Construction method and device of metal material performance prediction model, equipment and medium

Abstract

The invention discloses a construction method and device of a metal material performance prediction model, equipment and a medium. The method comprises the following steps: constructing an initial graph neural network; acquiring a three-dimensional volume element structure of the specified metal material and mechanical property data corresponding to the three-dimensional volume element structure; describing the three-dimensional volume element structure as an undirected graph structure; constructing a target data set by using the undirected graph structure and the mechanical property data; and performing iterative training on the initial graph neural network through the target data set to obtain a metal material performance prediction model. Therefore, the undirected graph structure of the three-dimensional volume element structure is used as a bridge, the metal material performance prediction model between the complex three-dimensional volume element structure and the mechanical performance data is established through the graph neural network, the model can rapidly and precisely predict the macroscopic performance of the metal material, the cost is low, and the method is easy to implement. And the requirements of rapid iteration and real-time feedback of the macroscopic performance of the material in engineering design can be met.

Description

Technical Field

[0001] This application relates to the field of materials technology, and in particular to methods, apparatus, equipment and media for constructing predictive models of metallic material properties. Background Technology

[0002] Metallic materials, as one of the most important structural materials in modern industry, are widely used in aerospace, machinery manufacturing, transportation, construction, and other production and daily life fields. In the practical application of metallic materials, their macroscopic mechanical properties are directly affected by their microstructure, including grain size, grain distribution, grain boundary characteristics, and phase types. Therefore, establishing a quantitative relationship between macroscopic mechanical properties and microstructure is of great significance for the practical application of metallic materials.

[0003] Currently, the Crystal Plasticity Finite Element Method (CPFEM) can be used to establish a quantitative relationship between the macroscopic mechanical properties and microstructure of metallic materials. Based on physical mechanisms such as dislocation slip, this method combines the constitutive relation of single crystals with the geometric modeling of polycrystalline aggregates. It can realistically reflect the interaction between grains, the evolution of local stress and strain fields, and the development of texture, thereby predicting the macroscopic stress-strain response and anisotropic characteristics of materials, that is, predicting macroscopic mechanical properties.

[0004] However, while CPFEM theoretically offers high physical fidelity, its practical applications suffer from complex data processing and high costs. Furthermore, CPFEM calculations involve solving nonlinear constitutive equations and creating fine meshes, resulting in massive computational scale and long computation time, making it difficult to meet the demands of rapid iteration and real-time feedback of material macroscopic mechanical properties in engineering design.

[0005] Therefore, how to quickly, efficiently and accurately predict the macroscopic mechanical properties of metallic materials and reduce costs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, one aspect of this application provides a method for constructing a predictive model for the properties of metallic materials, the method comprising: Construct the initial graph neural network; Obtain the three-dimensional volumetric element structure of a specified metallic material and the corresponding mechanical property data of the three-dimensional volumetric element structure; The three-dimensional volumetric metastructure is described as an undirected graph structure; Using the undirected graph structure and the mechanical performance data, construct the target dataset; The initial graph neural network is iteratively trained using the target dataset to obtain a prediction model for the properties of metallic materials.

[0007] Optionally, obtaining the three-dimensional volumetric element structure of the specified metallic material and the corresponding mechanical property data includes: Obtain the microstructure characteristics of the specified metallic material; wherein, the microstructure characteristics include grain characteristics and phase characteristics; Based on the aforementioned microstructural features, the three-dimensional volumetric element structure is generated; The three-dimensional volumetric element structure is simulated to obtain the mechanical performance data.

[0008] Optionally, describing the three-dimensional volumetric element structure as an undirected graph structure includes: Extract the grains of the three-dimensional volumetric element structure and the relevant parameters of the grains; The grains are used as nodes in the undirected graph structure; The relevant parameters are combined into a feature vector, and the feature vector is injected into the node; Edges are constructed between adjacent grains and between phase boundaries of adjacent phases to obtain the undirected graph structure.

[0009] Optionally, the mechanical property data includes elastic modulus and yield strength; the target dataset includes a first dataset and a second dataset, the first dataset being composed of the undirected graph structure and the elastic modulus, and the second dataset being composed of the undirected graph structure and the yield strength; the initial graph neural network includes a first graph neural network and a second graph neural network; the metallic material property prediction model includes an elastic modulus prediction model and a yield strength prediction model.

