A high-entropy alloy composition design method and system covering the entire periodic table of elements

Abstract

The application relates to a high-entropy alloy component design method and system covering the whole element periodic table, and the method comprises the following steps: collecting general material database resources in advance, extracting the core features and crystal graphs of the materials, and training a pre-trained large model; collecting and obtaining performance data of multi-element alloys, and constructing a probe test set; the probe test set is divided into multiple subspaces according to element types and phase structures; the pre-trained large model is used for alloy performance prediction in each subspace, and high-performance subspaces are screened out according to the prediction results; an optimization algorithm is used for alloy component optimization in each high-performance subspace; the restrictions on raw material elements and target alloy characteristics are added to ensure the solid solution of the target alloy; the verification is carried out through a theoretical calculation and an experimental verification method, and if the verification is passed, the high-entropy alloy component is finally obtained. Compared with the prior art, the application has the advantages of reducing the data cost of high-entropy alloy component design, being accurate and being able to effectively guide experiments.

Description

Technical Field

[0001] This invention relates to the field of high-entropy alloy composition design technology, and in particular to a method and system for designing high-entropy alloy compositions covering the entire periodic table. Background Technology

[0002] High-entropy alloys are a new type of alloy material with a mixing entropy greater than 1.5R. They are typically composed of more than four metallic elements and have attracted attention in aerospace and nuclear industries due to their excellent mechanical properties, high thermal stability, corrosion resistance, and radiation resistance. However, as the number of alloy components increases, their chemical complexity rises, making it difficult to screen high-performance high-entropy alloy matrix materials through theoretical calculations. With the development of artificial intelligence technology, it has become possible to use deep neural networks to predict the properties and perform reverse design of high-entropy alloys with potential high mechanical and thermal transport properties.

[0003] Traditional machine learning-assisted high-entropy alloy composition design techniques (such as the inventions with publication numbers CN113870957A, CN114678086A, and CN113870957A) face a key technical bottleneck: the high cost of acquiring data on the mechanical, thermal, and heat transport properties of high-entropy alloys. Most studies collect training data from published literature, but these data sources are complex, highly non-standardized, small in volume, and have simple elemental compositions. Most data only involve about ten elements suitable for alloy smelting, allowing for fine-tuning of element proportions within existing high-entropy alloy compositions to achieve better ratios. This approach makes it difficult to discover potential new high-entropy alloys and lacks sufficient depth of exploration of the periodic table. However, high-throughput experimental preparation and verification of high-entropy alloy performance incurs high costs for raw materials, labor, and equipment maintenance. Furthermore, the "short-range order, long-range disorder" structural characteristics of high-entropy alloys, as demonstrated by first-principles simulations, necessitate sufficiently large lattice models, leading to high computational costs. For some high-order physical properties, such as lattice thermal conductivity, high-throughput theoretical simulations of high-entropy alloys are almost impossible to achieve.

[0004] Through decades of effort by experimental and computational chemists, the academic community has gradually accumulated a wealth of general-purpose materials data resources, including open-source materials databases such as the Materials Project and AFLOW. These databases contain over 10,000 general-purpose materials databases based on pure metals and intermetallic compounds, semiconductors, metal oxides, and other compounds. They include first-principles calculations of physicochemical properties such as elastic constants, various elastic moduli, Poisson's ratio, heat capacity, vibrational entropy, phonon free energy, lattice thermal conductivity, electronic thermal conductivity, Debye temperature, and Green's Eisen constant. These data typically cover a wide range of elements and complex and diverse crystal structures, and with appropriate feature extraction methods, they can meet the diverse needs of alloy design. Although high-entropy alloys are not included in these datasets, various crystals in solids follow similar physical laws. Using appropriate out-of-distribution generalization (OODG) techniques, the effective information extracted from these general datasets can be used for property prediction and reverse design of high-entropy alloys, further accelerating the development of potential high-performance high-entropy alloys. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art, such as a narrow range of selectable elements, limited available training data, and poor model training stability, and to provide a high-entropy alloy composition design method and system that covers the entire periodic table.

