CN-121999930-A - Multi-stage progressive identification method for parameters of dual-phase steel crystal plasticity constitutive model

CN121999930ACN 121999930 ACN121999930 ACN 121999930ACN-121999930-A

Abstract

The invention provides a multi-stage progressive identification method of parameters of a dual-phase steel crystal plasticity constitutive model, which relates to the field of dual-phase steel materials and deep learning, and comprises the following steps of S1, determining parameters to be identified based on a dual-phase steel physical base crystal plasticity constitutive model; the method comprises the steps of S2, constructing a parameter identification model based on deep learning based on parameters to be identified, S3, decoupling the parameters of the constitutive model to be identified, and S4, carrying out multistage progressive identification on the parameters of the constitutive model to be identified. According to the multistage progressive recognition method provided by the invention, the influence degree of the parameters to be recognized in the constitutive model on the dislocation density and the strength of the material is arranged from high to low, and the parameters are sequentially recognized, so that the coupling between the parameters is effectively reduced, the sensitivity of the model to the parameters with low influence is enhanced, the sample data volume required by training the neural network is reduced, and the neural network can be stabilized under a small sample.

Inventors

CHEN LEI
LI XIAOLONG
WANG YINQIANG
WANG SHUO
QI WEN
JIN MIAO
Cai Xingzhou
ZHANG QINGLING

Assignees

燕山大学

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. The multi-stage progressive identification method for the parameters of the crystal plasticity constitutive model of the dual-phase steel is characterized by comprising the following steps of: s1, determining parameters to be identified based on a physical base crystal plasticity constitutive model of the dual-phase steel, wherein the parameters comprise two phases: 、、、、、、、、、 ; s2, constructing a parameter identification model based on deep learning based on the parameters to be identified in the step S1; s3, decoupling the constitutive model parameters to be identified, wherein the method comprises the following substeps: S31, decoupling initial input layer data of a parameter identification model, and sequentially identifying movable dislocation density and immovable dislocation density of two phases; s32, based on the two-phase movable dislocation and the immovable dislocation density, sequentially identifying the relevant parameters of the immovable dislocation density evolution from large to small according to the influence degree of each parameter in the dislocation density evolution model And Identifying and then to And Identifying and then to Identifying; Wherein, the For the movable dislocation rest coefficient, Dynamic recovery coefficient of the immovable dislocation, For the movable dislocation density increment coefficient, Annihilation coefficients for movable dislocation densities; s4, carrying out multistage progressive recognition on the constitutive model parameters to be recognized: Ensuring that the neural network trained in each stage has the same parameter settings except for input and output; ensuring that the training sample library and the verification data set are the same in each stage; identifying the parameters of which the coupling degree reaches a first set threshold value with the initial input quantity, adding the identified parameters into a neural network input layer, retraining, identifying the parameters of which the coupling degree reaches a second set threshold value with the identified parameters, and the like, so as to realize the determination of the parameters to be identified.
2. The multi-stage progressive identification method of parameters of a dual-phase steel crystal plastic constitutive model according to claim 1, wherein the identification sequence of the parameters in the step S4 is as follows: In the first stage, the input quantity is 、、、 Output is 、 ; In the second stage, the input quantity is 、、、、、、 Output is 、 ; In the third stage, the input quantity is 、、、、、、 Output is 、 ; Fourth stage, input quantity is 、、、、、、、、 Output is 、 ; Fifth stage, input quantity is 、、、、、、、、 Output is 、 ; In the sixth stage, the input quantity is 、、、、、、、、、、 Output is 、 ; Seventh stage, input quantity is 、、 Output is ; Eighth stage, input quantity is 、、、 Output is 。
3. The multi-stage progressive identification method of parameters of a dual-phase steel crystal plastic constitutive model according to claim 1, wherein the dual-phase steel physical base crystal plastic constitutive model comprises: Calculating the shear strain rate of the slip system by adopting a phenomenological velocity-dependent flow equation: ; in the formula, As a reference shear strain rate for a slip system, In order to achieve the splitting stress of the sliding system, For the length Cheng Zuli caused by the dislocation, In order to account for short-range drag caused by defects, Is a strain rate sensitivity coefficient; consider the length Cheng Zuli caused by movable and non-movable dislocations, respectively: ; in the formula, For the hardening coefficient caused by the movable dislocation, For the hardening coefficient caused by the immovable dislocation, For the length Cheng Zuli caused by the movable dislocations, Length Cheng Zuli for the immovable dislocation; The relationship between dislocation density and long-range resistance is: ; ; in the formula, In order to achieve a shear modulus, the polymer is, For the vectors of the number of Burgers, Is the interaction intensity coefficient matrix of dislocation between each sliding system, For the dislocation density of the movable dislocation, Dislocation density as an immovable dislocation; the evolution model of movable dislocation and immovable dislocation density is: ; ; in the formula, For the value-added coefficient of the movable dislocation, As the mean free path of the mobile dislocations, Wherein As the dislocation annihilation coefficient, As the coefficient of quiescence of the movable dislocation, Dynamic recovery coefficients for the immovable dislocations; in order for the movable dislocations to be trapped as the mean free path of the immovable dislocations, the calculation method comprises the following steps: ; 。
