A physical-informed machine learning approach for a hyperelastic multi-deformation mode constitutive model

Through physical information machine learning methods, the deep regression network and physical information network are constructed, and the multi-deformed mode constitutive relationship of superelastic materials is automatically derived, which solves the problem of poor prediction effect of multi-deformed mode in the existing technology, and realizes the multiple deformation modes accurate response prediction and improved model applicability.

CN118428209BActive Publication Date: 2024-11-19NINGXIA UNIVERSITY
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410492042.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-23
Publication Date
2024-11-19
Estimated Expiration
2044-04-23

AI Technical Summary

Technical Problem

The prior art is difficult to construct a constitutive model of superelastic materials suitable for multiple deformation modes, resulting in poor prediction of other deformation modes by models established in one mode, and it is difficult to achieve accurate prediction of nonlinear material response in multi-deform mode. .

Method used

Using physical information machine learning method, deep regression networks and physical information networks are constructed through experimental data sets in multi-deform mode, and constitutive relationships in multi-deform mode are automatically derived, and explicit expressions reflecting the large deformation behavior of superelastic materials are generated. , realize nonlinear material response prediction for various deformation modes.

Benefits of technology

Accurate prediction of the response of hyperelastic materials in multi-deformation mode is achieved, the prediction accuracy and applicability of the constitutive model are improved, overfitting data in a specific mode is avoided, and the generalization ability and applicability of the model are enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118428209B_ABST
    Figure CN118428209B_ABST
Patent Text Reader

Abstract

The present invention discloses a physical information machine learning method for a hyperelastic multi-deformation mode constitutive model, comprising the following steps: step 1: experimental data set acquisition, obtaining the stretching ratio, shear ratio and nominal stress under multiple deformation modes; step 2: physical information network construction, used to predict the mapping relationship between two deformation tensor invariants and strain energy functions under different deformation modes calculated according to the stretching ratio and the shear ratio; step 3: deep regression network construction, used to generate an explicit expression of a constitutive model reflecting the large deformation behavior of hyperelastic materials under multiple deformation modes; step 4: network training based on the experimental data set of hyperelastic materials under multiple deformation modes, joint training of the physical information network and the deep regression network; generating an explicit expression of a constitutive model reflecting the large deformation behavior of hyperelastic materials under multiple deformation modes. The present invention realizes accurate prediction of nonlinear material responses under multiple deformation modes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of multi-deformation mode constitutive modeling of hyperelastic materials, and in particular to a physical information machine learning method for a hyperelastic multi-deformation mode constitutive model. Background Art

[0002] Hyperelastic materials are a type of nonlinear elastic material with large deformation capability, and their constitutive properties are completely determined by their strain energy function. The main methods for studying the strain energy function form of hyperelastic materials are:

[0003] The first method is to determine first and then identify. Based on their own assumptions, researchers use basic theories to determine the basic form of the function, and then use optimization methods to identify the undetermined parameters in the strain energy function. The second method is to determine and identify at the same time. Based on experimental data, according to the changing characteristics of the experimental curves under different deformation modes and different deformation ranges, the corresponding function form is selected, and then optimization methods are used to identify the undetermined parameters in the strain energy function.

[0004] The effectiveness of these methods depends on the subjective experience of researchers, who need to select the function form based on the changing characteristics of experimental data. When applying hyperelastic materials, engineers must first select a constitutive model and conduct relevant mechanical tests, such as uniaxial tension and compression tests, biaxial tension and compression tests, shear tests, etc. Based on the experimental data set obtained, combined with the optimization algorithm, a suitable constitutive model can be selected and the model parameters of the selected constitutive model can be determined.

[0005] In addition, it is worth noting that most current studies establish corresponding constitutive models based on experimental results under one deformation mode. However, for nonlinear materials, the constitutive model established based on one mode may have a poor prediction effect on other deformation modes, or even be inapplicable. For example, the constitutive relationship and its parameters determined only by uniaxial tension experimental data cannot well predict the nonlinear law between deformation and stress under other deformation modes such as compression and shear. How to construct and select constitutive relationships applicable to multiple deformation modes is still an important issue that needs to be solved in this field. Summary of the invention

[0006] In order to overcome the above technical problems, the purpose of the present invention is to provide a physical information machine learning method for a hyperelastic multi-deformation mode constitutive model. By using a machine learning method, the constitutive relationship under the multi-deformation mode is automatically derived from an experimental data set consisting of multiple deformation modes, so as to achieve accurate prediction of the nonlinear material response under multiple deformation modes.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] A physical information machine learning method for a hyperelastic multi-deformation mode constitutive model comprises the following steps;

