Metal flow stress prediction method and system based on multi-physics field attention

Abstract

The embodiment of the invention discloses a metal flow stress prediction method and system based on multi-physics field attention, and the method comprises the steps: carrying out the material characteristic coding of the material components and microstructure parameters of a metal material through employing a material characteristic embedding layer of an improved Kolmogov-Arnod network KAN, and obtaining the material characteristics of the metal material; obtaining a characteristic vector of the metal material; carrying out weighted fusion on a plurality of physical field parameters corresponding to isothermal strain measurement by adopting a multi-physical field attention layer of an improved KAN to obtain a high-order feature vector; and performing flow stress prediction on the high-order feature vector and the feature vector based on the temperature value of the isothermal strain measurement by adopting the improved KAN mixed sensing layer to obtain the flow stress of the metal material under the isothermal strain measurement. The accuracy of flow stress prediction can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine learning and material science, and relates to a metal rheological stress prediction method and system based on multi-physical field attention. BACKGROUND

[0002] The rheological stress of a metal material is a key parameter for describing its resistance to deformation during plastic deformation. Accurate prediction of the rheological stress is crucial for material design, optimization of plastic processing techniques (such as forging, rolling, extrusion, etc.), and the accuracy of finite element numerical simulation. Meanwhile, the rheological stress is not a material constant, but is influenced by the complex nonlinearity of the material's intrinsic properties (such as chemical composition, microstructure) and external deformation conditions (such as temperature, strain, strain rate).

[0003] In related technologies, the prediction of rheological stress is mainly divided into two categories: physical-based constitutive models and data-based machine learning models. Both have the following key limitations in predicting the rheological stress of metal materials: Physical-based constitutive models, such as the Arrhenius equation and the Johnson-Cook (JC) model, describe the relationship between rheological stress and deformation conditions through mathematical formulas. Although these models have clear physical meaning, their expressions are relatively fixed and often cannot accurately capture the complex effects of subtle changes in material composition and microstructure on rheological stress. Meanwhile, data-based machine learning models, such as the Kolmogorov-Arnold Network (KAN) and the Multilayer Perceptron (MLP), can learn complex nonlinear relationships from large amounts of experimental data, overcoming the limitations of constitutive models to some extent. However, these models also have limitations such as limited feature interaction ability, lack of physical information, and parameter redundancy. Therefore, how to efficiently and accurately predict rheological stress has become an important problem that needs to be solved in the industry. SUMMARY

[0004] To address the above problems, the embodiments of the present application provide a metal rheological stress prediction method and system based on multi-physical field attention, which aims to improve the accuracy of rheological stress prediction based on the intelligent prediction framework provided by the improved KAN, which provides deep integration of "material perception" and "process perception".

[0005] The technical solutions of the embodiments of the present application are implemented as follows: In a first aspect, the embodiments of the present application provide a metal rheological stress prediction method based on multi-physical field attention, which comprises: The material property embedding layer of the improved Kolmogorov-Arnold network KAN is used to code the material composition and microstructure parameters of the metal material to obtain a characteristic vector of the metal material; The multi-physical field attention layer of the improved KAN is used to weight and fuse a plurality of physical field parameters corresponding to the isothermal strain measurement to obtain a high-order feature vector; The hybrid perception layer of the improved KAN is used to predict the rheological stress of the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement to obtain the rheological stress of the metal material under the isothermal strain measurement.

[0006] In some embodiments, the plurality of physical field parameters includes temperature, strain, and strain rate; the multi-physical field attention layer of the improved KAN is used to weight and fuse a plurality of physical field parameters corresponding to the isothermal strain measurement to obtain a high-order feature vector, including: The two-layer perception machine of the multi-physical field attention layer of the improved KAN is used to weight and fuse the temperature, the strain, and the strain rate when predicting the rheological stress to obtain a temperature weight, a strain weight, and a rate weight; The first product between the temperature and the temperature weight, the second product between the strain and the strain weight, and the third product between the strain rate and the rate weight are fused to obtain the high-order feature vector.

[0007] In some embodiments, before the hybrid perception layer of the improved KAN is used to predict the rheological stress of the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement to obtain the rheological stress of the metal material under the isothermal strain measurement, the method further includes: A material composition query database is used to determine an isostatic temperature value corresponding to the material composition of the metal material; The hybrid perception layer of the improved KAN is used to predict the rheological stress of the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement to obtain the rheological stress of the metal material under the isothermal strain measurement, including: Based on the temperature value of the isothermal strain measurement and the isostatic temperature value, an edge activation function is determined; The N nodes in the hybrid perception layer of the improved KAN that match the characteristic vector are used to perform a nonlinear transformation on the high-order feature vector and the characteristic vector based on the edge activation function to obtain the rheological stress of the metal material under the isothermal strain measurement; Wherein, N is determined by the characteristic vector.

[0008] In some embodiments, the determining the edge activation function based on the temperature value of the isothermal strain measurement and the iso-stress temperature value comprises: In the case that the temperature value of the isothermal strain measurement is less than the iso-stress temperature value, adopting a 3-order B-spline basis function as the edge activation function; In the case that the temperature value of the isothermal strain measurement is greater than or equal to the iso-stress temperature value, adopting a 3-order Gaussian radial basis function as the edge activation function.

[0009] In some embodiments, after the rheological stress prediction of the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement by using the improved KAN hybrid perception layer, the method further comprises: In response to identifying an extreme physical field parameter, activating the extreme working condition correction layer of the improved KAN; wherein the extreme physical field parameter is a parameter whose value exceeds a preset normal value range in the plurality of physical field parameters; Using a residual network in the extreme working condition correction layer of the improved KAN to quantify the difference between the value of the extreme physical field parameter and the normal value range, and the influence of the difference on the rheological stress, to obtain a correction value; Using the correction value to correct the rheological stress to obtain the corrected rheological stress.

