Method and system for determining a composition of an alloy to be designed that meets predetermined requirements, and computer storage medium

Abstract

Embodiments of the present application relate to the technical field of computer materials science, in particular to a method and system for determining a composition of an alloy to be designed satisfying predetermined requirements, and a computer storage medium. The method of the embodiments of the present application trains a composition model and a physical parameter model independently respectively, so that the composition model and the physical parameter model do not have the same systematic deviation, and uses the two models to predict the alloy performance of a first candidate alloy composition, and comprehensively scores the first candidate alloy composition according to the two prediction results, so that through the mutual verification of the composition model and the physical parameter model, the composition of the alloy to be designed determined according to the comprehensive score can be supported from two aspects of data statistical law and physical law, and the reliability can be effectively verified.

Description

Technical Field

[0001] Embodiments of this application relate to the field of computer materials science and technology, specifically to a method and system for determining the composition of an alloy to be designed that meets predetermined requirements, as well as a computer storage medium. Background Technology

[0002] The statements herein are provided merely as background information in connection with this application and do not necessarily constitute prior art.

[0003] When designing alloys, it is necessary to determine the requirements that the alloy must meet based on the actual application scenario, and then determine the alloy composition according to the predetermined requirements. However, the compositional range of alloys is very wide, making it difficult to find an alloy composition that can meet the predetermined requirements. Summary of the Invention

[0004] A brief overview of this application is provided below to offer a basic understanding of certain aspects thereof. It should be understood that this overview is not an exhaustive summary of the application. It is not intended to identify key or essential parts of the application, nor is it intended to limit its scope. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0005] In a first aspect, embodiments of this application provide a method for determining the composition of an alloy to be designed that meets predetermined requirements, comprising the following steps: S10: Determine the training dataset according to the predetermined requirements. The training dataset includes the composition data, condition parameter data, physical parameter data, and performance data of multiple sets of sample alloys. S20: Based on the training dataset, train the component model and the physical parameter model respectively to obtain the trained component model and physical parameter model; The compositional data and conditional parameter data of multiple sets of sample alloys are used as inputs to the compositional model, and the performance data of multiple sets of sample alloys are used as outputs to the compositional model; the physical parameter data of multiple sets of sample alloys are used as inputs to the physical parameter model, and the performance data of multiple sets of sample alloys are used as outputs to the physical parameter model. S30: Determine the composition of the first candidate alloy based on the predetermined requirements of the alloy to be designed and the preset physical constraints; S40: Based on the first candidate alloy composition and composition model, determine the first predicted performance value of the first candidate alloy composition; based on the physical parameters and physical parameter model of the first candidate alloy composition, determine the second predicted performance value of the first candidate alloy composition. S50: Determine the comprehensive score of the first candidate alloy composition based on the first and second performance prediction values ​​of the first candidate alloy composition and the predetermined requirements of the alloy to be designed. S60: Based on the first candidate alloy composition and its comprehensive score, update the first candidate alloy composition and determine the updated effective candidate alloy composition and its comprehensive score. S70: Determine the composition of the alloy to be designed based on the effective candidate alloy compositions and their comprehensive scores.

[0006] Compositional models rely heavily on statistical regularities derived from large amounts of training data, thus offering higher prediction accuracy for alloy properties with well-covered compositional spaces. Physical parameter models, on the other hand, tend to predict alloy properties with sparsely covered compositional spaces based on physical laws. The method in this application trains the compositional model and the physical parameter model independently, preventing them from exhibiting identical systematic biases. These two models are then used to predict the alloy properties of a first candidate alloy composition. Furthermore, the first candidate alloy composition is comprehensively scored based on the two prediction results. This mutual verification between the compositional and physical parameter models ensures that the final alloy composition determined by the comprehensive score is supported by both statistical and physical laws, effectively validating its reliability. Moreover, pre-determining the first candidate alloy composition using preset physical constraints, then predicting its properties and updating it, not only guarantees that the final alloy composition conforms to physical laws but also avoids wasting time and computational resources on candidate alloy compositions that do not conform to physical laws during model prediction and updating, thereby improving the efficiency of alloy composition determination.

[0007] Secondly, embodiments of this application also provide a system for determining the composition of an alloy to be designed that meets predetermined requirements, comprising: The data acquisition module is configured to determine the training dataset according to predetermined requirements. The training dataset includes composition data, condition parameter data, physical parameter data, and performance data of multiple sets of sample alloys. The processor is configured as follows: Based on the training dataset, the component model and the physical parameter model are trained separately to obtain the trained component model and the physical parameter model. The compositional data and conditional parameter data of multiple sets of sample alloys are used as inputs to the compositional model, and the performance data of multiple sets of sample alloys are used as outputs to the compositional model; the physical parameter data of multiple sets of sample alloys are used as inputs to the physical parameter model, and the performance data of multiple sets of sample alloys are used as outputs to the physical parameter model. Based on the predetermined requirements of the alloy to be designed and the preset physical constraints, the composition of the first candidate alloy is determined; Based on the composition of the first candidate alloy and the composition model, determine the first predicted performance value of the first candidate alloy composition; based on the physical parameters of the first candidate alloy composition and the physical parameter model, determine the second predicted performance value of the first candidate alloy composition. The comprehensive score of the first candidate alloy composition is determined based on the first and second performance prediction values ​​of the first candidate alloy composition and the predetermined requirements of the alloy to be designed. Based on the first candidate alloy composition and its comprehensive score, the first candidate alloy composition is updated to determine the updated effective candidate alloy composition and its comprehensive score. The composition of the alloy to be designed is determined based on the effective candidate alloy compositions and their comprehensive scores.

[0008] Thirdly, embodiments of this application provide a computer storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the method provided in any embodiment of the first aspect of this application. Attached Figure Description

[0009] Other objects and advantages of this application will become apparent from the following description of embodiments of this application with reference to the accompanying drawings, and will help to provide a comprehensive understanding of this application.

