Material deformation behavior high-robustness prediction method and system based on bionic intelligent constitutive structure

Abstract

The invention provides a material deformation behavior high-robustness prediction method and system based on a bionic intelligent constitutive structure, and belongs to the technical field of solid mechanics and computational intelligence cross, and the method comprises the steps: obtaining a multi-modal data sample under multiple materials and multiple scenes; the method comprises the following steps: constructing an initial BP neural network constitutive model, taking a multi-modal data sample as input, taking stress as output, simulating a stem cell hierarchical differentiation mechanism by using a bionic hierarchical architecture to obtain an optimized genetic algorithm, optimizing an initial weight and an initial threshold of the initial BP neural network constitutive model by using the optimized genetic algorithm to obtain a transition weight and a transition threshold, and constructing the stem cell hierarchical differentiation mechanism based on the transition weight and the transition threshold. Optimizing the transition weight and the transition threshold through a gradient descent algorithm to obtain an improved BP neural network constitutive model; and predicting the deformation behavior of the to-be-predicted material based on the improved BP neural network constitutive model. According to the method, the deformation behaviors of multiple materials in multiple scenes can be quickly and accurately predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solid mechanics and computing intelligence, and particularly relates to a material deformation behavior high-robustness prediction method and system based on a bionic intelligent constitutive model. BACKGROUND

[0002] As a mathematical model for describing the stress-strain relationship of materials under external load, the constitutive model is an important means to study the rheological behavior of materials. Traditional constitutive models construct a mathematical mapping of macroscopic mechanical behavior through empirical formula fitting or internal variable theory. In recent years, neural networks have been widely used in the construction of constitutive models due to their strong non-linear mapping ability. Data-driven methods are used to replace traditional mathematical modeling to capture the non-linear mechanical behavior of materials under complex loads.

[0003] However, existing researches on constitutive models based on neural networks are mainly limited to single material systems or single physical field application scenarios, and their applicability under multi-material and multi-physical field coupling conditions has not been fully explored. For example, models represented by back propagation neural networks have complex non-linear transient shock regions in the deformation behavior of materials under multi-physical field coupling.

[0004] In view of the above-mentioned core problem that the existing constitutive model is difficult to effectively characterize the transient shock behavior of materials under multi-material and multi-physical field coupling conditions, the existing technology has the following specific defects: Traditional constitutive models have the problems of complex model structure, multiple material parameters, difficulty in accurate calculation, and poor prediction accuracy due to the difficulty in characterizing complex deformation behavior under multi-field coupling. Constitutive models based on neural networks, especially BP neural networks, have the problems of difficulty in handling complex non-linear transient shock behavior under multi-physical field coupling, easy to fall into local optimal solution in the training process, low convergence efficiency, and poor prediction robustness. SUMMARY

[0005] Therefore, it is necessary to provide a material deformation behavior high-robustness prediction method and system based on a bionic intelligent constitutive model to solve the problems of difficulty in handling complex non-linear transient shock behavior under multi-physical field coupling, easy to fall into local optimal solution in the training process, low convergence efficiency, poor prediction robustness, and difficulty in effectively predicting the deformation behavior of materials.

[0006] In order to solve the above-mentioned technical problems, in a first aspect, the present application provides a material deformation behavior high-robustness prediction method based on a bionic intelligent constitutive model, comprising: Obtaining multi-modal data samples under multi-material and multi-scenario conditions; An initial BP neural network constitutive model is constructed, the multi-modal data samples are taken as inputs, and stresses are taken as outputs, a bionic hierarchical architecture is used to simulate a stem cell hierarchical differentiation mechanism, an optimized genetic algorithm is obtained, the initial weights and initial thresholds of the initial BP neural network constitutive model are optimized by using the optimized genetic algorithm to obtain transition weights and transition thresholds, the transition weights and transition thresholds are optimized by using a gradient descent algorithm, and an improved BP neural network constitutive model is obtained. The improved BP neural network constitutive model is used to predict the deformation behavior of a material to be predicted.

[0007] In a possible implementation, before the multi-modal data samples under the multiple materials and multiple scenes are acquired, the method further includes: Text data, table data, and image data are extracted. The text data is subjected to text analysis, the unstructured text data is converted into a pure text sequence, the pure text sequence is scanned sentence by sentence, and a structured parameter key-value pair is extracted. The table data is located by using a declarative query rule, the table data is crawled and analyzed, and a structured parameter mapping table is extracted. The image data is subjected to image enhancement, a stress-strain curve region is demarcated by using an image segmentation model, discrete data point sequences are generated through coordinate axis calibration and pixel mapping, and the structured parameter key-value pair, the structured parameter mapping table, and the discrete data point sequences are unified into a multi-dimensional tensor data set, and the multi-modal data samples are obtained.

