Research on characteristic quantity and intelligent classification prediction of metal magnetic memory detection signal

Abstract

Metal magnetic memory (MMM) is an innovative, nondestructive testing method. It can detect both stress concentrations and macroscopic defects. The three-dimensional force-magnetic coupling model was established by the ANSYS simulation software, the evolution process of different defect depths was studied in detail, and the change of the signal characteristic was analyzed. The results showed that the variation trend and amplitude characteristic of MMM signals resulted in obvious differences among different defect types. Meanwhile, the impacts caused by the defect parameters and the type are complex, which cannot be decoupled or calculated by a certain formula. The accuracy of the simulation data was verified by experiments. To solve the classification prediction problem in MMM detection, the signal peak and valley Hp-v, the signal width W, the gradient Ky, and the peak energy Hy were selected as characteristic parameters to evaluate different defect types according to the change in the signal waveform. Finally, using these vectors as the input variables, the radial basis function neural network (RBFNN) pre-classification test model was established to realize the classification recognition of pit defects, crack defects, and porosity defects. The results show that the accuracy of the training and test sets, and it is feasible to use this model to complete the intelligent classification of defects.

Keywords

• metal magnetic memory detection
• stress concentration
• defect classification
• radial basis function neural network
• prediction