[0009] Step 1: Acquisition of experimental data sets, obtaining the stretching ratio, shear ratio and nominal stress under various deformation modes, which are used to generate the input data of the physical information network in step 2 and the deep regression network in step 3;

[0010] Step 2: Building a physical information network; used to predict the mapping relationship between the two deformation tensor invariants I1, I2 and the strain energy function W(I1, I2) under different deformation modes calculated according to the stretching ratio and shear ratio;

[0011] Step 3: Building a deep regression network to generate explicit expressions of the constitutive model that reflects the large deformation behavior of hyperelastic materials under multiple deformation modes;

[0012] Step 4: Perform network training based on the experimental data set of hyperelastic materials under multiple deformation modes obtained in step 1, and jointly train the physical information network in step 2 and the deep regression network in step 3; finally, generate an explicit expression of the constitutive model that reflects the large deformation behavior of hyperelastic materials under multiple deformation modes.

[0013] The step 1 is specifically as follows:

[0014] Firstly, a mechanical test experiment was carried out on an incompressible hyperelastic rectangular plate, including uniaxial tension and compression test, biaxial tension and compression test, and mechanical test experiment under various deformation modes of shear, to obtain an experimental data set, which includes the main expansion ratio obtained under uniaxial tension and compression mode. Uniaxial stress under combination Main stretch ratio under biaxial tension and compression mode Combined biaxial stress Main stretch ratio in shear mode Nominal shear stress under combination The superelastic material composed of is the principal expansion ratio obtained from uniaxial tension and compression experiments, is the principal expansion ratio obtained from biaxial tension and compression experiments, The shear ratio obtained from the shear test is is the uniaxial stress, is the biaxial stress, is the shear stress;

[0015] Calculate the deformation tensor invariants in uniaxial tension and compression modes based on hyperelasticity theory

[0016]

[0017] Calculate the deformation tensor invariants in biaxial tension and compression modes based on hyperelasticity theory

[0018]

[0019] Based on the hyperelastic theory, the shear mode Deformation Tensor Invariants

[0020]

[0021] Among them, I 1TC and I 2TC are the two deformation tensor invariants of the right Cauchy Green deformation tensor under uniaxial tension and compression; I 1BT and I 2BT are the two deformation tensor invariants of the right Cauchy Green deformation tensor under biaxial tension and compression; I 1SS and I 2SS are the two deformation tensor invariants of the right Cauchy Green deformation tensor under shear deformation.

[0022] The step 2 is specifically as follows:

[0023] Building a physical information network net , to predict the strain energy function W(I1,I2) under multiple deformation modes;

[0024] W pred (I1,I2)=W net (I1,I2;w,b) (4)

[0025] Among them, W pred are the predicted values ​​of strain energy functions of the two invariants I1 and I2 under the multi-deformation mode, and W net represents a fully connected neural network, and w represents a fully connected neural network W net The weight of b represents the fully connected neural network W net Bias of

[0026] First, for uniaxial tension and compression modes, W net The input is two deformation tensor invariants I in uniaxial tension and compression mode 1TC ,I 2TC ; For biaxial tension and compression mode, W net The input I1 and I2 are two deformation tensor invariants I under biaxial tension and compression modes. 1BT ,I 2BT ; For shear mode, W net The input I1, I2 are two deformation tensor invariants I in the shear mode. 1SS ,I 2SS , W netThe output is the constitutive relation W under multiple deformation modes. pred (I1,I2);

[0027] Combined with the automatic differentiation calculation capability of neural networks, the following parameters of the constitutive model under uniaxial tension and compression, biaxial tension and compression, and shear modes are calculated:

[0028]

[0029] Furthermore, the constitutive model parameters under uniaxial tension and compression modes are combined to calculate the different principal expansion ratios. Predicted uniaxial stress values ​​under

[0030]

[0031] Furthermore, the constitutive model parameters under biaxial tension and compression mode are combined to calculate the different principal expansion ratios. Predicted biaxial stress values ​​under combination

[0032]

[0033] Furthermore, the principal expansion ratios of different Predicted shear stress under

[0034]

[0035] The total loss function Loss of the physical information network is defined as the main expansion ratio of the measurement point The true value of the uniaxial tensile and compressive stress at and the predicted value P predTC Mean square error Loss TC , biaxial main telescopic ratio The true value of the biaxial tensile and compressive stresses at and the predicted value P predBT Mean square error Loss BT And the main expansion ratio of the measuring point The true value of the shear stress at and the predicted value P predSS Mean square error Loss SS The linear sum of

