Method and system for enhancing stress-strain data of metal material

Abstract

The invention belongs to the technical field of material science and artificial intelligence crossing, and particularly relates to a metal material stress-strain data enhancement method and system. The invention provides a metal material stress strain data enhancement method. The method comprises the following steps: 1) constructing a training data set; 2) constructing an initial model of the conditional generative adversarial network, which comprises a generator and a discriminator, and constructing a total loss function of the generator, which comprises a metal full-stage dynamic coupling physical constraint loss function; 3) training the initial model by using the training data set to obtain a training optimization model; and 4) expanding stress-strain data of the metal material by using the optimization model obtained in the step 3) to form an enhanced data set. According to the data enhancement method based on the multi-physical constraint condition generative adversarial network, the physical constraint adaptability of the output data is high, and the physical coincidence rate gt of the data is generated; and the method not only accords with the physical law, but also has the engineering practicability, and overcomes the defect of pseudo data generation in the traditional method.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of materials science and artificial intelligence, and specifically relates to a method and system for enhancing stress and strain data of metallic materials. Background Technology

[0002] Stress-strain data for metallic materials is crucial for engineering design, but obtaining it experimentally faces challenges such as high cost, small sample sizes, and scarcity of data under extreme conditions. A single experiment can cost over ten thousand yuan; extreme conditions (such as 1200℃ high temperature, 10 s...) are particularly challenging. -1 The scarcity of data at high strain rates leads to poor generalization ability of machine learning models.

[0003] Existing augmentation methods have significant drawbacks: traditional interpolation cannot expand the data distribution, statistical models are difficult to adapt to the nonlinearity of metal yield-hardening, and traditional GANs lack metal-specific physical constraints, easily generating pseudo-data such as "yield point shift" and "stress drop during hardening," which cannot meet engineering requirements. Traditional stress-strain data augmentation methods suffer from pseudo-data output and poor adaptability to physical constraints.

[0004] Therefore, there is an urgent need to integrate the mechanical properties of metals with deep learning enhancement methods to ensure that the generated data not only conforms to physical laws but also has engineering applicability. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for enhancing stress and strain data of metallic materials. This invention is based on a generative adversarial network with multiple physics constraints (CGAN) for enhancing stress and strain data of metallic materials. The generated data has a physical conformity rate of >95% and an engineering confidence level of Q≥0.95. It can effectively supplement metal data under extreme working conditions and is suitable for metallic material design, structural analysis and finite element simulation scenarios.

[0006] The first aspect of this invention provides a method for enhancing stress-strain data of metallic materials, comprising the following steps: 1) Obtain the original stress and strain data of the metal, and obtain the training dataset by preprocessing the original stress and strain data; 2) Construct an initial model of the conditional generative adversarial network, which includes a generator and a discriminator, and construct the generator total loss function in the initial model of the conditional generative adversarial network; The generator's total loss function includes the metal's full-stage dynamic coupling physical constraint loss function; 3) Train the initial model of the Conditional Generative Adversarial Network (CGN) using the training dataset, and obtain the optimized CGN model after training; 4) Use the conditional generative adversarial network optimization model obtained in step 3) to expand the stress-strain data of metallic materials to form an enhanced dataset.

[0007] In one embodiment, the metal full-stage dynamic coupling physical constraint loss function in step 2) is constructed based on multiple metal constraints of the metal full-stage dynamic coupling. The multiple metal constraints include monotonicity constraints, convexity constraints, energy conservation constraints, and yield point constraints.

[0008] The full metal stage includes all stages of the metal at different temperatures and strain rates.

[0009] The loss function of the dynamic coupling physical constraint of the metal throughout the entire stage is shown in Equation (1);

[0010] Formula (1), in, To constrain the basic weights; For phased dynamic weighting, For weight adjustment factor, It is the gradient smoothing factor. For the convexity weights of the elastic phase, For hardening recognition factor, This is the energy deviation tolerance coefficient. The average energy of the original metal data. In response, For stress, For the first j Stress at -1 data point For the first j Stress at each data point For the first j Strain at each data point For the first j Strain at -1 data point This represents the yield strain of the metal (the boundary strain between the elastic stage and the yield-hardening stage). The yield stress of the metal (corresponding to) (Stress).

