Data-driven material performance cross-scale prediction method and system fusing grain deformation and dislocation mechanisms

Abstract

This invention discloses a data-driven method and system for cross-scale prediction of material properties, integrating grain deformation and dislocation mechanisms, belonging to the field of materials computational science and performance prediction. The method first acquires microstructural characterization data of materials and extracts micro-deformation parameters and fine-grain characteristics. Then, a representative volumetric element model is constructed, and through finite element simulation embedding a crystal plastic constitutive model, stress-strain data and macroscopic mechanical responses of each grain under different strains are obtained. Next, a cross-scale dataset is constructed by correlating macro- and micro-data, and a machine learning model is trained using this dataset to establish a quantitative mapping relationship from microstructure characteristics to mechanical properties. Finally, the trained model is used to achieve second-level prediction of new material properties, and reverse design of the microstructure is performed using model interpretability and optimization algorithms. This invention integrates the mechanistic nature of physical simulation with the efficiency of machine learning, achieving rapid and reliable cross-scale prediction and design of material properties.

Description

Technical Field

[0001] This invention relates to the field of materials computational science and performance prediction technology, and in particular to a data-driven method and system for cross-scale prediction of materials properties that integrates grain deformation and dislocation mechanisms, for realizing rapid cross-scale prediction and reverse design of metallic materials from microstructure to macroscopic mechanical properties. Background Technology

[0002] The macroscopic mechanical properties of metallic materials, such as strength, plasticity, and toughness, fundamentally depend on their microscopic structure, including grain size, shape, orientation (texture), grain boundary characteristics, and the density and distribution of defects such as dislocations. Accurately understanding and predicting the quantitative relationship between microstructure evolution and macroscopic properties is crucial for new material development and component forming process optimization in fields such as aerospace, automotive manufacturing, and energy equipment.

[0003] The theory of crystal plasticity and its numerical implementation method—the Crystal Plasticity Finite Element Method (CPFEM)—provides a powerful physical framework for connecting microstructure and macroscopic response. CPFEM can simulate the deformation process of polycrystalline aggregates under external loads based on the crystallographic information (such as slip systems) and deformation mechanisms (such as dislocation slip) of materials, reproducing complex phenomena such as anisotropy and texture evolution. However, CPFEM faces two major bottlenecks in practical applications: First, its computational cost is extremely high. Simulating the response of a representative volumetric unit containing thousands of grains under complex loading may take days or even weeks, which seriously hinders its application in high-throughput material screening or process parameter optimization. Second, traditional CPFEM research focuses on mechanism revelation and phenomenon reproduction. Although it can qualitatively correlate certain microscopic features (such as strong texture) with macroscopic behaviors (such as anisotropy), it is difficult to systematically extract, quantify, and correlate a large number of micro- and mesoscopic parameters (such as local stress concentration, dislocation density distribution, and grain boundary compatibility of each grain). Therefore, it is impossible to construct an efficient and reliable quantitative prediction model that directly translates microstructure input to macroscopic performance output.

[0004] In recent years, data-driven machine learning methods have brought a new paradigm to the prediction of material properties. By training models to learn complex patterns in historical data, rapid performance prediction can be achieved. However, purely "black box" data-driven models often suffer from poor interpretability, weak extrapolation ability, and strong dependence on the quality and quantity of training data due to the lack of physical constraints. The physical rationality of their predictions and their reliability under unknown operating conditions are often questioned.

[0005] Therefore, how to deeply integrate the physical mechanism advantages of CPFEM with the high-efficiency computational advantages of machine learning to build a cross-scale prediction tool that is both physically reliable and fast and efficient, and further realize intelligent design of micro-organisms based on performance objectives, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to overcome the above-mentioned defects of the prior art and provide a data-driven method and system for cross-scale prediction of material properties that integrates grain deformation and dislocation mechanism. This method establishes a fast and quantitative mapping relationship from microstructure characteristics to macro / micro mechanical properties by coupling high-fidelity crystal plasticity finite element simulation and machine learning, and supports reverse design of microstructure.