[0010] Optionally, the step of iteratively training the initial graph neural network using the target dataset to obtain a metal material property prediction model includes: The elastic modulus prediction model is obtained by iteratively training the first graph neural network using the first dataset. The yield strength prediction model is obtained by iteratively training the second graph neural network using the second dataset.

[0011] Optionally, after iteratively training the initial graph neural network using the target dataset to obtain the metal material property prediction model, the method further includes: Obtain the performance parameters of the metal material performance prediction model; wherein, the performance parameters include parameter determination coefficient, root mean square error, and mean absolute error; Determine whether the performance parameters meet the corresponding threshold conditions; the threshold conditions include that the parameter determination coefficient is greater than the coefficient threshold, the root mean square error is less than the first threshold, and the mean absolute error is less than the second threshold. If the conditions are met, the metal material performance prediction model will be used as the target metal material performance prediction model. If the conditions are not met, return to the step of iteratively training the initial graph neural network using the target dataset, and execute subsequent steps.

[0012] Optionally, if the performance parameter meets the corresponding threshold condition, the method further includes: Obtain a test dataset that includes tests on undirected graph structures and tests on mechanical properties; The test undirected graph structure is input into the metal material performance prediction model to obtain the predicted mechanical properties; Determine whether the target parameter value is less than a preset value; the target parameter value is used to characterize the degree of difference between the predicted mechanical properties and the tested mechanical properties; the smaller the target parameter value, the smaller the degree of difference. If it is less than, the metal material performance prediction model shall be used as the target metal material performance prediction model; If it is not less than, return to the step of iteratively training the initial graph neural network using the target dataset, and execute subsequent steps.

[0013] Another aspect of this application provides an apparatus for constructing a predictive model for the properties of metallic materials, the apparatus comprising: The initial network building block is used to build the initial graph neural network; The target acquisition module is used to acquire the three-dimensional volumetric element structure of a specified metallic material and the mechanical property data corresponding to the three-dimensional volumetric element structure. The description module is used to describe the three-dimensional volumetric metastructure as an undirected graph structure; A dataset construction module is used to construct a target dataset using the undirected graph structure and the mechanical performance data; The iterative training module is used to iteratively train the initial graph neural network using the target dataset to obtain a prediction model for the properties of metallic materials.

[0014] Another aspect of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of a method for constructing a predictive model for the properties of metallic materials.

[0015] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for constructing the predictive model of the properties of the metallic material.

[0016] The beneficial effects of the construction method, apparatus, equipment and medium for the prediction model of metallic material properties provided in this application are as follows: using the undirected graph structure of the three-dimensional volume element structure as a bridge, a prediction model of metallic material properties between the complex three-dimensional volume element structure and mechanical property data is established through graph neural network. This model can predict the macroscopic properties of metallic materials quickly and with high accuracy. It is low in cost and can meet the needs of rapid iteration and real-time feedback of the macroscopic properties of materials in engineering design. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for constructing a performance prediction model for metallic materials provided in an embodiment of this application; Figure 2(a) is a schematic diagram of a three-dimensional volume element structure provided in an embodiment of this application; Figure 2(b) is a schematic diagram of a grain distribution corresponding to the three-dimensional volume element structure in Figure 2(a); Figure 2(c) is a mechanical property curve corresponding to the three-dimensional volume element structure in Figure 2(a); Figure 2(d) is a schematic diagram of an undirected graph structure corresponding to Figure 2(a); Figure 3(a) is a schematic diagram of the prediction result of the elastic modulus provided in an embodiment of this application; Figure 3(b) is a schematic diagram of the prediction result of yield strength provided by an embodiment of this application; Figure 4 A schematic diagram of a device for constructing a predictive model for the properties of metallic materials provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0018] The attached diagram is labeled as follows: 40 is the initial network construction module, 41 is the target acquisition module, 42 is the description module, 43 is the dataset construction module, 44 is the iterative training module, 50 is the memory, 51 is the processor, 52 is the display screen, 53 is the input / output interface, 54 is the communication interface, 55 is the power supply, 56 is the communication bus, 501 is the computer program, 502 is the operating system, and 503 is the data. Detailed Implementation