[0011] In one embodiment, the generator's total loss function is as shown in formula (2): Formula (2), in, The total loss of the generator. This represents the loss function for the dynamic coupling physical constraints of the metal throughout all stages. For training cycles; And / or, the discriminator adversarial loss function is as shown in Equation (3): Equation (3), in, .

[0012] In one embodiment, an adaptive learning rate optimizer is used to train the model in step 3). After generating data by training the model in step 3), the engineering confidence score is used to determine whether the conditional generative adversarial network model can be trained to completion, thus obtaining the optimized conditional generative adversarial network model.

[0013] In one embodiment, the formula for calculating the engineering confidence level is as shown in formula (4): Formula (4) in, The compliance rate of the monotonicity constraint. The compliance rate of convexity constraints. The compliance rate of the energy conservation constraint. The compliance rate of the yield point constraint; In one embodiment, when the engineering confidence level is greater than or equal to 0.95, the conditional generative adversarial network model can be trained to completion, resulting in an optimized conditional generative adversarial network model.

[0014] In one embodiment, the preprocessing in step 1) includes deduplicating and sorting the original stress and strain data, then generating data points using cubic spline interpolation with an error correction factor, and integrating the generated data points with the original stress and strain data to obtain a training dataset. In one embodiment, the cubic spline interpolation formula is as shown in formula (5): Formula (5), in, The average strain interval of the original data. , , , These are the interpolation coefficients. x For the interpolated strain, x i For the original data points, strain, For the corresponding x The stress.

[0015] In one embodiment, step 1) further includes standardizing the training dataset, the standardization process being adapted to the properties of the metal; The data standardization formula is shown in equation (6).

[0016] in, The target metal's elastic modulus, To reference the elastic modulus of metals, For stress, The original data value. These are standardized data values.

[0017] A second aspect of the present invention also provides a stress-strain data enhancement system for metallic materials, the stress-strain data enhancement system for metallic materials comprising: The data acquisition module is used to acquire the original stress and strain data of the metal, as well as to acquire the training dataset; The data preprocessing module, connected to the data acquisition module, is used to preprocess the acquired raw stress and strain data; The model building module is used to build the initial model of the conditional generative adversarial network and embed the loss function, which facilitates the subsequent training and optimization of the generator and discriminator in the initial model of the conditional generative adversarial network. The model optimization module is used to train and optimize the initial model of the conditional generative adversarial network using the training dataset, so as to obtain the optimized model of the conditional generative adversarial network after training. The data augmentation module takes strain data as input and outputs an augmented dataset through a conditional generative adversarial network (GAN) optimization model. This achieves the goal of expanding the stress-strain data of metallic materials using a GAN optimization model.

[0018] In one embodiment, the metal material stress-strain data enhancement system further includes an evaluation module for calculating engineering confidence and determining whether the conditional generative adversarial network model can be trained to completion. And / or, the metal material stress-strain data enhancement system further includes a standardization module connected to the data preprocessing module, used to standardize the preprocessed data or training dataset, and to de-standardize the model output data.

[0019] A third aspect of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to perform the aforementioned method for enhancing stress-strain data of metallic materials.

[0020] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the above-described method for enhancing stress-strain data of metallic materials.

[0021] The fifth aspect of the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the above-described method for enhancing stress and strain data of metallic materials.

[0022] The method for enhancing stress-strain data of metallic materials provided by this invention has the following advantages: 1) The method for enhancing stress and strain data of metallic materials provided by the present invention is based on the generation of adversarial networks (CGAN) with multiple physical constraints. The output data has strong physical constraint adaptability, which not only conforms to physical laws (high physical conformity rate) but also has engineering applicability.

