Prediction method of microstructure and mechanical properties of alloy castings based on machine learning

Abstract

The present invention provides a method for predicting the microstructure mechanical properties of alloy castings based on machine learning, which relates to the technical field of mechanical property prediction of castings, including: obtaining multiple groups of microstructure characteristic parameters and corresponding mechanical parameters of aluminum alloy cast battery pack frames, generating a sample data set, constructing a neural network model based on the data set, taking the microstructure characteristics as input and the mechanical parameters as labels for training. The microstructure characteristic parameters of each microstructure area of ​​the aluminum alloy casting to be predicted are input into the trained model, and the predicted mechanical parameters are output to determine the initial strength-toughness index. The strength-toughness correction factor is calculated based on the tissue heterogeneity index and the defect sensitivity threshold, and the initial strength-toughness index is corrected to obtain a comprehensive strength-toughness index. The mechanical properties of each microstructure area are predicted based on the comprehensive strength-toughness index, thereby achieving an accurate evaluation of the heterogeneous mechanical properties of the automotive aluminum alloy cast battery pack frame.

Description

Technical Field

[0001] The present invention relates to the technical field of prediction of mechanical properties of castings, and in particular to a method for predicting the microstructure mechanical properties of alloy castings based on machine learning. Background Art

[0002] During the production process of aluminum alloy castings, the formation and evolution of the microstructure has a significant impact on the mechanical properties of the final product. Traditional mechanical property prediction methods are mostly based on the homogeneous material model. This method assumes that the material properties of the entire casting are uniform and cannot accurately reflect the microstructural differences that exist in actual production. This uniformity assumption may lead to large errors between the predicted results and the actual performance in many cases. Especially in the application of key components such as aluminum alloy battery pack frames, microstructural details such as dendrite secondary arm spacing, porosity and β phase area ratio have a direct impact on the mechanical properties of the material such as toughness, yield strength and elastic modulus. However, due to the lack of effective means to quantify and utilize these microstructural characteristics, many castings fail to meet the expected performance standards in actual applications, which in turn affects the safety and durability of the overall structure.

[0003] In addition, existing casting performance prediction technologies are often unable to dynamically adapt to changes in the microstructure during the casting process. These technologies usually rely on static data and empirical formulas, and lack a feedback mechanism for real-time organizational characteristics, resulting in the inability to effectively predict and correct defects and organizational heterogeneity during the casting process. This limitation not only affects the design and optimization process of materials, but also hinders the application of data-driven methods, making it impossible to achieve precise control of the performance of aluminum alloy castings. With the development of intelligent manufacturing and big data technology, there is an urgent need for a novel method that combines machine learning with microstructural characteristics to achieve data-based dynamic prediction and conduct a comprehensive and accurate analysis of the performance of castings, thereby filling the shortcomings of traditional prediction technologies.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting the microstructure and mechanical properties of alloy castings based on machine learning to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for predicting the microstructure and mechanical properties of alloy castings based on machine learning, comprising the following steps:

[0008] S1, obtain multiple sets of microstructure characteristic parameters and corresponding mechanical parameters of automotive aluminum alloy casting battery pack frames, map each set of microstructure characteristic parameters and mechanical parameters one by one, and generate a sample data set, wherein the microstructure characteristic parameters include dendrite secondary arm spacing, Phase area ratio, porosity and small-angle misorientation ratio of grain boundaries, and mechanical parameters including elastic modulus, yield strength and Poisson's ratio;

[0009] S2, based on the data in the sample data set, a neural network model is constructed, the microstructure characteristic parameters in the sample data set are used as the input of the model, and the mechanical parameters in the sample data set are used as labels to train the neural network model to obtain a mechanical parameter prediction model;

[0010] S3, based on the casting structure difference, the predicted automotive aluminum alloy casting battery pack frame is divided into microstructure regions, and the microstructure characteristic parameters of each microstructure region of the divided alloy casting are collected, the microstructure characteristic parameters including the dendrite secondary arm spacing, Phase area ratio, porosity and grain boundary small angle misorientation ratio;

[0011] S4, inputting the microstructural characteristic parameters of the alloy casting to be predicted into the trained mechanical parameter prediction model, the model outputting a predicted elastic modulus, a predicted yield strength, and a predicted Poisson's ratio, and determining an initial toughness index based on the predicted elastic modulus, the predicted yield strength, and the predicted Poisson's ratio;

[0012] S5, determining the microstructure heterogeneity index based on the standard deviation of the predicted yield strength of all microstructure regions, determining the defect sensitivity threshold based on the maximum defect size, calculating the strength-toughness correction factor based on the microstructure heterogeneity index and the defect sensitivity threshold, and using the strength-toughness correction factor to correct the initial strength-toughness index to obtain a comprehensive strength-toughness index that reflects the mechanical properties of the microstructure region;

[0013] S6, construct a finite element model based on the alloy casting to be predicted, spatially align the microstructure area of ​​the alloy casting with the model geometric topology, and perform three-dimensional meshing. Map the comprehensive strength and toughness index of each microstructure area to the grid node at the corresponding position, and determine the comprehensive strength and toughness index of other grid nodes based on the spatial distance attenuation law to complete the prediction of the mechanical properties of the entire alloy casting.

[0014] Furthermore, the method for generating the sample data set is: mapping each set of microstructure characteristic parameters and mechanical parameters one by one to form a data chain, and recording the formed data chain as the sample data set;

[0015] Based on the data in the sample data set, a neural network model is established. The long short-term memory network LSTM model is selected as the basic model. The activation function and optimization algorithm are selected. The Tanh function is selected as the activation function and Adam is selected as the optimization algorithm of the LSTM model.

[0016] The Tanh function expression is as follows:

[0017] ;

[0018] Where, Represents the Tanh function, independent variable represents the weighted sum of the neuron's inputs, that is, the result of the weighted summation of the inputs received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons; wherein the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the number of batches is set to 256, and the number of hidden layer neurons is 32;

[0019] The sample data set is randomly divided into a training set and a test set, and a mechanical parameter prediction model is constructed based on a long short-term memory network. The microstructure characteristic parameters in the training set are used as input and the mechanical parameters are used as labels to train the model to obtain a trained mechanical parameter prediction model. The microstructure characteristic parameters in the test set are substituted into the trained model to obtain the corresponding prediction results; the error between the prediction result and the actual value in the test set is calculated; it is determined whether the error meets a preset error threshold; if so, the trained model, i.e., the mechanical parameter prediction model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error, and determination coefficient between the prediction result and the actual value in the test set;

[0020] The process of randomly dividing the dataset into training and test sets is as follows: randomly sort the sample dataset, use 80% of the sorted sample dataset as the training set, and the remaining 20% ​​as the test set;

[0021] The input of the trained mechanical parameter prediction model is the microstructure characteristic parameters, and the output is the mechanical parameters.