[0010] In some embodiments, the construction process of the improved KAN comprises: Obtaining a pre-training set and a new material training set; wherein the pre-training set comprises: a plurality of typical metal materials and the rheological stress of each typical metal material under isothermal strain measurement; the new material training set comprises: a plurality of new metal materials and the rheological stress of each new metal material under isothermal strain measurement; the plurality of typical metal materials are a plurality of benchmark metal materials with a clear composition range and a rheological property data quantity greater than a preset threshold; the plurality of new metal materials are a plurality of metal materials containing new alloy elements and having a rheological property data quantity less than the preset threshold; Using the pre-training set to train the hybrid perception layer of the Kolmogorov-Arnold network (KAN) to obtain an intermediate KAN; Using the new material training set to fine-tune the material characteristic embedding layer and the multi-physical field attention layer of the intermediate KAN to obtain the improved KAN.

[0011] In a second aspect, the embodiments of the present application provide a metal rheological stress prediction system based on multi-physical field attention, which comprises: The material property coding module is configured to code material properties of a metal material by using a material property embedding layer of an improved Kolmogorov-Arnold network (KAN) to obtain a characteristic vector of the metal material. The weighted fusion module is configured to fuse multiple physical field parameters corresponding to the isothermal strain measurement by using a multi-physical field attention layer of the improved KAN to obtain a high-order feature vector. The rheological stress prediction module is configured to predict the rheological stress of the metal material under the isothermal strain measurement by using a hybrid perception layer of the improved KAN to obtain the high-order feature vector and the characteristic vector based on a temperature value of the isothermal strain measurement.

[0012] The technical scheme provided by the embodiments of the present application has at least the following beneficial effects: In the method execution process, first, the material property embedding layer of the improved KAN is used to code the material properties of the metal material by using the material composition and the microstructure parameters to obtain the characteristic vector of the metal material, that is, the material property embedding layer of the improved KAN is used to nonlinearly map the material composition and the microstructure parameters of the metal material into a highly condensed and reusable characteristic vector, which provides strong parameter support for the subsequent rheological stress prediction; then, the multi-physical field attention layer of the improved KAN is used to fuse the multiple physical field parameters corresponding to the isothermal strain measurement to obtain the high-order feature vector; that is, the multi-physical field attention layer of the improved KAN is used to adaptively weight and fuse the key physical field parameters in the isothermal strain measurement process, which can learn the contribution of different physical field parameters to the flow stress under a specific deformation condition, so as to extract the high-order feature vector that best represents the current thermodynamic state, thereby greatly enhancing the ability of the improved KAN to capture complex nonlinear coupling effects; finally, the hybrid perception layer of the improved KAN is used to predict the rheological stress of the metal material under the isothermal strain measurement based on the temperature value of the isothermal strain measurement to obtain the rheological stress of the metal material under the isothermal strain measurement. In this way, the hybrid perception layer of the improved KAN is used to intelligently integrate the "characteristic vector" of the material itself and the "high-order feature vector" of the deformation process guided by the temperature value to accurately predict the rheological stress. Therefore, based on the intelligent prediction framework of the improved KAN that provides "material perception" and "process perception", the accuracy of the rheological stress prediction can be significantly improved. At the same time, the improved KAN has inherent interpretability and learning ability for multi-physical field coupling mechanism, which can provide a powerful digital tool and deep physical insight for accelerating new material research and development and optimizing extreme manufacturing processes.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only and are not restrictive of the technical solutions provided by the embodiments of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings. Figure 1 A flowchart of a multi-physical field attention metal rheological stress prediction method provided by the embodiments of the present application; Figure 2 A schematic diagram of a hybrid activation mechanism using a combination of "3-order B-spline basis function + 3-order Gaussian radial basis function" provided by the embodiments of the present application; Figure 3 A schematic diagram of an improved KAN architecture provided by the embodiments of the present application; Figure 4 A schematic diagram of a multi-physical field attention metal rheological stress prediction system provided by the embodiments of the present application; Figure 5 A schematic diagram of a composition of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0015] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments of the present application. The following embodiments are used to illustrate the present application, but not to limit the scope of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the scope of protection of the present application.

[0016] In the following description, "some embodiments" are described, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict.

[0017] It should be noted that the terms "first", "second", "third" involved in the embodiments of the present application are only to distinguish similar objects, and do not represent a specific order of the objects. It can be understood that "first", "second", "third" can be interchanged in a specific order or sequence as allowed, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0018] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which embodiments belong. It should also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0019] Embodiment one: Referring to Figure 1 As shown in the figure, a flowchart of a multi-physical field attention metal rheological stress prediction method provided by the embodiments of the application is shown, wherein: Step 101, using a material property embedding layer of an improved Kolmogorov-Arnold network KAN, the material composition and microstructure parameters of the metal material are coded to obtain the characteristic vector of the metal material.

[0020] In some embodiments of the application, on the basis of the known existing network KAN, relevant network layers or modified parameters are added to obtain the improved Kolmogorov-Arnold network KAN; wherein the improved KAN can use a multi-module collaborative architecture to predict the rheological stress. For example, in some embodiments of the application, the improved KAN can be composed of: a material property embedding layer (coding layer), a multi-physical field attention layer (weighted fusion layer), a hybrid perception layer (rheological stress prediction layer), and an extreme working condition correction layer.

[0021] It should be noted that the core concept of KAN is based on the Kolmogorov-Arnold representation theorem, which states that any multivariate continuous function can be represented as a finite combination of univariate functions. In simple terms, any complex multi-dimensional relationship can be accurately expressed by skillfully combining a series of one-dimensional functions.

[0022] In some embodiments of the application, the material composition of the metal material can be represented by the content of its internal components, such as the content of C, Si, and Mn; at the same time, the microstructure parameters of the metal material can be represented by grain size, phase composition, precipitate distribution, and microstructure morphology. The application does not make specific limitations on this.

