A Conditional Variational Autoencoder Algorithm for Reconstructing Defect Data of Magnetic Flux Leakage

Magnetic flux leakage (MFL) testing is widely used in pipeline nondestructive testing. The acquisition quality of MFL signal is an important prerequisite for the accuracy of pipeline detection and evaluation. However, data loss often occurs in the process of detection, due to the interference of external environmental factors. Therefore, it is important to reconstruct the missing data. This paper analyzes four types of MFL data loss, and presents a conditional variational autoencoder algorithm to reconstruct defect data of MFL for these types of loss. The algorithm effectiveness is tested by comparing with traditional variational autoencoder method. The results demonstrate that the proposed method can improve the accuracy of reconstructing defect data loss.