Research On Multi-magnetic Parameter Fusion And Fatigue Life Prediction Methods Of Ferromagnetic Materials

Ferromagnetic metal materials are widely used in engineering and are a basic component of major mechanical equipment,such as aviation equipment,railways,and large-scale amusement facilities.During the long-term service,ferromagnetic materials are often subjected to various complex loads and uncertain factors,which may cause fatigue or even fracture.Magnetic Barkhausen Noise(MBN)technology is a convenient and effective non-destructive testing method and it can qualitatively characterize and detect the microstructure changes and fatigue damage of ferromagnetic materials.However,the stochastic characteristic of MBN signal makes its magnetic characteristic parameters with low robustness and universality,which affects the detection accuracy of material fatigue damage.Therefore,it is of great practical significance to study the multimagnetic parameter fusion and fatigue life prediction of ferromagnetic material in complex environments,which provides a reliable basis for the safety and reliable operation of major mechanical equipment.Due to the complexity,uncertainty and dispersion of the material fatigue failure,as well as the randomness and sensitivity of MBN signal,research on multi-parameter fusion of MBN signal and material fatigue life prediction based on uncertainty assessment methods are carried out.The main research content and technical route are as follows:(1)In view of the existence of many uncertain factors in the material fatigue life prediction and the shortcomings of the current deterministic life prediction methods,and based on the information and state of knowledge at different stages from data to related material testing and fatigue cycle testing,it was developed that multi-magnetic parameter fusion of MBN signal and probabilistic fatigue life prediction framework using comprehensive uncertainty analysis and Bayesian information updating and uncertainty to quantify the observation uncertainty caused by the randomness of ferromagnetic signals,and the model uncertainty resulting from model type,input,parameter setting and modeling training.(2)A multi-magnetic parameter fusion method based on the analysis of observation uncertainty is studied.Based on Principal Component Analysis(PCA)and Feature Correlation Analysis(FCA),an optimal weight combination feature selection method is proposed,which provides a fast database feature selection method for multiparameter fusion and modeling.This method is mainly used to select independent features that contain the main information of the original data signal under the premise of considering different feature dimensions.Moreover,the randomness of the MBN signal is characterized as the observation uncertainty.Based on the modeling of observation uncertainty using Bayesian variational autoencoder,the analysis characterization,definition and elimination of MBN signal randomness is studied.Finally,the purpose of effective fusion of the multi-magnetic characteristic parameters is achieved and the fused magnetic characteristic parameters can improve the accuracy of the fatigue life prediction of the material.(3)A fatigue life prediction method based on the analysis of model uncertainty is studied.On the basis of modeling the uncertainty using Bayesian neural network and Bayesian linear regression,the qualitative description and evaluation of the degree of uncertainty in fatigue life prediction is realized.At the same time,based on the uncertainty analysis method of Bayes reasoning and information update and the standard of probability model selection method,the interaction and influence of the prior distribution,likelihood function,and posterior distribution of the Bayesian model parameters are analyzed and verified.Considering that the effectiveness of the probability model is limited by prior knowledge and data volume,a method for training the posterior distribution of parameters under the condition of no prior information and small data set is proposed,which optimizes the setting of the prior probability.The accuracy of Bayesian model prediction results is further improved,and the advantages of the proposed prior training method is verified in the fatigue life prediction experiment.