[0023] In some embodiments of the application, the material composition and microstructure parameters of the metal material are input into the material property embedding layer of the improved KAN for material property coding; wherein the material property embedding layer can perform corresponding material property coding through 1-dimensional convolution and residual network, that is, the characteristic vector of the metal material can be obtained through the following formula (1) : Formula (1); in, This is the original vector of material properties corresponding to the material composition and microstructure parameters of the metallic material; The weights are for 1-dimensional (1D) convolutions, which can be determined through the training process of the improved KAN. To Perform residual connection calculations; To use weights 1D convolutional layers for input Perform feature extraction and transformation; It is a linear rectifier function (activation function).

[0024] In this way, by using the improved KAN material property embedding layer as an encoder, the material composition and microstructure parameters of the metallic material are taken as input and encoded through a learnable function, that is, the material specificity is converted into a reusable property vector, thereby realizing cross-material parameter transfer.

[0025] Step 102: Using the improved KAN multiphysics attention layer, the multiple physical field parameters corresponding to the isothermal strain measurement are weighted and fused to obtain a high-order feature vector.

[0026] In some embodiments of this application, isothermal strain measurement refers to the process of accurately measuring the deformation (i.e., strain) of a material (such as the metallic material involved in this application) under stress (loading) conditions at a constant temperature. The temperature in isothermal strain measurement is a specific value, such as -50℃, 25℃, 200℃, etc.

[0027] In some embodiments of this application, the multiple physical field parameters corresponding to isothermal strain measurement may include: temperature (T), strain (T), and strain (T). and strain rate ( Correspondingly, step 102 above can be implemented through steps 1021 and 1022. Figure 1 (Not yet realized in China) Step 1021: Using the two-layer perceptron of the improved KAN multiphysics attention layer, the importance of temperature, strain and strain rate in predicting rheological stress is weighted and assigned accordingly to obtain temperature weight, strain weight and rate weight.

[0028] In some embodiments of the present application, the two-layer perceptron of the improved multi-physical field attention layer of KAN generally refers to a feedforward neural network with a single hidden layer, which is the core module of the attention mechanism of the multi-physical field attention layer and is mainly used to learn complex and nonlinear interaction relationships.

[0029] In some embodiments of the present application, the weight distribution theory of the two-layer perceptron of the improved multi-physical field attention layer of KAN can be implemented based on the information entropy principle, i.e., the importance index of each physical field parameter is defined based on the information entropy principle, thereby realizing the quantification of the influence of each physical field parameter on the rheological stress, and further providing strong theoretical support for the attention mechanism of the multi-physical field attention layer.

[0030] In some embodiments of the present application, the two-layer perceptron of the improved multi-physical field attention layer of KAN is first used to quantize the importance index of temperature, strain and strain rate based on the information entropy function, i.e., the importance of temperature in predicting the rheological stress , the importance of strain in predicting the rheological stress , and the importance of strain rate in predicting the rheological stress ; then, the weight , of the two-layer perceptron of the multi-physical field attention layer is used for weight distribution, and the temperature weight , the strain weight , and the rate weight are obtained, as shown in the following formula (2): Formula (2); wherein, is an activation function for outputting attention weights, is a function of the original correlation strength of , , , , by using the weight .

[0031] In some embodiments of the present application, the temperature weight , the strain weight , and the rate weight satisfy the following formula (3): Formula (3).

[0032] Step 1022, fusing the first product between the temperature and the temperature weight, the second product between the strain and the strain weight, and the third product between the strain rate and the rate weight to obtain the high-order feature vector.

[0033] In some embodiments of the present application, a plurality of physical field parameters and corresponding weights can be fused or spliced to form a high-order feature vector That is, as shown in the following formula (4): Formula (4).

[0034] In this way, the improved KAN multi-physical field attention layer, i.e., the improved KAN relationship modeler, which handles the complex nonlinear relationship between a plurality of physical field parameters (strain, strain, and strain rate, etc.) by means of the "attention mechanism" within the two-layer perception machine, can learn the specific function form of the mutual coupling between the strain and the strain rate at a fixed temperature, and accordingly perform weighting (attention score) to achieve a more physically intuitive "weighted fusion", thereby obtaining a more accurate high-order feature vector.

[0035] Step 103, using the improved KAN hybrid perception layer to perform rheological stress prediction on the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement to obtain the rheological stress of the metal material under the isothermal strain measurement.

[0036] In some embodiments of the present application, the improved KAN hybrid perception layer, i.e., as the main body of the improved KAN, based on the temperature value of the isothermal strain measurement, comprehensively analyzes the characteristic vector output by the material characteristic embedding layer and the high-order feature vector output by the multi-physical field attention layer, i.e., performs rheological stress prediction, to obtain the rheological stress of the metal material under the isothermal strain measurement.

[0037] In some embodiments of the present application, the rheological stress refers to the true stress required by the metal material to maintain its continuous deformation (flow) during plastic deformation.

[0038] Here, the temperature value of the isothermal strain measurement is a fixed value.

[0039] In some embodiments of the present application, before using the improved KAN hybrid perception layer to perform rheological stress prediction on the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement, the number of nodes of the improved KAN hybrid perception layer can be dynamically adjusted based on the type of the metal material represented by the characteristic vector, such as: when the metal material is a high-alloy steel, the number of nodes of the improved KAN hybrid perception layer can be determined as 30; when the metal material is an aluminum alloy, the number of nodes of the improved KAN hybrid perception layer can be determined as 20.

[0040] It should be noted that the number of nodes of the improved KAN mixed perception layer changes, and the width of the improved KAN mixed perception layer changes accordingly.

[0041] In some embodiments of the present application, before performing the above-mentioned step 103, the present application can also perform the following step A: Step A, using a material composition query database to determine the temperature value corresponding to the material composition of the metal material.

[0042] In some embodiments of the present application, the temperature value corresponds to the temperature point at which the grain strength of the metal material is equal to its grain boundary strength.

[0043] In some embodiments of the present application, the material composition query database can be a known database, which internally stores the temperature value corresponding to the material composition of each of a plurality of different metal materials.

[0044] Here, after performing the above-mentioned step A, the above-mentioned step 103 can be implemented by the following steps 1031 and 1032 (not shown): Figure 1 Step 1031, determining the edge activation function based on the temperature value of the isothermal strain measurement and the temperature value.

