Study On The Constitutive Model Of GH4169 Alloy Under High Temperature And High Strain Rate Based On Genetic Algorithm And Neural Network

GH4169 nickel-based superalloy is widely used in modern aero-engines to manufacture key components such as aviation turbine blades and engines.The environment in which these components are located is more and more complex and harsh,so it is inevitable to withstand high temperature and high impact during service,under these complex loads,metal parts are prone to loss of accuracy,failure or even fracture.In order to ensure good stability and reliability of GH4169 in complex environments,it is necessary to study the mechanical properties of GH4169 alloy at high temperature and high strain rate(impact).It is of great engineering significance to establish an accurate and reliable constitutive relation to describe the dynamic mechanical properties changes under extreme conditions(high temperature and high strain rate).The main work of this paper includes:(1)A new method based on improved selection operator genetic algorithm to optimize artificial neural network to predict flow stress of metal in complex service environment is proposed.Based on the experimental data of 304 stainless steel flow stress in the strain range of 0.1-0.5,temperature change of 293K?873K,and strain rate range of 0.001s^-1-100s^-1,the flow stress prediction model of 304 stainless steel was constructed using this method,and compared with decision tree,linear regression and unimproved genetic neural network model,and finally verified the accuracy of the improved algorithm with mean absolute error MAE and coefficient of determination R²as test parameters.(2)The shock compression experiment of GH4169 at temperature(293K?873K)and strain rate(1000s-1?6000s-1)was carried out by using separate Hopkinson compression bar experimental device,and the stress-strain of GH4169 under a wide range of temperature and strain rate was obtained.The experimental data provides data support for the parameter fitting of the Johnson-Cook(J-C)constitutive model and the constitutive training of the neural network,at the same time,it can be analyzed from the experimental results that the flow stress of the GH4169 alloy is related to the strain rate and temperature,the flow stress decreases with the increase of temperature,and the flow stress increases with the increase of the strain rate.(3)The parameters in the J-C constitutive equation were fitted by the experimental data under a wide range of temperature and strain rate,and the J-C constitutive model of the GH4169 alloy based on the physical model was established,the model can better reflect the overall trend of the plastic deformation stage;at the same time,the constitutive model of GH4169 at high temperature and high strain rate was obtained by training the improved neural network method(IGNN)in(1).The prediction accuracy of the two constitutive models on the flow stress in the data and out of the data is compared and analyzed.(4)A neural network training visualization interface is developed for the research of IGNN in the GH4169 constitutive model.Through the integration of the IGNN algorithm and the adjustment of training parameters through the visualization interface,a usable model with higher fitting accuracy can be constructed faster and the training is simplified,improve work efficiency,and the tool can be applied to the construction of constitutive models of other alloy materials.The results of this study show that the flow stress and deformation parameters are highly nonlinear at high temperature and high strain rate,resulting in a certain decrease in the accuracy of the J-C constitutive model with the increase of temperature and strain rate,but it can more accurately reflect the changing trend of the plastic deformation stage;compared with the J-C constitutive model,A new method based on improved selection operator genetic algorithm to optimize artificial neural network(IGNN)can be well trained to obtain the relationship between the stress and deformation parameters due to the introduction of a nonlinear activation function,the nonlinear relationship results in better accuracy than the J-C constitutive model under the corresponding conditions,but the extrapolation generalization ability has declined.