[0010] Figure 1 The average deviation between the yield strength and the target yield strength of the effective candidate alloy composition determined by the method according to the embodiments of this application is shown, as well as the average deviation between the yield strength and the target yield strength of the alloy composition determined by the targetless random sampling method. Figure 2 The figures show the proportion of the deviation between the yield strength of the effective candidate alloy composition determined by the method according to the embodiments of this application and the target yield strength that meets the deviation requirement, as well as the proportion of the deviation between the yield strength of the alloy composition determined by the target random sampling method that meets the deviation requirement.

[0011] It should be noted that the accompanying drawings are not necessarily drawn to scale, but are shown only in a schematic manner without affecting the reader's understanding. Detailed Implementation

[0012] Exemplary embodiments of this application will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of actual implementations are described in the specification. However, it should be understood that many implementation-specific decisions must be made in the development of any such actual embodiment to achieve the developer's specific goals, such as meeting constraints related to the system and business, and these constraints may vary depending on the implementation. Furthermore, it should be understood that while development work can be very complex and time-consuming, such development work is merely a routine task for those skilled in the art who benefit from the content of this application.

[0013] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the equipment structure and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0014] In related technologies, machine learning prediction models are often used to determine the composition of alloys that meet predetermined requirements. However, the inventors of this application have discovered that existing methods for determining the composition of alloys that meet predetermined requirements using machine learning prediction models lack an inherent reliability verification mechanism, resulting in the reliability of the determined alloy composition not being guaranteed.

[0015] Based on this, embodiments of this application provide a method for determining the composition of an alloy to be designed that meets predetermined requirements, the method comprising the following steps: S10: Determine the training dataset according to the predetermined requirements. The training dataset includes the composition data, condition parameter data, physical parameter data, and performance data of multiple sets of sample alloys. S20: Based on the training dataset, train the component model and the physical parameter model respectively to obtain the trained component model and physical parameter model; The compositional data and conditional parameter data of multiple sets of sample alloys are used as inputs to the compositional model, and the performance data of multiple sets of sample alloys are used as outputs to the compositional model; the physical parameter data of multiple sets of sample alloys are used as inputs to the physical parameter model, and the performance data of multiple sets of sample alloys are used as outputs to the physical parameter model. S30: Determine the composition of the first candidate alloy based on the predetermined requirements of the alloy to be designed and the preset physical constraints; S40: Based on the first candidate alloy composition and composition model, determine the first predicted performance value of the first candidate alloy composition; based on the physical parameters and physical parameter model of the first candidate alloy composition, determine the second predicted performance value of the first candidate alloy composition. S50: Determine the comprehensive score of the first candidate alloy component based on the first and second performance prediction values ​​of the first candidate alloy component; S60: Based on the first candidate alloy composition and its comprehensive score, update the first candidate alloy composition and determine the updated effective candidate alloy composition and its comprehensive score. S70: Determine the composition of the alloy to be designed based on the effective candidate alloy compositions and their comprehensive scores.

[0016] Compositional models rely heavily on statistical regularities derived from large amounts of training data, thus offering higher prediction accuracy for alloy properties with well-covered compositional spaces. Physical parameter models, on the other hand, tend to predict alloy properties with sparsely covered compositional spaces based on physical laws. The method in this application trains the compositional model and the physical parameter model independently, preventing them from exhibiting identical systematic biases. These two models are then used to predict the alloy properties of a first candidate alloy composition. Furthermore, the first candidate alloy composition is comprehensively scored based on the two prediction results. This mutual verification between the compositional and physical parameter models ensures that the final alloy composition determined by the comprehensive score is supported by both statistical and physical laws, effectively validating its reliability. Moreover, pre-determining the first candidate alloy composition using preset physical constraints, then predicting its properties and updating it, not only guarantees that the final alloy composition conforms to physical laws but also avoids wasting time and computational resources on candidate alloy compositions that do not conform to physical laws during model prediction and updating, thereby improving the efficiency of alloy composition determination.

[0017] In some embodiments, the composition of the alloy can be the mole fraction of the various elements that make up the alloy.

[0018] In some embodiments, the predetermined requirements may include the predetermined alloy system type, composition rules, condition parameters, and target performance requirements.

[0019] In some embodiments, the alloy system type may include a high-entropy alloy system, a refractory high-entropy alloy system, a refractory metal alloy system, a transition metal alloy system, and an alloy system with one of the following elements as the base element: molybdenum, tungsten, tantalum, niobium, rhenium, iron, nickel, cobalt, titanium, aluminum, or copper.

[0020] In some embodiments, the composition rules may include equal atomic ratio rules, near-equal atomic ratio rules, rules where the content of matrix elements is greater than a predetermined ratio, rules where each component element is within a predetermined content range, rules where each component element satisfies a predetermined ratio relationship, rules where specified elements must be present or must not be present, and rules where the total content of one or more groups of elements meets a predetermined range.

[0021] In some embodiments, it can be understood that the aforementioned predetermined ratio, preset content range, preset ratio relationship, and preset range are all determined by those skilled in the art based on the actual situation.

[0022] In some embodiments, the condition parameters may include alloy preparation process parameters, heat treatment parameters, processing parameters, and performance testing parameters. The preparation process parameters may include melting temperature, sintering temperature, sintering time, cooling rate, and forming pressure; the heat treatment parameters may include solution treatment temperature, aging temperature, annealing temperature, holding time, and cooling rate; the processing parameters may include deformation temperature, deformation amount, and strain rate; and the performance testing parameters may include test temperature, loading rate, and strain rate. It is understood that the specific condition parameters used can be determined based on the types of parameters included in the training dataset and the target usage conditions of the alloy to be designed.

[0023] In some embodiments, the condition parameter can be the test temperature in the performance test parameters. The test temperature is the temperature at which the mechanical properties of the alloy are tested or predicted, and can also be understood as the performance evaluation temperature corresponding to the target service environment of the alloy. It can be understood that the alloy's performance refers to the performance exhibited by the alloy under the target service environment; therefore, the alloy to be designed needs to meet the target performance of the alloy at a predetermined test temperature.