[0023] 2) The generator's total loss function incorporates a metal-wide dynamic coupling physical constraint loss function, enhancing the design of both full-stage dynamic coupling constraints and specific metal-specific constraints. The full-stage dynamic coupling constraints achieve precise constraints across the entire metal elastic-yield-hardening stage through stage-by-stage weights and dynamic adjustment factors, addressing the poor adaptability of traditional fixed constraints. Metal-specific constraints include Gaussian smoothing for monotonic constraints, hardening type adaptation for convex constraints, high strain correction for energy constraints, and matching degree weighting for yield constraints, all customized for metal mechanical properties and possessing strong physical interpretability.

[0024] 3) Use engineering confidence level for engineering quality assessment: Use Q value to evaluate the quality of generated data from multiple dimensions to ensure that the output data meets the design accuracy of metal structure and avoid the defect of "physically reasonable but engineering useless".

[0025] 4) This invention addresses the problems of insufficient experimental data for metallic materials (such as TA15 titanium alloy and aluminum alloy) and poor adaptability of physical constraints in traditional reinforcement methods. It innovatively designs a dynamic coupling physical constraint loss function for the entire metal process, integrates interpolation preprocessing with error correction, a metal-specific CGAN model and a phased weighting strategy to ensure that the generated data conforms to the mechanical properties of the metal throughout the elastic-yield-hardening process.

[0026] 5) This invention solves the defects of traditional methods in generating pseudo data. The generated data has a physical conformity rate of >95% and an engineering confidence level Q≥0.95, which can effectively supplement metal data under extreme working conditions and is suitable for metal material design, structural analysis and finite element simulation scenarios.

[0027] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0028] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0029] Figure 1 This is a flowchart illustrating the method for enhancing stress and strain data of metallic materials in an embodiment of the present invention. Detailed Implementation Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0030] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0031] In the description of the embodiments of this invention, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this invention, "multiple" means two or more, unless otherwise explicitly defined.

[0032] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0033] In the description of the embodiments of this invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0034] In the description of the embodiments of the present invention, the term "multiple" refers to two or more (including two), similarly, "multiple groups" refers to two or more (including two groups), and "multiple pieces" refers to two or more (including two pieces).

[0035] In the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention according to the specific circumstances.

[0036] The following explanations of some terms used in this invention are provided to enable those skilled in the art to understand them.

[0037] This invention addresses the problem of insufficient experimental data for metallic materials (such as TA15 titanium alloy and aluminum alloy), and the shortcomings of traditional stress-strain data augmentation methods, which suffer from spurious output data and poor adaptability to physical constraints. The first aspect of this invention provides a method for enhancing stress-strain data in metallic materials, combining... Figure 1 The method involves the following steps: 1) Obtaining the original stress and strain data of the metal, and obtaining a training dataset by preprocessing the original stress and strain data; 2) Constructing an initial model of a conditional generative adversarial network (GAN), which includes a generator and a discriminator, and constructing the generator's total loss function in the initial model; the generator's total loss function includes the metal's full-stage dynamic coupling physical constraint loss function; 3) Training the initial model of the GAN using the training dataset to obtain the optimized model of the GAN after training; 4) Expanding the stress and strain data of the metal material using the optimized model of the GAN obtained in step 3).

[0038] This invention proposes a method for enhancing stress-strain data of metallic materials based on Generative Adversarial Networks (CGANs) with Multi-Physical Constraints. Addressing the issues of insufficient experimental data for metallic materials (such as TA15 titanium alloy and aluminum alloy) and poor physical constraint adaptability of traditional enhancement methods, this invention innovatively incorporates a dynamic coupling loss function for all stages of the metal process. This solves the shortcomings of traditional methods, such as generating pseudo-data and poor physical constraint adaptability. The generated data achieves a physical compliance rate >98% and an engineering confidence level Q ≥ 0.95, effectively supplementing metallic data under extreme conditions. It is suitable for metallic material design, structural analysis, and finite element simulation scenarios.

[0039] In this embodiment of the invention, the metal can be a titanium alloy or an aluminum alloy, and the stress-strain data enhancement method of this invention can be applied to all types of metal materials.

[0040] Combination Figure 1 The method for enhancing stress-strain data of metallic materials in this invention is described in detail below.

