Real-Time Inspection in Detection Magnetic Flux Leakage by Deep Learning Integrated with Concentrating Non-Destructive Principle and Electromagnetic Induction

One of the most common techniques of pipeline inspection is magnetic flux leakage (MFL). It is a non-destructive testing (NDT) method that employs magnetic sensitive sensors to detect MFL of faults on pipelines' internal and external surfaces. This research proposed a novel technique in real-time detection of MFL with pattern recognition in non-destructive principle using deep learning architectures. Here, the MFL signal has been collected as a large data sequence which has to be trained and validated using neural networks. Initially, the MFL has been detected using Faraday's law of electromagnetic induction (EMI) which is induced with Z-filter in electromagnetic (EM) decomposition. The collected signal of MFL has been classified using convolutional neural network (CNN), and this classified signal has been recognized by the patterns based on their threshold of the signal. By extracting and analyzing magnetic properties of MFL for a signal, the quantitative MFL has exceeded their threshold value from detected signals. Damage indices based on the link between enveloped MFL signal and the threshold value, as well as a generic damage index for MFL technique, were used to strengthen the quantitative analysis.