4. The multi-stage progressive recognition method of parameters of a dual-phase steel crystal plastic constitutive model according to claim 1, wherein the method for constructing the deep learning-based parameter recognition model in the step S2 comprises: S21, determining input and output parameters of a neural network; s22, establishing a sample library; s23, determining the structure and built-in parameters of the neural network.
5. The multi-stage progressive recognition method of the parameters of the crystal plastic constitutive model of the dual-phase steel according to claim 4, wherein in the step S21, the input parameters of the neural network comprise stress characteristic values, two-phase dislocation density characteristic values, volume fractions of ferrite phase and austenite phase in the dual-phase steel of a plastic section of a true stress-true strain curve of the dual-phase steel; The output parameters of the neural network include parameters to be identified in the constitutive model.
6. The multi-stage progressive recognition method of the parameters of the dual-phase steel crystal plastic constitutive model of claim 5, wherein the step S21 further comprises normalizing the determined input and output quantities: ; In the equation left side To the right of the equation for normalized data In order for the data to be normalized, And The maximum and minimum values in the normalized object data set, respectively.
7. The multi-stage progressive identification method of the parameters of the crystal plastic constitutive model of the dual-phase steel according to claim 5, wherein the stress characteristic value is a true stress value corresponding to a curve plastic segment strain ratio of 0.00, 0.03, 0.09, 0.18, 0.30, 0.45, 0.63, 0.84 and 1.00, and the dislocation density characteristic value is a ferrite and austenite phase dislocation density corresponding to a true strain ratio of 0.2, 0.4, 0.6, 0.8 and 1.0; the dislocation density calculating method comprises the following steps: ; Wherein, the ; In the formula, Is of half-peak width and is of the same order, For the diffraction angle corresponding to the diffraction peak, , And Are all obtained by XRD experiments.
8. The multi-stage progressive recognition method of parameters of the dual-phase steel crystal plastic constitutive model of claim 4, wherein in step S22, the method for establishing a sample library comprises establishing a unidirectional stretching numerical simulation model by combining a crystal plastic finite element method and a representative voxel method, and randomly generating in a set parameter value range 、、、、 The parameter combination is endowed with a polycrystal stretching finite element model, each grain in the polycrystal stretching finite element model is endowed with orientation at random, the stretching deformation behavior of a representative voxel is simulated, the characteristic value of a true stress-true strain curve and the characteristic value of dislocation density evolution curve are extracted, and the characteristic value of the true stress-true strain curve and the characteristic value of dislocation density evolution curve are combined with the phase profile during simulation to be used as the input quantity of a neural network; the set parameter value range comprises: Ferrite is MPa, austenite is MPa; Ferrite is Austenite is ; Ferrite is Austenite is ; Ferrite is Austenite is ; : Austenite is ; The phase ratio is ferrite Austenite is 。
9. The multi-stage progressive identification method of parameters of a dual-phase steel crystal plasticity constitutive model according to claim 8, wherein a Taylor model is adopted as a polycrystal homogenizing model by a polycrystal stretching finite element model, deformation gradients of all crystal grains are assumed to be consistent with macroscopic deformation gradients, and a change curve of dislocation density of ferrite phase and austenite phase with true strain is derived in the same manner: ; ; ; ; in the formula, And Is the stretching direction Macroscopic true stress and true strain on the surface, And Is the stretching direction True stress and true strain at integral points within the upper die, The total number of integration points; And To be dislocation densities of the austenitic and ferritic phases, And The total dislocation density of the respective integration points of the austenite and ferrite phases, And Points are two-phase points.
10. The multi-stage progressive recognition method of parameters of a dual-phase steel crystal plastic constitutive model of claim 4, wherein the neural network structure and built-in parameters in the step S23 are set to 7 layers, including an input layer, an output layer and 5 hidden layers, wherein the number of nodes of the input layer is the same as the number of corresponding input quantities, the number of nodes of the output layer is the same as the number of corresponding output quantities, the number of nodes of the hidden layers is 32-16-32-16-8, and the first, second and fifth hidden layers adopt tanh activation functions, and the third and fourth hidden layers adopt Leaky Relu activation functions.