[0007] The present invention adopts the following technical solution: On the one hand, this invention provides a data-driven method for cross-scale prediction of material properties that integrates grain deformation and dislocation mechanisms, comprising the following steps: S1: Obtain EBSD images of the target material and obtain EBSD microstructure characterization data; S2: Based on the EBSD microstructure characterization data, extract a first type of microscale features and a second type of microscale features; the first type of microscale features includes crystal orientation and phase distribution information; the second type of microscale features includes at least one of Schmidt factor SF, Taylor factor TF, grain size, and texture. S3: Based on the microstructure characterization data of the EBSD, a representative volume element RVE model containing the real grain structure is constructed using the first type of microscale features; the crystal plastic constitutive model is assigned to the RVE model, and finite element calculation is performed to obtain macroscopic mechanical response and grain-scale stress-strain distribution data; S4: Connect the stress-strain data and macroscopic mechanical response of each grain with its corresponding first-type and second-type microscale features to construct a cross-scale associated dataset; S5: Based on the cross-scale associated dataset, at least one feature is selected from the first type of microscale features and the second type of microscale features as input parameters to train a machine learning model and establish a mapping relationship from microscale features to material mechanical properties. S6: Using the trained machine learning model, predict the macroscopic mechanical properties and / or microscopic field distribution of the new material sample based on the microstructure characterization data.

[0008] In addition to any of the possible implementations described above, another implementation is provided in which, in step S1, the microstructure characterization data is the original EBSD data obtained through experimental measurement, or the simulation data generated through crystal plasticity finite element simulation.

[0009] In addition to any of the possible implementations described above, a further implementation is provided in which, in step S2, the first type of microscale feature further includes grain geometric boundary information; and the second type of microscale feature further includes at least one of dislocation density and grain boundary type.

[0010] In addition to any of the possible implementations described above, another implementation is provided, wherein step S3 specifically includes: S31: Based on the first type of microscale features, perform mesh generation to generate a finite element mesh model containing the real grain morphology; S32: Import the finite element mesh model into the finite element software and embed a crystal plastic constitutive model containing the dislocation density evolution mechanism into it; S33: Assign initial crystal orientation and material parameters to each grain in the model, apply boundary conditions and loads for simulation calculation; S34: Extract stress-strain data of each grain under different macroscopic strains and the overall macroscopic stress-strain response from the simulation results.

[0011] In addition to any of the possible implementations described above, another implementation is provided in which, in step S32, the crystal plastic constitutive model is embedded in the finite element solver through a user material subroutine.

[0012] In addition to any of the possible implementations described above, another implementation is provided in which, in step S5, the machine learning model is a deep neural network, a recurrent neural network, or a gradient boosting decision tree model; before training, the features in the cross-scale associated dataset are standardized or normalized.

[0013] In addition to any of the possible implementations described above, another implementation is provided in which the following is included after step S6: S7: Utilize model interpretability techniques to analyze key first-type and / or second-type microscale features that affect material properties; S8: Based on the analysis results, combined with optimization algorithms or generative models, reverse design of micro-organization is carried out with target performance as a constraint.

[0014] In addition to any of the possible implementations described above, another implementation is provided in which, in step S8, the optimization algorithm is Bayesian optimization and the generative model is a generative adversarial network.

[0015] On the other hand, the present invention also provides a data-driven cross-scale prediction system for material properties that integrates grain deformation and dislocation mechanisms. The system is used to implement the above-mentioned method and includes: The data acquisition and processing module is used to acquire microstructure characterization data of materials and extract first-type and second-type microscale features; The physical simulation calculation module is used to construct and run finite element simulations based on the first type of microscale features, embedding a crystal plastic constitutive model, to generate stress-strain data and macroscopic mechanical response for each grain; A cross-scale dataset construction module is used to associate the stress-strain data, macroscopic mechanical response, and corresponding first-type and second-type microscale features to form a structured dataset. The machine learning modeling and training module is used to select at least one feature from the first type of microscale features and the second type of microscale features as input parameters based on the structured dataset, construct and train a machine learning model to learn the mapping relationship from microscale features to material mechanical properties. The performance prediction and application module is used to load pre-trained machine learning models, quickly predict material properties based on newly input microstructure characterization data, and support model-interpretable analysis and reverse design.

[0016] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0017] The beneficial effects of this invention are as follows: 1. Deep integration of physical mechanisms and data-driven approaches: Using CPFEM simulation or experimental data as a high-quality, physically reliable data source ensures that the laws learned by the machine learning model have a solid physical basis, significantly improving the model's interpretability and extrapolation prediction capabilities.