[0008] In a possible implementation, the bionic hierarchical architecture is used to simulate the stem cell hierarchical differentiation mechanism, the optimized genetic algorithm is obtained, the initial weights and initial thresholds of the initial BP neural network constitutive model are optimized by using the optimized genetic algorithm to obtain the transition weights and transition thresholds, and the method includes the following steps. The initial weights and initial thresholds of the initial BP neural network constitutive model are encoded into real number vectors, a real number coded chromosome population is formed, and the real number coded chromosome population is divided into a stem cell population and a differentiated cell population. The stem cell population and the differentiated cell population are subjected to iterative co-evolution, when a first preset condition is met, an optimal chromosome is output, and the optimal chromosome is decoded to obtain the transition weights and transition thresholds of the initial BP neural network constitutive model. The first preset condition is that a preset population iteration number or a preset fitness convergence threshold is reached.

[0009] In a possible implementation, the stem cell population and the differentiated cell population are subjected to iterative co-evolution, and the method includes the following steps. ​retaining the individual with the highest fitness in the stem cell population by roulette wheel selection method; replacing the individual with the lowest fitness in the stem cell population with the individual with the highest fitness in the differentiated cell population when a preset genetic generation is met; eliminating half of the individuals in the stem cell population, extracting an equal amount of differentiated cells from the differentiated cell population, and injecting the differentiated cells into the stem cell population to reconstruct the population structure when the stem cell population meets a preset reconstruction condition; The preset reconstruction condition is that the stem cell population does not evolve within a preset number of iterations.

[0010] In a possible implementation, the retaining the individual with the highest fitness in the stem cell population by roulette wheel selection method comprises: calculating the fitness of individuals in the stem cell population, and applying an epigenetic marker to the individual with high fitness in the stem cell population to make the individual with high fitness be preferentially retained in genetic operation; The formula for calculating the fitness of an individual is as follows: (1) (2) wherein, is the index number of the individual in the population, is the probability of being selected by the individual, is the fitness value of the individual, is the population size, is a constant, and is a prediction error.

[0011] In a possible implementation, after the division into the stem cell population and the differentiated cell population, the method further comprises: configuring the stem cell population with genetic operators of low mutation rate and high crossover rate; configuring the differentiated cell population with genetic operators of high mutation rate and low crossover rate.

[0012] In a possible implementation, the optimizing the transition weight and the transition threshold by the adaptive gradient descent algorithm comprises: taking the transition weight and the transition threshold as a starting point; taking a fixed number of the data samples for small batch training each time, comprising: calculating the loss between the true value and the predicted value of stress according to the small batch training, calculating the gradient of the transition weight and the transition threshold by back propagation, calculating the step size by an adaptive algorithm, adjusting the transition weight and the transition threshold in the negative gradient direction by using the gradient and the step size; repeating the taking the fixed size of the data samples for small batch training, and obtaining the target weight and the target threshold when a second preset condition is met; The second preset condition is satisfied when the gradient norm is lower than a preset threshold or a preset number of gradient descent iterations is reached.

[0013] In another aspect, the present application also provides a material deformation behavior high-robust prediction system based on bionic intelligent constitutive, comprising: A data acquisition module is configured to acquire multi-modal data samples under multi-material and multi-scenario conditions. A model optimization module is configured to construct an initial BP neural network constitutive model, take the multi-modal data samples as input and take stress as output, simulate stem cell hierarchical differentiation mechanism by using a bionic hierarchical architecture, obtain an optimized genetic algorithm, and optimize initial weights and initial thresholds of the initial BP neural network constitutive model to obtain transition weights and transition thresholds by using the optimized genetic algorithm, and optimize the transition weights and the transition thresholds by using a gradient descent algorithm to obtain an improved BP neural network constitutive model. A prediction module is configured to predict deformation behavior of a material to be predicted based on the improved BP neural network constitutive model.

[0014] In a second aspect, the present application also provides an electronic device comprising a memory and a processor, wherein The memory is configured to store a program. The processor is coupled to the memory and is configured to execute the program stored in the memory to implement steps of the material deformation behavior high-robust prediction method based on bionic intelligent constitutive in any of the implementation manners described above.

[0015] In a third aspect, the present application also provides a computer-readable storage medium configured to store computer-readable programs or instructions, wherein the programs or instructions are executed by a processor to implement steps of the material deformation behavior high-robust prediction method based on bionic intelligent constitutive in any of the implementation manners described above.

