A magnetic flux leakage analysis model based on finite element neural network

摘要

The major drawback of the finite element method (FEM), which is commonly used in magnetic leakage field testing, is the high cost of calculation. In this paper, a finite element neural network (FENN), which embeds a finite element model in a neural network structure, is adopted that enables a faster and more accurate solution for the forward model compared with FEM. The second-order Newton method is introduced as a learning algorithm. The FENN model for magnetic flux leakage (MFL) testing is established and comparisons between the gradient decent algorithm and the second-order Newton method, under the circumstances of various defects, are presented. The vector plot of magnetic field intensity and the vertical components of magnetic flux density are analysed. The relevant results indicate that the second-order Newton method-based FENN can resolve the MFL forward model in parallel, which has the advantages of rapidness and stability, and it is a valid and feasible calculation method.