[0018] 2. Achieve high-performance prediction in seconds: Once the machine learning model is trained offline, for new microstructures, there is no need to perform time-consuming CPFEM calculations. Key performance indicators such as macroscopic stress-strain curves, yield strength, and uniform elongation can be obtained in seconds, as well as grain-scale stress / strain field distribution predictions, improving efficiency by several orders of magnitude.

[0019] 3. Constructing a complete cross-scale quantitative correlation: The parameters of microscopic and macroscopic mechanical responses are systematically extracted and correlated, and a structured dataset that can be used for machine learning is constructed, breaking through the bottleneck of traditional methods in cross-scale quantitative correlation.

[0020] 4. Supports intelligent microstructure design: Combining model interpretability technology and optimization algorithms, it can identify the key microstructure features that have the greatest impact on target performance, and reverse search the optimal microstructure morphology and texture that meet specific performance requirements (such as high strength and high plasticity), providing direct guidance for material design and process optimization.

[0021] 5. High versatility and flexibility: This method framework is applicable to different metal material systems (such as aluminum alloys, steel, and titanium alloys) and different deformation conditions. It can be adapted by changing the constitutive model parameters and adjusting the feature set, and has good universality.

[0022] 6. Taking BCC / FCC metal as an example, the yield strength, tensile strength and elongation of the section AA test are 1101MPa, 1528MPa and 27% respectively, while the yield strength, tensile strength and elongation predicted by the method of this invention are 1152MPa, 1504MPa and 29.5% respectively. The prediction errors of yield strength, tensile strength and elongation are 4.63%, 1.57% and 9.26% respectively, which shows that the prediction has a high accuracy. Attached Figure Description

[0023] Figure 1 The flowchart illustrates the overall process of the cross-scale prediction and optimization method for material properties provided in this embodiment of the invention.

[0024] Figure 2 This is a schematic diagram of a two-dimensional representative volume element (RVE) finite element mesh model constructed based on real EBSD data in an embodiment of the present invention.

[0025] Figure 3 The image shows the Mises stress distribution cloud map under different strains obtained by CPFEM simulation in an embodiment of the present invention.

[0026] Figure 4 This is a schematic diagram comparing the prediction results of the macroscopic stress-strain curves of the test set samples by the trained machine learning model in an embodiment of the present invention with the CPFEM simulation results. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0028] The accompanying drawings illustrate a layer structure according to an embodiment of the present invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0029] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0030] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0031] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0032] like Figure 1 As shown in the figure, an embodiment of the present invention provides a data-driven method for cross-scale prediction of material properties that integrates grain deformation and dislocation mechanisms, comprising the following steps: S1: Obtain EBSD images of the target material and obtain EBSD microstructure characterization data; S2: Based on the EBSD microstructure characterization data, extract a first type of microscale features and a second type of microscale features; the first type of microscale features includes crystal orientation and phase distribution information; the second type of microscale features includes at least one of Schmidt factor SF, Taylor factor TF, grain size, and texture. S3: Based on the microstructure characterization data of the EBSD, a representative volume element RVE model containing the real grain structure is constructed using the first type of microscale features; the crystal plastic constitutive model is assigned to the RVE model, and finite element calculation is performed to obtain macroscopic mechanical response and grain-scale stress-strain distribution data; S4: Connect the stress-strain data and macroscopic mechanical response of each grain with its corresponding first-type and second-type microscale features to construct a cross-scale associated dataset; S5: Based on the cross-scale associated dataset, at least one feature is selected from the first type of microscale features and the second type of microscale features as input parameters to train a machine learning model and establish a mapping relationship from microscale features to material mechanical properties. S6: Using the trained machine learning model, predict the macroscopic mechanical properties and / or microscopic field distribution of the new material sample based on the microstructure characterization data.

[0033] It should be noted that the modeling material and the prediction material should generally have the same composition, process, heat treatment, and similar crystal structure and phase composition.

[0034] In one specific embodiment, in step S1, the EBSD microstructure characterization data is the original EBSD data obtained through experimental measurement, or the simulated data generated through crystal plasticity finite element simulation.

