Prediction Of Plastic Compression Deformation And Optimization Of Machining Parameters For Ultrasonic Rolling Of Magnesium Alloy

As a lightweight structural metal material,magnesium alloy is widely used in aerospace,new energy vehicles,communication equipment,etc.However,magnesium alloy has low hardness,poor wear resistance and corrosion resistance.The workpiece made of magnesium alloy is prone to fatigue damage in the high intensity working environment.Ultrasonic rolling strengthening induces severe surface plastic deformation of magnesium alloy workpiece,which can improve the surface quality and mechanical properties.At the same time,the reduction of workpiece diameter caused by plastic deformation will affect the fitting relationship of parts during assembly.Therefore,it is important to construct the deformation prediction model based on ultrasonic rolling parameters and compressive deformation values,and to build the machining parameter optimization model according to the desired size,for reducing the experimental workload and obtaining the workpiece with strict size deviation requirements.AZ31B magnesium alloy was used as the experimental material in this paper.The feed rate,static pressure,rolling times and ultrasonic amplitude were used as the experimental factors to design 81 sets of ultrasonic rolling tests with 4 factors and 3 levels.The diameters of magnesium alloy specimens were measured by a universal tool microscope before and after rolling.The diameter difference was used as the plastic compression deformation resulted by the ultrasonic rolling.The deformation prediction model with BP neural network was constructed based on the training set data,and it was examined by the test set data.The results show that the absolute error between the predicted value and the experimental value varies from 0.04 to 4.94μm,and the maximum relative error is 25.78%.The error of the model is a little large,indicating that the deformation prediction accuracy is not ideal,although the prediction model constructed with single BP neural network has certain prediction ability.The weights and thresholds of the BP network were optimized with genetic algorithm(GA)and thinking evolutionary algorithm(MEA)to construct GA-BP and MEA-BP network prediction models to improve the prediction accuracy of the BP model.The mean absolute percentage error(MAPE),root mean square error(RMSE),and coefficient of determination(R~2)of the single BP,GA-BP,and MEA-BP network prediction models were investigated.The MAPE,RMSE,and R~2 of the single BP network model is 8.32%,2.34,and 0.946,respectively.The MAPE,RMSE,and R~2 of the GA-BP network model is 2.20%,0.682 and 0.980,respectively.The MAPE,RMSE,and R~2 of the MEA-BP network model is 1.73%,0.475 and 0.989,respectively.The MAPE and RMSE are the smallest,and the R~2 is the largest for the MEA-BP network prediction model,compared with that of the other two models.Therefore,the MEA-BP network model was used as the deformation prediction model for the ultrasonic rolling of magnesium alloy to obtain more reliable nonlinear mapping relationship between the machining parameters and deformation values.Based on the mapping relationship of MEA-BP prediction model and combined with GA,the MEA-BP-GA parameter optimization model is proposed to optimize the machining parameters corresponding to the desired shape variables.Through the parameter optimization and experimental verification of the target shape variables(10,25 and 40 microns),it is found that the corresponding actual rolling shape variables are 11.3,23.7 and 39 microns with relative errors of 13.30%,5.32%and 2.50%,respectively,which indicates that the MEA-BP-GA model is feasible for ultrasonic rolling of magnesium alloy,and the larger the desired shape variables are,the smaller the relative errors are.Finally,based on the prediction model and the parameter seeking model,the form variable prediction and parameter optimization system was constructed based on the GUI module of MATLAB,and the two-way prediction of parameters was realized.