[0016] The beneficial effects of the present application are: the material deformation behavior high-robustness prediction method based on the bionic intelligent constitutive provided by the present application firstly acquires multi-modal data samples under multiple materials and multiple scenes, then constructs an initial BP neural network constitutive model, takes the multi-modal data samples as input and takes stress as output, simulates the stem cell hierarchical differentiation mechanism by using the bionic hierarchical architecture, obtains an optimized genetic algorithm, uses the optimized genetic algorithm to optimize the initial weights and initial thresholds of the initial BP neural network constitutive model to obtain transition weights and transition thresholds, optimizes the transition weights and transition thresholds by using the gradient descent algorithm, obtains an improved BP neural network constitutive model, and predicts the deformation behavior of the material to be predicted based on the improved BP neural network constitutive model. Since the multi-modal data samples under multiple materials and multiple scenes are taken as the input of the model training, the complex nonlinear transient sudden change behavior of the material under the coupling action of multiple physical fields can be predicted, further, the traditional genetic algorithm is optimized by using the bionic hierarchical architecture to simulate the stem cell hierarchical differentiation mechanism, the optimized genetic algorithm is obtained, the weights and thresholds of the BP neural network constitutive model are optimized by using the optimized genetic algorithm, the optimization efficiency is improved, and the optimization strategy of the BP neural network constitutive model by using the multi-modal data samples under multiple materials and multiple scenes and the bionic hierarchical architecture to simulate the stem cell hierarchical differentiation mechanism makes it possible to quickly and accurately predict the deformation behavior of the material under the coupling action of multiple physical fields. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 An embodiment flowchart of the material deformation behavior high-robustness prediction method based on the bionic intelligent constitutive provided by the present application is shown in the figure. Figure 2 An embodiment schematic diagram of acquiring multi-modal data samples under multiple materials and multiple scenes provided by the present application is shown in the figure. Figure 3 Another embodiment schematic diagram of acquiring multi-modal data samples under multiple materials and multiple scenes provided by the present application is shown in the figure. Figure 4 An embodiment flowchart of S102 in the present application is shown in the figure. Figure 1 Figure 5 Another embodiment flowchart of S102 in the present application is shown in the figure. Figure 1 ​In S102 of the method, a stem cell hierarchical differentiation mechanism is simulated by using a bionic hierarchical architecture to obtain an optimized genetic algorithm, and the initial weights and initial thresholds of an initial BP neural network constitutive model are optimized by using the optimized genetic algorithm to obtain a transition weight and a transition threshold. Figure 6 The present application provides a neural network constitutive model for a material deformation behavior high-robust prediction system. Figure 1 In S102 of the method, a stem cell hierarchical differentiation mechanism is simulated by using a bionic hierarchical architecture to obtain an optimized genetic algorithm, and the initial weights and initial thresholds of an initial BP neural network constitutive model are optimized by using the optimized genetic algorithm to obtain a transition weight and a transition threshold. Figure 7 The present application provides a neural network constitutive model for a material deformation behavior high-robust prediction system. Figure 8 The present application provides a neural network constitutive model for a material deformation behavior high-robust prediction system. Figure 9 The present application provides a neural network constitutive model for a material deformation behavior high-robust prediction system. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0020] In the description of the embodiments of the present application, unless otherwise specified, the meaning of "a plurality of" is two or more. The association relationship of the associated objects is described as "and / or", which means that there can be three relationships, for example: A and / or B, which can represent the three cases of A alone, A and B together, and B alone.

[0021] The "first", "second", and the like described in the embodiments of the present application are only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the technical features limited by "first" and "second" can explicitly or implicitly include at least one of the features.

[0022] In this document, the term "embodiment" means that the specific features, structures, or characteristics described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0023] Before demonstrating the embodiments, the following terms will be explained.

[0024] Constitutive model: also known as the mechanical constitutive equation of a material, or the stress-strain model of a material, is a mathematical expression describing the mechanical properties (stress-strain-strength-time relationship) of a material. The stress-strain relationship of a material is very complex, exhibiting nonlinearity, viscoelasticity, plasticity, dilatation, anisotropy, etc., and is also affected by stress level, stress history, as well as the material's composition, state, and structure.

[0025] Stress refers to the internal force generated between different parts of an object when it deforms due to external factors (force, humidity, temperature field changes, etc.). The internal force per unit area is called stress.

[0026] Image enhancement refers to the purposeful emphasis on the overall or local characteristics of an image, making an originally unclear image clearer or emphasizing certain features of interest, amplifying the differences between features of different objects in the image, suppressing features of no interest, thereby improving image quality, enriching information, enhancing image interpretation and recognition, and meeting the needs of certain special analyses.

[0027] This invention provides a robust prediction method and system for material deformation behavior based on biomimetic intelligent constitutive models, which will be described in detail below.

[0028] Figure 1 This is a schematic flowchart of an embodiment of the highly robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model provided by the present invention, as shown below. Figure 1 As shown, robust prediction methods for material deformation behavior based on biomimetic intelligent constitutive models include: S101. Obtain multimodal data samples under multiple materials and multiple scenarios.

[0029] It should be noted that "multi-material" refers to the inclusion of different categories of engineering materials (such as metals, alloys, and composite materials), and "multi-scenario" refers to multi-physics coupling, which means that there are two or more physical fields interacting at the same time, resulting in a strong nonlinear material response.

[0030] In some embodiments of the present invention, the materials are 7075-T6 aluminum alloy and ultra-high-strength steel components. It should be noted that the selection of materials can be adjusted according to actual needs and is not limited to 7075-T6 aluminum alloy and ultra-high-strength steel components, which will not be elaborated here.

[0031] In some embodiments of the present invention, the physical field is electrically assisted stretching or cyclic loading. It should be noted that the selection of the physical field can be adjusted according to actual needs and is not limited to electrically assisted stretching and cyclic loading, which will not be elaborated here.