[0041] Obtain the training dataset (a) Data Acquisition and Preprocessing In this embodiment of the invention, the original stress of the metal is obtained ( ),strain( The original stress and strain data are loaded, deduplicated, and sorted in ascending order of strain. Then, high-density data points (interpolation points) are generated using cubic spline interpolation.

[0042] Specifically, high-density data points are generated using cubic spline interpolation with an error correction factor, where the data points are located near the yield point ( Enhanced interpolation weights (to adapt to metal data fluctuations), high strain regions ( Reduce the weights (to avoid over-extrapolation) to ensure the interpolation curve conforms to the elastic-yield-hardening trend of metals; the interpolation formula is:

[0043] in, The average strain interval of the original data. , , , These are the interpolation coefficients. For interpolated strain, For the corresponding x stress, Strain for the original data points; It is worth mentioning that, through data fusion confidence level Assess reliability ( The original number of data points. (Standard deviation of original stress) The interpolation parameters are readjusted periodically to ensure the quality of the input data.

[0044] In some embodiments, the generated high-density data points are 3 to 5 times the original data.

[0045] In this embodiment of the invention, the original data points and interpolation points are merged, and duplicates are removed again to generate the training dataset.

[0046] In this embodiment of the invention, the generated training dataset is standardized, and a metal stiffness correction coefficient is introduced during the standardization process. The standardization range is adjusted according to the differences in metal stiffness to make TA15 titanium alloy ( GPa), aluminum alloy ( Data on materials with different stiffnesses (GPa, etc.) are in similar numerical ranges to avoid the model's over-reliance on high-stiffness material data and improve generalization ability; For example, using metal property adaptation standardization: ( The target metal's elastic modulus, To reference the elastic modulus of metals, The mean strain, (Standard deviation of strain).

[0047] In this embodiment of the invention, the standardized parameters are retained for destandardization.

[0048] Specifically, the anti-standardization formula It accurately restores the physical dimensions of MPa (stress) and dimensionless (strain), ensuring that the generated data can be directly used for engineering calculations.

[0049] Construct an initial model of a conditional generative adversarial network (CGAN). In this embodiment of the invention, the CGAN model is defined as follows: 1) Generator

[0050] The core function of the generator is to learn the physical mapping relationship between strain and stress, and guide the direction of data generation through conditional input to avoid the randomness of unconstrained GANs.

[0051] In this embodiment of the invention, the generator Input random noise (e.g., dimension 100) and conditional strain Output virtual stress The network structure is as follows: Linear(101,128)→LeakyReLU(0.2)→Linear(128,256)→BatchNorm1d(256)→LeakyReLU(0.2)→Linear(256,1)→Tanh() In some embodiments, a multilayer perceptron (MLP) architecture is employed, with the input being "100-dimensional random noise". +1-dimensional conditional strain Virtual stress is generated through nonlinear transformation. .

[0052] Network layer design adapts to the nonlinear characteristics of materials: Hidden layers introduce nonlinear correction terms. This enhances the ability to fit nonlinear characteristics such as metal yield plateau and work hardening, and avoids generating linearized pseudo-data.

[0053] 2) Discriminator

[0054] In this embodiment of the invention, the discriminator Introducing metal physical feature enhancement terms into the input layer (Reflects the characteristics of metal stiffness variation), improves the ability to identify metal pseudo data (such as yield point shift, stress drop during hardening stage), and ensures that the generated data conforms to the essence of metal mechanics.

[0055] In this embodiment of the invention, the discriminator Input stress and conditional strain, output true / false probabilities, network structure is as follows: Linear(3,256)→LeakyReLU(0.2)→Linear(256,128)→LeakyReLU(0.2)→Linear(128,1)→Sigmoid().

[0056] In some embodiments, the MLP architecture is also used, with the input being "1-dimensional stress". +1-dimensional conditional strain The output data represents the probability (0~1) that the data is the real data.

[0057] Construction of dynamic coupling physical constraint loss function for metal throughout all stages In this embodiment of the invention, a triple mechanism of "phased dynamic weighting + multi-physical constraint fusion + adaptive adjustment" is used to solve the problem that traditional fixed constraints cannot adapt to the multi-stage characteristics of metals, and to achieve precise constraints on the physical characteristics of metals in the full deformation stage.