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0020] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0021] Figure 1 This is a flowchart illustrating a method for constructing a performance prediction model for metallic materials provided in an embodiment of this application. Figure 1 As shown, the method includes: S10: Construct the initial graph neural network; S11: Obtain the three-dimensional volumetric element structure of the specified metallic material and the corresponding mechanical property data of the three-dimensional volumetric element structure; In a specific embodiment, in order to obtain a model that can be directly used for metal performance prediction, an initial graph neural network (GNN) is pre-constructed. In an optional embodiment, the architecture of the constructed initial GNN may include a fully connected layer for graph structure feature preprocessing, two message passing layers, a global average pooling layer (graph readout layer), and two fully connected post-processing layers.

[0022] In addition, a target dataset needs to be constructed to train the initial GNN. Specifically, the three-dimensional volumetric element structure of the specified metallic material and the mechanical property data corresponding to each three-dimensional volumetric element structure are first obtained. Here, the three-dimensional volumetric element structure refers to selecting a smallest but "sufficiently large" volumetric unit in three-dimensional space to represent the microstructural features of the entire specified metallic material, which can be used for numerical simulation of mechanical properties, physical behavior, or failure mechanisms.

[0023] Figure 2(a) is a schematic diagram of a three-dimensional volumetric element structure provided in an embodiment of this application. As shown in Figure 2(a), it is a three-dimensional volumetric element structure of a specified metal material, that is, a volumetric unit constituting the specified metal material. Figure 2(a) also shows the phase portion of the specified metal material. This application does not limit the shape, size, etc. of the three-dimensional volumetric element structure.

[0024] In specific embodiments, the mechanical property data may include, but are not limited to, elastic modulus, yield strength, Brinell hardness, and impact toughness. Elastic modulus is used to measure the elastic properties of a metallic material, yield strength is used to measure the strength properties of a metallic material, Brinell hardness is used to measure the hardness properties of a metallic material, and impact toughness is used to measure the toughness properties of a metallic material.

[0025] S12: Describe the three-dimensional volumetric element structure as an undirected graph structure; S13: Construct the target dataset using undirected graph structure and mechanical performance data; In order to enable the pre-built initial GNN to fully learn the information in the three-dimensional volumetric metastructure, in an optional embodiment, the three-dimensional volumetric metastructure is described as an undirected graph structure. This allows the establishment of a correspondence between the undirected graph structure and the mechanical performance data, thereby obtaining the target dataset.

[0026] In other words, in specific embodiments, each three-dimensional volume element structure corresponds to different mechanical performance data, and each three-dimensional volume element structure can be described as an undirected graph structure. Thus, there is a correspondence between the undirected graph structure and the mechanical performance data.

[0027] S14: Using the target dataset, iteratively train the initial graph neural network to obtain a prediction model for the properties of metallic materials.

[0028] Furthermore, the initial GNN is iteratively trained using the target dataset to obtain a metal material performance prediction model that can directly predict the macroscopic properties of metallic materials. In an optional embodiment, the initial GNN can be trained using a GPU-accelerated computer system, which can significantly improve the model training speed.

[0029] In one optional embodiment, the metal property prediction module obtained through the embodiments of this application can achieve high-precision prediction of the mechanical properties of polycrystalline multiphase materials. In the prediction results, the average absolute error is less than 5%, and the coefficient of determination is greater than 0.9. At the same time, the time required for prediction by this prediction model is on the order of 0.1 seconds, which is much less than the time required for obtaining mechanical properties through crystal finite element simulation, thus improving the performance acquisition efficiency by two orders of magnitude.

[0030] Therefore, the method for constructing a metal material performance prediction model provided in this application uses an undirected graph structure of a three-dimensional volume element structure as a bridge, and establishes a metal material performance prediction model between a complex three-dimensional volume element structure and mechanical performance data through a graph neural network. This model can quickly and accurately predict the macroscopic performance of metal materials, is low in cost, and can meet the needs of rapid iteration and real-time feedback of material macroscopic performance in engineering design.

[0031] In one optional embodiment, obtaining the three-dimensional volumetric element structure of a specified metallic material and the corresponding mechanical property data includes: Obtain the microstructure characteristics of a specified metallic material; wherein, the microstructure characteristics include grain characteristics and phase characteristics; Based on the microstructure characteristics, a three-dimensional volumetric element structure is generated. Simulation of a three-dimensional volumetric element structure is performed to obtain mechanical performance data.