[0032] This embodiment covers aluminum alloys and steel, which belong to two different material systems and have different deformation behaviors, demonstrating the wide applicability of this application.

[0033] In some embodiments of the present invention, such as Figure 2 As shown, step S101, which acquires multimodal data samples under multiple materials and scenarios, can be used as sample data for constructing the constitutive model of 7075-T6 aluminum alloy. The dataset contains dynamic variation information of multiple physical field parameters such as strain (ε), stress (σ), current density (J(A / mm2)), and temperature (T).

[0034] In some embodiments of the present invention, such as Figure 3 As shown, step S101, which acquires multimodal data samples under multiple materials and scenarios, can be used as sample data for constructing the constitutive model of ultra-high strength steel. The dataset contains dynamic change information of multiple physical field parameters such as stress (σ), strain (ε) curves, and temperature (T) from cyclic loading-unloading-loading tests.

[0035] S102. Construct an initial BP neural network constitutive model. Using multimodal data samples as input and stress as output, simulate the stem cell hierarchical differentiation mechanism using a biomimetic hierarchical architecture to obtain an optimized genetic algorithm. Using the optimized genetic algorithm, optimize the initial weights and initial thresholds of the improved BP neural network constitutive model to obtain transition weights and transition thresholds. Optimize the transition weights and transition thresholds using a gradient descent algorithm to obtain an improved BP neural network constitutive model.

[0036] It should be noted that multimodal data refers to a dataset composed of two or more different modalities. Here, modality refers to the form or type of data, such as text, images, audio, video, sensor data, etc.

[0037] It should be noted that the initial range of the number of hidden layer nodes in the BP neural network is estimated using the empirical formula (Equation 1), and other parameters (training iterations: 2000, learning rate: 0.1, training accuracy: 0.00001, momentum factor: 0.01) are also considered. (1) In the formula, Represents the number of hidden layer nodes. m This represents the number of nodes in the input layer. n Represents the number of output layer nodes. α It is an empirical adjustment parameter. α ∈[1,10].

[0038] It should be noted that before using multimodal data samples as input, the input multimodal data samples need to be preprocessed. A linear transformation is used to map each parameter value to the [0,1] interval to eliminate the interference of dimensional differences on model training. The mathematical expression is shown in Equation 2: (2) In the formula, x Represents the original experimental data. x max and x min These represent the maximum and minimum values ​​in the corresponding data, respectively. x ' represents the normalized variable data.

[0039] S103. The deformation behavior of the material to be predicted is predicted based on the target BP neural network constitutive model.

[0040] Specifically, the target BP neural network constitutive model is input as a parameter vector of the material in the current scenario. It calculates the stress value by the input parameter vector and performs predictions through the integrated ABAQUS simulation tool.

[0041] It should be noted that the material-scene-stress correspondence rules established by clustering multimodal data samples under multiple materials and multiple scenarios enable the trained target BP neural network constitutive model to have accurate parameter matching and low prediction error.

[0042] In summary, this invention provides a robust prediction method for material deformation behavior based on biomimetic intelligent constitutive models. First, multimodal data samples are acquired across multiple materials and scenarios. Then, an initial BP neural network constitutive model is constructed. Using the multimodal data samples as input and stress as output, a biomimetic hierarchical architecture is used to simulate the hierarchical differentiation mechanism of stem cells, resulting in an optimized genetic algorithm. Using this optimized genetic algorithm, the initial weights and thresholds of the constructed initial BP neural network constitutive model are optimized to obtain transition weights and thresholds. The transition weights and thresholds are then optimized using a gradient descent algorithm to obtain an improved BP neural network constitutive model. Finally, the deformation behavior of the material to be predicted is predicted based on this improved BP neural network constitutive model. By using multimodal data samples from multiple materials and scenarios as input for model training, it is possible to predict the complex nonlinear transient behavior of materials under multi-physics coupling. Furthermore, by using a biomimetic hierarchical architecture to simulate the stem cell hierarchical differentiation mechanism, the traditional genetic algorithm is optimized to obtain an optimized genetic algorithm. The optimized genetic algorithm is then used to optimize the weights and thresholds of the BP neural network constitutive model, improving the optimization efficiency. The optimization strategy of using multimodal data samples from multiple materials and scenarios and the biomimetic hierarchical architecture to simulate the stem cell hierarchical differentiation mechanism for the BP neural network constitutive model enables the rapid and accurate prediction of the deformation behavior of materials under multi-physics coupling.

[0043] In some embodiments of the present invention, such as Figure 4 As shown, step S101 includes: S401. Extract text data, tabular data, and image data; S402. Perform text parsing on the text data, convert the unstructured text data into a plain text sequence, scan the plain text sequence sentence by sentence, and extract structured parameter key-value pairs; S403. Locate table data through declarative query rules, crawl and parse the table data, and extract the structured parameter mapping table; S404. Image enhancement is performed on the image data. The stress-strain curve region is calibrated through the image segmentation model, and a discrete data point sequence is generated through coordinate axis calibration and pixel mapping. S405. Unify the structured parameter key-value pairs, structured parameter mapping table, and discrete data point sequence into a multidimensional tensor dataset to obtain multimodal data samples.