[0022] Furthermore, the microstructure characteristic parameters include dendrite secondary arm spacing, Phase area ratio, porosity and grain boundary small angle misorientation ratio, where:

[0023] The dendrite secondary arm spacing is obtained by the metallographic sample intercept method: for an aluminum alloy sample corroded by Keller reagent, the vertical distance between the roots of adjacent secondary dendrite arms is measured under a 100× optical microscope field of view, and the arithmetic average of 30 sets of data is used as the dendrite secondary arm spacing of the sample;

[0024] described The phase area ratio was imaged using the backscattered electron mode of a scanning electron microscope. Software threshold segmentation calculation description Phase pixel ratio, and the analysis area is 5 2000× fields of view;

[0025] The porosity is obtained by three-dimensional reconstruction of X-ray tomography. The grayscale threshold is set to distinguish the matrix and the pores, and the porosity is calculated as the percentage of the pore volume to the total scan volume. The scanning resolution is ;

[0026] The small-angle misorientation ratio of the grain boundary is determined by electron backscatter diffraction technology, and the misorientation angle is set For small-angle grain boundaries, the proportion of the total length of the grain boundary is counted, and the scanning step length is .

[0027] Furthermore, the microstructure of the predicted automotive aluminum alloy casting battery pack frame is divided into different regions according to the casting structure differences. The specific logic is as follows:

[0028] Determining the orientation difference between adjacent pixels by electron backscatter diffraction orientation imaging And continuous length The area is taken as the dendrite trunk area; then the grain roundness is divided based on the gray gradient of the metallographic image. And the equivalent diameter is The area is regarded as the equiaxed crystal area; and then the energy spectrum analysis is used to identify The proportion of phase area in the phase-enriched zone is The transition zone is used as the transition zone; finally, the X-ray tomography data is combined to calibrate a spherical domain with a radius of 2.5 times the defect diameter centered on the pore as the defect-affected zone, forming four types of microstructural regions including the dendrite trunk region, equiaxed crystal region, transition region and defect-affected region.

[0029] Furthermore, the microstructure characteristic parameters of the alloy casting to be predicted are input into the trained mechanical parameter prediction model, and the model outputs the predicted elastic modulus. , predict yield strength and predicted Poisson's ratio ;

[0030] The initial toughness index is determined based on the predicted elastic modulus, predicted yield strength, and predicted Poisson's ratio according to the following formula:

[0031] ;

[0032] Where, is the initial toughness index, To predict the elastic modulus, is the reference elastic modulus, To predict the yield strength, is the reference yield strength, To predict Poisson's ratio, is the reference Poisson's ratio, is a natural constant, 、 and is the preset weight value and satisfies , The value range of is (0,1], and It is a material constant, and its specific value is determined according to the type of material.

[0033] Furthermore, the standard deviation of the predicted yield strength of all microstructure regions in the predicted automotive aluminum alloy cast battery pack frame is used as the non-uniformity index, denoted as The ratio of the maximum defect diameter in all microstructural regions of the predicted automotive aluminum alloy cast battery pack frame to the critical defect size of the material is used as the defect sensitivity threshold, denoted as ;

[0034] The strength and toughness correction factor is calculated based on the tissue heterogeneity index and the defect sensitivity threshold, according to the following formula:

[0035] ;

[0036] Where, is the toughness correction factor, is the tissue heterogeneity index, is the defect sensitivity threshold, and is the preset weight value, , and satisfies ;

[0037] The initial strength-toughness index is corrected using the strength-toughness correction factor to obtain a comprehensive strength-toughness index that reflects the mechanical properties of the microstructure region. The formula is as follows:

[0038] ;

[0039] Where, is the comprehensive toughness index, is the initial toughness index, is the toughness correction factor, is the preset weight of the toughness correction factor, and The value range is [0.2, 0.8].

[0040] Furthermore, a finite element model based on the alloy casting to be predicted is constructed, the microstructure area of ​​the alloy casting is spatially aligned with the model geometric topology, and three-dimensional meshing is performed, wherein the three-dimensional meshing is performed according to the following rules:

[0041] (1) At the junction of the dendrite region and the equiaxed crystal region, Local mesh encryption is used within the phase aggregation area and the range of 3 times the equivalent diameter around the defect, and the minimum unit size does not exceed 1 / 5 of the characteristic area size, ensuring that each microstructure partition contains at least 3 layers of mesh units;

[0042] (2) The mesh size change rate between adjacent microstructure regions is controlled within 1.2 times, and the smooth transfer of mechanical property parameters is achieved through the transition layer mesh;

[0043] (3) Based on the stress distribution preliminary analysis results of the finite element model, the bolt connection holes and weld heat-affected zones of the battery pack frame are identified as high-risk areas and double encryption is implemented, with the mesh density reaching twice that of non-critical areas;

[0044] (4) The spatial alignment error between the grid nodes and the tissue boundary identified by the metallographic image is kept less than , a curvature-adaptive hexahedron-dominated hybrid grid is used in the curved surface area;

[0045] (5) For the grid nodes in the non-microstructure area, the comprehensive strength and toughness index is determined according to the exponential decay law.