Description

Multi-stage progressive identification method for parameters of dual-phase steel crystal plasticity constitutive model Technical Field The invention relates to the technical field of dual-phase steel materials and deep learning, in particular to a multi-stage progressive identification method for parameters of a dual-phase steel crystal plasticity constitutive model. Background Modeling calculation of plastic behavior of materials is important to accelerate process design and reduce use cost in application process. The crystal plastic theory combines a micro deformation mechanism and a macro plastic behavior of metal, reveals an inherent physical mechanism of crystal material deformation, is a powerful tool for researching metal plastic behavior, and has been widely used. The crystal plasticity constitutive model can describe the deformation behavior of a metal crystal under various load conditions, but the effectiveness thereof depends on the accuracy of introducing parameters of the constitutive model. Therefore, identifying a set of appropriate material parameters within appropriate physical boundaries is an important prerequisite for efficient prediction using a crystalline plastic constitutive model. For simple models containing a small number of parameters, the parameters can typically be calibrated using trial and error or regression methods. However, the method is not suitable for a physical base crystal plastic constitutive model with more parameters and strong nonlinearity, the coupling between the parameters is strong, each parameter not only directly influences the evolution of internal variables in the model, but also indirectly influences the action of other parameters on the internal variables, and the mutual influence between the parameters brings great difficulty to the identification of the parameters. The dual-phase steel combines the advantages of ferrite phase and austenite phase, shows excellent mechanical property and corrosion resistance, but the material property is influenced by the component phase property and the component phase proportion, the mechanical property of each phase is difficult to be independently determined, and the different properties of the two phases cause complex stress strain distribution, so that the mechanical response of each phase is influenced, and a great challenge is brought to the identification of the physical base crystal plastic constitutive parameters of each component phase of the dual-phase steel. Aiming at the technical problems, a plurality of students try to adopt a deep learning method to solve the difficulty that the mechanical properties of each phase of the dual-phase steel are difficult to independently determine. Deep learning is one of the machine learning algorithms that has gained popularity due to its excellent modeling performance and wide application as a general approximator. Which can represent complex material behavior, and the constitutive response comes directly from training data. On the one hand, training a neural network requires a large amount of sample data, and extremely high cost is required for training a model by establishing a sample library through an experimental method, on the other hand, because of the strong coupling between parameters of a physical base crystal plasticity constitutive model, the influence of a single parameter on the material performance is difficult to separate, the neural network model is difficult to search for a proper gradient direction to optimize and assign corresponding weight to the gradient direction, so that the parameter identification effect is reduced, and the problem is difficult to solve even if the sample data volume is increased. Therefore, for the physical base crystal plasticity constitutive model of the dual-phase steel, a new parameter identification method is necessary to be developed to identify the strong coupling parameters. Disclosure of Invention In order to solve the defects in the prior art, the invention aims to provide a multi-stage progressive identification method for parameters of a dual-phase steel crystal plasticity constitutive model, which can realize staged identification of parameters by combining the determined parameters according to the influence of the parameters on the material performance and the coupling relation among the parameters based on limited input quantity, can effectively reduce strong coupling among the parameters, greatly improve the identification effect of a neural network and ensure the reliability of the identification result of the parameters. Specifically, in one aspect, the invention provides a multi-stage progressive identification method of parameters of a dual-phase steel crystal plasticity constitutive model, which comprises the following steps: s1, determining parameters to be identified based on a physical base crystal plasticity constitutive model of the dual-phase steel, wherein the parameters comprise two phases: 、、、、、、