[0032] In one optional embodiment, the true microstructure characteristics of a specified metallic material can be obtained experimentally. These microstructure characteristics include grain features and phase features. Grain features include, but are not limited to, grain size, grain orientation, and grain distribution, while phase features include, but are not limited to, phase name, phase type, and phase content.

[0033] After obtaining the microstructure characteristics, in one optional embodiment, a three-dimensional volumetric element structure is generated using Dream3D software based on these characteristics. The resulting three-dimensional volumetric element structure includes grain features and phase features, which can characterize the microstructure characteristics of a specified metallic material. As shown in Figure 2(a), the different colors blue and green represent different phase types, and yellow represents precipitated phases. Figure 2(b) is a schematic diagram of a grain distribution corresponding to the three-dimensional volumetric element structure in Figure 2(a). As shown in Figure 2(b), different colors represent grain structures with different orientation parameters, and adjacent grains are separated by grain boundaries.

[0034] In a specific embodiment, after obtaining the three-dimensional volume element structure, as an optional embodiment, the DAMASK software can be used to perform crystal plasticity finite element simulation on the three-dimensional volume element structure to obtain the mechanical property data of the specified metal material. Thus, the correspondence between the three-dimensional volume element structure and the mechanical property data can be constructed, thereby constructing the database required for training the GNN performance prediction model.

[0035] Figure 2(c) is a mechanical property curve corresponding to the three-dimensional volumetric element structure of Figure 2(a). In an optional embodiment, when the mechanical property data includes elastic modulus and yield strength, simulation of the three-dimensional volumetric element structure of Figure 2(a) can yield the mechanical property curve of Figure 2(c), where the horizontal axis represents strain and the vertical axis represents stress. The elastic modulus and yield strength of the specified metallic material can be directly read from this curve.

[0036] In one alternative embodiment, the three-dimensional volumetric metastructure is described as an undirected graph structure, including: Extract the grains of the three-dimensional volumetric element structure and the relevant parameters of the grains; The grains are used as nodes in the undirected graph structure; The relevant parameters are combined into a feature vector, and the feature vector is injected into the node; Edges are constructed between adjacent grains and between phase boundaries of adjacent phases to obtain an undirected graph structure.

[0037] Figure 2(d) is a schematic diagram of an undirected graph structure corresponding to Figure 2(a). For ease of understanding, the construction of the undirected graph structure will be explained below in conjunction with Figures 2(a), 2(b), and 2(d). In a specific embodiment, the grains of the three-dimensional volumetric element structure shown in Figures 2(a) and 2(b), as well as the relevant parameters of the grains, are extracted. These relevant parameters include, but are not limited to, grain orientation parameters, grain size parameters, and phase simulation parameters used in finite element simulation of the phases contained in the grains. Further, the extracted grains are used as nodes in the undirected graph structure, and the relevant parameters of the grains are combined into feature vectors. These feature vectors are then injected into the corresponding nodes to characterize the relevant information of the nodes. In a specific embodiment, the parameters of different grains can be queried by viewing different nodes.

[0038] Furthermore, the boundaries of adjacent grains, the phase boundaries of precipitated phases and the basic phase are used as edges in the undirected graph structure to realize undirected connections between different nodes, thereby completing the construction of the undirected graph structure. The undirected graph structure includes nodes composed of grains, and the nodes include relevant parameters of the grains. There are edges between adjacent grains, and there are also edges between the phase boundaries of adjacent phases.

[0039] Therefore, the three-dimensional volumetric metastructures shown in Figures 2(a) and 2(b) can be described as the undirected graph structure shown in Figure 2(d). The generated undirected graph structure can be used as input data for the initial GNN to iteratively train the initial GNN so that the initial GNN can learn the microstructure information of the metallic material in the undirected graph structure.

[0040] In one optional embodiment, the mechanical property data includes elastic modulus and yield strength. Correspondingly, the metallic material property prediction model includes an elastic modulus prediction model that can be used to predict elastic properties, and a yield strength prediction model that can be used to predict strength.