[0046] Furthermore, the comprehensive strength-toughness index of each microstructure region is mapped to the grid node at the corresponding position to obtain the microstructure region mapping node. Based on the spatial distance attenuation law, the comprehensive strength-toughness index of other grid nodes is determined according to the following formula:

[0047] ;

[0048] Where, For the The comprehensive toughness index of other grid nodes, is the index of other grid nodes, For the The comprehensive toughness index of the microstructure region mapping node, is the total number of nodes mapped in the micro-organization area, is the index of the microstructure region mapping node, is the weight coefficient, indicating the The micro-organizational region mapping node is The degree of influence of other grid nodes is related to the spatial distance;

[0049] Among them, determine The formula is as follows:

[0050] ;

[0051] Where, Mapping the 3D coordinates of nodes to the microstructure region , For the 3D coordinates of other mesh nodes ,

[0052] is a natural constant;

[0053] Based on the above steps, the comprehensive strength-toughness index of each grid node is determined to complete the prediction of the mechanical properties of the entire alloy casting.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The "machine learning-based prediction method for the microstructure mechanical properties of alloy castings" of the present invention effectively solves the limitations of the uniformity assumption in traditional prediction technologies, and significantly improves the accuracy of mechanical property prediction of castings by introducing microstructure characteristic parameters and tissue area division. This method can perform detailed analysis based on the actual casting tissue differences, and achieve accurate evaluation of the heterogeneous mechanical properties of automotive aluminum alloy cast battery pack frames. At the same time, the dynamic correction strategy adopted further enhances the model's adaptability to defects and tissue heterogeneity, making the prediction results more practical. Through the implementation of this method, not only can the design and manufacturing quality of aluminum alloy castings be improved, but also the potential safety risks caused by material property mismatch can be reduced, providing more reliable technical support for the automotive industry, thereby promoting technological progress and innovation in related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 Schematic diagram of the overall method flow of the present invention;

[0057] Figure 2 、 Figure 3 、 Figure 4 These are the fitting curves of predicted elastic modulus, predicted yield strength, predicted Poisson's ratio and initial toughness index respectively. DETAILED DESCRIPTION

[0058] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0059] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0060] Example:

[0061] See also Figure 1 , the present invention provides a technical solution:

[0062] A method for predicting the microstructure and mechanical properties of alloy castings based on machine learning, comprising the following steps:

[0063] S1, obtain multiple sets of microstructure characteristic parameters and corresponding mechanical parameters of automotive aluminum alloy casting battery pack frames, map each set of microstructure characteristic parameters and mechanical parameters one by one, and generate a sample data set, wherein the microstructure characteristic parameters include dendrite secondary arm spacing, Phase area ratio, porosity and small-angle misorientation ratio of grain boundaries, and mechanical parameters including elastic modulus, yield strength and Poisson's ratio;

[0064] S2, based on the data in the sample data set, a neural network model is constructed, the microstructure characteristic parameters in the sample data set are used as the input of the model, and the mechanical parameters in the sample data set are used as labels to train the neural network model to obtain a mechanical parameter prediction model;

[0065] In this embodiment, the method for generating the sample data set is: mapping each set of microstructure characteristic parameters and mechanical parameters one by one to form a data chain, and recording the formed data chain as the sample data set;

[0066] Based on the data in the sample data set, a neural network model is established. The long short-term memory network LSTM model is selected as the basic model. The activation function and optimization algorithm are selected. The Tanh function is selected as the activation function and Adam is selected as the optimization algorithm of the LSTM model.

[0067] The Tanh function expression is as follows:

[0068] ;

[0069] Where, Represents the Tanh function, independent variable represents the weighted sum of the neuron's inputs, that is, the result of the weighted summation of the inputs received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons; wherein the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the number of batches is set to 256, and the number of hidden layer neurons is 32;

[0070] The sample data set is randomly divided into a training set and a test set, and a mechanical parameter prediction model is constructed based on a long short-term memory network. The microstructure characteristic parameters in the training set are used as input and the mechanical parameters are used as labels to train the model to obtain a trained mechanical parameter prediction model. The microstructure characteristic parameters in the test set are substituted into the trained model to obtain the corresponding prediction results; the error between the prediction result and the actual value in the test set is calculated; it is determined whether the error meets a preset error threshold; if so, the trained model, i.e., the mechanical parameter prediction model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error, and determination coefficient between the prediction result and the actual value in the test set;

[0071] The process of randomly dividing the dataset into training and test sets is as follows: randomly sort the sample dataset, use 80% of the sorted sample dataset as the training set, and the remaining 20% ​​as the test set;

[0072] The input of the trained mechanical parameter prediction model is the microstructure characteristic parameters, and the output is the mechanical parameters.

[0073] The microstructural characteristic parameters include dendrite secondary arm spacing, Phase area ratio, porosity and grain boundary small angle misorientation ratio, where:

[0074] The dendrite secondary arm spacing is obtained by the metallographic sample intercept method: for an aluminum alloy sample corroded by Keller reagent, the vertical distance between the roots of adjacent secondary dendrite arms is measured under a 100× optical microscope field of view, and the arithmetic average of 30 sets of data is used as the dendrite secondary arm spacing of the sample;

[0075] described The phase area ratio was imaged using the backscattered electron mode of a scanning electron microscope. Software threshold segmentation calculation description Phase pixel ratio, and the analysis area is 5 2000× fields of view;

[0076] The porosity is obtained by three-dimensional reconstruction of X-ray tomography. The grayscale threshold is set to distinguish the matrix and the pores, and the porosity is calculated as the percentage of the pore volume to the total scan volume. The scanning resolution is ;

[0077] The small-angle misorientation ratio of the grain boundary is determined by electron backscatter diffraction technology, and the misorientation angle is set For small-angle grain boundaries, the proportion of the total length of the grain boundary is counted, and the scanning step length is .

[0078] The advantage of steps S1 and S2 lies in the construction of a detailed and high-quality sample dataset through systematic data acquisition and processing. These steps not only ensure an accurate mapping between microstructural characteristic parameters and corresponding mechanical parameters, but also train the mechanical parameter prediction model using advanced machine learning models such as long-short-term memory networks. This approach improves the accuracy and reliability of predictions. Compared to the static models and uniform material assumptions used in existing technologies, it can dynamically capture the complexity of the microstructure during the casting process, overcome the limitations of traditional methods, and enhance the practicality of the prediction results.

[0079] In this solution, the implementation of steps S1 and S2 significantly improves the overall effectiveness of the overall approach. By accurately constructing a sample dataset and efficiently training the neural network, subsequent microstructural zoning, mechanical property prediction, and strength and toughness index calculations can be built on a more solid foundation. This not only improves the accuracy of the entire prediction process but also provides more reliable input data for subsequent finite element analysis, ensuring the performance and safety of aluminum alloy castings in practical applications and promoting the widespread application of intelligent manufacturing technologies in the aluminum alloy casting field.

