Magnetic Memory Feature Extraction And Domain Observation Of 45 Steel Specimen In Fatigue Damage Process

Fatigue damage is a typical form of damage to mechanical components,and for ferromagnetic components,the fatigue damage process is often accompanied by changes in the surface magnetic signal and surface microscopic magnetic domain structure.As the surface magnetic signal can effectively reflect the early stress concentration state of a ferromagnetic component,magnetic memory detection techniques have been widely used in engineering for the qualitative analysis of damage to various types of ferromagnetic components.However,due to the differences in the initial magnetic state and service environment of ferromagnetic components,it is difficult to quantify the fatigue damage by directly using the acquired magnetic signals such as normal magnetic field strength Hp(y)and magnetic field strength gradient K.It is urgent to carry out the extraction of magnetic memory features of the fatigue damage process and to elaborate the changes of surface magnetic domain structure from a microscopic perspective.In this paper,a magnetic memory signal acquisition platform was set up for 45 steel specimen,and low perimeter fatigue experiments were carried out to obtain the normal magnetic field strength Hp(y)and the magnetic field strength gradient K.Based on the Hp(y)and K values,the integral value of the magnetic field strength gradient,the wavelet packet energy of the normal magnetic field strength,the peak value of the magnetic field strength gradient and the multi-scale entropy value of the normal magnetic field strength were further extracted.Based on the above four feature quantities,BP neural network and GA-BP neural network methods were selected to construct multi-feature quantity fusion evaluation models respectively.The magnetic domain morphology of the specimen surface in different fatigue damage states was observed by the powder pattern method and the magnetic force microscope method,and the magnetic domain morphology characteristics and the phase angle variation of the magnetic signal were described.The results show that with the increase of fatigue cycle cycles,the eigenvalues of E,K_max,MSE,?are gradually increased.Using the wavelet packet energy decomposition method to decompose the energy of the normal magnetic field intensity Hp(y),the energy of the low frequency band and the energy of the high frequency band after decomposition account for 98.79%and 1.21%of the total energy respectively.Through the comparative analysis of the two models,it is found that the error of the GA-BP neural network is within 3%,the BP neural network error over 10%,the accuracy of the GA-BP neural network is higher than the BP neural network.The magnetic domains on the surface of the specimen are observed by the powder pattern method,at the initial stage,a variety of domains such as 180°domains,90°domains and striped domains appear on the surface of the specimen,the magnetic domains also shift and deflect as the specimen is cycled over an increasing number of cycles.The magnetic domains on the surface of the specimen are observed by the magnetic force microscopy method,it is found that the magnetic domain structure on the surface of the specimen is mainly bar-shaped domains,and the domain width decreases and increasing phase angle of the magnetic signal on the surface as the number of cycles increases.The results of this paper can provide theoretical and methodological support for the quantitative assessment of fatigue damage in ferromagnetic components,and have certain prospects for engineering applications.Figure[41]table[7]reference[91]...