[0045] In some embodiments of the present application, the ratio between the temperature value of the isothermal strain measurement and the temperature value can be used to select an appropriate edge activation function.

[0046] It should be noted that the edge activation function is the core of the improved KAN, which is a one-dimensional function defined on the network connection edge of the KAN.

[0047] In some embodiments of the present application, the above-mentioned step 1031 can be implemented in the following two ways: Method one: in the case where the temperature value of the isothermal strain measurement is less than the temperature value, using a 3-order B-spline basis function as the edge activation function.

[0048] Method two: in the case where the temperature value of the isothermal strain measurement is greater than or equal to the temperature value, using a 3-order Gaussian radial basis function as the edge activation function.

[0049] ​In some embodiments of the present application, in the improved KAN mixed perception layer, a mixed activation mechanism can be adopted, that is, a dynamic edge activation function is selected, which is a 3-order B-spline basis function when the temperature value of the isothermal strain measurement is less than the low temperature of the equal strength temperature value, that is, the metal material is in the low temperature section, and which is a 3-order Gaussian radial basis function when the temperature value of the isothermal strain measurement is greater than or equal to the low temperature of the equal strength temperature value, that is, the metal material is in the high temperature section.

[0050] Here, see Figure 2 As shown in the figure, a mixed activation mechanism provided by the embodiment of the present application is combined with "3-order B-spline basis function + 3-order Gaussian radial basis function"; wherein k=1, k=2 and k=3 represent the order respectively. At the same time, g is the gating parameter of the improved KAN mixed perception layer, which is used to realize the automatic selection ability of the improved KAN mixed perception layer. As shown in the figure, Figure 2 As shown in the figure, when (1-g) corresponds to the low temperature section, the 3-order B-spline basis function is selected, and when g corresponds to the high temperature section, the 3-order Gaussian radial basis function is selected.

[0051] It should be noted that Figure 2 "3-order B-spline basis function + 3-order Gaussian radial basis function" shown in the figure also corresponds to the three hidden layers of the improved KAN mixed perception layer.

[0052] Here, the B-spline basis function is a local, piecewise polynomial function, that is, a function for fitting a smooth function, and the Gaussian radial basis function is a function for capturing local sharp features.

[0053] In this way, through the mixed activation mechanism of "3-order B-spline basis function + 3-order Gaussian radial basis function" combination, a complex function (B-spline basis function + Gaussian radial basis function) containing nonlinear transformation of different scale features (smooth and sharp) is fitted to form, so as to improve the expressiveness and flexibility of the improved KAN in predicting rheological stress.

[0054] Step 1032, using the N nodes in the improved KAN mixed perception layer that match the characteristic vector, based on the edge activation function, performing nonlinear transformation on the high-order feature vector and the characteristic vector to obtain the rheological stress of the metal material under the isothermal strain measurement.

[0055] Wherein, N is determined by the characteristic vector.

[0056] In some embodiments of the present application, N nodes in the improved KAN mixed perception layer are adopted, and based on a determination of an edge activation function (3-order B-spline basis function or 3-order Gaussian radial basis function), a high-order feature vector and a characteristic vector are simultaneously subjected to a nonlinear transformation to realize prediction of the flow stress of a metal material under isothermal strain measurement.

[0057] In this way, first, the key temperature value of the metal material is obtained according to the metal material composition, that is, the prior knowledge is deeply embedded in the neural network structure, that is, the improved KAN, realizing deep fusion of data driving and physical mechanism; then, by comparing the temperature value of the current isothermal strain measurement with the temperature value of the isothermal strain measurement, the activation function most suitable for the deformation physical mechanism in the temperature interval is intelligently selected, that is, the 3-order B-spline basis function is selected in the low-temperature section to fit the smooth deformation dominated by the crystal in the metal material, and the 3-order Gaussian radial basis function is selected in the high-temperature section to capture the localized deformation dominated by the crystal boundary in the metal material; in this way, the improved KAN can perceive the physical state (temperature relative to the isothermal temperature) of the metal material, and adjust the edge activation function in the improved KAN accordingly, thereby improving the accuracy, interpretability and generalization ability of the improved KAN for predicting the flow stress.

[0058] In some embodiments of the present application, after performing the above step 103, the embodiments of the present application can further perform the following steps B1 to step B3: Step B1, in response to identifying the extreme physical field parameter, activating the extreme working condition correction layer of the improved KAN.

[0059] Among them, the extreme physical field parameter is a parameter whose value exceeds the preset normal value range in the plurality of physical field parameters.

[0060] Step B3, using the residual network in the extreme working condition correction layer of the improved KAN, quantifying the difference between the value of the extreme physical field parameter and the normal value range, and the influence on the flow stress, to obtain a correction value.

[0061] Step B3, using the correction value to correct the flow stress, to obtain the corrected flow stress.

[0062] In some embodiments of the present application, when there are parameters exceeding the normal range in the plurality of physical field parameters under isothermal strain measurement, that is, the extreme physical field parameter (such as T>1200℃, or, >100s -1) can be activated, and correspondingly, after the extreme condition correction layer is activated, it can identify the error between the extreme physical field parameter and the corresponding parameter value under the normal condition (i.e. under the normal isothermal strain measurement). Here, the residual network in the extreme condition correction layer can be used to learn the error between the extreme physical field parameter and the normal value range , and then use the correct the rheological stress output by the hybrid perception layer of the improved KAN, so as to obtain the corrected rheological stress.

[0063] It should be noted that if there is a parameter that exceeds the normal range among the multiple physical field parameters under the isothermal strain measurement, it is defaulted that the current isothermal strain measurement is an extreme condition.

[0064] In this way, by means of the extreme condition correction layer of the improved KAN, a correction mechanism is provided in the rheological stress prediction link, which can further improve the accuracy of the improved KAN in predicting the rheological stress.

[0065] In some embodiments of the present application, the construction process of the improved KAN can be implemented in the following steps C1 to C3: Step C1, obtaining a pre-training set and a new material training set.