[0024] In some embodiments, the target performance can be the yield strength of the target.

[0025] In some embodiments, in step S10, the condition parameter data can be test temperature data.

[0026] In some embodiments, in step S10, the physical parameter data may include data on mixing entropy, atomic size difference, mixing enthalpy, Omega parameter, valence electron concentration, melting point difference, content of easily oxidizable elements, and theoretical density.

[0027] In some embodiments, in step S10, the performance data is yield strength data.

[0028] In some embodiments, in step S10, the system type of all sample alloys in the training dataset is consistent with the alloy system type in the predetermined requirements, the condition parameter data of all sample alloys is consistent with the condition parameters in the predetermined requirements, and the composition rules of all sample alloys are consistent with the composition rules in the predetermined requirements.

[0029] In such an embodiment, since the alloy system type, condition parameters, and composition rules of all sample alloys in the training dataset are consistent with the predetermined requirements, the systematic error of the model prediction caused by the difference in alloy system type, condition parameters, and composition rules can be eliminated, so that the model prediction can better reflect the relationship between the composition and properties of the alloy and the relationship between physical parameters and properties that meet the predetermined requirements of alloy system type, condition parameters, and composition rules.

[0030] In some embodiments, in step S10, the training dataset can be obtained through literature review and supplementary experiments.

[0031] In some embodiments, in step S20, the composition model is an attention network model based on a multi-head attention mechanism, so that the composition model can allocate attention according to the correlation between different elements and alloy properties, and can more accurately obtain the relationship between the composition and alloy properties of the alloy.

[0032] In some embodiments, in step S20, the physical parameter model is an attention network model based on a multi-head attention mechanism, so that the physical parameter model can allocate attention according to the correlation between different physical parameters and alloy properties, and can more accurately obtain the relationship between the physical parameters and alloy properties of the alloy.

[0033] In some embodiments, step S30 specifically includes: S31: according to the predetermined requirements of the alloy to be designed, selecting multiple sampling points in the composition space that meets the predetermined requirements, and randomly generating multiple initial candidate alloy compositions in the local composition space corresponding to each sampling point; S32: determining the physical parameters of the multiple initial candidate alloy compositions at each sampling point; S33: determining the first candidate alloy composition at each sampling point according to the preset physical constraints and the physical parameters of the multiple initial candidate alloy compositions.

[0034] In this embodiment, compared to selecting only one sampling point and generating initial candidate alloy compositions from the local composition space corresponding to that sampling point, this application selects multiple sampling points in a composition space that meets predetermined requirements, and randomly generates multiple initial candidate alloy compositions in the local composition space corresponding to each sampling point. This can improve the coverage and diversity of the initial candidate alloy compositions in the entire feasible composition space, so that the subsequent process of updating the first candidate alloy composition obtained from the initial candidate alloy composition screening can cover different composition regions. This reduces the risk of the update process being concentrated in a local region or trapped in a local optimum due to an imperfect initial sampling position, and increases the probability of discovering high-scoring candidate alloy compositions distributed in different composition regions.

[0035] In some embodiments, in step S31, within the component space that meets the predetermined requirements, those skilled in the art may use random sampling, Latin hypercube sampling, or low-difference sequence sampling methods to sample and select multiple sampling points, depending on the actual situation.

[0036] In some embodiments, step S31 specifically includes: S311: determining optional constituent elements according to the requirements of the predetermined alloy system type of the alloy to be designed; S312: determining a feasible composition space according to the requirements of the predetermined composition rules of the alloy to be designed and the optional constituent elements; S313: selecting multiple sampling points in the feasible composition space, and randomly generating multiple initial candidate alloy compositions in the local composition space corresponding to each sampling point. In some embodiments, in step S32, the physical parameters of the initial candidate alloy compositions and the corresponding preset physical constraints can be determined by those skilled in the art based on the system type, target performance, and application conditions of the alloy to be designed.

[0037] In some embodiments, such as step S32, the physical parameters may include: mixing entropy, atomic size difference, mixing enthalpy, Omega parameter, melting point difference, and easily oxidized elements. In some embodiments, such as step S33, when the alloy to be designed is a high-entropy alloy system, a refractory high-entropy alloy system, or a near-equiatomic-ratio multi-principal-element alloy system, the preset physical constraints may include the following six items: mixing entropy greater than 12.5 J / (mol·K), atomic size difference less than 6.6%, mixing enthalpy in the range of -15 to 5 kJ / mol, Omega parameter greater than 1.1, melting point difference less than 500K, and total content of easily oxidized elements less than 40%.

[0038] In some embodiments, in step S33, the initial candidate alloy composition whose physical parameters satisfy a predetermined number of preset physical constraints is determined as the first candidate alloy composition.

[0039] In some embodiments, the number of predetermined items is greater than or equal to 3.

[0040] In some embodiments, step S50 specifically includes: S51: determining the consistency score of the first candidate alloy component based on the first performance prediction value and the second performance prediction value of the first candidate alloy component; S52: determining the target matching score of the first candidate alloy component based on the first performance prediction value and the second performance prediction value of the first candidate alloy component and the target performance of the alloy to be designed; S53: determining the physical verification score of the first candidate alloy component based on the physical parameters of the first candidate alloy component and the predetermined physical constraints; S54: determining the comprehensive score of the first candidate alloy component based on the consistency score, the target matching score, and the physical verification score of the first candidate alloy component.

[0041] In such an embodiment, the consistency score can characterize the reliability of the first candidate alloy composition cross-validated by the composition model and the physical parameter model, the target matching score can characterize the degree of matching between the first and second performance prediction values ​​of the first candidate alloy composition and the target performance, and the physical verification score can characterize the degree of conformity between the first candidate alloy composition and physical laws. The comprehensive score of the first candidate alloy composition determined by these three scores can jointly verify the reliability of the first candidate alloy composition from three perspectives: the reliability of the cross-validation of the two models, the degree of matching with the target performance, and the degree of conformity with physical laws.