[0036] Loss = ω TC Loss TC +ω BT Loss BT +ω SS Loss SSTC (9)

[0037] in

[0038]

[0039]

[0040]

[0041] in, is the true value of the uniaxial tensile and compressive stress at the measuring point, P predTC is the predicted value of uniaxial tensile and compressive stress at the measuring point, N TC are the uniaxial tension and compression stress data points The number of sampling points under is the true value of the biaxial tensile and compressive stress at the measuring point, P predBT is the predicted value of biaxial tensile and compressive stress at the measuring point, N BT are the biaxial tension and compression stress data points The number of sampling points under ; is the true value of the shear stress at the measuring point, P predSS is the predicted value of the shear stress at the measuring point, N SS is the shear stress data point The number of sampling points under ω TC ,ω BT and ω SS Used to balance the convergence of various residual components.

[0042] The step three is specifically as follows:

[0043] Construct a deep regression network to resolve the deformation tensor invariants I1, I2 and the strain energy function W * (I1, I2), for deep regression networks, symbolic expressions are treated as binary trees, where each node represents a symbol of an expression in the available symbol library, i.e., the input variables are the deformation invariants (I1, I2) of the hyperelastic plate, constants (e.g., C 10 ,C 20 ,C 30 ) or operation (e.g., +, -, ×, log, exp…), the final output of the deep regression network is the predicted strain energy function W * Explicit expression of (I1,I2);

[0044] For each candidate W * (I1, I2), the fitting quality reward R = 1 / (1 + NRMSE) is calculated in the deep regression network, where NRMSE is calculated by the target σ W Root mean square error after deviation normalization:

[0045]

[0046] where σ W is the variance of the strain energy data set calculated by the target strain energy function, W pred (I1, I2) is composed of the physical information network W net Predicted strain energy function W under different loading conditions pred (I1,I2),W * (I1,I2) is the strain energy function predicted by the deep regression network.

[0047] The step 4 is specifically as follows:

[0048] When the loss function Loss of the physical information network and the reward R of the fitting quality of the deep regression network meet the conditions at the same time, stop the network training;

[0049] Different thresholds are set for the physical information network and the deep regression network. That is, when the loss function Loss of the physical information network is less than a certain threshold (such as 10 -4 ), and the fitting quality reward R exceeds a certain threshold (such as 99%) at the same time, the network training is stopped. Taking into account the differences in training efficiency and convergence effect between deep regression and physical information networks, an iterative alternating training scheme is adopted. First, the loss function Loss of the physical information network is reduced, and then the fitting quality reward R of the deep regression network is trained. The network joint training is realized through an iterative algorithm.

[0050] Through the above steps, the final output of the deep regression network is the explicit expression W of the strain energy function of the constitutive relation of the multiple deformation modes obtained by the constitutive modeling method. * (I1, I2), that is, a constitutive model is generated that can reflect the nonlinear behavior of the hyperelastic material plate under multiple deformation modes.

[0051] And the strain energy function of the constitutive model W * The model parameters in (I1, I2) are determined simultaneously by a deep regression network. The obtained hyperelastic constitutive model can be verified by comparing with the experimental data set.

[0052] Beneficial effects of the present invention:

[0053] The method proposed in the present invention realizes the experimental data sets obtained through mechanical testing experiments under multiple deformation modes, combines deep regression with the training of physical information networks, directly obtains the explicit expression of the strain energy function in the constitutive model of hyperelastic materials under multiple deformation modes, and simultaneously determines the model parameters in the constitutive model.

[0054] The present invention realizes a neural network automated modeling scheme for a hyperelastic constitutive model under multiple deformation modes. The constitutive model is completely obtained by neural network training, and the strain energy function expression of the model result is different from all traditional hyperelastic constitutive models. Automatically discovering new expressions of the hyperelastic constitutive model can predict complex hyperelastic behaviors.

[0055] In order to take into account the experimental data under different modes at the same time, the idea of ​​synchronously identifying constitutive parameters using experimental data of multiple deformation modes is adopted, that is, for multiple deformation mode training, the test data of all deformation modes are used as training data. Using machine learning methods, the constitutive relations under multiple deformation modes are automatically derived from the experimental data set composed of multiple deformation modes, so as to achieve the goal of achieving the best prediction accuracy of the predicted constitutive model for different modes at the same time.