[0006] The objective of this invention can be achieved through the following technical solutions: A method for designing high-entropy alloy compositions covering the entire periodic table includes: Pre-collect general materials database resources, extract the core features and crystal diagrams of the materials, and use them to train the hybrid graph neural network to obtain a pre-trained large model; Collect and acquire performance data of multi-element alloys, and construct a probe test set; The probe test set is divided into multiple subspaces according to element type and phase structure; the pre-trained large model is used to predict alloy properties in each subspace, and high-performance subspaces are selected based on the prediction results. The alloy composition was optimized using an optimization algorithm for each high-performance subspace. Each high-performance subspace after composition optimization was screened by adding restrictions on raw material elements and target alloy properties to ensure alloy solid solution; The selected alloy composition is verified by both theoretical calculation and experimental methods. If the verification is successful, it is taken as the final high-entropy alloy composition.

[0007] Furthermore, the general materials database resources include the elemental composition, crystal structure information, and corresponding physical and chemical properties of crystalline materials; The extracted core features include space group, crystal density, and element-weighted electronegativity, etc. In the process of extracting the crystal map, Pearson correlation coefficient is used for feature engineering to screen out sequence features that are strongly correlated with the target properties, and crystal map is constructed using lattice structure and interatomic topological relationships.

[0008] Furthermore, the hybrid graph neural network includes an MLP part and a CGCNN part. Based on the outputs of the MLP part and the CGCNN part, the output result is obtained after processing through a hybrid layer. During the training process of the hybrid graph neural network, the core features are input into the MLP part, and the crystal graph is input into the CGCNN part.

[0009] Furthermore, the large-scale training data does not necessarily need to include high-entropy alloys; it can be a general materials dataset. Here, a general materials dataset refers to data that can include semiconductors, pure metals, intermetallic compounds, metal oxides, and other compounds, without being limited to specific material types. Because these data are widely available, inexpensive, and cover a broad range of elements, extracting and transferring information from them can reduce data costs.

[0010] Furthermore, the probe test set data must be high-entropy alloys or other multi-element alloys with similar properties to high-entropy alloys (such as multi-element intermetallic compounds) to ensure the effectiveness of the testing and screening of each subspace and to ensure the accuracy of the model's prediction of the properties of high-entropy alloys.

[0011] Furthermore, the step of selecting high-performance subspaces based on prediction results specifically includes: The correlation coefficient between the alloy property prediction results of the pre-trained large model in the subspace and the corresponding true values ​​is calculated, and the subspace with the correlation coefficient greater than the preset correlation threshold is selected as the high-performance subspace.

[0012] Furthermore, during the alloy composition optimization process, if the prediction results of the pre-trained large model targeting multiple physicochemical properties all show high performance in the same subspace, then a non-dominated genetic algorithm is used to optimize the alloy composition of the subspace according to the physicochemical properties with high performance.

[0013] Furthermore, during the alloy composition optimization process, if in the same subspace, only the pre-trained large model with a single physical and chemical property as the objective has a high performance in prediction results, then the Bayesian optimization algorithm is used to optimize the alloy composition of the subspace according to the physical and chemical property with high performance.

[0014] Furthermore, the limitations on the raw material elements include: the radius difference between solid solution atoms is within 15%, as defined by the Hume-Rothery rule; The limitations on the target alloy properties include: Various atomic radius differences in the synthesis of the target high-entropy alloy δ and Mixed enthalpy change ΔH mix Within a specific range that ensures the solid solution of the element; Average number of valence electrons in synthetic alloys VEC The specific range it occupies matches the component optimization subspace.

[0015] Furthermore, the theoretical calculation verification methods include density functional theory and molecular dynamics theory. The density functional theory is used to calculate the elastic constants, elastic modulus, micro Vickers hardness, and lattice thermal conductivity of the alloy. The molecular dynamics theory is used to simulate the metal melting and crystallization process in order to predict the solid solution phase; The experimental verification method includes melting high-entropy alloys using vacuum arc melting, testing the thermal conductivity of the synthesized high-entropy alloys using laser thermal conductivity, and testing the hardness of the alloys using nanoindentation.