[0044] The sources of text data, tabular data, and image data can be the data obtained through experiments, or the data obtained through processing based on natural language processing engines and data mining algorithms, or from publicly available internet databases and academic literature, etc., which will not be elaborated here.

[0045] The text parsing method can be any entity recognition task model used in natural language processing, such as the sequence labeling model BiLSTM-CRF that combines Bidirectional Long Short-Term Memory Network (BiLSTM) and Conditional Random Field (CRF), which will not be elaborated here.

[0046] Declarative query rules can be any method that can locate structured data, such as XPath, JSONiq, JSONPath, XQuery, LINQ, PandasQL, etc., which will not be elaborated here.

[0047] Image enhancement can be any method that can enhance an image to selectively highlight features of interest or suppress (mask) certain unwanted features, such as image denoising, rotation correction, contrast enhancement, etc.

[0048] The image segmentation model can be any model capable of classifying each pixel in an image into a specific category. In this embodiment, the specific category can be: kground (background), Axes (coordinate axes), Gridlines (grid lines), Curve (target stress-strain curve), Text (coordinate axis labels, legends, etc.); and the model can be any of U-Net and its variants, ResNet classification branches, etc., which will not be elaborated here.

[0049] By using multimodal data samples from multiple materials and scenarios as input for model training, it is possible to predict the complex nonlinear transient behavior of materials under the coupling effect of multiple physics fields.

[0050] In some embodiments of the present invention, such as Figure 5 As shown, in step S102, a biomimetic hierarchical architecture is used to simulate the stem cell hierarchical differentiation mechanism, and the initial weights and initial thresholds of the initial BP neural network constitutive model are optimized to obtain transition weights and transition thresholds, including: The initial weights and initial thresholds of the initial BP neural network constitutive model are encoded as real number vectors to form a real number encoded chromosome population, which is then divided into stem cell population and differentiated cell population. The stem cell population and the differentiated cell population are iteratively co-evolved. When the first preset condition is met, the optimal chromosome is output. The optimal chromosome is then decoded to obtain the transition weights and transition thresholds of the initial BP neural network constitutive model.

[0051] Among them, satisfying the first preset condition is to reach a preset number of population iterations or a preset fitness convergence threshold.

[0052] In some embodiments of the present invention, after a preset number of population iterations or a preset fitness convergence threshold is reached, the two populations are merged, the individual with the highest fitness is selected as the optimal solution, and the final parameter configuration is output. The final parameter configuration (7075-T6 aluminum alloy) is shown in Table 1: Table 1: Parameter Configuration of 7075-T6 Aluminum Alloy

[0053] In some embodiments of the present invention, after a preset number of population iterations or a preset fitness convergence threshold is reached, the two populations are merged, the individual with the highest fitness is selected as the optimal solution, and the final parameter configuration is output. The final parameter configuration (ultra-high strength steel) is shown in Table 2: Table 2: Parameter Configuration of Ultra-High Strength Steel

[0054] In some embodiments of the present invention, the step of iterative co-evolution of stem cell populations and differentiated cell populations includes: The highest-fitting individuals in the stem cell population were preserved using a roulette wheel selection method. When the preset number of genetic generations is met, the individual with the highest fitness in the differentiated cell population replaces the individual with the lowest fitness in the stem cell population; When the stem cell population meets the preset reconstruction conditions, half of the individuals in the stem cell population are eliminated, and a new amount of differentiated cells equal to half of the individuals are extracted from the differentiated cell population and injected into the stem cell population to reconstruct the population structure.

[0055] The preset reconstruction condition is that the stem cell population does not evolve within a preset number of iterations.

[0056] It should be noted that replacing the individual with the lowest fitness in the stem cell population with the individual with the highest fitness in the differentiated cell population, triggered by chemokines, achieves bidirectional signal feedback between the stem cell population and the differentiated cell population. Here, bidirectional signal feedback refers to the mutual information transmission between the stem cell population and the differentiated cell population. Chemokines are a class of small molecule cytokines or signaling proteins secreted by cells, which enable the mutual information transmission between the stem cell population and the differentiated cell population.

[0057] It should be noted that population remodeling is achieved through the p53 apoptosis program. p53 is a tumor suppressor that can monitor cellular stress signals and determine cell fate. The apoptosis program refers to the process of cell apoptosis, a form of programmed cell death that eliminates useless or abnormal cells through strictly regulated molecular pathways to maintain tissue homeostasis. The p53 apoptosis program is mapped as follows: when the fitness of the stem cell population does not improve within consecutive generations, it is judged as premature convergence, triggering the p53 apoptosis program, eliminating 50% of individuals and injecting an equal amount of new solutions from the differentiated cell population. This operation simulates the function of biological apoptosis in eliminating abnormal cells and breaking through the local optimum trap.