[0058] In this embodiment of the invention, the loss function of dynamic coupling physical constraint of the metal in all stages is constructed based on multiple metal constraints of the metal in all stages; wherein the multiple metal constraints include monotonicity constraint, convexity constraint, energy conservation constraint, and yield point constraint; the metal in all stages includes elastic stage, yield-hardening stage and recrystallization stage.

[0059] In this embodiment of the invention, the function uses "basic weights" +Phase-based weighting +Dynamic adjustment factor Using this framework, four types of metal-specific constraints—monotonicity, convexity, energy conservation, and yield point—are integrated, and mathematical expectations are applied. The statistical constraints for batch data are established, and the specific formula for the loss function of dynamic coupling physical constraints for the entire metal process is as follows:

[0060] In the formula: To constrain the basic weights; For phased dynamic weighting, For weight adjustment factor, It is the gradient smoothing factor. For the convexity weights of the elastic phase, For hardening recognition factor, This is the energy deviation tolerance coefficient. The average energy of the original metal data. In response, For stress, For the first j Stress at -1 data point For the first j Stress at each data point For the first j Strain at each data point For the first j Strain at -1 data point This represents the yield strain of the metal (the boundary strain between the elastic stage and the yield-hardening stage). The yield stress of the metal (corresponding to) (Stress).

[0061] In some optional embodiments, the basic weights are constrained: , , , .

[0062] Phased dynamic weights Based on the differences in the physical mechanisms of metal deformation stages, dynamically allocate the weights of each constraint: elastic stage ( Elastic stage: The core constraints are monotonicity weight of 0.4 and convexity weight of 0.4. The remaining weights are allocated by energy weight and yield point weight, with energy weight of 0.2 and yield point weight of 0.0 (yield point is not activated), and the total weight is 1; ensuring that stress-strain is strictly linear and conforms to the elastic deformation law of metals; For yield strain. Yield-hardening stage ( The core constraint is that the yield point weight is 0.3, and the remaining weights are allocated by the monotonicity weight, convexity weight, and energy weight. The monotonicity weight is 0.3, the convexity weight is 0.3, and the energy weight is 0.1, with a total weight of 1. This matches the core characteristic of metals "explicit yielding" and avoids generating pseudo data with no yield plateau. The initial strain for recrystallization. Recrystallization stage ( The core constraint is the energy constraint (energy weight 0.4). The remaining weights are allocated by the monotonicity weight, convexity weight, and energy weight. The monotonicity weight is 0.3, the convexity weight is 0.2, and the yield point weight is 0.1. The total weight is 1, which meets the engineering requirements for energy absorption under high strain in metals and avoids generating non-physical data with negative energy growth. For maximum strain.

[0063] Dynamic adjustment factor :pass Achieve adaptive optimization of constraint weights: When the compliance rate of a certain constraint is lower than the target (e.g., the yield point compliance rate is only 90%, while the target is 98%), positive ( The constraint weight is automatically increased by 8%, ensuring that the model prioritizes the optimization of weak constraints and avoids the loss of physical properties caused by a single constraint dominating. To constrain the target compliance rate, This represents the current compliance rate.

[0064] Metal adapter design with four types of constraints: Monotonicity constraint ( ): By Gaussian filtering ( The first derivative is smoothed to suppress experimental noise interference near the metal yield point; a 5% tolerance (+0.05) is introduced to adapt to the small stress fluctuations caused by dislocation movement during the metal yielding stage and avoid excessive constraint. This is a gradient smoothing factor, used in some optional embodiments. This is the gradient smoothing factor.

[0065] Convexity constraint ( ):pass Dynamically setting the second derivative threshold for strongly hardened metals ( Higher convexity is required. ), weakly hardened metals ( No forced convexity ( When metal hour, .002. Precisely matches the work hardening characteristics of different metals; the elastic stage is achieved through... Constrain linear relationships (second derivative close to 0) to avoid generating nonlinear elastic data. For the convexity weights of the elastic stage. pass Calculations are used to distinguish between metal hardening ( ), medium hardening ( ), weak hardening ( Its characteristics allow it to adapt to the hardening constraint requirements of different metals.

