Identification Method Of Magnesium Alloy Crystal Plasticity Constitutive Parameters Based On Optimization Algorithm Coupled With Neural Network

The crystal plasticity constitutive model is the prerequisite and basis for crystal plasticity finite element simulation.The accuracy of the constitutive parameters directly affects the simulation accuracy of crystal plasticity finite elements.There are many crystal plasticity constitutive parameters and it is difficult to identify them.At present,for polycrystalline materials,optimization algorithms or deep learning algorithms are mainly used to identify crystal plasticity constitutive parameters.This requires a large amount of simulation iteration and experimental work and does not have universality.The identification of crystal plasticity constitutive parameters for each material requires a large amount of optimization work.Therefore,the efficiency of existing crystal plasticity constitutive parameter identification is low.Especially for densely packed hexagonal structured magnesium alloys,which not only have relatively complex slip systems but also accompanied by twinning deformation,their crystal plasticity constitutive parameters are more numerous and more difficult to identify,which has become a major problem restricting the practical application of crystal plasticity finite elements.Based on this,this paper proposes a crystal plasticity constitutive parameter identification method based on the optimization algorithm coupled with neural networks for typical magnesium alloy structures.This method obtains a general deep learning model for magnesium alloys through large data training in the early stage.In the later stage,it inputs the microstructure and texture information of specific materials and unidirectional tensile mechanical response and is supplemented by corresponding optimization algorithms.Without a large amount of simulation iteration and experimental work,crystal plasticity constitutive parameters can be obtained.Through experiments with two different magnesium alloys,this study aims to compare the differences between this method and existing methods in terms of parameter identification accuracy and efficiency.The research work and main conclusions are as follows:(1)Building a big data relationship between the microstructure and crystal plasticity constitutive parameters of magnesium alloys and their macroscopic mechanical response.First,a large number of virtual material combinations are designed through Latin hypercube design within a reasonable range,covering magnesium alloy grain geometry models,texture information and crystal plasticity constitutive parameters.The virtual grain geometry model and virtual texture information are parameterized.Then for each combination,the uniaxial tensile process is simulated by crystal plasticity finite element method to obtain the corresponding mechanical response.High-throughput data is established for each material combination and its corresponding mechanical response as the data basis for subsequent BP neural network training.(2)Building a neural network model for the crystal plasticity constitutive parameters of magnesium alloys and a neural network model for the parameterization of actual texture.First,the grain geometry model,texture information and crystal plasticity constitutive parameters of virtual materials are used as inputs to the BP neural network.The simulated mechanical response is used as the output of the BP neural network.The BP neural network model F is trained to establish the constitutive relationship in crystal plasticity finite element simulation.Compared with finite element simulation,the neural network model has faster calculation speed and does not require complex finite element modeling.The orientation density of textures in the virtual material database is used as input to the BP neural network and texture weight parameters are used as output to train BP neural network model Q.Neural network model Q realizes parameterization of actual textures into specific weight parameters and establishes a mapping relationship between virtual textures and actual textures.The recognition accuracy of neural network models F and Q exceeds 98%,indicating that the parameters selected for building the database do have a decisive impact on crystal plasticity finite element simulation and considering these parameters is a necessary condition for accurately identifying crystal plasticity constitutive parameters.(3)Identification of magnesium alloy crystal plasticity constitutive parameters based on optimization algorithm coupled with neural network.By using neural network models F and Q as surrogate models to avoid the disadvantages of low accuracy and low complexity of general surrogate models,Isight software is coupled with MATLAB software as a framework.The optimization algorithm calls the surrogate model to establish a crystal plasticity constitutive parameter identification model based on the optimization algorithm coupled with the neural network model.By inputting the microstructure and texture information and uniaxial tensile mechanical response parameters of specific materials into the parameter identification model,the corresponding crystal plasticity constitutive parameters can be quickly identified.The calibrated crystal plasticity constitutive parameters in random virtual material crystal plasticity finite element simulation tests have a mean square error between simulated and target mechanical performance curves lower than 20.For two actual magnesium alloys calibrated for their crystal plasticity constitutive parameters,their corresponding simulated mechanical performance curve mean square errors(9-18)are 1-2 orders of magnitude lower than those of traditional optimization algorithms(123-1025)proving that the method for identifying crystal plasticity constitutive parameters based on optimization algorithms coupled with neural networks has high calibration accuracy and high versatility.In summary,this paper improves and innovates the existing crystal plasticity constitutive parameter identification methods and proposes a magnesium alloy crystal plasticity constitutive parameter identification method based on optimization algorithm coupled with neural network.This method takes into account both the accuracy and efficiency of parameter identification and achieves accurate and fast calibration of magnesium alloy crystal plasticity constitutive parameters.It overcomes the shortcomings of existing constitutive parameter identification methods and has certain universality for magnesium alloy materials.This will help promote the further popularization and application of magnesium alloy crystal plasticity finite element.