Quantitative identification of microcracks through magnetic flux leakage testing based on improved back-propagation neural network

This paper aims to overcome the difficulties faced in the quantitative identification of microcracks. For this purpose, principal component analysis (PCA) was integrated with an improved neural network to establish a forecast model of the width and depth of microcracks. PCA significantly simplifies the network structure, as it removes data correlation and reduces the dimensions of the input sample. The improved neural network, denoted as the GA-BP neural network, is a back-propagation (BP) neural network optimised by a genetic algorithm (GA). This network can effectively avoid the trap of local minimum. The effect of the proposed model in the quantitative identification of microcracks was verified through a theoretical calculation based on a magnetic dipole model and a magnetic flux leakage (MFL) detection experiment for artificial groove microcracks. The research findings lay a scientific basis for the early quantitative identification of microcracks.