[0058] In some embodiments of the present invention, the highest-fitting individuals in the stem cell population are preserved using a roulette wheel selection method, including: The fitness of individuals in a stem cell population is calculated, and epigenetic markers are applied to individuals with high fitness to ensure their preferential retention in genetic manipulation. The formula for calculating individual fitness is as follows: (3) (4) In the formula, This refers to the index number of an individual in the population. The probability of an individual being selected. This represents the individual fitness value. For population size, It is a constant. This represents the prediction error.

[0059] To avoid getting stuck in local optima during model training, step S301, after dividing the model into stem cell and differentiated cell populations, also includes: Configure genetic operators with low mutation rate and high crossover rate for stem cell populations; Genetic operators with high mutation rates and low crossover rates are configured for differentiated cell populations.

[0060] In step S102, the transition weights and transition thresholds are optimized using an adaptive gradient descent algorithm, such as... Figure 6 As shown, it includes: S601, Use the transition weight and transition threshold as the starting point; S602. A fixed number of data samples are taken each time for mini-batch training, including: The loss between the actual stress value and the predicted stress value is calculated based on mini-batch training. The gradient of the transition weight and transition threshold is calculated through backpropagation. The step size is calculated using an adaptive algorithm. The transition weight and transition threshold are adjusted along the negative gradient direction using the gradient and step size. S603. Repeatedly take data samples of a fixed size for mini-batch training. When the second preset condition is met, the target weight and target threshold are obtained.

[0061] Among them, the second preset condition is that the gradient norm is lower than a preset threshold or the preset number of gradient descent iterations is reached.

[0062] In some embodiments of the present invention, when the gradient norm is lower than a preset threshold or after reaching a preset number of gradient descent iterations, the final mesh parameter configuration (7075-T6 aluminum alloy) is shown in Table 3: Table 3: Final Mesh Configuration for 7075-T6 Aluminum Alloy

[0063] Additionally, the mesh structure (7075-T6 aluminum alloy) has the following parameters: Input layer: temperature, strain, current density; Hidden layer nodes: 21; Output layer: stress; Mathematical expression: σ=f(T,ε,J), where T is temperature, ε is strain, and J is current density; Hidden layer: Sigmoid function (nonlinear mapping capability); Output layer: Purelin function (ensuring stress continuity); Loss function: Mean squared error (MSE); Data partitioning: 70% training set / 30% validation set.

[0064] In some embodiments of the present invention, when the gradient norm is lower than a preset threshold or after reaching a preset number of gradient descent iterations, the final mesh parameter configuration (ultra-high strength steel) is shown in Table 4: Table 4: Final Mesh Parameter Configuration for Ultra-High Strength Steel

[0065] Additionally, the mesh structure (ultra-high strength steel) has the following parameters: Input layer: temperature, strain; Hidden layer nodes: 15; Output layer: stress; Mathematical expression: T is the temperature. In response, The strain rate is represented by the following parameters: Hidden layer: Sigmoid function (for nonlinear mapping capability); Output layer: Purelin function (to ensure stress continuity); Loss function: Mean squared error (MSE); Data partitioning: 70% training set / 30% validation set.

[0066] In some embodiments of the present invention, in order to verify the highly robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model provided by the present invention, simulation is performed in the following manner: After saving the optimized parameters as an HDF5 file, the pre-trained weights and biases are read and the neural network parameters are loaded. The current strain increment from ABAQUS is received, and the total strain is updated. The forward propagation function is called to calculate the current stress value. Parameters such as current strain, temperature, and current density are recorded for use in the next increment step. A Python script is used to control the geometric definition, mesh generation, material property binding, and job submission for ABAQUS modeling. Finally, the stress-strain data in the simulation results is extracted, and the mean absolute error (MAE) and coefficient of determination (R²) are calculated and output.

[0067] It should be noted that during the simulation process: The VUHARD materials subroutine interface provided by ABAQUS is used to simulate the stem cell hierarchical differentiation mechanism using a biomimetic hierarchical architecture. The initial BP neural network constitutive model is optimized and encapsulated as a custom module in Fortran. Through the Python-C kernel communication mechanism, deep coupling between the algorithm logic and the ABAQUS solver is achieved. ABAQUS involves programming the calculation of the stress tensor based on the initial stress tensor and strain increment, employing a prediction / correction algorithm. During the correction phase, the flow stress is calculated at the current integration point based on strain, strain rate, and temperature.