[0066] The pre-training set includes: a plurality of typical metal materials and the rheological stress of each typical metal material under isothermal strain measurement; the new material training set includes: a plurality of new metal materials and the rheological stress of each new metal material under isothermal strain measurement; the plurality of typical metal materials are a plurality of reference metal materials with a clear composition range and a rheological property data amount greater than a preset threshold; the plurality of new metal materials are a plurality of metal materials containing new alloy elements and having a rheological property data amount less than the preset threshold.

[0067] Step C2, training the hybrid perception layer of the Kolmogorov-Arnold network KAN using the pre-training set to obtain an intermediate KAN.

[0068] Step C3, fine-tuning the material characteristic embedding layer and the multi-physical field attention layer of the intermediate KAN using the new material training set to obtain the improved KAN.

[0069] In some embodiments of the present application, the plurality of typical metal materials in the pre-training set can be: a plurality of mature (conventional) metal materials that have been industrialized and applied, or a plurality of benchmark metal materials with a clear composition range and a rheological property data amount greater than a preset threshold, i.e., a sufficient amount of rheological property data, such as carbon steel, aluminum alloy, titanium alloy, etc. At the same time, the plurality of new metal materials in the new material training set can be: a plurality of metal materials containing new alloy elements and having a rheological property data amount less than a preset threshold, i.e., insufficient or limited amount of rheological property data. Here, the plurality of new metal materials can also be metal materials containing more than a preset proportion of new alloy elements or having special microstructures (microstructures with abnormal grain size, special texture, or unconventional phase composition, etc.), or high-performance metal materials in the research and development stage or special application fields, such as nickel-based alloys, aluminum-based alloys, etc.

[0070] In some embodiments of the present application, the preset threshold can be determined according to actual conditions, which is not limited in the present application.

[0071] In some embodiments of the present application, first, the mixed perception layer (main part of KAN) of KAN is trained using the pre-training set (a plurality of typical metal materials with a rheological property data amount greater than a preset threshold, i.e., a sufficient amount of rheological property data), i.e., training the basic ability of KAN, to obtain an intermediate KAN; then, the material property embedding layer and the multi-physical field attention layer (which can also include an extreme working condition correction layer) of the intermediate KAN are fine-tuned using the new material training set (a plurality of new metal materials with a rheological property data amount less than a preset threshold, i.e., a limited but clear amount of rheological property data), i.e., for fine-tuning the adaptability of the intermediate KAN. That is, the basic ability of KAN is trained using a plurality of typical metal materials to obtain an intermediate KAN, and the adaptability of KAN (intermediate KAN) is fine-tuned using a plurality of new metal materials.

[0072] That is, in the process of constructing an improved KAN based on a known KAN, the following two stages are involved: 1. Cross-material transfer learning strategy - pre-training stage: training the basic model, i.e., KAN, using the pre-training set, such as data corresponding to 5 typical metal materials (carbon steel, aluminum alloy, titanium alloy, etc.), and freezing part of the shared parameters (such as the weights of the multi-physical field attention layer and the parameters of the material property embedding layer) during training to obtain an intermediate KAN.

[0073] 2. Fine-tuning stage: using the new material training set to update only the material property embedding layer, the multi-physical field attention layer, and the extreme working condition correction layer of the intermediate KAN. In this way, the data amount in the fine-tuning stage is reduced to 40% of the original scheme (i.e., the scheme of fine-tuning all network layers of the intermediate KAN).

[0074] In addition, in the process of constructing the improved KAN, an online incremental learning mechanism, a sliding window, can be involved, such as caching the latest 500 sets of production data in real time to form an incremental data set D_inc, and updating the parameters of the improved KAN; in the parameter updating stage, the "freeze-fine-tune" mode is also used for sampling, that is, in the pre-training stage, the material property embedding layer, the multi-physical field attention layer and the extreme working condition correction layer of the improved KAN are frozen, and only 30% of the edge activation parameters of the main body of the improved KAN, i.e., the hybrid perception layer, are updated, so that the time consumption of each update is less than 5 minutes.

[0075] It should be noted that in the process of constructing the improved KAN, i.e., in the pre-training and fine-tuning stage of KAN, the multi-objective loss function shown in the following formula (5) can be used as the loss function for training and fine-tuning: Formula (5); wherein, L_total is the total loss, MSE is the mean squared error (MSE), L_att is the attention weight coefficient constraint of the multi-physical field attention layer, which is used to encourage the weight of the key physical field parameter to be larger; L_trans is the cross-material parameter transfer loss of the material property embedding layer, which is used to minimize the difference between the new and old material parameters, L_extreme is the extreme working condition error penalty of the extreme working condition correction layer, which is used to give higher weight to the out-of-range data; , , are all empirical values.

[0076] In this way, with the known multiple typical metal materials and their sufficient rheological stress data as the pre-training set, and with a small amount of multiple new metal materials and their limited rheological stress data as the new material training set, on the basis of the existing KAN, through the "pre-training + fine-tuning" two-stage training strategy, the core mapping layer of the KAN trained by the data-rich benchmark materials (multiple typical metal materials) can establish the general physical law, and the characteristic adaptation layer of the KAN can be fine-tuned by a small amount of new material (multiple new metal materials) data, thereby realizing the efficient combination of knowledge transfer and small sample learning, significantly improving the prediction accuracy and generalization ability of the improved KAN constructed for new metal materials, and greatly reducing the experimental cost and research and development period.