[0016] This invention also provides a high-entropy alloy obtained using a high-entropy alloy composition design method covering the entire periodic table as described above, wherein the composition of the high-entropy alloy is as follows: W 18.74 Re 22.31 Pt 31.55 Pd 15.30 Ru 12.10 Re 14.62 Hf 28.05 Ta 28.42 Zr 14.33 Nb 14.58 or W 28.63 Re 14.50 Ta 28.18 Zr 14.21 Nb 14.47 ; The W 18.74 Re 22.31 Pt 31.55 Pd 15.30 Ru 12.10 The high-entropy alloy achieves a single-target high hardness of 816±49 HV, the W 28.63 Re 14.50 Ta 28.18 Zr 14.21 Nb 14.47The hardness of the refractory high-entropy alloy is 801±9 HV, and the thermal conductivity is 18.002±0.060 W / (m·K).

[0017] The present invention also provides a high-entropy alloy composition design system covering the entire periodic table, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the method described above.

[0018] Compared with the prior art, the present invention has the following advantages: (1) Starting with a general materials database, this invention develops a target space partitioning prediction enhancement strategy. It learns about crystal materials based on the general materials database through a hybrid graph neural network. After constructing a probe dataset based on the performance data of multi-element alloys and dividing it into multiple subspaces, it selects high-performance subspaces that can be clearly identified by the model based on the test results of the pre-trained model. Then, it optimizes the alloy composition in a targeted manner and adds restrictions on the raw material elements and target alloy properties. Experimental verification is carried out. This process realizes the successful distributional out-generalization of the model from general materials to the field of high-entropy alloys, and extends the element space for high-entropy alloy composition optimization to the entire periodic table. The whole workflow greatly reduces the data cost of high-entropy alloy composition design and creates conditions for accelerating the search for new high-performance high-entropy alloy systems.

[0019] (2) The present invention develops a hybrid graph neural network model, which combines the traditional invariant graph neural network architecture with a multilayer perceptron to improve the graph neural network’s ability to predict higher-order physical properties. During the prediction process, the extracted core features such as crystal density and electronegativity are input into the multilayer perceptron. The crystal graph is input into the invariant graph neural network, which enhances the prediction accuracy of the mechanical and thermal transport properties of high-entropy alloys.

[0020] (3) This invention utilizes pre-trained models and reverse design strategies to develop three novel high-entropy alloy systems with high hardness and thermal conductivity, and conducts experimental verification. The experimental results of hardness and thermal conductivity of the three novel refractory high-entropy alloys show a high degree of consistency with the AI ​​algorithm and theoretical calculation expectations. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the principle of a high-entropy alloy composition design method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating a high-entropy alloy composition design method provided in an embodiment of the present invention. Figure 3 This is an elemental distribution, material type, and elemental type analysis diagram of a mechanical property database (mechanical data portion of the Materials Project) provided in an embodiment of the present invention. Figure 4 This is an elemental distribution, material type, and elemental type analysis diagram of a thermal transport performance database (room temperature lattice thermal conductivity data in AFLOW) provided in an embodiment of the present invention. Figure 5 This is a diagram of a hybrid graph neural network model architecture provided in an embodiment of the present invention; Figure 6 This is a diagram illustrating the performance improvement of a hybrid graph neural network compared to the original CGCNN network in an embodiment of the present invention. Figure 7 This is a performance evaluation diagram of a model for predicting the bulk modulus and lattice thermal conductivity of an alloy, provided in an embodiment of the present invention. Figure 8 This is a performance demonstration diagram of a high-entropy alloy partitioning prediction model provided in an embodiment of the present invention; Figure 9 The images provided in this embodiment of the invention show the phase composition and surface morphology of three alloys, namely XRD patterns and SEM and EDS images. Figure 10 This invention provides nanoindentation curves and micro-indentation photographs of three alloys in an embodiment of the invention. Figure 11 This is a scatter plot showing the relative positions of experimental data on hardness and total thermal conductivity of three alloys provided in this embodiment of the invention in various types of high-entropy alloys and multi-principal element alloys. Figure 12 This is a schematic diagram of the overall process of a high-entropy alloy composition design method covering the entire periodic table provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0025] It should be noted that, in the implementation of this invention, the multi-objective optimization of metals is not limited to their mechanical and thermal transport properties, but may include, but is not limited to, other physical and chemical properties such as electrical, thermal, optical, corrosion, and catalytic properties; the multi-objective optimization of metals is not limited to two properties, but may include three or more; the range of raw material elements for metals does not necessarily have to cover the entire periodic table, and the covered element range can be any subset of the entire periodic table, and may even include some non-metallic elements commonly used in alloy smelting; the general-purpose material database used is not limited to Materials Project and AFLOW, but can be any other general-purpose material database that meets the characteristics described in the invention; the optimization algorithm used is not limited to NSGA-II (Non-dominated Sorting Genetic Algorithm II) and TPE (Tree-structured Parzen). The estimator (based on a tree-structured Parsons estimator) can be Bayesian optimized, or other optimization algorithms; the prediction model is not limited to hybrid graph neural networks, but can also be other machine learning algorithms or neural networks, or even molecular dynamics methods with general machine learning interatomic potentials; the partitioning method is not limited to division according to the types of constituent elements and phase structure, but can be divided according to the actual needs of its own research system, but other partitioning methods that meet the characteristic of "removing poorly performing regions after partitioning" are acceptable.