[0042] In some embodiments, in step S51, the ratio of the smaller of the first performance prediction value and the second performance prediction value to the larger value is determined as the consistency score of the first candidate alloy composition.

[0043] In this embodiment, the consistency score is set as the ratio of the smaller of the first and second performance prediction values ​​to the larger value. This allows the consistency score to clearly reflect the difference between the first and second performance prediction values, directly reflecting the reliability of the first candidate alloy composition cross-validated by the composition model and the physical parameter model. When the consistency score is close to 1, meaning the first and second performance prediction values ​​are similar, it indicates that the first candidate alloy composition is supported by both statistical and physical laws. When the consistency score is much less than 1, meaning the first and second performance prediction values ​​differ significantly, it indicates that the first candidate alloy composition may be located in the extrapolation region of the model or that there is data noise, resulting in low reliability.

[0044] In some embodiments, step S52 specifically includes: S521: determining the mean of the first performance prediction value and the second performance prediction value; S522: determining the ratio of the mean to the target performance of the alloy to be designed, wherein the ratio is determined as the target matching score of the first candidate alloy composition.

[0045] In this embodiment, the target matching score is set as the ratio of the mean of the first performance prediction value and the second performance prediction value to the target performance. This allows the target matching score to clearly and explicitly reflect the degree of matching between the performance prediction value of the first candidate alloy component and the target performance, thereby measuring the gap between the performance prediction value of the first candidate alloy component and the target performance. Furthermore, since the target matching score considers both the first performance prediction value obtained from the composition model and the second performance prediction value obtained from the physical parameter model, the reliability of the target matching score is enhanced.

[0046] In some embodiments, step S53 specifically includes: S531: determining the number of preset physical constraints satisfied by the physical parameters of the first candidate alloy composition; S532: determining the degree to which the physical parameters of the first candidate alloy composition satisfy each preset physical constraint; S533: determining the physical verification score of the first candidate alloy composition based on the number of preset physical constraints satisfied by the physical parameters of the first candidate alloy composition and the degree to which the physical parameters of the first candidate alloy composition satisfy each preset physical constraint.

[0047] In such an embodiment, the physical verification score of the first candidate alloy composition is determined by comprehensively considering the number of preset physical constraints satisfied by the physical parameters of the first candidate alloy composition, as well as the degree to which each preset physical constraint is satisfied. This can more accurately reflect the physical rationality of the first candidate alloy composition.

[0048] In some embodiments, step S54 specifically includes: S541: setting weighting coefficients; S542: determining the comprehensive score of the first candidate alloy composition based on the consistency score, target matching score, and physical verification score of the first candidate alloy composition and the set weighting coefficients.

[0049] In this embodiment, by setting weighting coefficients, different weighting coefficients can be assigned to the consistency score, target matching score, and physical verification score based on different research objectives when determining the comprehensive score. This allows the comprehensive score to reflect the reliability, target performance matching degree, and physical rationality of the first candidate alloy composition with varying emphasis, depending on the research objective. For example, for alloy designs still in the exploratory stage, the weights of the consistency score and physical verification score can be increased, resulting in a comprehensive score that better reflects the reliability and physical rationality of the first candidate alloy composition, allowing for the selection of alloy compositions with high reliability and physical rationality. For mature alloy design systems with substantial prior knowledge, the weight of the target matching score can be increased, resulting in a comprehensive score that better reflects the degree of matching between the first candidate alloy composition and the target performance, thus enabling the pursuit of high-performance alloy compositions.

[0050] In some embodiments, in step S60, for each sampling point, the first candidate alloy composition is updated based on the first candidate alloy composition and the comprehensive score of the first candidate alloy composition, and the updated effective candidate alloy composition for each sampling point is determined; wherein, the number of first candidate alloy compositions for each sampling point is equal.

[0051] In this embodiment, the first candidate alloy composition for each sampling point is updated independently to determine the effective candidate alloy composition corresponding to each sampling point. This approach allows for the acquisition of high-scoring candidate alloy compositions in different local composition spaces, improving the compositional diversity of the final determined scheme. Furthermore, the updating process for the first candidate alloy composition at each sampling point can be performed in parallel, thereby improving update efficiency. Additionally, the number of first candidate alloy compositions at each sampling point is the same, which helps reduce the differences in the updating process for each sampling point, while also improving update efficiency.

[0052] In some embodiments, step S60 specifically includes: S61: For each sampling point, determine the observation dataset. Each observation dataset includes multiple first candidate alloy compositions for the corresponding sampling point and a comprehensive score of the multiple first candidate alloy compositions. S62: For each observation dataset, train the corresponding Gaussian process surrogate model based on the first candidate alloy composition and its comprehensive score in the observation dataset; S63: Set the acquisition function according to each Gaussian process proxy model; S64: Based on the acquisition function, determine the second candidate alloy composition from the composition space that meets the predetermined requirements, and determine the comprehensive score of the second candidate alloy composition according to steps S40-S54; S65: The improvement of the overall score for determining the second candidate alloy composition relative to the highest overall score in the observation dataset before the start of this iteration; S66: Add the second candidate alloy composition and its comprehensive score to the corresponding observation dataset, repeat steps S61-S65, iterate the Gaussian process surrogate model, and determine a second candidate alloy composition, a comprehensive score of the second candidate alloy composition, and an improvement of the comprehensive score of the second candidate alloy composition in each iteration. For each sampling point, if the improvement of the comprehensive score of the second candidate alloy composition obtained in each iteration does not meet the preset conditions in the first predetermined number of iterations, return to step S31, change the sampling point in the composition space that meets the predetermined requirements of the alloy to be designed, reselect multiple initial candidate alloy compositions, and repeat steps S32-S65 until the preset maximum number of iterations is reached or there is a candidate alloy composition in the corresponding observation dataset that meets the target comprehensive score; Among them, when the preset maximum number of iterations is reached, the candidate alloy component with the highest comprehensive score in the observation dataset is determined as a valid candidate alloy component, or when there is a candidate alloy component in the observation dataset whose comprehensive score reaches the target comprehensive score, the candidate alloy component whose comprehensive score reaches the target comprehensive score is determined as a valid candidate alloy component.