[0077] The metal flow stress prediction based on multi-physical field attention provided by the embodiments of the application first adopts the improved KAN material property embedding layer to perform material property coding on the material composition and microstructure parameters of the metal material to obtain a characteristic vector of the metal material, that is, the improved KAN material property embedding layer is used to nonlinearly map the material composition and microstructure parameters of the metal material into a highly condensed and reusable characteristic vector, which provides strong parameter support for subsequent flow stress prediction; then the improved KAN multi-physical field attention layer is used to perform weighted fusion on a plurality of physical field parameters corresponding to isothermal strain measurement to obtain a high-order feature vector; that is, the improved KAN multi-physical field attention layer adaptively performs importance weighting and fusion on key physical field parameters in the isothermal strain measurement process, which can learn the contribution degree of different physical field parameters to the flow stress under a specific deformation condition, so as to extract a high-order feature vector that best represents the current thermodynamic state, thereby greatly enhancing the ability of the improved KAN to capture complex nonlinear coupling effects; finally, the improved KAN hybrid perception layer is used to perform flow stress prediction on the high-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement to obtain the flow stress of the metal material under the isothermal strain measurement. In this way, the improved KAN hybrid perception layer is used to guide the temperature value to intelligently integrate the “characteristic vector” of the material itself and the “high-order feature vector” of the deformation process to complete accurate prediction of the flow stress. In this way, on the basis of the intelligent prediction framework provided by the improved KAN, which provides deep fusion of “material perception” and “process perception”, the accuracy of flow stress prediction can be significantly improved. At the same time, the improved KAN has inherent interpretability and learning ability for multi-physical field coupling mechanisms, which can provide a powerful digital tool and deep physical insight for accelerating new material research and development and optimizing extreme manufacturing processes.

[0078] As a person skilled in the art, it should be known that the original KAN model, MLP, Arrhenius and the like in the prior art have the following key limitations in metal material flow stress prediction: 1. Multi-physical field coupling blind area: the original KAN model regards temperature, strain and strain rate as equal inputs and does not distinguish the influence weight of different parameters under different working conditions (for example, the influence of temperature on softening behavior is much greater than that of strain at high temperature), which leads to insufficient capture of key factors under complex working conditions.

[0079] 2. Material specificity limitation: the model in the prior art is designed only for a single material, and when predicting new materials (such as from carbon steel to aluminum alloy), the entire model needs to be retrained, which cannot utilize the knowledge transfer of existing materials and has poor adaptability.

[0080] 3. Dynamic scene hysteresis: Industrial field data is generated in real time (such as batch material composition fluctuations, equipment parameter drift), and the model in the prior art needs to be retrained offline with full data, and cannot update parameters online, making it difficult to cope with dynamic production environments.

[0081] 4. Insufficient robustness in extreme conditions: Under extreme conditions such as ultra-high temperature (temperature > 1200℃), rapid cooling and heating (temperature change rate > 50℃ / s), the model in the prior art lacks a targeted correction mechanism, and the prediction error increases by more than 30% compared to normal conditions.

[0082] Therefore, in the metal rheological stress prediction method based on multi-physical field attention provided by the embodiments of the present application, an improved KAN is disclosed, which adopts a four-module collaborative architecture, and three core modules are added to the existing known KAN, namely: material property embedding layer, multi-physical field attention layer, and extreme condition correction layer. Correspondingly, the architecture of the improved KAN can be referred to as shown in Figure 3 The improved KAN mainly consists of four network layers, which are: 1. Material property embedding layer, used to receive material composition (including the content of each element) and microstructure parameters (such as grain size, phase composition, precipitate distribution, and microstructure morphology, etc.) of metal materials, and to process the received material composition and microstructure parameters through internal residual modules, 1D convolution (Conv1D), and linear rectifier function to form reusable property vectors .

[0083] 2. Multi-physical field attention layer, used to receive multiple physical field parameters corresponding to isothermal strain measurement, namely: temperature (T), strain ( ), and strain rate ( ), and to perform importance weighting and fusion through an internal attention layer, i.e., two perception machines, to obtain high-order feature vectors.

[0084] 3. Hybrid perception layer, which is the main body of the improved KAN, which dynamically adjusts the activation function required by the improved KAN (3-order Gaussian radial basis function for high-temperature segment, 3-order B-spline basis function for low-temperature segment) through a hybrid activation mechanism combining "3-order B-spline basis function + 3-order Gaussian radial basis function", and performs linear transformation on the property vectors output by the material property embedding layer and the high-order feature vectors output by the multi-physical field attention layer to obtain the rheological stress of the metal material under isothermal strain measurement.

[0085] 4. Extreme condition correction layer, which needs to be activated under relevant conditions, i.e., when there are parameters that exceed the normal range among the multiple physical field parameters under isothermal strain measurement, i.e., when there are extreme physical field parameters (such as T> 1200℃, or, > 100 s -1 ) can be activated, and the residual network in the extreme condition correction layer can be used to learn the error between the extreme physical field parameters and the normal numerical range , and the rheological stress output by the mixed perception layer of the improved KAN is corrected to obtain the corrected rheological stress.

[0086] Thus, based on the Kolmogorov-Arnold theorem, the present application provides an adaptive KAN based on multi-physical field attention and cross-material migration (corresponding to the improved KAN provided in the embodiments of the present application), which can break through the limitations of traditional models in multi-physical field coupling analysis, cross-material adaptability, dynamic scene response, and extreme condition robustness. By constructing an enhanced KAN that integrates multi-dimensional intelligent mechanisms, high-precision and high-adaptability prediction of metal material rheological stress is achieved.

[0087] In addition, in the embodiments of the present application, the improved KAN disclosed in the present application is compared with the original KAN, MLP, Arrhenius model, and ablation models such as “original KAN + multi-physical field attention layer” and “original KAN + material property embedding layer” in the following test scenarios in terms of performance (update time consumption, cross-material prediction accuracy, MSE, and average absolute relative error (AARE) of each model): Test scenario 1: cross-material prediction, trained with carbon steel data and predicted with aluminum alloy rheological stress; Test scenario 2: dynamic update test, simulating 100 new data sets per hour in an industrial site to evaluate the real-time accuracy of each model; Test scenario 3: extreme condition test, such as prediction error under the condition of 1250℃ high temperature, = 200 s -1

[0088] The corresponding specific experimental data are as follows: In test scenario 1, i.e., cross-material adaptability enhancement-rheological stress prediction of new materials, the training data amount of the improved KAN provided in the present application is reduced by 60% compared with the original KAN, and the prediction accuracy is as high as 92% (the original KAN needs to be retrained, and the accuracy is only 78%).

