Classification and evaluation for nearside/backside defect via magnetic flux leakage: A dual probe design with SVM and PSO intelligence algorithms

Abstract

In automated magnetic flux leakage (MFL) applications, the defect location on the nearside/backside of the tested specimen is often unknown, posing challenges for evaluating defect parameters due to varying calibration relationships. To address this, we propose a combined scheme of defect classification followed by parameter evaluation for testing steel plates with nearside/backside defects. The innovative design of dual probes, featuring lift-off difference, is used for signal acquisition and analysis. An efficient defect classification is achieved through a PSO-SVM neural network, and defect parameter evaluation involves solving the inverse problem using a particle swarm intelligence algorithm. MFL signal from steel plates with circular hole defects of different sizes were tested using a self-developed system and dual probes. Finite element analysis, alongside experimental data, explores the change in MFL signal. Defect classification and parameter evaluation, based on experimental signals and simulations, confirm the effectiveness of scheme in quantitatively assessing defect parameters with an evaluation accuracy within 0.3 mm. This scheme serves as a reference for intelligent testing and evaluation using MFL signals.