[0041] In a specific embodiment, the target dataset may include a first dataset and a second dataset. The first dataset includes an undirected graph structure and an elastic modulus, while the second dataset includes an undirected graph structure and a yield strength. That is, when constructing the target dataset, the elastic modulus and yield strength of the specified metallic material are obtained through simulation. Then, the correspondence between the undirected graph structure and the elastic modulus is constructed to obtain the first dataset, and the correspondence between the undirected graph structure and the yield strength is constructed to obtain the second dataset.

[0042] Furthermore, the pre-constructed initial GNN also includes two components, specifically a first neural network and a second neural network. Based on this, as an optional implementation, the initial graph neural network is iteratively trained using the target dataset to obtain a metal material performance prediction model, including: Using the first dataset, the first graph neural network is iteratively trained to obtain the elastic modulus prediction model; Using the second dataset, the second graph neural network is iteratively trained to obtain the yield strength prediction model.

[0043] Therefore, by iteratively training the GNN using different datasets, models with different performance predictions can be obtained. In fact, in specific embodiments, mechanical performance parameters may also include elongation for measuring plasticity, Brinell hardness for measuring hardness, and fatigue limit for measuring fatigue performance.

[0044] Correspondingly, models for predicting different performance can be generated. Similarly, in a specific embodiment, the correspondence between an undirected graph structure and mechanical performance data of different dimensions is constructed as a training dataset to train the GNN, thereby obtaining a model for predicting mechanical performance of the corresponding dimension.

[0045] It should be noted that, in an optional embodiment, a correspondence between the undirected graph structure, elastic modulus, and yield strength can also be constructed. That is, a correspondence between the undirected graph structure and the dimensional mechanical property data can be constructed, thereby training an initial GNN to obtain a model that can predict multi-dimensional mechanical properties. This application does not limit whether to train one or multiple models.

[0046] In an optional embodiment, after iteratively training the initial graph neural network using the target dataset to obtain a metal material property prediction model, the method further includes: Obtain the performance parameters of the metal material performance prediction model; among which, the performance parameters include the parameter determination coefficient, root mean square error, and mean absolute error; Determine whether the performance parameters meet the corresponding threshold conditions; the threshold conditions include that the parameter determination coefficient is greater than the coefficient threshold, the root mean square error is less than the first threshold, and the mean absolute error is less than the second threshold. If the conditions are met, the metal material performance prediction model will be used as the target metal material performance prediction model. If the conditions are not met, return to the step of iteratively training the initial graph neural network using the target dataset, and then execute the subsequent steps.

[0047] In a specific embodiment, in order to improve the accuracy of the metal material performance prediction model, the initial GNN is iteratively trained using the target dataset to obtain the performance parameters of the metal material performance prediction model. Specifically, these parameters may include the parameter determination coefficient, root mean square error, and mean absolute error.

[0048] The coefficient of determination (COD) represents the model's ability to explain data variation. A COD closer to 1 indicates a better model fit, meaning the model can effectively explain data changes. Therefore, in a specific implementation, when the COD is greater than a threshold, it signifies that the COD dimension, used to measure the model's fit, meets expectations.

[0049] Both the root mean squared error (RMSE) and the mean absolute error (MAE) are considered to be smaller when they indicate better model performance, meaning the model has less prediction error. Therefore, in a specific implementation, when the RMSE is less than a first threshold and the MAE is less than a second threshold, it indicates that the model has met expectations in evaluating these two dimensions.

[0050] Therefore, in a specific embodiment, in order to ensure that the trained metal material performance prediction model can accurately predict the macroscopic mechanical properties of metals, the performance parameters of the model are monitored in real time. When the performance parameters meet the corresponding threshold conditions, the current model is guaranteed to meet the expectations. At this time, the currently trained metal material performance prediction model can be used as the final target metal material performance prediction model for metal prediction.

[0051] If any performance parameter fails to meet the corresponding threshold condition, it indicates that the current model's performance is not as expected. In this case, iterative training of the metal material performance prediction model using the target dataset is necessary. An example will be provided below for easier understanding.