[0089] In test scenario 2, the MSE of the improved KAN provided in the present application is reduced by 28% compared with the original KAN, and the AARE is reduced to 3.2%.

[0090] ​​In the test scenario 3, i.e. extreme working condition, the AARE of the improved KAN provided by the application at 1250 DEG C is reduced by 42% compared with the original KAN, which is better than MLP (reduced by 65% compared with MLP).

[0091] Therefore, the metal flow stress prediction method based on multi-physical field attention provided by the embodiments of the application can automatically identify key influence parameters under different working conditions by designing a multi-physical field attention mechanism (a multi-physical field attention layer) in the improved KAN, thereby improving the prediction accuracy of a complex coupling scene by more than 15%; the material property embedding layer is introduced to realize cross-material knowledge transfer, so that the training efficiency of the improved KAN is improved by more than 60% when predicting a new material; and the extreme working condition correction layer is constructed to reduce the prediction error of the improved KAN under extreme conditions by more than 40%. At the same time, the application can further develop an online incremental learning unit to support real-time updating of dynamic data and shorten the parameter adjustment response time to the minute level.

[0092] That is, the metal flow stress prediction method based on multi-physical field attention provided by the embodiments of the application is based on the following theories: 1. Multi-physical field weight distribution theory, which can define the importance index of multiple physical field parameters based on the information entropy principle, thereby further quantifying the influence weight of the physical field parameters on the flow stress by means of the importance index, and providing theoretical support for the attention mechanism.

[0093] 2. Material knowledge transfer theory: a cross-material shared feature space is constructed through the mapping relationship between material properties (composition, microstructure) and flow behavior, so that the model parameters of different materials can be reused (for example, the sharing degree of high-temperature softening rules of aluminum alloy and magnesium alloy is 70%).

[0094] Based on the above theories, the improved KAN of the embodiments of the application has the following innovation points: 1. Architecture innovation point, i.e. the fusion architecture of the multi-physical field attention layer and the KAN: the physical field parameters are differentially processed through dynamic weight distribution, which is different from the equal input mode of the original KAN; material property embedding and cross-material transfer mechanism: the material properties are encoded into reusable vectors, breaking through the single material limitation of traditional models.

[0095] 2. Method innovation point - parameter updating strategy of online incremental learning: the "freeze-fine-tune" mode is adopted to realize dynamic data adaptation, which is different from the offline full-volume retraining in the prior art; extreme working condition self-adaptive correction mechanism: the residual network is used to compensate the error under extreme conditions, solving the problem of insufficient robustness of the conventional model.

[0096] Based on the same inventive concept, the embodiment of the present application further provides a metal rheological stress prediction system based on multi-physical field attention, which is shown in Figure 4 The metal rheological stress prediction system 400 based on multi-physical field attention includes: A material property coding module 401 is configured to code material properties of a metal material by using a material property embedding layer of an improved Kolmogorov-Arnold network (KAN), to obtain a characteristic vector of the metal material. A weighted fusion module 402 is configured to fuse a plurality of physical field parameters corresponding to isothermal strain measurement by using a multi-physical field attention layer of the improved KAN, to obtain a high-order feature vector. A rheological stress prediction module 403 is configured to predict rheological stress of the high-order feature vector and the characteristic vector based on a temperature value of the isothermal strain measurement by using a hybrid perception layer of the improved KAN, to obtain rheological stress of the metal material under the isothermal strain measurement.

[0097] It should be noted that the above system embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the system embodiment of the present application, please refer to the description of the method embodiment of the present application.

[0098] It should be noted that in the embodiment of the present application, if the above-mentioned metal rheological stress prediction method based on multi-physical field attention is realized in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing an electronic device to execute all or part of the method described in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk, and various storage media that can store program codes. Thus, the embodiments of the present application are not limited to any specific hardware and software combination.

[0099] Correspondingly, the embodiment of the present application provides a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the metal rheological stress prediction method based on multi-physical field attention in any of the above embodiments. Correspondingly, the embodiment of the present application further provides a computer program product for implementing the steps of the metal rheological stress prediction method based on multi-physical field attention in any of the above embodiments when the computer program product is executed by a processor of an electronic device.

[0100] Based on the same technical concept, the embodiment of the present application provides an electronic device for implementing the metal flow stress prediction method based on multi-physical field attention described in the above method embodiment. Figure 5 The hardware entity schematic diagram of the electronic device provided by the embodiment of the present application is shown in Figure 5 As shown, the electronic device 500 includes a memory 510 and a processor 520, the memory 510 stores a computer program executable on the processor 520, and the processor 520 implements the steps in the metal flow stress prediction method based on multi-physical field attention of any of the embodiments of the present application when executing the program.

[0101] The memory 510 is configured to store instructions and applications executable by the processor 520, and can also cache data to be processed by the processor 520 and each module in the electronic device (for example, image data, audio data, voice communication data and video communication data), which can be realized by FLASH or RAM.

[0102] The processor 520 implements the steps of the metal flow stress prediction method based on multi-physical field attention of any of the above when executing the program. The processor 520 generally controls the overall operation of the electronic device 500.

[0103] The above processor can be at least one of an ASIC, a DSP, a DSPD, a PLD, an FPGA, a CPU, a controller, a microcontroller, and a microprocessor. It can be understood that the electronic device realizing the function of the above processor can also be other, and the embodiment of the present application does not make specific limitation.

[0104] The computer storage medium / memory can be a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Ferromagnetic Random Access Memory (FRAM), a Flash Memory, a magnetic surface storage, an optical disc, a Compact Disc Read-Only Memory (CD-ROM), or the like memory; or can be various electronic devices including one or any combination of the above memories, such as a mobile phone, a computer, a tablet device, a personal digital assistant, and the like.

[0105] It should be noted that the above description of the storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding.

[0106] It should be understood that the "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that the size of the sequence number of each process in various embodiments of the present application does not mean the execution order, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The sequence number of the above embodiments of the present application is only for description, not representing the advantages and disadvantages of the embodiments.