[0066] Energy conservation constraint ( ): Introducing a high strain correction factor ( To improve the accuracy of energy calculation (error <5%), the integral weight is increased for sparse data points with high strain in metals. This is the energy deviation tolerance coefficient. The average energy of the raw metal data is obtained by... Allowing for small energy deviations avoids model oscillations due to calculation errors, which meets the actual needs of engineering.

[0067] Yield point constraint ( ): By yield matching degree The weighted squared error is calculated as follows: when the yield point deviation is <5%, the matching degree is ≈0.95 and the penalty term is relatively small; when the deviation is >10%, the matching degree is ≈0.9 and the penalty term is automatically increased by 2 times to ensure that the yield characteristics of the generated data are consistent with those of real metal, which can be used for strength verification.

[0068] In the formula, the mathematical expectation The average value of the batch data is taken to make the loss statistically significant, avoid the interference of noise from a single sample, and ensure the stability of the constraint. , Automatic differentiation calculation using PyTorch reduces gradient noise by ≥60% after Gaussian smoothing, avoiding data fluctuations near the metal yield point from interfering with constraint determination.

[0069] This invention improves the accuracy of output data by incorporating a dynamic coupling physical constraint loss function for the entire metal process, integrating interpolation preprocessing with error correction, a metal-specific CGAN model, and a phased weighting strategy, thereby ensuring that the generated data conforms to the mechanical properties of the metal throughout the elastic-yield-hardening process.

[0070] Model training optimization and data generation Adversarial loss optimization: In this embodiment of the invention, adversarial loss optimization during model training includes: introducing a gradient penalty term. Limiting the discriminator gradient norm avoids model collapse due to fluctuations in metal experimental data (such as stress discrepancies), ensuring stable learning of metal data distribution.

[0071] Specifically, the discriminator adversarial loss function formula is as follows:

[0072] in, .

[0073] Generator Total Loss Balancing: In this embodiment of the invention, the generator total loss function during model training is:

[0074] in, The total loss of the generator. This represents the loss function for the dynamic coupling physical constraints of the metal throughout all stages. For training cycles; Physical constraint weights In the early stages of training (epoch < 500), the epoch value is approximately 1, with strong constraints ensuring physical plausibility; in the later stages (epoch > 1000), the epoch value is approximately 0.5, balancing adversarial learning and physical constraints, and increasing data diversity (e.g., supplementing with 1200℃ high temperature, 1 second...). -1 Data on extreme working conditions such as high strain rates.

[0075] Model optimization and quality evaluation: The model is trained using an adaptive learning rate optimizer. ( When the loss decreases slowly, the learning rate is automatically reduced to accelerate model convergence; after data generation, the learning rate is adjusted using engineering confidence scores. Multi-dimensional assessment The condition is judged as qualified at the specified time to ensure that the output data meets the accuracy requirements of metal structure design, and the trained conditional generative adversarial network optimization model is obtained. The adaptive learning rate optimizer can be any one of the following: AdaGrad optimizer, RMSprop optimizer, Adam optimizer, and AdaDelta optimizer.

[0076] Engineering confidence level The standard for calculating the constraint compliance rate is as follows: for The percentage of data points For the elastic stage And hardening stage The percentage of data points for The percentage of data points for The percentage of data points.

[0077] in, The first derivative of stress with respect to strain. The second derivative of stress with respect to strain. This represents the integrated energy of the stress-strain curve.

[0078] In this embodiment of the invention, the process of generating the target dataset includes: inputting the target strain, such as an equally spaced strain sequence (0~1.0, interval 0.001) and random noise into the obtained conditional generative adversarial network optimization model, generating virtual stress data, restoring the physical dimensions through denormalization, saving it as an Excel file, and obtaining the target dataset.

[0079] In some embodiments, the generated data strain range is extended to 1.3 times that of the original data to supplement extreme condition data.