[0068] Specifically, a high-performance computing module was written in Fortran to integrate core computational functions such as dynamic loading of parameters, stress updates, and state variable evolution of the trained neural network model. This model achieves deep coupling with the finite element kernel through the ABAQUS user subroutine interface, employing a radial return algorithm to iteratively solve the stress-strain response, thereby simulating the deformation behavior of complex materials. Based on the singular value spectrum of the Jacobian matrix output by the neural network, the global load increment step size is adjusted in real time; simultaneously, a strain path tracking mechanism is established, and the continuity of the constitutive model is improved by using BP neural network units to remember historical deformation forms. A CUDA kernel-optimized neural network inference layer is developed, utilizing GPU shared memory to reuse material state variables. In the trained neural network model, the highest-accuracy neural network constitutive model is selected, and the optimized neural network internal parameters are saved in an HDF5 file for use in the finite element code. A self-written program reads and generates Fortran subroutines to perform numerical simulations of structures subjected to thermal loads. By extracting the internal variables of the neural network, they are passed as subroutines in Fortran through functional equations. The flow stress integration process of the neural network constitutive model is shown below. Figure 7 As shown.

[0069] To better implement the robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model in this invention, based on the robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model, correspondingly, as follows: Figure 8 As shown, this embodiment of the invention also provides a robust prediction system for material deformation behavior based on biomimetic intelligent constitutive models. The robust prediction system 800 for material deformation behavior based on biomimetic intelligent constitutive models includes: Data acquisition module 801 is used to acquire multimodal data samples under multiple materials and multiple scenarios; The model optimization module 802 is used to construct an initial BP neural network constitutive model. It takes multimodal data samples as input and stress as output, and uses a biomimetic hierarchical architecture to simulate the stem cell hierarchical differentiation mechanism to obtain an optimized genetic algorithm. Using the optimized genetic algorithm, the initial weights and initial thresholds of the initial BP neural network constitutive model are optimized to obtain transition weights and transition thresholds. The transition weights and transition thresholds are optimized by the gradient descent algorithm to obtain an improved BP neural network constitutive model. The prediction module 803 predicts the deformation behavior of the material to be predicted based on an improved BP neural network constitutive model.

[0070] The above embodiment provides a robust prediction system 800 for material deformation behavior based on biomimetic intelligent constitutive model, which can realize the technical solution described in the embodiment of the robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model. The specific implementation principle of each module or unit can be found in the corresponding content in the embodiment of the robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model, which will not be repeated here.

[0071] like Figure 9 As shown, the present invention also provides an electronic device 900. The electronic device 900 includes a processor 901, a memory 902, and a display 903. Figure 9 Only some components of the electronic device 900 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0072] In some embodiments, processor 901 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 902 or process data, such as a robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model in this invention.

[0073] In some embodiments, processor 901 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 901 may be local or remote. In some embodiments, processor 901 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0074] In some embodiments, memory 902 may be an internal storage unit of electronic device 900, such as a hard disk or memory of electronic device 900. In other embodiments, memory 902 may also be an external storage device of electronic device 900, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 900.

[0075] Furthermore, the memory 902 may include both internal storage units of the electronic device 900 and external storage devices. The memory 902 is used to store application software and various types of data installed on the electronic device 900.

[0076] In some embodiments, display 903 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 903 is used to display information from electronic device 900 and to display a visual user interface. Components 901-903 of electronic device 900 communicate with each other via a system bus.

[0077] In one embodiment, when processor 901 executes a robust prediction program for material deformation behavior based on a biomimetic intelligent constitutive model stored in memory 902, the following steps can be implemented: Acquire multimodal data samples from multiple materials and scenarios; An initial BP neural network constitutive model is constructed, with multimodal data samples as input and stress as output. A biomimetic hierarchical architecture is used to simulate the stem cell hierarchical differentiation mechanism. The initial weights and thresholds of the initial BP neural network constitutive model are optimized to obtain transition weights and thresholds. The transition weights and thresholds are then optimized using a gradient descent algorithm to obtain the target BP neural network constitutive model. The deformation behavior of the material to be predicted is predicted based on the target BP neural network constitutive model.

[0078] It should be understood that when the processor 901 executes a robust prediction program for material deformation behavior based on a biomimetic intelligent constitutive model in the memory 902, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0079] Furthermore, this embodiment of the invention does not specifically limit the type of electronic device 900 mentioned. Electronic device 900 can be a portable electronic device such as a mobile phone, tablet computer, or laptop computer. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the invention, electronic device 900 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0080] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the robust prediction method for material deformation behavior based on biomimetic intelligent constitutive model provided in the above-described method embodiments.

[0081] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0082] The above provides a detailed description of a robust prediction method and system for material deformation behavior based on biomimetic intelligent constitutive model provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for high-robust prediction of material deformation behavior based on bionic intelligent constitutive, characterized in that, The method comprises the following steps: acquiring multi-modal data samples under multiple materials and multiple scenarios; constructing an initial BP neural network constitutive model, taking the multi-modal data samples as input and stress as output, simulating stem cell hierarchical differentiation mechanism by using a biomimetic hierarchical architecture, obtaining an optimized genetic algorithm, and using the optimized genetic algorithm to optimize the initial weights and initial thresholds of the initial BP neural network constitutive model to obtain transition weights and transition thresholds, and optimizing the transition weights and transition thresholds by using a gradient descent algorithm to obtain an improved BP neural network constitutive model; predicting the deformation behavior of a material to be predicted based on the improved BP neural network constitutive model.