[0107] It should be noted that, in the present document, the terms "comprising", "containing", or any other similar term are intended to encompass non-exclusive inclusions, such that a process, method, article, or apparatus that comprises a list of elements does not necessarily include those elements only, but can include other elements not expressly listed, or can include elements inherent in such process, method, article, or apparatus. Without more limitations, an element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0108] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative, for example, the division of the units is only a logical functional division, and actual implementation can have another division manner, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be through some interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0109] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units; they can be located in one place or distributed on multiple network units; and part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0110] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be realized in the form of hardware or in the form of hardware plus software functional unit.

[0111] Alternatively, the integrated unit of the present application, if implemented in the form of a software function module and sold or used as an independent product, can also be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing an apparatus to perform all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes: mobile storage devices, ROM, magnetic or optical disks, and various other media that can store program codes.

[0112] The methods disclosed in the several method embodiments provided by the present application can be combined arbitrarily without conflict to obtain new method embodiments.

[0113] The features disclosed in the several method or device embodiments provided by the present application can be combined arbitrarily without conflict to obtain new method embodiments or device embodiments.

[0114] The above merely provides the implementation manners of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for predicting rheological stress in metals based on multiphysics attention, characterized in that, The method includes: A modified Kolmogorov-Arnold network (KAN) material property embedding layer is used to encode the material composition and microstructure parameters of the metallic material to obtain the property vector of the metallic material. An improved KAN multiphysics attention layer is used to weight and fuse multiple physical field parameters corresponding to isothermal strain measurements to obtain a high-order feature vector. Using the improved KAN hybrid sensing layer, based on the temperature value of the isothermal strain measurement, the rheological stress is predicted on the higher-order feature vector and the characteristic vector to obtain the rheological stress of the metal material under the isothermal strain measurement.

2. The method according to claim 1, characterized in that, The multiple physical field parameters include: temperature, strain, and strain rate; the improved KAN multiphysics attention layer weighted and fused the multiple physical field parameters corresponding to the isothermal strain measurement to obtain a high-order feature vector, including: Using the two-layer perceptron of the improved KAN multiphysics attention layer, the importance of temperature, strain and strain rate in predicting rheological stress is weighted, and the corresponding temperature weight, strain weight and rate weight are obtained. The higher-order feature vector is obtained by fusing the first product between the temperature and the temperature weight, the second product between the strain and the strain weight, and the third product between the strain rate and the rate weight.

3. The method according to claim 1, characterized in that, The method, which employs the improved KAN hybrid sensing layer and predicts rheological stress based on the temperature value of the isothermal strain measurement, and obtains the rheological stress of the metallic material under the isothermal strain measurement, further includes: The equivalent strength temperature value corresponding to the material composition of the metal material is determined by using a material composition query database. The hybrid sensing layer employing the improved KAN, based on the temperature value measured by the isothermal strain measurement, performs rheological stress prediction on the higher-order eigenvector and the characteristic vector to obtain the rheological stress of the metallic material under the isothermal strain measurement, including: Based on the temperature values ​​measured by the isothermal strain and the isothermal temperature values, the edge activation function is determined; Using N nodes in the hybrid sensing layer of the improved KAN that match the characteristic vector, and based on the edge activation function, a nonlinear transformation is performed on the higher-order feature vector and the characteristic vector to obtain the rheological stress of the metal material under the isothermal strain measurement. Wherein, N is determined by the characteristic vector.

4. The method according to claim 3, characterized in that, The determination of the edge activation function based on the temperature value measured by the isothermal strain and the isothermal temperature value includes: When the temperature value of the isothermal strain measurement is less than the isothermal temperature value, a third-order B-spline basis function is used as the edge activation function; When the temperature value of the isothermal strain measurement is greater than or equal to the isothermal temperature value, a third-order Gaussian radial basis function is used as the edge activation function.

5. The method according to claim 1, characterized in that, The method, which employs the improved KAN hybrid sensing layer and predicts the rheological stress of the metal material under the isothermal strain measurement based on the temperature value of the isothermal strain measurement, further includes: In response to the identification of extreme physical field parameters, the extreme condition correction layer of the improved KAN is activated; wherein, the extreme physical field parameters are the parameters among the plurality of physical field parameters whose values ​​exceed a preset normal value range; The residual network in the extreme condition correction layer of the improved KAN is used to quantify the difference between the extreme physical field parameters and the normal value range, and the effect on the rheological stress to obtain the correction value. The rheological stress is corrected using the correction value to obtain the corrected rheological stress.

6. The method according to any one of claims 1 to 5, characterized in that, The construction process of the improved KAN includes: Obtain a pre-training set and a new material training set; wherein, the pre-training set includes: multiple typical metallic materials and the rheological stress of each typical metallic material under isothermal strain measurement; the new material training set includes: multiple novel metallic materials and the rheological stress of each novel metallic material under isothermal strain measurement; the multiple typical metallic materials are multiple reference metallic materials with clearly defined composition ranges and rheological property data exceeding a preset threshold; the multiple novel metallic materials are multiple metallic materials containing novel alloying elements and with rheological property data less than the preset threshold; Using the pre-training set, the hybrid perception layer of the Kolmogorov-Arnold network (KAN) is trained to obtain the intermediate KAN; Using the new material training set, the material property embedding layer and multiphysics attention layer of the intermediate KAN are fine-tuned to obtain the improved KAN.

7. A metal rheological stress prediction system based on multiphysics attention, characterized in that, The system includes: The material property encoding module is used to encode the material composition and microstructure parameters of the metallic material using the material property embedding layer of the improved Kolmogorov-Arnold network (KAN) to obtain the property vector of the metallic material. The weighted fusion module is used to perform weighted fusion of multiple physical field parameters corresponding to isothermal strain measurements using an improved KAN multiphysics attention layer to obtain a high-order feature vector. The rheological stress prediction module is used to perform rheological stress prediction on the higher-order feature vector and the characteristic vector based on the temperature value of the isothermal strain measurement using the improved KAN hybrid sensing layer, so as to obtain the rheological stress of the metal material under the isothermal strain measurement.

Images