ANN Models Of Flow Stress And Hot Deformation Behavior In Sb And RE Agonized AZ31 Magnesium Alloy

During hot deformation, the microstructure in materials goes through a serious of dynamic changes, which has important influences on the flow behavior of materials properties of materials. So, studying of microstructure evolution during hot forming and establishment flow stress model considered the effect of microstructure the process parameters and controlling the service quality of work pieces. Because the flow stress and process parameters of AZ31 magnesium alloy containing Sb and RE during deformation are non-linear relationship, establishing the model is very difficult with conventional methods. Artificial neural networks, which are universal approximations for continuous functions, are used in this paper. The hot deformation behaviors in AZ31 magnesium alloy containing Sb and RE were studied and artificial neural network (ANN) model was established to predict the flow stress of AZ31 magnesium alloy containing Sb and RE in this paper.The deformation behavior of AZ31 magnesium alloy containing Sb and RE was investigated by compression tests with Gleeble-3500 thermal system. The tests were performed in the temperature range between 250℃and 400℃, and strain rates between 0.01and 10s^-1 . During compressing at elevated temperature, the relationship between peak stress in form of hyperbolic sine and strain rate and reciprocal of deforming temperature is closely power law, and then materials constants of AZ31 magnesium alloy containing Sb and RE during compressing at elevated temperature are obtained by employing Zener-Holloman parameter. The mixed addition of RE and Sb can increase the deformation energy of AZ31 alloy. The energy of dynamic recovery is also increased.The processing map theory, which based on dynamic material model, was introduced in investigation on deformation behavior. According to the data from compressing at elevated temperature, the processing maps of AZ31 magnesium alloy containing Sb and RE, based on dynamic material model (DMM), were plotted at strain of 0.4and 0.6, respectively. According to the change of efficiency of power dissipationηwith processing parameter and the relationship betweenηand microstructure , the optimal hot work parameter of AZ31 magnesium alloy containing Sb and RE is of at temperature of 350～400°C, strain rate 0.01 s^-1～0.1s^-1.The model based on BP network was established according to the data from tests in order to predict the flow stress of AZ31 magnesium alloy containing Sb and RE. The inputs of the model were temperature, strain rate and strain, and the output was flow stress. L-M algorithm and two hidden layers were used in the BP network models. Then the prediction performance of the BP model was tested by experiment data. The results of the study show that the BP network has very good nonlinear approach abilities. The training correlation coefficient of the model is 0.997. The predictions of BP network well approximate the experimental data, and the maximum average relative error of BP network is 5.63%.The BP neural network is more accurate than the method of regression.