with each other through a consistency strategy so that the complementary properties of multiview MFL information can be explored. Moreover, they also collaborate with the traditional deterministic decision maker so that the decision bias can be reduced. Finally, the experiments are performed on real world pipeline network in northern China, which show that our EIC-Net improves the accuracy of irregular defect size estimation by 10.7% compared to other advanced methods and has strong practicality.