[0035] In one specific embodiment, in step S2, the first type of microscale feature further includes grain geometric boundary information; the second type of microscale feature further includes at least one of dislocation density and grain boundary type.

[0036] In one specific embodiment, step S3 specifically includes: S31: Based on the first type of microscale features, perform mesh generation to generate a finite element mesh model containing the real grain morphology; S32: Import the finite element mesh model into the finite element software and embed a crystal plastic constitutive model containing the dislocation density evolution mechanism into it; S33: Assign initial crystal orientation and material parameters to each grain in the model, apply boundary conditions and loads for simulation calculation; S34: Extract stress-strain data of each grain under different macroscopic strains and the overall macroscopic stress-strain response from the simulation results.

[0037] In one specific embodiment, in step S32, the crystal plastic constitutive model is embedded in the finite element solver through a user material subroutine.

[0038] In one specific embodiment, in step S5, the machine learning model is a deep neural network, a recurrent neural network, or a gradient boosting decision tree model; before training, the features in the cross-scale associated dataset are standardized or normalized.

[0039] In one specific embodiment, step S6 is followed by: S7: Utilize model interpretability techniques to analyze key first-type and / or second-type microscale features that affect material properties; S8: Based on the analysis results, combined with optimization algorithms or generative models, reverse design of micro-organization is carried out with target performance as a constraint.

[0040] In one specific embodiment, in step S8, the optimization algorithm is Bayesian optimization, and the generative model is a generative adversarial network.

[0041] This invention provides a data-driven cross-scale prediction system for material properties that integrates grain deformation and dislocation mechanisms. The system is used to implement the above-described method and includes: The data acquisition and processing module is used to acquire microstructure characterization data of materials and extract first-type and second-type microscale features; The physical simulation calculation module is used to construct and run finite element simulations based on the first type of microscale features, embedding a crystal plastic constitutive model, to generate stress-strain data and macroscopic mechanical response for each grain; A cross-scale dataset construction module is used to associate the stress-strain data, macroscopic mechanical response, and corresponding first-type and second-type microscale features to form a structured dataset. The machine learning modeling and training module is used to select at least one feature from the first type of microscale features and the second type of microscale features as input parameters based on the structured dataset, construct and train a machine learning model to learn the mapping relationship from microscale features to material mechanical properties. The performance prediction and application module is used to load pre-trained machine learning models, quickly predict material properties based on newly input microstructure characterization data, and support model-interpretable analysis and reverse design. Example 1: A data-driven method for predicting and optimizing material properties by integrating CPFEM and machine learning

[0042] Step 1: Data Preparation and Feature Extraction The goal of this step is to construct a dataset for training machine learning models. Data sources include high-fidelity CPFEM "virtual experiments".

[0043] 1. Initial Sample Acquisition: Collect or generate a batch of die-cast aluminum alloy samples with different microstructure characteristics. For example, by changing the casting or heat treatment process, obtain multiple samples with grain sizes ranging from tens to hundreds of micrometers and varying texture strengths. Perform electron backscatter diffraction (EBSD) scans on a specific cross-section (e.g., cross-section AA) of each sample to obtain raw data containing information on grain orientation and grain boundaries.

[0044] 2. Microscale Feature Extraction: The raw EBSD data was processed using MTEX ​​or similar crystallographic analysis tools. For each grain, the following microscale features were extracted: a. First type of microscale features: crystal orientation and phase distribution information.

[0045] b. Second type of microscale features: dislocation density, grain boundary type, Schmidt factor (SF), Taylor factor (TF), grain size, texture type and intensity.

[0046] These features constitute the "input features" (describing the initial microstate) of a machine learning model.

[0047] 3. Construct RVE and perform CPFEM simulation to generate label data: a. Mesh Generation: Based on the cross-sectional EBSD data, the contour of each grain is extracted and imported into a mesh generation tool (such as the MATLAB-based MTEX ​​toolkit or dedicated software) to perform two-dimensional Delaunay triangular mesh generation. The mesh size is set to approximately 1 / 10 of the average grain size (e.g., 2-5 micrometers). The generated mesh model is as follows: Figure 2 As shown, each unit belongs to a specific grain.