[0080] S3, based on the casting structure difference, the predicted automotive aluminum alloy casting battery pack frame is divided into microstructure regions, and the microstructure characteristic parameters of each microstructure region of the divided alloy casting are collected, the microstructure characteristic parameters including the dendrite secondary arm spacing, Phase area ratio, porosity and grain boundary small angle misorientation ratio;

[0081] In this embodiment, the microstructure of the predicted automotive aluminum alloy cast battery pack frame is divided into regions according to the casting structure differences. The specific logic is as follows:

[0082] Determining the orientation difference between adjacent pixels by electron backscatter diffraction orientation imaging And continuous length The area is taken as the dendrite trunk area; then the grain roundness is divided based on the gray gradient of the metallographic image. And the equivalent diameter is The area is regarded as the equiaxed crystal area; and then the energy spectrum analysis is used to identify The proportion of phase area in the phase-enriched zone is The transition zone is used as the transition zone; finally, the X-ray tomography data is combined to calibrate a spherical domain with a radius of 2.5 times the defect diameter centered on the pore as the defect-affected zone, forming four types of microstructural regions including the dendrite trunk region, equiaxed crystal region, transition region and defect-affected region.

[0083] Dendrite secondary arm spacing is a key parameter to characterize the fineness of casting solidification structure. Smaller dendrite secondary arm spacing indicates faster solidification cooling rate and higher degree of grain refinement, which will significantly improve the yield strength and fatigue life of the material. Its mechanism of action includes: small dendrite arm spacing can increase grain boundary area, hinder dislocation movement, and improve strength through Hall-Petch effect; uniform dendrite structure can delay crack propagation and increase fracture toughness by 10%-20%. For example, in A356 aluminum alloy, SDAS is from down to It can increase the yield strength by about 15%.

[0084] Phase is a common brittle intermetallic compound in aluminum alloys. Its content and distribution directly affect the strength and toughness balance of the material: Phase ratio When the content of α is greater than 8%, it can play a role in dispersion strengthening and improve the yield strength; when it exceeds 8%, it is easy to form a continuous network structure, becoming a crack source, resulting in a sudden drop of 30%-50% in elongation and impact toughness. Quantification through image analysis The phase ratio can predict the brittle fracture tendency of the material.

[0085] Pores are unavoidable defects in the casting process. Their volume fraction and size distribution have a decisive influence on the mechanical properties: Experiments show that for every 1% increase in porosity, the tensile strength decreases by an average of 5%-8% and the elastic modulus decreases by 3%-5%. When the maximum pore diameter is When the fatigue life is reduced to 1 / 10 of the defect-free material.

[0086] The percentage of small-angle misorientation at grain boundaries reflects the distribution of misorientation between grains. EBSD analysis reveals that a high percentage of small-angle grain boundaries leads to higher creep and stress corrosion resistance, as they effectively hinder dislocation slip. A low percentage of large-angle grain boundaries reduces grain boundary strengthening, leading to a decrease in yield strength. Research has shown that by controlling the solidification rate, the percentage of small-angle grain boundaries can be increased from 25% to 50%, correspondingly increasing the room-temperature strength of aluminum alloys by approximately 12%.

[0087] The advantage of step S3 is that by finely dividing the microstructure of the aluminum alloy cast battery pack frame to be predicted, it can fully extract and quantify the microstructure characteristic parameters of different regions. This refined regional division not only improves the accuracy of collecting microstructure characteristic parameters, but also makes the subsequent mechanical property prediction more realistic. Compared with the homogeneous assumption commonly used in existing technologies, this method significantly enhances the sensitivity and responsiveness to microstructural changes.

[0088] In this solution, the implementation of step S3 significantly facilitates the overall implementation of the proposed approach, providing more accurate input data for subsequent mechanical parameter predictions. This targeted analysis of different microstructural regions enables the overall prediction model to better reflect the actual performance of aluminum alloy castings, thereby improving the reliability and effectiveness of mechanical property predictions. This refined analysis will ultimately help optimize the design and manufacturing processes of aluminum alloy castings, enhancing their performance stability and safety in practical applications.

[0089] S4, inputting the microstructural characteristic parameters of the alloy casting to be predicted into the trained mechanical parameter prediction model, the model outputting a predicted elastic modulus, a predicted yield strength, and a predicted Poisson's ratio, and determining an initial toughness index based on the predicted elastic modulus, the predicted yield strength, and the predicted Poisson's ratio;

[0090] In this embodiment, the microstructure characteristic parameters of the alloy casting to be predicted are input into the trained mechanical parameter prediction model, and the model outputs the predicted elastic modulus. , predict yield strength and predicted Poisson's ratio ;

[0091] The initial toughness index is determined based on the predicted elastic modulus, predicted yield strength, and predicted Poisson's ratio according to the following formula:

[0092] ;

[0093] Where, is the initial toughness index, To predict the elastic modulus, is the reference elastic modulus, , To predict the yield strength, is the reference yield strength, , To predict Poisson's ratio, is the reference Poisson's ratio, , is a natural constant, 、 and is the preset weight value, , , , and is the material constant, , . Weight value and Respectively represent the relative contribution of the predicted elastic modulus and predicted yield strength to the toughness of the alloy casting. In this formula, set The main reason is that yield strength usually plays a more critical role in the mechanical properties of materials. Yield strength directly determines the plastic deformation ability of the material when subjected to external loads, and can effectively reflect the load-bearing capacity and durability of the material in practical applications. In contrast, the elastic modulus mainly describes the stiffness of the material within a small deformation range. Although its influence cannot be ignored, the influence of yield strength is more significant in terms of the ultimate strength and toughness performance of the material. Therefore, by giving A higher weight value can more accurately reflect the importance of yield strength in improving the strength and toughness of materials. This weight setting helps to better predict the mechanical behavior of materials in engineering applications and ensure that the materials can exhibit better performance under actual use conditions.

[0094] This formula is used to calculate the initial toughness index, and its rationality is reflected in many aspects. First, the ratio form is used and , which enables the formula to be relatively compared between different materials, eliminating the impact of unit and scale differences. This standardization process enhances the versatility and applicability of the formula. or The larger the value, the better the mechanical properties of the material in both elastic and yield behavior, and the better the stability and integrity of the structure under higher loads, thereby improving the initial strength and toughness index. Secondly, by introducing weight values, the different contributions of elastic modulus and yield strength to toughness are balanced, reflecting their relative importance in material performance evaluation. In addition, the introduction of material constants enables the model to be adjusted according to the characteristics of specific materials, improving the accuracy of predictions. The relative change of Poisson's ratio is introduced in the form of an exponential, that is, , not only considers the stress-strain characteristics of the material during deformation, but also reflects the nonlinear effect of Poisson's ratio on the material's toughness. Therefore, this formula integrates multiple material properties to evaluate the material's strength and toughness in a scientific and rational manner, with high practicality and guiding significance.