[0080] A second aspect of this invention provides a system for enhancing stress and strain data of metallic materials, comprising: a data acquisition module for acquiring raw stress and strain data of the metal and acquiring a training dataset; a data preprocessing module connected to the data acquisition module for preprocessing the acquired raw stress and strain data; a model building module for constructing an initial model of a conditional generative adversarial network (GAN) and embedding a loss function to facilitate subsequent training and optimization of the generator and discriminator in the initial GAN ​​model; a model optimization module for training and optimizing the initial GAN ​​model using the training dataset to obtain a trained optimized GAN model; and a data enhancement module for inputting strain data of the points to be predicted and outputting enhanced strain data through the optimized GAN model, thereby achieving the purpose of expanding the stress and strain data of metallic materials using the optimized GAN model.

[0081] In some embodiments, the metal material stress-strain data augmentation system further includes an evaluation module for calculating engineering confidence and determining whether the conditional generative adversarial network model can be trained to completion.

[0082] In some embodiments, the metal material stress-strain data augmentation system further includes a standardization module connected to the data preprocessing module, used to standardize the preprocessed data or training dataset, and to de-standardize the model output data.

[0083] Unless otherwise defined, the technical terms used in the following embodiments have the same meaning as commonly understood by those skilled in the art. Unless otherwise specified, the experimental reagents used in the following embodiments are conventional biochemical reagents; the raw materials, instruments, and equipment used in the following embodiments can all be obtained commercially or through existing methods; unless otherwise specified, the amounts of experimental reagents used are the amounts used in conventional experimental operations; unless otherwise specified, the experimental methods are conventional methods. It should be further noted that the following descriptions are merely exemplary and not intended to limit the specific scope of the invention. Moreover, the comparative examples below are selected to compare with the technical solutions of the present invention to demonstrate the advancement of the present invention, and do not necessarily represent prior art in this technical field.

[0084] Example 1 Using TA15 titanium alloy (800℃~1200℃, 0.001 s) -1 ~1 s -1 Taking stress and strain data enhancement as an example, the stress and strain data enhancement method based on conditional generative adversarial networks specifically includes the following steps: 1) Raw data processing: TA15 titanium alloy was loaded at 1000℃ for 0.1 s.-1 The original data under the working condition (100 points, strain 0~1.0, stress 75 MPa~120 MPa) was deduplicated, retaining 95 points. Cubic spline interpolation with error correction was then performed to generate 300 points. After merging and deduplication, 378 training samples were obtained. The data fusion confidence score was... (Qualified). Data standardization processing, standardization. It is compatible with titanium alloy steel.

[0085] 2) Model initialization: The generator has an input dimension of 101 (100+1) and an output dimension of 1; the discriminator has an input dimension of 2 (1+1) and an output dimension of 1; the optimizer is Adam.

[0086] 3) Loss function settings: , , , , , , GPa, MPa; 4) Model training: Performance of generated data after 5000 epochs of iterative training: Physical compliance rate: (Monotonicity) (convexity) (energy), (Yield point); Engineering confidence level (qualified); Compared with real data: correlation coefficient 0.94, Wasserstein distance 0.12, yield point deviation <3%, fully meeting the design requirements of aerospace titanium alloy components.

[0087] 5) Input the equally spaced strain sequences (0~1.0, interval 0.001) and random noise (random noise dimension is 100) into the optimal model obtained after training to obtain the target dataset generated by the optimal model.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for enhancing stress-strain data of metallic materials, characterized in that, Includes the following steps: 1) Obtain the original stress and strain data of the metal, and obtain the training dataset by preprocessing the original stress and strain data; 2) Construct an initial model of the conditional generative adversarial network, which includes a generator and a discriminator, and construct the generator total loss function in the initial model of the conditional generative adversarial network; The generator's total loss function includes the metal's full-stage dynamic coupling physical constraint loss function; 3) Train the initial model of the Conditional Generative Adversarial Network (CGN) using the training dataset, and obtain the optimized CGN model after training; 4) Use the conditional generative adversarial network optimization model obtained in step 3) to expand the stress-strain data of metallic materials to form an enhanced dataset.