[0053] In this embodiment, compared to merging the first candidate alloy compositions of all sampling points and optimizing them uniformly, this application optimizes the first candidate alloy composition corresponding to each sampling point independently. This allows different Gaussian process surrogate models to learn the "alloy composition-comprehensive score" relationship in different local composition spaces and update the candidate alloy compositions independently within their respective local composition spaces. This reduces the impact of differences in the "alloy composition-comprehensive score" relationship between different local composition regions on the fitting accuracy of the same Gaussian process surrogate model. Furthermore, it prevents local regions with more sampling data or higher comprehensive scores from prematurely dominating the entire search process, ensuring the optimization process continuously searches multiple different composition regions and reducing the risk of the search process concentrating on a single local optimum. This allows for the separate acquisition of high-scoring candidate alloy compositions in different local composition regions, improving the compositional diversity of the final determined scheme. Moreover, the optimization processes corresponding to each sampling point are independent and can be performed in parallel, which is beneficial for improving optimization efficiency and the utilization efficiency of computational resources.

[0054] Furthermore, if the improvement of the comprehensive score of the second candidate alloy composition obtained in each iteration does not meet the preset conditions in the first predetermined number of iterations, it indicates that the reliability of the first candidate alloy composition in the observation dataset corresponding to the sampling point is low. Therefore, the process returns to step S31 to change the sampling point and reselect the initial candidate alloy composition in order to find new possible effective candidate alloy compositions, thereby improving the efficiency of the update process and finding as many possible effective candidate alloy compositions as possible.

[0055] In some embodiments, the first predetermined number of iterations, the preset maximum number of iterations, the preset conditions, and the target comprehensive score can all be determined by those skilled in the art based on the actual situation.

[0056] In some embodiments, in step S64, within a second predetermined number of iterations, a global search is performed in the entire composition space corresponding to the Gaussian process surrogate model according to the acquisition function to determine the second candidate alloy composition; from the next iteration after completing the second predetermined number of iterations, the search is only performed in the composition space where the function value of the acquisition function is greater than a preset threshold to determine the second candidate alloy composition. In such embodiments, the acquisition function can fully search the entire composition space in the early stage of the iteration to avoid missing possible candidate compositions, and can focus on searching high-potential composition spaces in the later stage of the iteration to more efficiently determine the second candidate alloy composition with a higher comprehensive score.

[0057] In some embodiments, the preset threshold can be determined by those skilled in the art based on the actual situation.

[0058] In some embodiments, the second predetermined number of iterations can be determined by those skilled in the art based on the actual situation.

[0059] In some embodiments, in step S64, starting from the next iteration after completing the second predetermined number of iterations, in the composition space where the function value of the acquisition function is greater than a preset threshold, the candidate alloy composition corresponding to the maximum function value of the acquisition function is determined as the second candidate alloy composition.

[0060] In such an embodiment, selecting the candidate alloy component with the largest function value from the component space where the function value of the acquisition function is greater than a preset threshold as the second candidate alloy component helps to select the candidate alloy component with a high comprehensive score more efficiently and improve the optimization efficiency.

[0061] To further illustrate the effectiveness of the method for determining the composition of an alloy to be designed that meets predetermined requirements according to the embodiments of this application, the following examples are used for illustration.

[0062] Example: The alloy to be designed is a molybdenum alloy, and its predetermined requirements include: the alloy system type is a refractory metal alloy system; the composition rule is a near-equal atomic ratio rule, in which each element accounts for 15-25%; the test temperature is 1200K; and the target yield strength is 600MPa.

[0063] In step S10, the training dataset is determined, which includes composition data, test temperature data, physical parameter data, and yield strength data of 874 sample molybdenum alloys.

[0064] The composition data of the sample molybdenum alloy includes the mole fraction of each element that makes up the sample molybdenum alloy; the physical parameter data of the sample molybdenum alloy includes the mixing entropy, atomic size difference, mixing enthalpy, Omega parameter, valence electron concentration, melting point difference, content of easily oxidized elements, and theoretical density.

[0065] Among them, the test temperature data of 874 molybdenum alloy samples were all 1200K, the composition rules were all near-equal atomic ratio rules, and the proportion of each element was 15-25%.

[0066] In step S20, a composition model is trained based on the training dataset. The composition model is an attention network model based on a multi-head attention mechanism, which includes three attention layers, each with eight attention heads. The input of the composition model is the composition data and test temperature data of the molybdenum alloy sample, and the output is the yield strength data of the molybdenum alloy sample. A physical parameter model is also trained based on the training dataset. The physical parameter model is an attention network model based on a multi-head attention mechanism, which includes three attention layers, each with eight attention heads. The input of the physical parameter model is the physical parameter data of the molybdenum alloy sample, and the output is the yield strength data of the molybdenum alloy sample.

[0067] In step S31, based on the predetermined requirements of the molybdenum alloy to be designed, 50 sampling points are selected in the composition space of Mo, W, Ta, Nb and Re, and 1000 initial candidate alloy compositions are randomly generated in the local composition space corresponding to each sampling point.

[0068] In step S32, the mixing entropy, atomic size difference, mixing enthalpy, Omega parameter, valence electron concentration, melting point difference, and content of easily oxidized elements for each initial candidate alloy composition at each sampling point are determined.

[0069] In step S33, initial candidate alloy compositions that meet at least three of the following physical constraints are identified as first candidate alloy compositions: mixing entropy greater than 12.5 J / (mol·K); atomic size difference less than 6.6%; mixing enthalpy in the range of -15 to 5 kJ / mol; Omega parameter greater than 1.1; melting point difference less than 500K; total content of easily oxidized elements less than 40%; and 200 first candidate alloy compositions are retained for each sampling point.