[0095] It is a comprehensive indicator used to characterize the strength and toughness properties of alloy castings. It combines multiple mechanical performance parameters such as elastic modulus, yield strength and Poisson's ratio, and reflects the mechanical behavior and deformation resistance of the material under the microstructural characteristics. The higher the value, the stronger the material's ability to resist deformation and damage when subjected to external loads, and thus has better durability and reliability in practical applications.

[0096] The independent variables include the predicted elastic modulus, yield strength, and Poisson's ratio, which are related to the dependent variable There is a significant correlation between them because they jointly affect the mechanical behavior of the material. Specifically: the elastic modulus reflects the rigidity of the material. The higher the elastic modulus, the smaller the deformation of the material when subjected to stress, and the corresponding increase in strength and toughness. Yield strength refers to the stress value at which the material begins to undergo plastic deformation. The higher the yield strength, the less likely the material is to undergo permanent deformation in actual use. A higher yield strength means that the material can withstand greater stress without failure during load-bearing. Poisson's ratio describes the relative deformation of a material in other directions when it is subjected to stress in a certain direction. Changes in Poisson's ratio affect the strain state and deformation capacity of the material, thereby affecting strength and toughness. Therefore, when predicting the elastic modulus, , predict yield strength or predicted Poisson's ratio When increasing, The corresponding part will also increase.

[0097] Table 1: Initial strength and toughness index statistics

[0098]

[0099] See also Figure 2-Figure 4 ,In this data analysis, according to the statistical data in Table 1, it can be seen from the table that when Ep increases from 65GPa to 83 GPa, It increases from 0.827 to 1.246, showing a clear upward trend. This indicates that with the increase of elastic modulus, the rigidity of the alloy material increases, which can better resist external stress and reduce the deformation of the material during use. This phenomenon is common in engineering materials, indicating that the rigidity and toughness of the material are closely related. The yield strength is predicted to increase from 180 MPa to 275 MPa. The Poisson's ratio also increased accordingly, from 0.827 to 1.246. This result further verifies the importance of yield strength in material properties. The higher the yield strength, the less likely the material is to undergo plastic deformation when subjected to load, which enhances the overall toughness of the material. This trend is very conducive to achieving higher safety and reliability in design and application. In the predicted data, the Poisson's ratio increased from 0.31 to 0.405, and at the same time It also shows a positive change from 0.827 to 1.246. The change of Poisson's ratio affects the deformation behavior of the material. A higher Poisson's ratio usually means that the material becomes more flexible when under tension and can better disperse stress, thereby improving the toughness of the material. The improvement of this indicator is particularly important in practical applications, especially in scenarios such as impact resistance and wear resistance. Through the above analysis, we can clearly see that the initial strength and toughness index There is a positive correlation between each independent variable. This correlation can be explained by the following points: the higher the elastic modulus of a material, the stronger its deformation ability under external force, and thus the strength and toughness index The higher the yield strength, the greater the yield strength. This relationship is particularly critical when designing structural materials for high-load applications. Increasing yield strength directly enhances the material's ability to withstand ultimate loads, thereby increasing the initial strength-toughness index. This is crucial for extending the service life of alloy castings. Changes in Poisson's ratio play an important regulatory role in calculating strength-toughness. It significantly influences the material's plastic deformation behavior and can improve its toughness performance in practical applications.

[0100] The advantage of step S4 is that by inputting microstructural characteristic parameters into a trained mechanical parameter prediction model, the corresponding predicted elastic modulus, yield strength, and Poisson's ratio can be accurately output. This approach enables the model to perform personalized mechanical property predictions based on detailed microstructural characteristics. Compared with existing technologies, it avoids the use of simplifications or homogenization assumptions, ensuring more realistic and reliable prediction results.

[0101] In this solution, step S4 significantly facilitates the overall implementation of the overall approach, providing key mechanical parameter data for subsequent strength and toughness analysis and finite element modeling. These specific prediction results enable more accurate assessment of the mechanical properties of aluminum alloy castings, thereby enhancing material design optimization and production process control. This will help improve the safety and reliability of aluminum alloy castings in practical applications and promote the advancement of intelligent manufacturing technology.

[0102] S5, determining the microstructure heterogeneity index based on the standard deviation of the predicted yield strength of all microstructure regions, determining the defect sensitivity threshold based on the maximum defect size, calculating the strength-toughness correction factor based on the microstructure heterogeneity index and the defect sensitivity threshold, and using the strength-toughness correction factor to correct the initial strength-toughness index to obtain a comprehensive strength-toughness index that reflects the mechanical properties of the microstructure region;

[0103] In this embodiment, the standard deviation of the predicted yield strength of all microstructure regions in the automotive aluminum alloy casting battery pack frame to be predicted is used as the non-uniformity index, denoted as The ratio of the maximum defect diameter in all microstructural regions of the predicted automotive aluminum alloy cast battery pack frame to the critical defect size of the material is used as the defect sensitivity threshold, denoted as ;

[0104] Defects in casting slices are measured using a scanning electron microscope. The largest defect area is identified and its diameter is recorded as the maximum defect diameter. The critical defect size of the material is obtained from experimental data or literature. The ratio of the maximum defect diameter to the critical defect size is used as the defect sensitivity threshold, which reflects the defect sensitivity of the material under load conditions. A smaller ratio indicates that the material can tolerate larger defects while still maintaining good performance, indicating a low sensitivity to defects. A larger ratio means that the material is more sensitive to defects and may be more likely to fail during load-bearing. Therefore, the defect sensitivity threshold is an important indicator for evaluating material reliability and safety.

[0105] The strength and toughness correction factor is calculated based on the tissue heterogeneity index and the defect sensitivity threshold, according to the following formula:

[0106] ;

[0107] Where, is the toughness correction factor, is the tissue heterogeneity index, is the defect sensitivity threshold, and is the preset weight value, , In materials science, defects are often the main factor leading to material failure, especially under dynamic loads or fatigue conditions, where the presence of defects can significantly reduce the tensile strength and impact toughness of the material. The larger the value, the more sensitive the casting is to defects, which has a profound impact on the safety and reliability of the material. Therefore, a higher weight value is given to it.