[0056] By using experimental data of multiple deformation modes as training data, the model can more comprehensively learn the response characteristics of the material under various strain conditions to more accurately reflect the constitutive behavior of the material. This helps to avoid overfitting data under specific modes and improve the generalization ability of the model. This method enables the model to train and understand data under different deformation modes, so it can be more widely applied to different types of materials and engineering scenarios. This enhances the applicability and practicality of the model.

[0057] In summary, the method of simultaneous identification using experimental data of multiple deformation modes can significantly improve the prediction accuracy and applicability of the constitutive model, and provide more reliable material performance predictions for engineering practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Schematic diagram of the flow of the physical-informed machine learning approach for the hyperelastic multi-deformation mode constitutive model.

[0059] Figure 2 Schematic diagram of the hyperelastic constitutive and deformation analysis of a material based on a hyperelastic test case of this method. DETAILED DESCRIPTION

[0060] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0061] The physical information machine learning method of the hyperelastic multi-deformation mode constitutive model of the present invention realizes a neural network automated modeling method of the hyperelastic constitutive model based on an experimental data set.

[0062] The following is a detailed description of the present invention in steps. Figure 1 As shown:

[0063] Step 1: Acquisition of experimental data sets, obtaining the stretch ratio, shear ratio and nominal stress under various deformation modes. This is used to generate the input data of the physical information network in step 2 and the deep regression network in step 3;

[0064] Step 2: Building a physical information network; used to predict the mapping relationship between the two deformation tensor invariants I1, I2 and the strain energy function W(I1, I2) under different deformation modes calculated according to the stretching ratio and shear ratio;

[0065] Step 3: Building a deep regression network to generate explicit expressions of the constitutive model that reflects the large deformation behavior of hyperelastic materials under multiple deformation modes;

[0066] Step 4: Perform network training based on the experimental data set of hyperelastic materials under multiple deformation modes obtained in step 1, and jointly train the physical information network in step 2 and the deep regression network in step 3; finally, generate an explicit expression of the constitutive model reflecting the large deformation behavior of the hyperelastic material under multiple deformation modes;

[0067] The step 1 is specifically as follows:

[0068] Firstly, a mechanical test experiment was carried out on an incompressible hyperelastic rectangular plate, and mechanical test experiments were carried out under various deformation modes such as uniaxial tension and compression test, biaxial tension and compression test, and shear test to obtain an experimental data set. The experimental data set includes the main expansion ratio obtained under uniaxial tension and compression mode. Uniaxial stress under combination Main stretch ratio under biaxial tension and compression mode Combined biaxial stress Main stretch ratio in shear mode Nominal shear stress under combination Here is the principal expansion ratio obtained from uniaxial tension and compression experiments, is the principal expansion ratio obtained from biaxial tension and compression experiments, The shear ratio obtained from the shear test is is the uniaxial stress, is the biaxial stress, is the shear stress;

[0069] Taking the superelastic plate as an example, a flat plate specimen was prepared for testing, and uniaxial tension and compression tests, biaxial tension and compression tests, and shear tests were carried out to obtain different principal expansion ratios under uniaxial tension and compression modes. And the main expansion ratio Uniaxial stress under Different principal stretch ratios in biaxial tension and compression modes And different main expansion ratio Combined biaxial stress Main stretch ratio in shear mode Main expansion ratio Shear stress under Experimental datasets of constructed hyperelastic materials;

[0070] As attached Figure 1 As shown: Specifically, a rectangular superelastic plate specimen with a length of x (m) and a width of y (m) was subjected to uniaxial tension and compression tests, biaxial tension and compression tests, and shear tests to obtain different principal expansion ratios under different uniaxial tension and compression modes. And different main expansion ratio Uniaxial stress under Different principal stretch ratios in biaxial tension and compression modes And different main expansion ratio Combined biaxial stress Main stretch ratio in shear mode Main expansion ratio Shear stress under combination Experimental data set of hyperelastic materials under multiple deformation modes composed of etc.

[0071] Based on the hyperelastic theory, different principal expansion ratios are calculated in uniaxial tension and compression modes. The corresponding deformation tensor invariant under the condition. Taking the hyperelastic plate as an example, different principal expansion ratios The corresponding deformation tensor invariant is calculated as follows:

[0072]

[0073] Based on the hyperelastic theory, different principal expansion ratios in biaxial tension and compression modes are calculated The corresponding deformation tensor invariant under the combined parameters. Taking the hyperelastic plate as an example, different principal expansion ratios The calculation formula of the corresponding deformation tensor invariant under the combined parameters is as follows:

[0074]

[0075] Calculation of the principal expansion ratio in shear mode based on hyperelasticity theory The corresponding deformation tensor invariant under the parameter. Taking the hyperelastic plate as an example, different principal expansion ratios The corresponding deformation tensor invariant is calculated as follows:

[0076] I 1SS =3+γ 2 ,I 2SS =3+γ 2 (13)

[0077] Among them, I 1TC and I 2TC are the two deformation tensor invariants of the right Cauchy Green deformation tensor under uniaxial tension and compression; I 1BT and I 2BT are the two deformation tensor invariants of the right Cauchy Green deformation tensor under biaxial tension and compression; I 1SS and I 2SS are the two deformation tensor invariants of the right Cauchy Green deformation tensor under shear deformation;

[0078] The step 2 is specifically as follows:

[0079] First, we build a fully connected neural network W net (w,b) predicted strain energy function W pred (I1, I2), where the neural network inputs the deformation tensor invariants I1, I2 of the hyperelastic material under the multi-variable mode, and outputs the predicted value W of the strain energy function under the multi-deformation mode pred (I1,I2), as attached Figure 1 shown.

[0080] W pred (I1,I2)=W net (I1,I2;w,b) (14)

[0081] Among them, W pred are the predicted values ​​of strain energy functions of the two deformation tensor invariants I1 and I2 under the predicted multiple deformation mode, W net represents a fully connected neural network, and w represents a fully connected neural network W net The weight of b represents the fully connected neural network W net Bias of

[0082] First, for uniaxial tension and compression modes, W net The input is two deformation tensor invariants I in uniaxial tension and compression mode 1TC ,I 2TC ; For biaxial tension and compression mode, W net The input I1 and I2 are two deformation tensor invariants I under biaxial tension and compression modes. 1BT ,I 2BT ; For shear mode, W net The input I1, I2 are two deformation tensor invariants I in the shear mode. 1SS ,I 2SS .W net The output is the strain energy function W under multiple deformation modes. pred (I1,I2).

[0083] Combined with the automatic differentiation calculation capability of neural networks, the following parameters of the constitutive model under uniaxial tension and compression, biaxial tension and compression, and shear mode are calculated:

[0084]

[0085] Furthermore, the constitutive model parameters under uniaxial tension and compression modes are combined to calculate the different principal expansion ratios. Predicted uniaxial stress values ​​under

[0086]

[0087] Furthermore, the constitutive model parameters under biaxial tension and compression mode are combined to calculate the different principal expansion ratios. Predicted biaxial stress values ​​under combination

[0088]

[0089] Furthermore, the principal expansion ratios of different Predicted shear stress under

[0090]

[0091] The total loss function Loss of the physical information network is defined as the main expansion ratio of the single axis at the measurement point The true value of the uniaxial stress at and the predicted value P predTC Mean square error Loss TC , biaxial main telescopic ratio The true value of the biaxial stress at and the predicted value P predBT Mean square error Loss BT And the main expansion ratio of the measuring point The true value of the shear stress at and the predicted value P predSS Mean square error Loss SS The linear sum of

[0092] Loss = ω TC Loss TC +ω BT Loss BT +ω SS Loss SSTC (19)

[0093] in

[0094]

[0095] in, is the true value of the uniaxial stress at the measuring point, PpredTC is the predicted value of the uniaxial stress at the measuring point, N TC is the uniaxial stress data point The number of sampling points under ; is the true value of the biaxial stress at the measuring point, P predBT is the predicted value of the biaxial stress at the measuring point, N BT is the biaxial stress data point The number of sampling points under ; is the true value of the shear stress at the measuring point, P predSS is the predicted value of the shear stress at the measuring point, N SS is the shear stress data point The number of sampling points under ω TC ,ω BT and ω SS Used to balance the convergence of various residual components.

[0096] The step three is specifically as follows:

[0097] Construct a deep regression network to resolve the deformation tensor invariants I1, I2 and the strain energy function W * The explicit expression between (I1,I2) is shown in the following Figure 1 For deep regression networks, symbolic expressions are treated as binary trees, where each node represents a symbol in the available symbol library, i.e., the input variables are the deformation tensor invariants (I1, I2) of the hyperelastic plate, constants (e.g., C 10 ,C 20 ,C 30 ) or operations (e.g., +, -, ×, log, exp, ...), in this representation, input variables and constants are called terminal nodes or symbols (without children), operations that accept a single argument (e.g., sin, log, ...) are unary symbols (with one child node), and operations that accept two arguments (e.g., +, -, ×, =, ...) are binary symbols. By first considering each node in depth and then from left to right, a one-dimensional list can be computed. In this representation, operators are placed in front of the corresponding operands in the expression, reducing the need for parentheses. Using prefix notation and treating symbols (called tokens) as categories allows any expression to be treated as a pure sequence of category vectors. The token sequence here is generated by using an RNN, and in order to greatly reduce the search space, the probability of certain tokens can be zeroed based on the context encoded in the generated expression tree. Additionally, we adopt a priori that enforces expression sizes to be less than 30 tokens long, contain no more than 2 levels of nested trigonometric operations (e.g., disallow cos(f.t+sin(x / x0+tan(x))) , but still allow cos(f.t+sin(x))) ), and do not contain self-nesting of exponential and logarithmic operators (e.g., disallow ) and disallow useless unary inverse operations (e.g., disallow e log(x) ), where x is the input variable, i.e., the deformation tensor invariant of the hyperelastic plate (I1, I2). Finally, the deep regression network outputs the predicted strain energy function W * An explicit expression for (I1,I2).