2. The method of claim 1, wherein, Before acquiring the multi-modal data samples under multiple materials and multiple scenarios, the method further comprises the following steps: extracting text data, table data and image data; performing text analysis on the text data, converting unstructured text data into pure text sequences, and extracting structured parameter key-value pairs by scanning the pure text sequences sentence by sentence; locating the table data by using a declarative query rule, and extracting a structured parameter mapping table by crawling and analyzing the table data; performing image enhancement on the image data, calibrating stress-strain curve regions by using an image segmentation model, and generating a discrete data point sequence through coordinate axis calibration and pixel mapping; unifying the structured parameter key-value pairs, the structured parameter mapping table and the discrete data point sequence into a multi-dimensional tensor data set to obtain the multi-modal data samples.

3. The method of claim 1, wherein, The method of simulating stem cell hierarchical differentiation mechanism by using a biomimetic hierarchical architecture, obtaining an optimized genetic algorithm, and using the optimized genetic algorithm to optimize the initial weights and initial thresholds of the initial BP neural network constitutive model to obtain transition weights and transition thresholds comprises the following steps: encoding the initial weights and initial thresholds of the initial BP neural network constitutive model into real number vectors to form a real number coded chromosome population, and dividing the chromosome population into stem cell populations and differentiated cell populations; iteratively evolving the stem cell populations and the differentiated cell populations, and outputting an optimal chromosome when a first preset condition is met, and decoding the optimal chromosome to obtain transition weights and transition thresholds of the initial BP neural network constitutive model; the first preset condition is that a preset population iteration number or a preset fitness convergence threshold is reached.

4. The method of claim 3, wherein, The method of iteratively evolving the stem cell populations and the differentiated cell populations comprises the following steps: retaining the individual with the highest fitness in the stem cell populations by using a roulette selection method; when a preset number of generations is met, replacing the individual with the lowest fitness in the stem cell populations with the individual with the highest fitness in the differentiated cell populations; when the stem cell populations meet a preset reconstruction condition, eliminating half of the individuals in the stem cell populations, extracting an equal number of differentiated cells from the differentiated cell populations, and injecting the differentiated cells into the stem cell populations to reconstruct the population structure; the preset reconstruction condition is that the stem cell populations do not evolve within a preset iteration number.

5. The method of claim 4, wherein, The method of retaining the individual with the highest fitness in the stem cell populations by using a roulette selection method comprises the following steps: computing individual fitness of stem cell population, imposing epigenetic marker on high fitness individual in the stem cell population to make the high fitness individual be preferentially reserved in genetic operation; the formula of computing individual fitness is as follows: wherein, is an index number of an individual in the population, is a probability of the individual being selected, is a fitness value of the individual, is a population size, is a constant, is a prediction error.

6. The method of claim 3, wherein, after the division into stem cell population and differentiated cell population, further comprising: configuring genetic operator with low mutation rate and high crossover rate for the stem cell population; configuring genetic operator with high mutation rate and low crossover rate for the differentiated cell population.

7. The method of claim 1, wherein, the optimization of the transition weight and the transition threshold value through the adaptive gradient descent algorithm comprises: taking the transition weight and the transition threshold value as a starting point; taking a fixed number of the data samples for small batch training each time, comprising: calculating the loss between the stress true value and the stress prediction value according to the small batch training, calculating the gradient of the transition weight and the transition threshold value through back propagation, calculating the step size through adaptive algorithm, adjusting the transition weight and the transition threshold value in the negative gradient direction by using the gradient and the step size; repeating the taking of the fixed size of the data samples for small batch training, and the target weight and the target threshold value are obtained when the second preset condition is met; the second preset condition is that the gradient norm is lower than the preset threshold or the preset gradient descent iteration number is reached.

8. A system for high-robust prediction of material deformation behavior based on bionic intelligent constitutive, characterized in that, comprising: a data acquisition module, configured to acquire multi-modal data samples under multi-material and multi-scenario; a model optimization module, configured to construct an initial BP neural network constitutive model, take the multi-modal data samples as input and take stress as output, simulate stem cell hierarchical differentiation mechanism by using a bionic hierarchical architecture, obtain an optimized genetic algorithm, optimize initial weight and initial threshold value of the initial BP neural network constitutive model to obtain transition weight and transition threshold value, optimize the transition weight and the transition threshold value through a gradient descent algorithm to obtain an improved BP neural network constitutive model; a prediction module, configured to predict deformation behavior of a material to be predicted based on the improved BP neural network constitutive model. 9.An electronic device, comprising a memory and a processor, wherein, the memory is configured to store a program; the processor, coupled with the memory, is configured to execute the program stored in the memory to implement steps of the method for high-robustness prediction of material deformation behavior based on bionic intelligent constitutive model according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, a computer readable program or instruction for storing, which can implement steps of the method for high-robustness prediction of material deformation behavior based on bionic intelligent constitutive model according to any one of claims 1 to 7 when executed by a processor.

Images