[0108] It integrates the structural heterogeneity and defect sensitivity of the material, and reflects the modification effect of the material on strength and toughness in practical applications. Reflects the uniformity of the microstructure. The larger the value, the more uneven the microstructure of the casting, which may lead to instability in the mechanical properties of the material. Microstructural heterogeneity often leads to local stress concentration, thus affecting the overall strength and toughness of the material. It indicates the ratio of the maximum defect diameter to the critical defect size, which directly reflects the sensitivity of the material to defects. The value means that the defect size is close to the critical value, which will significantly reduce the reliability of the casting.

[0109] The initial strength-toughness index is corrected using the strength-toughness correction factor to obtain a comprehensive strength-toughness index that reflects the mechanical properties of the microstructure region. The formula is as follows:

[0110] ;

[0111] Where, is the comprehensive toughness index, is the initial toughness index, is the toughness correction factor, is the preset weight of the toughness correction factor, and .

[0112] Comprehensive strength index It is a dimensionless index that quantitatively characterizes the mechanical properties of the microstructure of alloy castings, and reflects the actual strength and toughness level after correction of structural inhomogeneity and defects. When enlarged, it means that tissue defects or heterogeneity worsen, leading to When hour, , which indicates that the comprehensive strength and toughness are equal to the initial strength and toughness when there is no microstructural inhomogeneity and defects. The increase, The value of will also increase, making The introduction of the square root form keeps the comprehensive strength index within a reasonable range. This design ensures Always less than or equal to , which is consistent with the changing trend of actual material properties.

[0113] The initial strength-toughness index is corrected using the strength-toughness correction factor to obtain a comprehensive strength-toughness index that aims to more accurately reflect the mechanical properties of the microstructure area. Indicates the basic strength and toughness of the material when it is not affected by micro defects, and the correction factor This method takes into account microstructural heterogeneity and its impact on performance. Pre-set weighting coefficients are used to adjust the influence of the correction factor on strength and toughness, ensuring that the revised index reflects the inhibitory effect of microscopic features on material properties while maintaining a scientifically reasonable correction range. In this way, the comprehensive strength-toughness index can more realistically reflect the mechanical properties of materials in practical applications, providing an important basis for material design and evaluation.

[0114] The rationality of correcting all initial strength-toughness indices based on the same correction factor is reflected in the fact that the characteristics of the material microstructure region usually have a certain similarity, especially within the same material system. This method can ensure that when calculating the comprehensive strength-toughness index, all regions are affected by the same microscopic defects and structural inhomogeneities, thereby providing a consistent basis for correction. By introducing a preset weight coefficient, the influence of the microstructure on the strength and toughness can be balanced, making the corrected strength-toughness index more representative and reflecting the comprehensive performance of the material under actual use conditions. This unified correction method not only simplifies the calculation process, but also enhances the comparability and reliability of the results, so that the predicted comprehensive strength-toughness index can more effectively reflect the mechanical properties of the alloy material.

[0115] The advantage of step S5 is that by calculating the structural heterogeneity index and defect sensitivity threshold, a more comprehensive and accurate assessment of the strength and toughness of the alloy casting can be made. This process takes into account the standard deviation of the yield strength of the microstructural region and the influence of defects, allowing the final comprehensive strength and toughness index to more accurately reflect the material's performance. Compared with existing technologies, this avoids the relatively simple homogeneous material model and can better capture the influence of the material's complex internal microstructure on mechanical properties.

[0116] In this solution, step S5 significantly facilitates the implementation of the overall approach, as it provides the necessary adjustments and correction factors for a detailed analysis of mechanical properties. This refined strength and toughness assessment improves the accuracy of subsequent finite element analysis, making the final predictions more reliable. This will help optimize the design and manufacturing process of aluminum alloy castings, ensuring their performance and safety in practical applications, thereby improving the stability and reliability of the entire casting process.

[0117] S6, constructing a finite element model based on the alloy casting to be predicted, spatially aligning the microstructure region of the alloy casting with the model geometric topology, and performing three-dimensional meshing, mapping the comprehensive strength-toughness index of each microstructure region to the mesh node at the corresponding position, and determining the comprehensive strength-toughness index of other mesh nodes based on the spatial distance attenuation law, thereby completing the prediction of the mechanical properties of the entire alloy casting;

[0118] In this embodiment, a finite element model based on the alloy casting to be predicted is constructed, the microstructure region of the alloy casting is spatially aligned with the model geometric topology, and three-dimensional meshing is performed, wherein: The three-dimensional meshing is performed according to the following rules:

[0119] (1) At the junction of the dendrite region and the equiaxed crystal region, The local grid is encrypted within the range of 3 times the equivalent diameter of the phase aggregation area and the defect periphery, and the minimum unit size does not exceed 1 / 5 of the characteristic area size to ensure that each microstructure partition contains at least 3 layers of grid units; the characteristic area refers to the junction of the dendrite area and the equiaxed crystal area, Phase aggregation area and within 3 times equivalent diameter around defects;

[0120] (2) The mesh size change rate between adjacent microstructure regions is controlled within 1.2 times, and the smooth transfer of mechanical property parameters is achieved through the transition layer mesh;

[0121] (3) Based on the stress distribution preliminary analysis results of the finite element model, the bolt connection holes and weld heat-affected zones of the battery pack frame are identified as high-risk areas and double encryption is implemented, with the mesh density reaching twice that of non-critical areas; the non-critical areas include the central flat plate area of ​​the frame, the non-load-bearing bracket connection surface, and the non-stressed side walls of the heat sink;

[0122] (4) The spatial alignment error between the grid nodes and the tissue boundary identified by the metallographic image is kept less than , a curvature-adaptive hexahedron-dominated hybrid grid is used in the curved surface area;

[0123] (5) For the grid nodes in the non-microstructure area, the comprehensive strength and toughness index is determined according to the exponential decay law.