[0098] For each candidate W * (I1, I2), the fitting quality reward R = 1 / (1 + NRMSE) is calculated in the deep regression network, where NRMSE is calculated by the target σ W Root mean square error after deviation normalization:

[0099]

[0100] where σ W is the variance of the strain energy data set calculated by the target strain energy function, W pred (I1, I2) is composed of the physical information network W net The predicted strain energy function under multiple deformation modes, W * (I1,I2) is the strain energy function predicted by the deep regression network.

[0101] The step 4 is specifically as follows:

[0102] When the loss function Loss of the physical information network and the reward R of the fitting quality of the deep regression network meet the conditions at the same time, stop the network training;

[0103] Different thresholds are set for the physical information network and the deep regression network. That is, when the loss function Loss of the physical information network is less than a certain threshold (such as 10 -4 ), and the fitting quality reward R exceeds a certain threshold (such as 99%) at the same time, stop network training. Considering the differences in training efficiency and convergence effect between deep regression and physical information networks, an iterative alternating training scheme can be adopted to first reduce the loss function Loss of the physical information network, and then train the fitting quality reward R of the deep regression network, and realize network joint training through an iterative algorithm.

[0104] Through the above steps, the final output of the deep regression network is the explicit expression W of the strain energy function of the hyperelastic constitutive obtained by the constitutive modeling method. * (I1, I2), that is, a hyperelastic constitutive model that reflects the nonlinear behavior of the hyperelastic material plate is generated. And the strain energy function W of the constitutive model * The model parameters in (I1, I2) are determined simultaneously by a deep regression network. The obtained hyperelastic constitutive model can be verified by comparing with the experimental data set.

[0105] In summary, the method proposed in the present invention realizes the experimental data set obtained through mechanical testing experiments under multiple deformation modes, combines deep regression with the training of physical information networks, directly obtains the explicit expression of the strain energy function in the constitutive model of hyperelastic materials under multiple deformation modes, and simultaneously determines the model parameters in the constitutive model.

[0106] Step 4: Neural network training to find the optimal network parameters;

[0107] Based on the experimental data set of hyperelastic materials under multiple deformation modes obtained in step 1, network training is performed, and the physical information network in step 2 and the deep regression network in step 3 are jointly trained;

[0108] When the loss function Loss of the physical information network and the reward R of the fitting quality of the deep regression network meet the conditions at the same time, stop the network training;

[0109] Different thresholds are set for the physical information network and the deep regression network. That is, when the loss function Loss of the physical information network is less than a certain threshold (such as 10 -4 ), and the fitting quality reward R exceeds a certain threshold (such as 99%) at the same time, stop network training. Considering the differences in training efficiency and convergence effect between deep regression and physical information networks, an iterative alternating training scheme can be adopted to first reduce the loss function Loss of the physical information network, and then train the fitting quality reward R of the deep regression network, and realize network joint training through an iterative algorithm.

[0110] Through the above steps, the final output of the deep regression network is the strain energy function W of the hyperelastic constitutive model under multiple deformation modes obtained by this constitutive modeling method. * The explicit expression of (I1, I2) generates a hyperelastic constitutive model that reflects the nonlinear behavior of the hyperelastic material plate under multiple deformation modes. And the strain energy function W of the constitutive model * The model parameters in (I1, I2) are determined simultaneously by a deep regression network. The obtained hyperelastic constitutive model can be verified by comparing with the experimental data set.

[0111] Example:

[0112] The invention discloses an application of the physical information machine learning method of the hyperelastic multi-deformation mode constitutive model in the hyperelastic constitutive and deformation analysis of a certain material.