[0048] b. Finite Element Modeling: Import the mesh model into Abaqus / Standard finite element software. Write a user material subroutine (VUMAT) to implement a crystal plastic constitutive model suitable for face-centered cubic aluminum alloys. This model contains 12 {111} <110> For slip systems, the hardening law employs a physical model that considers the evolution of dislocation density, where the flow stress is related to the square root of the dislocation density. Key material parameters are calibrated through literature or macroscopic tensile experiments, such as: elastic constants C11 = 108 GPa, C12 = 61 GPa, C44 = 28 GPa; initial critical shear stress τ0 = 60 MPa; initial hardening modulus h0 = 400 MPa; saturated rheological stress τs = 120 MPa; dislocation evolution related parameters y0 = 0.1, k = 0.05.

[0049] c. Simulation Setup: Apply periodic boundary conditions to the RVE model to simulate the macroscopic uniaxial tensile process. Apply displacement load in the X direction with a strain increment step of 0.002 until the total strain reaches 0.15.

[0050] The simulated Mises stress distribution cloud diagrams under different strains are shown below. Figure 3 As shown.

[0051] d. Data Extraction: After the simulation is complete, extract the following data from the results file: * Macroscopic label: The average stress (macroscopic stress) and average strain (macroscopic strain) of the entire RVE at each increment step constitute the macroscopic stress-strain curve.

[0052] * Microscopic Labels: At several specific macroscopic strain points (e.g., ε=0.02, 0.05, 0.10), output the stress, strain, and dislocation density data for all elements. Subsequently, data aggregation is performed on a grain-by-grain basis: for each grain, calculate the average and maximum Mises stress (as stress concentration factor), the average equivalent plastic strain, and the average dislocation density for all elements within it.

[0053] The calculated "average equivalent plastic strain", "average dislocation density", and "maximum stress concentration factor" constitute part of the "output label" of the machine learning model (describing the microscopic state during deformation), while the macroscopic stress value is another key output label.

[0054] Step 2: Construct a cross-scale dataset All the data obtained in step one are correlated and integrated. Each "data sample" corresponds to the state of a specific grain in a specific RVE (i.e., a specific microstructure) under a specific macroscopic strain.

[0055] Input feature vector: the initial microscale features of the grain (some or all of the following: crystal orientation, phase type, grain size, texture type and intensity, dislocation density value, Schmid factor value, Taylor factor value).

[0056] Output tags: Microfield data of the grain under the current macroscopic strain state (mean equivalent plastic strain, mean dislocation density, maximum Mises stress) or macroscopic stress value of the entire RVE under the current macroscopic strain.

[0057] By performing the above processing on all grains across multiple RVEs and multiple strain levels, a large dataset can be constructed. For example, with 10 different RVEs, each RVE will generate a stress-strain curve after tension. Each stress-strain curve can select multiple mechanical property parameters; for example, the simplest option is to select only one mechanical property value, yield strength. Each RVE will generate one yield strength after tension. Each RVE contains approximately 500 grains, and each grain has 5 microscale feature parameters. The yield strength of this one RVE can be divided into 500 yield strength components, generating 10 × 500 = 5000 data samples. Each sample includes 5 input parameters (5 microscale features) and 1 output parameter (1 mechanical property parameter). The dataset is then randomly divided into training, validation, and test sets in a 70%:15%:15% ratio.

[0058] Step 3: Machine Learning Model Training and Validation This embodiment trains models for two prediction tasks separately: Task 1: Predict the grain-level microfield. Using the initial characteristics of the grain as input, predict the average equivalent strain, average dislocation density, and maximum stress of the grain under a certain macroscopic strain. The Gradient Boosting Decision Tree (GBDT) model is selected due to its excellent performance on structured tabular data and good interpretability. The model is trained on the training set, and hyperparameters (such as tree depth, learning rate, and number of trees) are adjusted using grid search on the validation set. The final model achieves a determination coefficient (R²) of over 0.92 between its predictions and CPFEM calculations on the test set.

[0059] Task 2: Directly Predict Macroscopic Stress-Strain Curves. The initial microscale features of all grains in the entire RVE (such as grain size distribution histograms, average orientation intensity, and other statistics) are used as the overall input, with the goal of mapping them to a complete stress-strain curve. A recurrent neural network (RNN), specifically a long short-term memory network (LSTM), is used to predict the stress sequence corresponding to the strain sequence. A well-trained LSTM model can accurately predict the macroscopic response of the new tissue, such as... Figure 4 As shown, its predicted curve is in high agreement with the CPFEM simulation curve.