[0124] The comprehensive strength-toughness index of each microstructure region is mapped to the grid node at the corresponding position to obtain the microstructure region mapping node. Based on the spatial distance attenuation law, the comprehensive strength-toughness index of other grid nodes is determined according to the following formula:

[0125] ;

[0126] Where, For the The comprehensive toughness index of other grid nodes, is the index of other grid nodes, For the The comprehensive toughness index of the microstructure region mapping node, is the total number of nodes mapped in the micro-organization area, is the index of the microstructure region mapping node, is the weight coefficient, indicating the The micro-organizational region mapping node is The degree of influence of other grid nodes is related to the spatial distance;

[0127] When predicting the mechanical properties of alloy castings, the rationality of using the spatial distance attenuation law to determine the comprehensive strength and toughness index of other grid nodes is reflected in its ability to effectively simulate the degree of influence of the microstructure area on the surrounding grid nodes. Specifically, it can be seen from the formula that the weight coefficient is related to the distance between the microstructure area mapping node and other grid nodes. The closer the distance, the greater the weight, which makes the contribution of adjacent nodes to the comprehensive strength and toughness index more significant. This spatial attenuation model is consistent with physical reality, that is, the influence of microstructure is usually local, and the degree of influence gradually weakens with increasing distance. Therefore, when calculating the comprehensive strength and toughness index of the j-th grid node, integrating the influence of all mapping nodes by weighted averaging can more accurately reflect its actual performance, thereby providing a scientific basis for the prediction of the overall mechanical properties of alloy castings.

[0128] Among them, determine The formula is as follows:

[0129] ;

[0130] Where, Mapping the 3D coordinates of nodes to the microstructure region , For the 3D coordinates of other mesh nodes , is a natural constant;

[0131] Based on the above steps, the comprehensive strength-toughness index of each grid node is determined to complete the prediction of the mechanical properties of the entire alloy casting.

[0132] The advantage of step S6 is that by constructing a finite element model-based spatial registration of microstructural regions and three-dimensional meshing, it can effectively correlate complex microstructural characteristics with mechanical properties. This process enables accurate simulation of the mechanical responses of different tissue regions at the microscale, thereby improving the accuracy and reliability of the prediction results. Compared with existing technologies, this method can more comprehensively consider the diversity and local variation within the material, thus overcoming the limitations of traditional homogeneous models.

[0133] In this approach, implementing step S6 significantly facilitates the overall solution, as it provides the necessary geometric and physical foundation for the final mechanical property predictions. Through precise spatial registration and meshing, combined with the calculated comprehensive strength-toughness index, the performance of aluminum alloy castings under actual use conditions can be more accurately reflected. This detailed analysis will help optimize material design and processing, improve the performance and safety of aluminum alloy castings in demanding applications such as automotive, and promote the advancement of intelligent manufacturing.

[0134] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0135] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0136] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0137] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for predicting the microstructure and mechanical properties of alloy castings based on machine learning, characterized in that: The specific steps include: S1, obtain multiple sets of microstructure characteristic parameters and corresponding mechanical parameters of automotive aluminum alloy casting battery pack frames, map each set of microstructure characteristic parameters and mechanical parameters one by one, and generate a sample data set, wherein the microstructure characteristic parameters include dendrite secondary arm spacing, Phase area ratio, porosity and small-angle misorientation ratio of grain boundaries, and mechanical parameters including elastic modulus, yield strength and Poisson's ratio; S2, based on the data in the sample data set, a neural network model is constructed, the microstructure characteristic parameters in the sample data set are used as the input of the model, and the mechanical parameters in the sample data set are used as labels to train the neural network model to obtain a mechanical parameter prediction model; S3, based on the casting structure difference, the predicted automotive aluminum alloy casting battery pack frame is divided into microstructure regions, and the microstructure characteristic parameters of each microstructure region of the divided alloy casting are collected, the microstructure characteristic parameters including the dendrite secondary arm spacing, Phase area ratio, porosity and grain boundary small angle misorientation ratio; S4, inputting the microstructural characteristic parameters of the alloy casting to be predicted into the trained mechanical parameter prediction model, the model outputting a predicted elastic modulus, a predicted yield strength, and a predicted Poisson's ratio, and determining an initial toughness index based on the predicted elastic modulus, the predicted yield strength, and the predicted Poisson's ratio; S5, determining the microstructure heterogeneity index based on the standard deviation of the predicted yield strength of all microstructure regions, determining the defect sensitivity threshold based on the maximum defect size, calculating the strength-toughness correction factor based on the microstructure heterogeneity index and the defect sensitivity threshold, and using the strength-toughness correction factor to correct the initial strength-toughness index to obtain a comprehensive strength-toughness index that reflects the mechanical properties of the microstructure region; S6, constructing a finite element model based on the alloy casting to be predicted, spatially aligning the microstructure region of the alloy casting with the model geometric topology, and performing three-dimensional meshing, mapping the comprehensive strength-toughness index of each microstructure region to the mesh node at the corresponding position, and determining the comprehensive strength-toughness index of other mesh nodes based on the spatial distance attenuation law, thereby completing the prediction of the mechanical properties of the entire alloy casting; Input the microstructural characteristic parameters of the alloy casting to be predicted into the trained mechanical parameter prediction model, and the model outputs the predicted elastic modulus , predict yield strength and predicted Poisson's ratio ; The initial toughness index is determined based on the predicted elastic modulus, predicted yield strength, and predicted Poisson's ratio according to the following formula: ， Where, is the initial toughness index, To predict the elastic modulus, is the reference elastic modulus, To predict the yield strength, is the reference yield strength, To predict Poisson's ratio, is the reference Poisson's ratio, is a natural constant, 、 and is the preset weight value and satisfies , The value range of is (0,1], and It is a material constant, and its specific value is determined according to the type of material.

2. The method for predicting the microstructure and mechanical properties of alloy castings based on machine learning according to claim 1, characterized in that: The method for generating the sample data set is: mapping each set of microstructure characteristic parameters and mechanical parameters one by one to form a data chain, and recording the formed data chain as the sample data set; Based on the data in the sample data set, a neural network model is established. The long short-term memory network LSTM model is selected as the basic model. The activation function and optimization algorithm are selected. The Tanh function is selected as the activation function and Adam is selected as the optimization algorithm of the LSTM model. The Tanh function expression is as follows: ， Where, Represents the Tanh function, independent variable represents the weighted sum of the neuron's inputs, that is, the result of the weighted summation of the inputs received by the neuron from the previous layer; at the same time, the hyperparameters of the LSTM model are set, and the hyperparameters of the LSTM model include: the number of network layers, the number of iterations, the learning rate, the batch size, the number of training times, the number of batches, and the number of hidden layer neurons; wherein the number of network layers is set to a 3-layer network structure, the number of iterations is set to 200, the learning rate is set to 0.001, the batch size is set to 32, the number of training times is set to 100, the number of batches is set to 256, and the number of hidden layer neurons is 32; The sample data set is randomly divided into a training set and a test set, and a mechanical parameter prediction model is constructed based on a long short-term memory network. The microstructure characteristic parameters in the training set are used as input and the mechanical parameters are used as labels to train the model to obtain a trained mechanical parameter prediction model. The microstructure characteristic parameters in the test set are substituted into the trained model to obtain the corresponding prediction results; the error between the prediction result and the actual value in the test set is calculated; it is determined whether the error meets a preset error threshold; if so, the trained model, i.e., the mechanical parameter prediction model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error, and determination coefficient between the prediction result and the actual value in the test set; The process of randomly dividing the dataset into training and test sets is as follows: randomly sort the sample dataset, use 80% of the sorted sample dataset as the training set, and the remaining 20% ​​as the test set; The input of the trained mechanical parameter prediction model is the microstructure characteristic parameters, and the output is the mechanical parameters.