[0052] Figure 3(a) is a schematic diagram of the prediction result of elastic modulus provided by an embodiment of this application, and Figure 3(b) is a schematic diagram of the prediction result of yield strength provided by an embodiment of this application. In an optional embodiment, if the coefficient threshold corresponding to the parameter determination coefficient is 0.85, as shown in Figure 3(a), the parameter determination coefficient of the elastic modulus prediction model in the above embodiment is greater than 0.91. As shown in Figure 3(b), the determination coefficient of the yield strength prediction model in the above embodiment is greater than 0.94. Therefore, the elastic modulus prediction model and the yield strength prediction model meet expectations in the performance dimension of parameter determination coefficient.

[0053] In another optional embodiment, if the performance parameters meet the corresponding threshold conditions, the method for constructing the performance prediction model of metallic materials further includes: Obtain a test dataset that includes tests on undirected graph structures and tests on mechanical properties; The undirected graph structure will be tested and input into a metal material property prediction model to obtain predicted mechanical properties. Determine whether the target parameter value is less than the preset value; the target parameter value is used to characterize the degree of difference between the predicted mechanical properties and the tested mechanical properties; the smaller the target parameter value, the smaller the degree of difference. If it is less than 1, the metal material performance prediction model will be used as the target metal material performance prediction model. If it is not less than, return to the steps of iteratively training the initial graph neural network using the target dataset, and execute subsequent steps.

[0054] In one alternative embodiment, the target dataset can be divided into a training dataset and a test dataset. In the actual model training process, although the training dataset and the test dataset have certain similarities, the metal material performance prediction model trained may show better performance on the training dataset, that is, the performance parameters all meet the corresponding threshold conditions, but the performance on the test dataset may not be good.

[0055] Therefore, to further improve the prediction accuracy of the metal material performance prediction model, in one optional embodiment, the model is tested using a test dataset. Specifically, the test dataset includes test undirected graph structures and test mechanical properties. This test dataset comes from metal materials different from the training dataset; that is, it contains data knowledge that the metal material performance prediction model obtained from the current iteration of training has not learned.

[0056] At this point, the undirected graph structure from the test dataset is input into the metal material property prediction model to obtain the corresponding predicted mechanical properties. Furthermore, by comparing the predicted mechanical properties with the tested mechanical properties, the model's performance can be evaluated; that is, the model's generalization and extrapolation capabilities can be assessed.

[0057] In one optional embodiment, a target parameter value can be calculated based on the predicted mechanical properties and the tested mechanical properties. This target parameter value is used to characterize the degree of difference between the predicted mechanical properties and the tested mechanical properties. The smaller the target parameter value, the lower the degree of difference between the predicted mechanical properties and the tested mechanical properties, indicating that the current model has a more accurate predictive ability.

[0058] After calculating the target parameter value, if the target parameter value is less than the preset value, it indicates that the current prediction result is relatively accurate, and the current metal material performance prediction model can be used as the target metal material performance prediction model. If the target parameter value is not less than the preset value, it indicates that the performance of the current metal material performance prediction model has not met expectations, and iterative training of the metal material performance prediction model using the target training dataset is required.

[0059] As shown in Figure 3(a), the horizontal axis represents the test elastic modulus obtained after the elastic modulus prediction model processes the test undirected graph structure, and the vertical axis represents the test elastic model obtained through simulation. Since the horizontal and vertical axes are data of the same dimension, the more concentrated the blue points in the figure are on the middle line, the better the performance of the model.

[0060] Similarly, in Figure 3(b), the horizontal axis represents the test yield strength obtained after the yield strength prediction model processes the test undirected graph structure, and the vertical axis represents the test yield strength obtained through simulation. Since the horizontal and vertical axes are data of the same dimension, the more concentrated the green points in the figure are on the middle line, the better the performance of the model.

[0061] In the above embodiments, the method for constructing a prediction model for the properties of metallic materials has been described in detail. This application also provides an embodiment of a device for constructing a prediction model for the properties of metallic materials.

[0062] Figure 4 This is a schematic diagram of the structure of a device for constructing a predictive model of metallic material properties provided in an embodiment of this application, as shown below. Figure 4 As shown, the device includes: Initial network building block 40 is used to build the initial graph neural network; Target acquisition module 41 is used to acquire the three-dimensional volume element structure of a specified metallic material and the mechanical property data corresponding to the three-dimensional volume element structure; Description module 42 is used to describe the three-dimensional volumetric metastructure as an undirected graph structure; Dataset building module 43 is used to build the target dataset using undirected graph structure and mechanical performance data; The iterative training module 44 is used to iteratively train the initial graph neural network using the target dataset to obtain a prediction model for the properties of metallic materials.