[0026] Example 1 This embodiment provides a method focusing on two physical properties of high-entropy alloys: hardness and lattice thermal conductivity. The elemental range covers the entire periodic table. Starting with data on bulk modulus and lattice thermal conductivity at 300K from the Materials Project and AFLOW databases, the extracted information is generalized to the high-entropy alloy domain. This accurately predicts the magnitude sequence relationship of properties among different alloy types, and an optimization algorithm is used to search for alloy compositions with high bulk modulus and lattice thermal conductivity. Since alloy bulk modulus is monotonically positively correlated with hardness, high bulk modulus and high hardness are almost equivalent. After obtaining the optimized composition, simulation calculations were performed for verification, and experimental preparation and testing were conducted.

[0027] Figure 1 and Figure 2 These are a schematic diagram of the principle and a flowchart of the process of this work. Figure 1 This demonstrates the core steps of the proposed solution—the core logic of the target space partitioning prediction enhancement strategy, which involves testing the general underlying model in subspaces of different target materials, finding the best-performing subspace, and then designing the composition within it. Figure 2This is a workflow diagram. This solution mainly includes element screening, property prediction (mainly training and partition prediction of the underlying general model), component optimization, and experimental verification.

[0028] Specifically, such as Figure 12 As shown, this embodiment provides a method for designing high-entropy alloy compositions covering the entire periodic table, including: S1: Pre-collect general materials database resources, extract the core features and crystal diagrams of materials, use them for training hybrid graph neural networks, and obtain a pre-trained large model; S2: Collect and acquire performance data of multi-element alloys and construct a probe test set; S3: Divide the probe test set into multiple subspaces according to element type and phase structure; use a pre-trained large model to predict alloy properties in each subspace, and select high-performance subspaces based on the prediction results; S4: Optimize the alloy composition for each high-performance subspace using an optimization algorithm; S5: Screening is performed by adding restrictions on raw material elements and target alloy properties to each high-performance subspace after composition optimization to ensure alloy solid solution; S6: The alloy composition obtained from the screening is verified by theoretical calculation and experimental methods. If the verification is successful, it is taken as the final high-entropy alloy composition.

[0029] Step S1 specifically includes collecting general materials database resources, cleaning the data, extracting core features and crystal diagrams, and using a hybrid graph neural network as the underlying model architecture to train a pre-trained large model of material properties.

[0030] The general materials database resource includes the elemental composition, crystal structure information, and physical and chemical properties of crystalline materials, such as mechanical, electrical, thermal, and heat transport properties. This general materials database must cover all commonly used metallic elements in the periodic table and non-metallic elements commonly used in alloy smelting to ensure a broad space of available chemical elements for optimizing the composition of high-entropy alloys. The database should have a particularly large amount of data, generally more than 5,000 records, to ensure the accuracy of the pre-trained model. Figure 3 and Figure 4 The document showcases details such as the element coverage and material classification of the two open-source databases used in the work.

[0031] In other words, the general materials dataset does not necessarily need to include high-entropy alloys; it can contain data on semiconductors, pure metals, intermetallic compounds, metal oxides, and other compounds, without being limited to specific material types. Because these data are widely available, inexpensive, and cover the entire periodic table, extracting and transferring information from them can reduce data costs; noise introduced by impurity data is removed through a partitioning optimization step.

[0032] The core features extracted include space group, unit cell density, and element-weighted electronegativity. The importance of the features is calculated using the Pearson correlation coefficient, and features with high importance are selected.