[0113] For a hyperelastic material specimen prepared from a hyperelastic human brain tissue, uniaxial tension and compression, biaxial tension and compression, and shear tests were carried out to obtain different principal expansion and contraction ratios under uniaxial tension and compression modes. Uniaxial stress under Different principal stretch ratios in biaxial tension and compression modes Combined biaxial stress Main stretch ratio in shear mode Shear stress under combination The experimental data set of hyperelastic materials is shown in the attached Figure 2 As shown in the point in (a).

[0114] Through the physical information machine learning method of the hyperelastic multi-deformation mode constitutive model of the present invention, combined with deep regression and physical information network training, the explicit expression of the strain energy function in the hyperelastic constitutive model is directly obtained as follows:

[0115] W * =a(I1-3) 2 +b(I1-3)+c(I1-3)(I2-3)

[0116] At the same time, the model parameters in the constitutive model are determined as a=2.6277; b=2.4656; c=1;

[0117] Based on the training of the network model, the hyperelastic constitutive model can be further calculated under different deformation modes. The results are shown in the attached figure. Figure 2 Compared with the experimental data set, as shown in the attached Figure 2 As shown, it can be seen that the constitutive model can predict the experimental data rules obtained from mechanical tests, and the obtained hyperelastic constitutive model is verified.

[0118] The hyperelastic multi-deformation mode constitutive model obtained based on network model training can be further combined with finite element analysis to calculate the displacement distribution of the material under different deformation modes, as shown in the attached figure. Figure 2 As shown in (b) in the figure, a quantitative analysis of the hyperelastic deformation mechanics problem of the material in multiple deformation modes under different loads is carried out.

Claims

1. A physical information machine learning method for a hyperelastic multi-deformation mode constitutive model, characterized in that: The steps include: Step 1: Acquisition of experimental data sets, obtaining the stretching ratio, shear ratio and nominal stress under various deformation modes, which are used to generate the input data of the physical information network in step 2 and the deep regression network in step 3; Step 2: Building a physical information network to predict the mapping relationship between the two deformation tensor invariants I1, I2 and the strain energy function W(I1, I2) under different deformation modes calculated according to the stretching ratio and shear ratio; Step 3: Building a deep regression network to generate explicit expressions of the constitutive model that reflects the large deformation behavior of hyperelastic materials under multiple deformation modes; Step 4: Perform network training based on the experimental data set of hyperelastic materials under multiple deformation modes obtained in step 1, and jointly train the physical information network in step 2 and the deep regression network in step 3; finally, generate an explicit expression of the constitutive model reflecting the large deformation behavior of the hyperelastic material under multiple deformation modes; The step 1 is specifically as follows: Firstly, a mechanical test experiment was carried out on an incompressible hyperelastic rectangular plate, including uniaxial tension and compression test, biaxial tension and compression test, and mechanical test experiment under various deformation modes of shear, to obtain an experimental data set, which includes the main expansion ratio obtained under uniaxial tension and compression mode. Uniaxial stress under combination Main stretch ratio in biaxial tension and compression mode Combined biaxial stress Main stretch ratio in shear mode Nominal shear stress under combination The superelastic material composed of is the principal expansion ratio obtained from uniaxial tension and compression experiments, is the principal expansion ratio obtained from biaxial tension and compression experiments, The shear ratio obtained from the shear test is is the uniaxial stress, is the biaxial stress, is the shear stress; Calculate the deformation tensor invariants in uniaxial tension and compression modes based on hyperelasticity theory Calculate the deformation tensor invariants in biaxial tension and compression modes based on hyperelasticity theory Based on the hyperelastic theory, the shear mode Deformation Tensor Invariants Among them, I 1TC and I 2TC are the two deformation tensor invariants of the right Cauchy Green deformation tensor under uniaxial tension and compression; I 1BT and I 2BT are the two deformation tensor invariants of the right Cauchy Green deformation tensor under biaxial tension and compression; I 1SS and I 2SS are the two deformation tensor invariants of the right Cauchy Green deformation tensor under shear deformation; Through the above steps, the final output of the deep regression network is the strain energy function W of the constitutive relation of the multiple deformation modes obtained by the constitutive modeling method. * The explicit expression of (I1, I2) generates a constitutive model that reflects the nonlinear behavior of the hyperelastic material plate under multiple deformation modes; And the strain energy function of the constitutive model W * The model parameters in (I1, I2) are determined simultaneously by a deep regression network; the obtained hyperelastic constitutive model is verified by comparison with the experimental data set.