[0063] Furthermore, the apparatus for constructing a predictive model for the properties of metallic materials provided in this application embodiment also includes: The structural feature acquisition module is used to acquire the microstructure features of a specified metallic material; the microstructure features include grain features and phase features. The metastructure generation module is used to generate three-dimensional volume metastructures based on microstructural features. The simulation module is used to simulate three-dimensional volumetric structures to obtain mechanical performance data.

[0064] The extraction module is used to extract the grains of the three-dimensional volumetric element structure, as well as the relevant parameters of the grains; The node determination module is used to identify grains as nodes in an undirected graph structure. The injection module is used to combine relevant parameters into a feature vector and inject the feature vector into the node. The edge building module is used to build edges between adjacent grains and between phase boundaries of adjacent phases to obtain an undirected graph structure.

[0065] The first training module is used to iteratively train the first graph neural network using the first dataset to obtain the elastic modulus prediction model. The second training module is used to iteratively train the second graph neural network using the second dataset to obtain the yield strength prediction model.

[0066] The performance parameter acquisition module is used to acquire the performance parameters of the metal material performance prediction model; among which, the performance parameters include the parameter determination coefficient, root mean square error, and mean absolute error. The first processing module is used to determine whether the performance parameters meet the corresponding threshold conditions. The threshold conditions include that the parameter determination coefficient is greater than the coefficient threshold, the root mean square error is less than the first threshold, and the mean absolute error is less than the second threshold. If the conditions are met, the metal material performance prediction model is used as the target metal material performance prediction model. If the conditions are not met, the iterative training module is called.

[0067] The test dataset acquisition module is used to acquire test datasets, including test datasets for undirected graph structures and test datasets for mechanical properties. The testing module is used to input the test undirected graph structure into the metal material property prediction model to obtain the predicted mechanical properties. The second processing module is used to determine whether the target parameter value is less than the preset value. The target parameter value is used to characterize the degree of difference between the predicted mechanical properties and the tested mechanical properties. The smaller the target parameter value, the smaller the degree of difference. If it is less than the preset value, the metal material performance prediction model is used as the target metal material performance prediction model. If it is not less than the preset value, the iterative training module is called.

[0068] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 5 As shown, the electronic device includes: a memory 50 for storing computer programs; The processor 51 is used to execute computer programs to implement the steps of the method for constructing a prediction model of metallic material properties as described in the above embodiments.

[0069] The electronic devices provided in this embodiment may include, but are not limited to, laptops or desktop computers.

[0070] The processor 51 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 51 may be implemented using at least one of the following hardware forms: Digital Signal Processor (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 51 may also include a main processor and a coprocessor. The main processor, also known as the Central Processing Unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 51 may integrate a Graphics Processing Unit (GPU), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 51 may also include an Artificial Intelligence (AI) processor, which is used to handle computational operations related to machine learning.

[0071] The memory 50 may include one or more computer-readable storage media, which may be non-transitory. The memory 50 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In this embodiment, the memory 50 is used to store at least the following computer program 501, which, after being loaded and executed by the processor 51, is capable of implementing the relevant steps of the method for constructing the metal material performance prediction model disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 50 may also include an operating system 502 and data 503, etc., and the storage method may be temporary storage or permanent storage. The operating system 502 may include Windows, Unix, Linux, etc. The data 503 may include, but is not limited to, the relevant data involved in the method for constructing the metal material performance prediction model.

[0072] In some embodiments, the electronic device may further include a display screen 52, an input / output interface 53, a communication interface 54, a power supply 55, and a communication bus 56.

[0073] Those skilled in the art will understand that Figure 5 The structures shown do not constitute a limitation on electronic devices and may include more or fewer components than those shown.

[0074] The electronic device provided in this application includes a memory and a processor. When the processor executes the program stored in the memory, it can implement the method for constructing the metal material performance prediction model in the above embodiments.