[0033] In the process of extracting sequence features and crystal diagrams, Pearson correlation coefficient is used for feature engineering to screen out sequence features that are highly correlated with the target properties, and crystal diagrams are constructed using lattice structure and interatomic topological relationships.

[0034] like Figure 5 As shown, hybrid graph neural networks combine the traditional invariant graph neural network architecture with multilayer perceptrons to improve the prediction ability of graph neural networks for higher-order physical properties. The hybrid graph neural network algorithm framework is presented here. Figure 5 The model architecture mainly consists of a combination of an MLP (Multilayer Perceptron) part and a CGCNN (Crystal Graph Convolutional Neural Network) part. The output result is obtained by jointly processing the outputs of the MLP part and the CGCNN part through the Hadamard product in the hybrid layer. During the training process of the Hybrid Graph Neural Network (HGCNN), the core features are input into the MLP part, and the crystal graph is input into the CGCNN part.

[0035] In this embodiment, step S1 collects data resources from the general materials database Materials Project and AFLOW. The database contains the elemental composition, crystal structure information, and physical and chemical properties of crystalline materials, such as mechanical, electrical, thermal, and heat transport properties. Clean the data by removing mechanically unstable and kinetically unstable thermal transport data that do not meet the Berne criterion, and also remove outliers with negative shear modulus and bulk modulus exceeding 1000 GPa. Core features such as crystal density and electronegativity are extracted using scripts. Pearson correlation coefficients are used for feature engineering to select features strongly correlated with the target properties, and a crystal map is constructed. A hybrid graph neural network (MLP) is used as the underlying model architecture to train a large-scale pre-trained model for material properties. The core features are input into the MLP part, and the crystal map is input into the CGCNN part. The bulk modulus and lattice thermal conductivity models are each trained for 1000 epochs. The training results are... Figure 7 As shown in the figure, since the overall performance predicted by shear modulus is not as good as that of bulk modulus, this embodiment chooses bulk modulus as the leading physical quantity for predicting hardness. In step S2, the performance data of multi-element alloys includes the performance data of high-entropy alloys and highly symmetric intermetallic compounds with different phases and element types.

[0036] In this embodiment, performance data of multi-element alloys were selected from peer-reviewed papers to construct a probe test set. These data include high-entropy alloys with different phases and element types, as well as some highly symmetric intermetallic compounds. All of these data were strictly modeled and restored according to the structure, properties, and modeling methods of the original materials.

[0037] All selected papers are peer-reviewed SCI papers, and the data is highly accurate.

[0038] The probe dataset must be data on high-entropy alloys or other multi-element alloys similar to high-entropy alloys (including but not limited to intermetallic compounds) to ensure the effectiveness of the verification and screening of each subspace and to ensure the accuracy of the model's prediction of the properties of high-entropy alloys. The types of material properties contained in the data are consistent with those in the general material dataset in step S1, such as the bulk modulus and lattice thermal conductivity of the materials. The amount of data can be 1 to 2 orders of magnitude smaller than that in the general material dataset in step S1, but it should be ensured that each subspace in S3 has at least more than 7 test data points.

[0039] In step S3, selecting the high-performance subspace based on the prediction results specifically includes: The correlation coefficient between the alloy property prediction results of the pre-trained large model in the subspace and the corresponding true values ​​is calculated, and the subspace with the correlation coefficient greater than the preset correlation threshold is selected as the high-performance subspace.

[0040] In this embodiment, the probe test set is divided into nine categories according to element type and phase structure: ternary FCC (Face-centered Cubic), ternary BCC (Body-centered Cubic), ternary HCP (Hexagonal Close-Packed), up to pentagonal FCC, BCC, and HCP. The pre-trained large model is tested in the subspace, and the correlation coefficient between the predicted and actual values ​​is selected. r 2 The high-performance subspace with a value greater than 0.75 is identified in this step, known as the target space partitioning prediction enhancement strategy. This is a crucial step in our approach, used to remove subspaces severely affected by data points from impurity materials. The results of this step are presented in... Figure 8 middle.

[0041] In this step, the basis for subspace partitioning does not conflict with the elemental composition of the alloy, thus not narrowing the range of material elements that the model can predict.