[0060] Step 4: Rapid Performance Prediction and Reverse Engineering Applications 1. Second-level prediction: For a brand-new die-cast aluminum alloy component, only the EBSD data of one cross-section is required. Microscale features are extracted using the method in step one, without any finite element calculations, and directly input into the pre-trained GBDT and LSTM models. On a standard workstation, the predicted macroscopic stress-strain curve, yield strength, tensile strength, and stress concentration and dislocation evolution trends of each grain during deformation can be obtained within seconds.

[0061] 2. Key Feature Analysis: The trained GBDT model was analyzed using the SHAP (SHapley Additive exPlanations) interpretability tool. This allowed for the quantification of the contribution of each input feature (such as crystal orientation, phase type, grain size, texture type and intensity, dislocation density, Schmidt factor, and Taylor factor) to the prediction of "maximum grain stress" or "macroscopic yield strength." The results showed that for this aluminum alloy, grain size and the average Taylor factor were the two most critical features affecting yield strength, followed by large-angle grain boundary density.

[0062] 3. Microstructure Reverse Design: Based on the above analysis, performance targets are set: for example, a yield strength ≥ 250 MPa and uniform elongation ≥ 8%. Using a Bayesian optimization algorithm, microscale features (such as average grain size and texture intensity) are used as design variables, and a trained LSTM model is used as a performance evaluator (surrogate model). The Bayesian optimization algorithm intelligently samples and searches within the feature space, iterating hundreds of times to recommend one or more sets of optimal microstructure feature parameters that simultaneously meet the strength and plasticity requirements (e.g., average grain size ~35 μm, with moderately strong cubic texture). This result can be directly used to guide alloy composition adjustment or the setting of casting / heat treatment process parameters. Example 2: A material property cross-scale prediction system

[0063] Based on the above method, this embodiment also provides a system, which includes the following modules implemented by software and / or hardware: Data acquisition and processing module: Configured to import microstructure data in formats such as EBSD, call tool libraries such as MTEX ​​for parsing, and automatically extract first-type and second-type microscale features.

[0064] Physical simulation calculation module: Based on the first type of microscale features, it is configured to receive processed data, automatically call the mesh generation script to generate RVE mesh, call finite element software such as Abaqus and its VUMAT subroutine, submit and monitor CPFEM simulation tasks, and automatically parse and extract macroscopic and microscopic data from the result file.

[0065] Cross-scale dataset construction module: Configured to automatically associate, align and stitch together the data output by the physics simulation module and the features extracted by the data acquisition module according to preset rules to form a structured database or data file.

[0066] Machine learning modeling and training module: Provides interfaces for various machine learning algorithms (such as DNN, GBDT, RNN), and supports automated execution of data preprocessing (standardization, serialization), model building, hyperparameter tuning, training and validation processes.

[0067] Performance Prediction and Application Module: Loads pre-trained model files, provides graphical or API interfaces for users to input microstructure data of new materials, and instantly returns performance prediction results. It integrates interpretable tool libraries such as SHAP and can connect with algorithms such as Bayesian optimization and generative adversarial networks to achieve reverse engineering capabilities.

[0068] The system can be deployed on a local server or cloud computing platform to provide users with integrated material performance prediction and design services. Example 3: An electronic device An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it performs the steps of the method described in Embodiment 1. The electronic device may be a high-performance workstation, a server, or a computing cluster.

[0069] The above description of the embodiments is only for the purpose of helping to understand the method and core idea of ​​this application; at the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0070] Certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that hardware manufacturers may use different names to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. The terms "comprising" and "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising / including but not limited to". "Approximately" means that within an acceptable margin of error, those skilled in the art can solve the technical problem and substantially achieve the technical effect within a certain margin of error. The following descriptions in the specification are preferred embodiments for carrying out this application; however, these descriptions are for the purpose of illustrating the general principles of this application and are not intended to limit the scope of this application. The scope of protection of this application shall be determined by the appended claims.

[0071] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0072] It should be understood that the term "and / or" used in this article 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 article generally indicates that the preceding and following related objects have an "or" relationship.