[0075] It should be noted that although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

Claims

1. A method for constructing a performance prediction model for metallic materials, characterized in that, The method includes: Construct the initial graph neural network; Obtain the three-dimensional volumetric element structure of a specified metallic material and the corresponding mechanical property data of the three-dimensional volumetric element structure; The three-dimensional volumetric metastructure is described as an undirected graph structure; Using the undirected graph structure and the mechanical performance data, construct the target dataset; The initial graph neural network is iteratively trained using the target dataset to obtain a prediction model for the properties of metallic materials.

2. The method for constructing a performance prediction model for metallic materials as described in claim 1, characterized in that, The acquisition of the three-dimensional volumetric element structure of the specified metallic material and the corresponding mechanical property data includes: Obtain the microstructure characteristics of the specified metallic material; wherein, the microstructure characteristics include grain characteristics and phase characteristics; Based on the aforementioned microstructural features, the three-dimensional volumetric element structure is generated; The three-dimensional volumetric element structure is simulated to obtain the mechanical performance data.

3. The method for constructing a performance prediction model for metallic materials as described in claim 1, characterized in that, The description of the three-dimensional volumetric element structure as an undirected graph structure includes: Extract the grains of the three-dimensional volumetric element structure and the relevant parameters of the grains; The grains are used as nodes in the undirected graph structure; The relevant parameters are combined into a feature vector, and the feature vector is injected into the node; Edges are constructed between adjacent grains and between phase boundaries of adjacent phases to obtain the undirected graph structure.

4. The method for constructing a performance prediction model for metallic materials as described in claim 1, characterized in that, The mechanical property data includes elastic modulus and yield strength; the target dataset includes a first dataset and a second dataset, the first dataset being composed of the undirected graph structure and the elastic modulus, and the second dataset being composed of the undirected graph structure and the yield strength; the initial graph neural network includes a first graph neural network and a second graph neural network; the metal material property prediction model includes an elastic modulus prediction model and a yield strength prediction model.

5. The method for constructing a performance prediction model for metallic materials as described in claim 4, characterized in that, The step of iteratively training the initial graph neural network using the target dataset to obtain a metal material property prediction model includes: The elastic modulus prediction model is obtained by iteratively training the first graph neural network using the first dataset. The yield strength prediction model is obtained by iteratively training the second graph neural network using the second dataset.

6. The method for constructing a performance prediction model for metallic materials as described in claim 1, characterized in that, After iteratively training the initial graph neural network using the target dataset to obtain the metal material property prediction model, the method further includes: Obtain the performance parameters of the metal material performance prediction model; wherein, the performance parameters include parameter determination coefficient, root mean square error, and mean absolute error; Determine whether the performance parameters meet the corresponding threshold conditions; the threshold conditions include that the parameter determination coefficient is greater than the coefficient threshold, the root mean square error is less than the first threshold, and the mean absolute error is less than the second threshold. If the conditions are met, the metal material performance prediction model will be used as the target metal material performance prediction model. If the conditions are not met, return to the step of iteratively training the initial graph neural network using the target dataset, and execute subsequent steps.

7. The method for constructing a performance prediction model for metallic materials as described in claim 6, characterized in that, If the performance parameter meets the corresponding threshold condition, the method further includes: Obtain a test dataset that includes tests on undirected graph structures and tests on mechanical properties; The test undirected graph structure is input into the metal material performance prediction model to obtain the predicted mechanical properties; Determine whether the target parameter value is less than a preset value; the target parameter value is used to characterize the degree of difference between the predicted mechanical properties and the tested mechanical properties; the smaller the target parameter value, the smaller the degree of difference. If it is less than, the metal material performance prediction model shall be used as the target metal material performance prediction model; If it is not less than, return to the step of iteratively training the initial graph neural network using the target dataset, and execute subsequent steps.

8. A device for constructing a predictive model for the properties of metallic materials, characterized in that, The device includes: The initial network building block is used to build the initial graph neural network; The target acquisition module is used to acquire the three-dimensional volumetric element structure of a specified metallic material and the mechanical property data corresponding to the three-dimensional volumetric element structure. The description module is used to describe the three-dimensional volumetric metastructure as an undirected graph structure; A dataset construction module is used to construct a target dataset using the undirected graph structure and the mechanical performance data; The iterative training module is used to iteratively train the initial graph neural network using the target dataset to obtain a prediction model for the properties of metallic materials.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for constructing the metal material performance prediction model according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method for constructing the metal material performance prediction model according to any one of claims 1 to 7.

Images