2. The method for enhancing stress-strain data of metallic materials according to claim 1, characterized in that, The dynamic coupling physical constraint loss function for the entire metal stage described in step 2) is constructed based on multiple metal constraints throughout the entire metal stage; The multiple metal constraints include monotonicity constraints, convexity constraints, energy conservation constraints, and yield point constraints.

3. The method for enhancing stress-strain data of metallic materials according to claim 1 or 2, characterized in that, The loss function of the dynamic coupling physical constraint of the metal throughout the entire stage is shown in Equation (1). Formula (1), in, To constrain the basic weights; For phased dynamic weighting, For weight adjustment factors, It is the gradient smoothing factor. For the convexity weights of the elastic phase, For hardening recognition factor, This is the energy deviation tolerance coefficient. The average energy of the original metal data. In response, For stress, For the first Stress at each data point For the first Stress at each data point For the first Strain at each data point For the first Strain at each data point For the yield strain of the metal, This represents the yield stress of the metal.

4. The method for enhancing stress-strain data of metallic materials according to any one of claims 1-3, characterized in that, The generator's total loss function is shown in formula (2): Official (2), in, The total loss of the generator. This represents the loss function for the dynamic coupling physical constraints of the metal throughout all stages. For training cycles; And / or, the discriminator adversarial loss function is as shown in Equation (3): Official (3), in, .

5. The method for enhancing stress-strain data of metallic materials according to any one of claims 1-3, characterized in that, In step 3), an adaptive learning rate optimizer is used to train the model; After generating data by training the model in step 3), the engineering confidence score is used to determine whether the conditional generative adversarial network model can be trained to completion, thus obtaining the optimized conditional generative adversarial network model.

6. The method for enhancing stress-strain data of metallic materials according to claim 5, characterized in that, The formula for calculating the confidence level of the project is shown in formula (4): Official (4), in, The compliance rate of the monotonicity constraint. The compliance rate of convexity constraints. The compliance rate of the energy conservation constraint. The compliance rate of the yield point constraint; Preferably, when the engineering confidence level is greater than or equal to 0.95, the conditional generative adversarial network model can be trained to obtain an optimized conditional generative adversarial network model.

7. The method for enhancing stress-strain data of metallic materials according to any one of claims 1-6, characterized in that, The preprocessing described in step 1) includes deduplicating and sorting the original stress and strain data, then using cubic spline interpolation with error correction factor to generate data points, and integrating the generated data points with the original stress and strain data to obtain the training dataset. Optionally, the cubic spline interpolation formula is shown in formula (5). Official (5), in, The average strain interval of the original data. , , , These are the interpolation coefficients. x For the interpolated strain, x i For the original data points, strain, For the corresponding x The stress.

8. The method for enhancing stress-strain data of metallic materials according to any one of claims 1-7, characterized in that, Step 1) also includes standardizing the training dataset, the standardization process being adapted to the properties of the metal; The data standardization formula is shown in equation (6). in, The target metal's elastic modulus, For reference, the elastic modulus of metals, For stress, The original data value. These are standardized data values.

9. A stress-strain data enhancement system for metallic materials, characterized in that, The stress-strain data enhancement system for metallic materials includes: The data acquisition module is used to acquire the original stress and strain data of the metal, as well as to acquire the training dataset; The data preprocessing module, connected to the data acquisition module, is used to preprocess the acquired raw stress and strain data; The model building module is used to build the initial model of the conditional generative adversarial network and embed the loss function, which facilitates the subsequent training and optimization of the generator and discriminator in the initial model of the conditional generative adversarial network. The model optimization module is used to train and optimize the initial model of the conditional generative adversarial network using the training dataset, so as to obtain the optimized model of the conditional generative adversarial network after training. The data augmentation module is used to take input strain data and output augmented datasets through a conditional generative adversarial network optimization model.

10. The stress-strain data enhancement system for metallic materials according to claim 9, characterized in that, The metal material stress-strain data enhancement system also includes an evaluation module for calculating engineering confidence and determining whether the conditional generative adversarial network model can be trained to completion. And / or, the metal material stress-strain data enhancement system further includes a standardization module connected to the data preprocessing module, used to standardize the preprocessed data or training dataset, and to de-standardize the model output data.

Images