[0070] If there are fewer than 200 first candidate alloy components after screening by physical constraints at a certain sampling point, new initial candidate alloy components are generated in the local composition space corresponding to that sampling point, and screened by physical constraints until 200 first candidate alloy components are obtained.

[0071] In step S40, the first candidate alloy composition and the test temperature of 1200K are input into the composition model to obtain the first performance prediction value of the first candidate alloy composition; the physical parameters of the first candidate alloy composition are input into the physical parameter model to obtain the second performance prediction value of the first candidate alloy composition.

[0072] In step S51, for each first candidate alloy composition, the smaller of its first performance prediction value and second performance prediction value is divided by the larger value, and the resulting ratio is determined as the consistency score of the first candidate alloy composition, with the consistency score ranging from 0 to 1.

[0073] In step S52, for each first candidate alloy composition, the mean of its first performance prediction value and second performance prediction value is determined, and the ratio of the mean value to the target yield strength of 600 MPa is determined. The obtained ratio is the target matching score of the first candidate alloy composition, and the target matching score ranges from 0 to 1.

[0074] In step S53, for each first candidate alloy composition, the number of physical constraints satisfied by its physical parameters and the degree to which each physical constraint is satisfied are determined. Based on the number of physical constraints satisfied by the physical parameters of the first candidate alloy composition and the degree to which each physical constraint is satisfied, the physical verification score of the first candidate alloy composition is determined. The physical verification score ranges from 0 to 1.

[0075] In step S54, the weighting coefficient for consistency score is set to 0.3, the weighting coefficient for target matching score is set to 0.4, and the weighting coefficient for physical verification score is set to 0.3. Based on the set weighting coefficients and the consistency score, target matching score, and physical verification score of each first candidate alloy component, the comprehensive score of each first candidate alloy component is determined.

[0076] In step S61, for each sampling point, multiple observation datasets are determined. Each observation dataset includes 200 first candidate alloy compositions for the corresponding sampling point and their comprehensive scores.

[0077] In step S62, for each observation dataset, a corresponding Gaussian process surrogate model is trained based on the first candidate alloy composition and its comprehensive score in the observation dataset.

[0078] In step S63, the acquisition function is set according to each Gaussian process proxy model.

[0079] In step S64, the second candidate alloy composition is selected in the composition space of the Gaussian process surrogate model using the acquisition function, and the comprehensive score of the second candidate alloy composition is determined according to steps S40-S54.

[0080] In the first 100 iterations, a global search is performed in the entire composition space corresponding to the Gaussian process surrogate model based on the acquisition function to determine the second candidate alloy composition. From the 101st iteration to the 300th iteration, the search is only performed in the composition space where the function value of the acquisition function is greater than a preset threshold, and the candidate alloy composition corresponding to the maximum function value of the acquisition function is determined as the second candidate alloy composition.

[0081] In step S65, the overall score for determining the second candidate alloy composition is an improvement over the highest overall score in the observation dataset prior to this iteration.

[0082] In step S66, the second candidate alloy composition and its comprehensive score are added to the corresponding observation dataset. Steps S61-S65 are repeated to iterate the Gaussian process surrogate model. In each iteration, a second candidate alloy composition, the comprehensive score of the second candidate alloy composition, and the improvement of the comprehensive score of the second candidate alloy composition are determined.

[0083] For each sampling point, if the improvement of the comprehensive score of the second candidate alloy composition obtained in each of the 20 consecutive iterations does not meet the preset conditions, return to step S31, change the sampling point in the composition space that meets the predetermined requirements, reselect multiple initial candidate alloy compositions, and repeat steps S32-S65 until 300 iterations are reached or the comprehensive score of the obtained second candidate alloy composition reaches the target comprehensive score.

[0084] Among them, the candidate alloy component with the highest comprehensive score in the observation dataset after the 300th iteration is determined as a valid candidate alloy component, or when there is a candidate alloy component in the observation dataset whose comprehensive score reaches the target comprehensive score, the candidate alloy component whose comprehensive score reaches the target comprehensive score is determined as a valid candidate alloy component; and one valid candidate alloy component is obtained for each sampling point.

[0085] In step S70, all valid candidate alloy compositions are ranked according to their comprehensive scores, and K-means clustering is used to remove duplicates. From these, the top 20 valid candidate compositions are selected as the composition schemes for the molybdenum alloy to be designed. For example, one determined composition scheme for the molybdenum alloy to be designed is 25%Mo, 22%Nb, 20%Ta, 18%W, and 15%Re.

[0086] See Figure 1 , Figure 1 The diagram illustrates the average deviation between the yield strength and the target yield strength of valid candidate alloy compositions determined by the method according to embodiments of this application, and the average deviation between the yield strength and the target yield strength of alloy compositions determined using a targetless random sampling method. Figure 1 In this context, the smaller the average deviation between the yield strength of an alloy composition and the target yield strength, the closer the yield strength of that alloy composition is to the target yield strength. From... Figure 1 As can be seen, the average deviation between the yield strength and the target yield strength of the effective candidate alloy composition determined by the method of this application embodiment is reduced by about 30.67% compared with the alloy composition determined by the random sampling method without a target. The yield strength performance of the effective candidate alloy composition determined by the method of this application embodiment is greatly improved to be close to the target yield strength performance.

[0087] See Figure 2 , Figure 2 The diagram illustrates the proportion of effective candidate alloy compositions determined by the method according to embodiments of this application whose yield strengths meet the target yield strength requirements, and the proportion of alloy compositions determined using a targetless random sampling method whose yield strengths meet the target yield strength requirements. Figure 2 As can be seen from the embodiments of this application, the effective candidate alloy composition determined by the method has a greater number of deviations between its yield strength and the target yield strength compared to the alloy composition determined by the random sampling method without a target. Overall, the yield strength of the effective candidate alloy composition is closer to the target yield strength.

