Research On Mechanical Property Parameter Estimation Of Ferromagnetic Materials Based On DAE

Mechanical property parameter is one of the important indexes that must be considered in the production of ferromagnetic materials,and it is also the standard for quality evaluation of ferromagnetic materials.Traditionally,mechanical properties of ferromagnetic materials,such as yield strength and tensile strength,were obtained by lossy off-line stretching.In this paper,based on four nondestructive testing methods,the characteristic optimization of electromagnetic parameters and the estimation of yield strength,tensile strength and elongation of ferromagnetic materials are studied The main contents are summarized as follows:(1)The barkhausen noise method,multi-frequency eddy current method,the incremental permeability,tangential magnetic field harmonic analysis of four kinds of nondestructive testing methods and the basic principle of electromagnetic parameters of the electromagnetic signal characteristics are introduced.The correlation between the characteristics of electromagnetic parameters is analyzed,and the results show that there is redundancy among the electromagnetic parameters,which indicates that it is necessary to select the characteristics of electromagnetic parameters before estimating the mechanical properties.(2)To solve the problem of slow estimation speed and poor real-time performance of BP neural network,an algorithm combining sequential backward selection and least square method is proposed.This algorithm considers the relationship between the electromagnetic parameter feature subset and the mechanical property parameters,which not only removes redundant and unimportant electromagnetic parameter features,but also greatly reduces the estimation time of the model while ensuring the estimation accuracy.(3)Aiming at the problem of redundancy in the electromagnetic parameters extracted by four kinds of sensors,an algorithm model based on random forest is proposed.The results of the proposed algorithm are compared with the traditional neural network models,and the results show that the proposed algorithm has good generality,and the three mechanical parameters of each steel can be estimated with high precision.(4)Traditional feature engineering requires manual experience to extract and select features.This paper designs a feature extraction and modeling algorithm based on deep sparse autoencoder.Experiment results show that compared with other feature extraction methods,deep learning method can automatically extract more essential and representative features of the original barkhausen signal.The experimental results show that the algorithm combining the characteristic selection of electromagnetic parameters with linear regression greatly improves the estimation speed of the model.The model based on random forest improves the estimation accuracy.Deep learning is effective in feature extraction of electromagnetic signals.