3. The method for predicting the microstructure and mechanical properties of alloy castings based on machine learning according to claim 1, characterized in that: The microstructural characteristic parameters include dendrite secondary arm spacing, Phase area ratio, porosity and grain boundary small angle misorientation ratio, where: The dendrite secondary arm spacing is obtained by the metallographic sample intercept method: for an aluminum alloy sample corroded by Keller reagent, the vertical distance between the roots of adjacent secondary dendrite arms is measured under a 100× optical microscope field of view, and the arithmetic average of 30 sets of data is used as the dendrite secondary arm spacing of the sample; described The phase area ratio was imaged using the backscattered electron mode of a scanning electron microscope. Software threshold segmentation calculation description Phase pixel ratio, and the analysis area is 5 2000× fields of view; The porosity is obtained by three-dimensional reconstruction of X-ray tomography. The grayscale threshold is set to distinguish the matrix and the pores, and the porosity is calculated as the percentage of the pore volume to the total scan volume. The scanning resolution is ; The small-angle misorientation ratio of the grain boundary is determined by electron backscatter diffraction technology, and the misorientation angle is set For small-angle grain boundaries, the proportion of the total length of the grain boundary is counted, and the scanning step length is .

4. The method for predicting the microstructure and mechanical properties of alloy castings based on machine learning according to claim 1, characterized in that: According to the differences in casting structures, the microstructure area division of the predicted automotive aluminum alloy casting battery pack frame is carried out. The specific logic is as follows: Determining the orientation difference between adjacent pixels by electron backscatter diffraction orientation imaging And continuous length The area serves as the dendrite trunk area; Then the grain roundness is divided based on the gray gradient of the metallographic image And the equivalent diameter is The area is regarded as the equiaxed crystal area; and then the energy spectrum analysis is used to identify The proportion of phase area in the phase-enriched zone is The transition zone is used as the transition zone; finally, the X-ray tomography data is combined to calibrate a spherical domain with a radius of 2.5 times the defect diameter centered on the pore as the defect-affected zone, forming four types of microstructural regions including the dendrite trunk region, equiaxed crystal region, transition region and defect-affected region.

5. The method for predicting the microstructure and mechanical properties of alloy castings based on machine learning according to claim 1, characterized in that: The standard deviation of the predicted yield strength of all microstructure regions in the predicted automotive aluminum alloy cast battery pack frame is used as the non-uniformity index, denoted as The ratio of the maximum defect diameter in all microstructural regions of the aluminum alloy casting battery pack frame to be predicted to the critical defect size of the material is used as the defect sensitivity threshold, which is recorded as ; The strength and toughness correction factor is calculated based on the tissue heterogeneity index and the defect sensitivity threshold, according to the following formula: ， Where, is the toughness correction factor, is the tissue heterogeneity index, is the defect sensitivity threshold, and is the preset weight value, , and satisfies ; The initial strength-toughness index is corrected using the strength-toughness correction factor to obtain a comprehensive strength-toughness index that reflects the mechanical properties of the microstructure region. The formula is as follows: ， Where, is the comprehensive toughness index, is the initial toughness index, is the toughness correction factor, is the preset weight of the toughness correction factor, and The value range is [0.2, 0.8].

6. The method for predicting the microstructure and mechanical properties of alloy castings based on machine learning according to claim 1, characterized in that: Construct a finite element model based on the alloy casting to be predicted, spatially align the microstructure area of ​​the alloy casting with the model geometry topology, and perform three-dimensional meshing, wherein: Three-dimensional meshing is performed according to the following rules: (1) At the junction of the dendrite region and the equiaxed crystal region, Local mesh encryption is used within the range of three times the equivalent diameter of the phase aggregation area and the defect periphery, with the minimum unit size not exceeding 1 / 5 of the characteristic area size, ensuring that each microstructure partition contains at least three layers of mesh units; (2) The mesh size change rate between adjacent microstructure regions is controlled within 1.2 times, and the smooth transfer of mechanical property parameters is achieved through the transition layer mesh; (3) Based on the stress distribution preliminary analysis results of the finite element model, the bolt connection holes and weld heat-affected zones of the battery pack frame are identified as high-risk areas and double encryption is implemented, with the mesh density reaching twice that of non-critical areas; (4) The spatial alignment error between the grid nodes and the tissue boundary identified by the metallographic image is kept less than , a curvature-adaptive hexahedron-dominated hybrid grid is used in the curved surface area; (5) For the grid nodes in the non-microstructure area, the comprehensive strength and toughness index is determined according to the exponential decay law.

7. The method for predicting the microstructure and mechanical properties of alloy castings based on machine learning according to claim 1, characterized in that: The comprehensive strength-toughness index of each microstructure region is mapped to the grid node at the corresponding position to obtain the microstructure region mapping node. Based on the spatial distance attenuation law, the comprehensive strength-toughness index of other grid nodes is determined according to the following formula: ， Where, For the The comprehensive toughness index of other grid nodes, is the index of other grid nodes, For the The comprehensive toughness index of the microstructure region mapping node, is the total number of nodes mapped in the micro-organization area, is the index of the micro-tissue region mapping node, is the weight coefficient, indicating the The micro-organizational region mapping node is The degree of influence of other grid nodes is related to the spatial distance; Among them, determine The formula is as follows: ， Where, Mapping the 3D coordinates of nodes to the microstructure region , For the 3D coordinates of other mesh nodes , is a natural constant; Based on the above steps, the comprehensive strength-toughness index of each grid node is determined to complete the prediction of the mechanical properties of the entire alloy casting.

Images