[0088] Embodiments of this application also provide a system for determining the composition of an alloy to be designed that meets predetermined requirements, comprising: a data acquisition module configured to determine a training dataset according to predetermined requirements, the training dataset including composition data, condition parameter data, physical parameter data, and performance data of multiple sets of sample alloys; The processor is configured as follows: Based on the training dataset, the component model and the physical parameter model are trained separately to obtain the trained component model and the physical parameter model. The compositional data and conditional parameter data of multiple sets of sample alloys are used as inputs to the compositional model, and the performance data of multiple sets of sample alloys are used as outputs to the compositional model; the physical parameter data of multiple sets of sample alloys are used as inputs to the physical parameter model, and the performance data of multiple sets of sample alloys are used as outputs to the physical parameter model. Based on the predetermined requirements of the alloy to be designed and the preset physical constraints, the composition of the first candidate alloy is determined; Based on the composition of the first candidate alloy and the composition model, determine the first predicted performance value of the first candidate alloy composition; based on the physical parameters of the first candidate alloy composition and the physical parameter model, determine the second predicted performance value of the first candidate alloy composition. The comprehensive score of the first candidate alloy composition is determined based on the first and second performance prediction values ​​of the first candidate alloy composition. Based on the first candidate alloy composition and its comprehensive score, the first candidate alloy composition is updated to determine the updated effective candidate alloy composition and its comprehensive score. The composition of the alloy to be designed is determined based on the effective candidate alloy compositions and their comprehensive scores.

[0089] Compositional models rely heavily on statistical regularities derived from large amounts of training data, thus offering higher prediction accuracy for alloy properties with well-covered compositional spaces. Physical parameter models, on the other hand, tend to predict alloy properties with sparsely covered compositional spaces based on physical laws. The system in this application trains the compositional model and the physical parameter model independently, preventing them from exhibiting identical systematic biases. These two models are then used to predict the alloy properties of a first candidate alloy composition. Furthermore, the system comprehensively scores the first candidate alloy composition based on both predictions. This mutual verification between the compositional and physical parameter models ensures that the final alloy composition determined by the comprehensive score is supported by both statistical and physical laws, effectively validating its reliability. Moreover, pre-determining the first candidate alloy composition using preset physical constraints, then predicting its properties and updating it, not only guarantees that the final alloy composition conforms to physical laws but also avoids wasting time and computational resources on candidate alloy compositions that do not conform to physical laws during model prediction and updating, thereby improving the efficiency of alloy composition determination.

[0090] Embodiments of this application also provide a computer storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the method provided in any embodiment of the first aspect of this application.

[0091] Regarding the embodiments of this application, it should also be noted that, without conflict, the embodiments of this application and the features in the embodiments can be combined with each other to obtain new embodiments.

[0092] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for determining the composition of an alloy to be designed that meets predetermined requirements, characterized in that, It includes the following steps: S10: Determine the training dataset according to the predetermined requirements. The training dataset includes composition data, condition parameter data, physical parameter data, and performance data of multiple sets of sample alloys. S20: Based on the training dataset, train the component model and the physical parameter model respectively to obtain the trained component model and the physical parameter model; The composition data and condition parameter data of the multiple sets of sample alloys are used as inputs to the composition model, and the performance data of the multiple sets of sample alloys are used as outputs to the composition model; the physical parameter data of the multiple sets of sample alloys are used as inputs to the physical parameter model, and the performance data of the multiple sets of sample alloys are used as outputs to the physical parameter model. S30: Determine the composition of the first candidate alloy according to the predetermined requirements of the alloy to be designed and the preset physical constraints; S40: Based on the first candidate alloy composition and the composition model, determine the first predicted performance value of the first candidate alloy composition; based on the physical parameters of the first candidate alloy composition and the physical parameter model, determine the second predicted performance value of the first candidate alloy composition. S50: Determine the comprehensive score of the first candidate alloy component based on the first and second performance prediction values ​​of the first candidate alloy component and the predetermined requirements of the alloy to be designed. S60: Based on the first candidate alloy composition and its comprehensive score, update the first candidate alloy composition to determine the updated effective candidate alloy composition and its comprehensive score. S70: Determine the composition of the alloy to be designed based on the effective candidate alloy compositions and their comprehensive scores.

2. The method according to claim 1, characterized in that, The S30 steps specifically include: S31: According to the predetermined requirements of the alloy to be designed, multiple sampling points are selected in the composition space that meets the predetermined requirements, and multiple initial candidate alloy compositions are randomly generated in the local composition space corresponding to each sampling point. S32: Determine the physical parameters of the plurality of initial candidate alloy compositions for each of the sampling points; S33: Determine the first candidate alloy composition for each sampling point based on the preset physical constraints and the physical parameters of the plurality of initial candidate alloy compositions.

3. The method according to claim 2, characterized in that, In step S32, the physical parameters include: mixing entropy, atomic size difference, mixing enthalpy, Omega parameter, melting point difference, and easily oxidized elements.

4. The method according to claim 3, characterized in that, In step S33, when the alloy to be designed is a high-entropy alloy system, a refractory high-entropy alloy system, or a near-equiatomic ratio multi-principal element alloy system, the preset physical constraints include the following six items: The mixing entropy is greater than 12.5 J / (mol·K), the atomic size difference is less than 6.6%, the mixing enthalpy ranges from -15 to 5 kJ / mol, the Omega parameter is greater than 1.1, the melting point difference is less than 500K, and the total content of easily oxidized elements is less than 40%.

5. The method according to any one of claims 2-4, characterized in that, In step S33, the initial candidate alloy composition whose physical parameters satisfy a predetermined number of the preset physical constraints is determined as the first candidate alloy composition.

6. The method according to claim 5, characterized in that, The S50 steps specifically include: S51: Determine the consistency score of the first candidate alloy component based on the first performance prediction value and the second performance prediction value of the first candidate alloy component; S52: Determine the target matching score of the first candidate alloy component based on the first performance prediction value and the second performance prediction value of the first candidate alloy component and the target performance of the alloy to be designed. S53: Determine the physical verification score of the first candidate alloy composition based on the physical parameters of the first candidate alloy composition and the predetermined physical constraints. S54: Determine the comprehensive score of the first candidate alloy component based on its consistency score, target matching score, and physical verification score.