2. The physical information machine learning method for a hyperelastic multi-deformation mode constitutive model according to claim 1 is characterized in that: The step 2 is specifically as follows: Building a physical information network net , to predict the strain energy function W(I1,I2) under multiple deformation modes; W pred (I1,I2)=W net (I1,I2;w,b) (4) Among them, W pred are the predicted values ​​of strain energy functions of the two invariants I1 and I2 under the multi-deformation mode, and W net represents a fully connected neural network, and w represents a fully connected neural network W net The weight of b represents the fully connected neural network W net Bias of First, for uniaxial tension and compression modes, W net The input is two deformation tensor invariants I in uniaxial tension and compression mode 1TC ,I 2TC ; For biaxial tension and compression mode, W net The input I1 and I2 are two deformation tensor invariants I under biaxial tension and compression modes. 1BT ,I 2BT ; For shear mode, W net The input I1, I2 are two deformation tensor invariants I in the shear mode. 1SS ,I 2SS , W net The output is the constitutive relation W under multiple deformation modes. pred (I1,I2); Combined with the automatic differentiation calculation capability of neural networks, the following parameters of the constitutive model under uniaxial tension and compression, biaxial tension and compression, and shear modes are calculated: Furthermore, the constitutive model parameters under uniaxial tension and compression modes are combined to calculate the different principal expansion ratios. Predicted uniaxial stress values ​​under Furthermore, the constitutive model parameters under biaxial tension and compression mode are combined to calculate the different principal expansion ratios. Predicted biaxial stress values ​​under combination Furthermore, the principal expansion ratios of different Predicted shear stress under The total loss function Loss of the physical information network is defined as the main expansion ratio of the measurement point The true value of the uniaxial tensile and compressive stress at and the predicted value P predTC Mean square error Loss TC , biaxial main telescopic ratio The true value of the biaxial tensile and compressive stresses at and the predicted value P predBT Mean square error Loss BT And the main expansion ratio of the measuring point The true value of the shear stress at and the predicted value P predSS Mean square error Loss SS The linear sum of Loss=ω TC Loss TC +ω BT Loss BT +ω SS Loss SSTC (9) in in, is the true value of the uniaxial tensile and compressive stress at the measuring point, P predTC is the predicted value of uniaxial tensile and compressive stress at the measuring point, N TC are the uniaxial tension and compression stress data points The number of sampling points under is the true value of the biaxial tensile and compressive stress at the measuring point, P predBT is the predicted value of biaxial tensile and compressive stress at the measuring point, N BT are the biaxial tension and compression stress data points The number of sampling points under is the true value of the shear stress at the measuring point, P predSS is the predicted value of the shear stress at the measuring point, N SS is the shear stress data point The number of sampling points under ω TC ,ω BT and ω SS Used to balance the convergence of various residual components.

3. The physical information machine learning method for a hyperelastic multi-deformation mode constitutive model according to claim 2 is characterized in that: The step three is specifically as follows: Construct a deep regression network to resolve the deformation tensor invariants I1, I2 and the strain energy function W * (I1, I2). For the deep regression network, the symbolic expression is regarded as a binary tree, where each node represents a symbol of the expression in the available symbol library, that is, the input variables are the deformation invariants (I1, I2) of the hyperelastic plate, constants or operations, and the final output of the deep regression network is the predicted strain energy function W * Explicit expression of (I1,I2); For each candidate W * (I1, I2), the fitting quality reward R = 1 / (1 + NRMSE) is calculated in the deep regression network, where NRMSE is calculated by the target σ W Root mean square error after deviation normalization: where σ W is the variance of the target strain energy function, W pred (I1, I2) is composed of the physical information network W net Predicted strain energy function W under different loading conditions pred (I1,I2),W * (I1,I2) is the strain energy function predicted by the deep regression network.

4. The physical information machine learning method for a hyperelastic multi-deformation mode constitutive model according to claim 3 is characterized in that: The step 4 is specifically as follows: When the loss function Loss of the physical information network and the reward R of the fitting quality of the deep regression network meet the conditions at the same time, stop the network training; Different thresholds are set for the physical information network and the deep regression network. That is, when the loss function Loss of the physical information network is less than a certain threshold and the fitting quality reward R exceeds a certain threshold at the same time, the network training is stopped and an iterative alternating training scheme is adopted to first reduce the loss function Loss of the physical information network, and then train the fitting quality reward R of the deep regression network. The network joint training is achieved through an iterative algorithm.

Citation Information

Patent Citations

  • Method for constructing and simulating hyperelastic constitutive model with stress abrupt rise characteristic under small strain

    CN116502480A

  • Hyperelastic constitutive model method based on deep regression and physical information network

    CN118428211A