[0073] The foregoing description illustrates and describes several preferred embodiments of this application. However, as previously stated, it should be understood that this application is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the application concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of this application should be within the protection scope of the appended claims.

Claims

1. A data-driven method for cross-scale prediction of material properties that integrates grain deformation and dislocation mechanisms, characterized in that, Includes the following steps: S1: Obtain EBSD images of the target material and obtain EBSD microstructure characterization data; S2: Based on the EBSD microstructure characterization data, extract the first type of microscale features and the second type of microscale features; The first type of microscale features includes crystal orientation and phase distribution information; the second type of microscale features includes at least one of Schmidt factor SF, Taylor factor TF, grain size, and texture. S3: Based on the microstructure characterization data of the EBSD, a representative volume element RVE model containing the real grain structure is constructed using the first type of microscale features; the crystal plastic constitutive model is assigned to the RVE model, and finite element calculation is performed to obtain macroscopic mechanical response and grain-scale stress-strain distribution data; S4: Connect the stress-strain data and macroscopic mechanical response of each grain with its corresponding first-type and second-type microscale features to construct a cross-scale associated dataset; S5: Based on the cross-scale associated dataset, at least one feature is selected from the first type of microscale features and the second type of microscale features as input parameters to train a machine learning model and establish a mapping relationship from microscale features to material mechanical properties. S6: Using the trained machine learning model, predict the macroscopic mechanical properties and / or microscopic field distribution of the new material sample based on the microstructure characterization data.

2. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 1, characterized in that, In step S1, the EBSD microstructure characterization data are either the original EBSD data obtained through experimental measurement or the simulated data generated through crystal plasticity finite element simulation.

3. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 1, characterized in that, In step S2, the first type of microscale features further includes grain geometric boundary information; the second type of microscale features further includes at least one of dislocation density and grain boundary type.

4. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 1, characterized in that, Step S3 specifically includes: S31: Based on the first type of microscale features, perform mesh generation to generate a finite element mesh model containing the real grain morphology; S32: Import the finite element mesh model into the finite element software and embed a crystal plastic constitutive model containing the dislocation density evolution mechanism into it; S33: Assign initial crystal orientation and material parameters to each grain in the model, apply boundary conditions and loads for simulation calculation; S34: Extract stress-strain data of each grain under different macroscopic strains and the overall macroscopic stress-strain response from the simulation results.

5. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 4, characterized in that, In step S32, the crystal plastic constitutive model is embedded into the finite element solver through a user material subroutine.

6. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 1, characterized in that, In step S5, the machine learning model is a deep neural network, a recurrent neural network, or a gradient boosting decision tree model; before training, the features in the cross-scale associated dataset are standardized or normalized.

7. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 1, characterized in that, Step S6 is followed by: S7: Utilize model interpretability techniques to analyze key first-type and / or second-type microscale features that affect material properties; S8: Based on the analysis results, combined with optimization algorithms or generative models, reverse design of micro-organization is carried out with target performance as a constraint.

8. The data-driven cross-scale prediction method for material properties that integrates grain deformation and dislocation mechanisms as described in claim 7, characterized in that, In step S8, the optimization algorithm is Bayesian optimization, and the generative model is a generative adversarial network.

9. A data-driven cross-scale prediction system for material properties that integrates grain deformation and dislocation mechanisms, characterized in that, The system is used to implement the method as described in any one of claims 1-8, the system comprising: The data acquisition and processing module is used to acquire microstructure characterization data of materials and extract first-type and second-type microscale features; The physical simulation calculation module is used to construct and run finite element simulations based on the first type of microscale features, embedding a crystal plastic constitutive model, to generate stress-strain data and macroscopic mechanical response for each grain; A cross-scale dataset construction module is used to associate the stress-strain data, macroscopic mechanical response, and corresponding first-type and second-type microscale features to form a structured dataset. The machine learning modeling and training module is used to select at least one feature from the first type of microscale features and the second type of microscale features as input parameters based on the structured dataset, construct and train a machine learning model to learn the mapping relationship from microscale features to material mechanical properties. The performance prediction and application module is used to load pre-trained machine learning models, quickly predict material properties based on newly input microstructure characterization data, and support model-interpretable analysis and reverse design.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-8.

Images