7. The method according to claim 6, characterized in that, In step S51, The ratio of the smaller of the first performance prediction value and the second performance prediction value to the larger value is determined as the consistency score of the first candidate alloy composition.

8. The method according to claim 6, characterized in that, The S52 steps specifically include: S521: Determine the mean of the first performance prediction value and the second performance prediction value; S522: Determine the ratio of the mean to the target performance of the alloy to be designed, and the ratio is determined as the target matching score of the first candidate alloy composition.

9. The method according to claim 6, characterized in that, The S53 steps specifically include: S531: Determine the number of preset physical constraints that the physical parameters of the first candidate alloy composition satisfy; S532: Determine the degree to which the physical parameters of the first candidate alloy composition satisfy each of the preset physical constraints; S533: Determine the physical verification score of the first candidate alloy component based on the number of preset physical constraints satisfied by the physical parameters of the first candidate alloy component, and the degree to which the physical parameters of the first candidate alloy component satisfy each preset physical constraint.

10. The method according to claim 6, characterized in that, The S54 steps specifically include: S541: Set the weighting coefficients; S542: Determine the comprehensive score of the first candidate alloy component based on its consistency score, target matching score, physical verification score, and the set weighting coefficient.

11. The method according to claim 2 or 10, characterized in that, In step S60, for each sampling point, the first candidate alloy composition is updated based on the first candidate alloy composition and the comprehensive score of the first candidate alloy composition, and the updated effective candidate alloy composition for each sampling point is determined. The number of first candidate alloy components is equal at each of the sampling points.

12. The method according to claim 11, characterized in that, The S60 steps specifically include: S61: For each of the sampling points, an observation dataset is determined, and each observation dataset includes multiple first candidate alloy components corresponding to the sampling point and a comprehensive score of the multiple first candidate alloy components; S62: For each of the observation datasets, train a corresponding Gaussian process proxy model based on the first candidate alloy composition and its comprehensive score in the observation dataset; S63: Set the acquisition function according to each Gaussian process proxy model; S64: Based on the acquisition function, determine the second candidate alloy composition from the composition space that meets the predetermined requirements, and determine the comprehensive score of the second candidate alloy composition according to steps S40-S54; S65: Determine the improvement of the comprehensive score of the second candidate alloy composition relative to the highest comprehensive score in the observation dataset before the start of this iteration; S66: Add the second candidate alloy composition and its comprehensive score to the corresponding observation dataset, repeat steps S61-S65, iterate the Gaussian process surrogate model, and determine a second candidate alloy composition, the comprehensive score of the second candidate alloy composition, and the improvement of the comprehensive score of the second candidate alloy composition in each iteration; For each sampling point, if the improvement of the comprehensive score of the second candidate alloy composition obtained in each iteration does not meet the preset condition in the iteration of the first predetermined number of iterations, return to step S31, change the sampling point in the composition space that meets the predetermined requirements of the alloy to be designed, reselect multiple initial candidate alloy compositions, and repeat steps S32-S65 until the preset maximum number of iterations is reached or there is a candidate alloy composition in the corresponding observation dataset whose comprehensive score reaches the target comprehensive score; Specifically, when the preset maximum number of iterations is reached, the candidate alloy component with the highest comprehensive score in the observation dataset is determined as a valid candidate alloy component; or when there is a candidate alloy component in the observation dataset whose comprehensive score reaches the target comprehensive score, the candidate alloy component whose comprehensive score reaches the target comprehensive score is determined as a valid candidate alloy component.

13. The method according to claim 12, characterized in that, In step S64, Within the second predetermined number of iterations, a global search is performed in the entire composition space corresponding to the Gaussian process proxy model according to the acquisition function to determine the second candidate alloy composition; Starting from the next iteration after completing the second predetermined number of iterations, the search is conducted only in the composition space where the function value of the acquisition function is greater than a preset threshold, in order to determine the second candidate alloy composition.

14. The method according to claim 13, characterized in that, In step S64, Starting from the next iteration after completing the second predetermined number of iterations, in the composition space where the function value of the acquisition function is greater than the preset threshold, the candidate alloy composition corresponding to the maximum function value of the acquisition function is determined as the second candidate alloy composition.

15. A system for determining the composition of an alloy to be designed that meets predetermined requirements, characterized in that, It includes: The data acquisition module is configured to determine a training dataset according to the predetermined requirements. The training dataset includes composition data, condition parameter data, physical parameter data, and performance data of multiple sets of sample alloys. The processor is configured as follows: Based on the training dataset, the component model and the physical parameter model are trained respectively to obtain the trained component model and the physical parameter model. The composition data and condition parameter data of the multiple sets of sample alloys are used as inputs to the composition model, and the performance data of the multiple sets of sample alloys are used as outputs to the composition model; the physical parameter data of the multiple sets of sample alloys are used as inputs to the physical parameter model, and the performance data of the multiple sets of sample alloys are used as outputs to the physical parameter model. The composition of the first candidate alloy is determined based on the predetermined requirements of the alloy to be designed and the preset physical constraints. Based on the first candidate alloy composition and the composition model, a first performance prediction value of the first candidate alloy composition is determined; based on the physical parameters of the first candidate alloy composition and the physical parameter model, a second performance prediction value of the first candidate alloy composition is determined. Based on the first and second performance prediction values ​​of the first candidate alloy composition and the predetermined requirements of the alloy to be designed, a comprehensive score for the first candidate alloy composition is determined. Based on the first candidate alloy composition and its comprehensive score, the first candidate alloy composition is updated to determine the updated effective candidate alloy composition and its comprehensive score. The composition of the alloy to be designed is determined based on the effective candidate alloy compositions and their comprehensive scores.

16. A computer storage medium, characterized in that, It stores a computer program thereon, which is executed by a processor to implement the method as described in any one of claims 1-14.

Images