[0042] In step S4, within the high-performance subspace selected in step S3, the alloy composition is optimized using an optimization algorithm. If the prediction model for multiple physicochemical properties as targets has high performance in the same subspace (especially for bulk modulus and lattice thermal conductivity), then the non-dominated genetic algorithm NSGA-II with bulk modulus and lattice thermal conductivity as optimization targets is used to perform multi-objective optimization of the alloy composition in that subspace. If only a single property has high performance in a certain subspace, then the Bayesian optimization algorithm TPE is used to perform single-objective optimization in that subspace. During the optimization process, commonly used alloying elements in the periodic table were divided into several element subgroups with a radius difference of less than 15% between them, in order to reduce the amount of calculation required for composition optimization and to exclude element combinations that are obviously insoluble.

[0043] All elements in the element subgroup n ( n = 3, 4, 5) element combination results are used as the element space for component optimization and are input as discrete variables into the NSGA-II algorithm or TPE Bayesian optimization algorithm; the proportion of each element is used as a continuous variable and input into the optimization algorithm to perform single-objective or multi-objective optimization of alloy composition.

[0044] Specifically, the objective function of the NSGA-II algorithm is: In the formula, 𝑓 𝐵 It is a pre-trained model for predicting bulk modulus, 𝑓 𝐿𝑇𝐶 This is a pre-trained model for predicting lattice thermal conductivity, where 𝐸𝑖 represents the element type, and 𝑝 𝑖 is the proportion of each element, and is the material structure (FCC, BCC, or HCP). Bulk modulus and lattice thermal conductivity are the optimization targets; bulk modulus is used as the prediction target because bulk modulus and hardness have an approximately monotonically positive correlation in materials.

[0045] The objective of this embodiment is to search for alloys with high hardness and high thermal conductivity; therefore, bulk modulus and lattice thermal conductivity are chosen as optimization targets. The NSGA-II algorithm here is optimized to adapt to a search oriented towards elemental composition space. If the target alloy should consist of… n Composed of various elements, the selectable element pool has a total of m Such elements, then this n All combinations of elements are treated as a set. ,in For example, for pentagonal alloys, if the optional element pool has... m There are 100 elements, so there are a total of 100 elements. There are several combinations, and these combinations are mapped to a set of... A sequence of elements: 1, 2, 3, …, Each integer in this sequence represents a quintuple (e.g., 1 represents...). ; It is an element combination, such as W-Re-Pt-Pd-Ru). This sequence is made continuous and used as the DNA for selection, crossover, and mutation in the NSGA-II algorithm. Substituting it into the algorithm can optimize and obtain the optimal alloy composition.

[0046] In step S5, restrictions are added on the raw material elements and the properties of the target alloy to ensure that it can be solidified. The restriction on the raw material elements is based on the Hume-Rothery rule, which stipulates that the radius difference between solid solution atoms is within 15%. The limitation on the properties of the target alloy is the difference in various atomic radii in the synthesis of the target high-entropy alloy. δ Mixed enthalpy change ΔH mix It needs to be within a specific range to ensure elemental solid solution; the average number of valence electrons in the synthesized alloy. VEC The specific range it occupies should match the component optimization subspace.

[0047] In this embodiment, the atomic radius difference of the target high-entropy alloy δ The value is less than 0.045. ΔH mix When the value ranges from -22 to 5 kJ / mol, the raw material metals in the alloy are in solid solution; when the average valence electron number of the high-entropy alloy is... VEC When the average valence electron count is greater than 0.8, the alloy forms a stable FCC phase; when the average valence electron count is greater than 0.8, the alloy forms a stable FCC phase. VEC When the value is less than 6.8, the alloy forms a stable BCC phase.

[0048] In step S6, during the theoretical calculation and experimental verification of the target alloy properties, the theoretical calculation and verification methods include density functional theory and molecular dynamics theory. Density functional theory is used to calculate the mechanical properties of the alloy, such as elastic constants, elastic modulus, and micro Vickers hardness, as well as thermal transport properties such as lattice thermal conductivity. Molecular dynamics is used to simulate the metal melting and crystallization process to predict the solid solution phase. The experiment uses vacuum arc melting to melt the high-entropy alloy, and laser thermal conductivity is used to test the thermal conductivity of the synthesized high-entropy alloy. The hardness of the alloy is tested using nanoindentation.

[0049] In cases where high hardness and high lattice thermal conductivity are the optimization targets, the screened BCC high-entropy alloy systems include two categories: W and / or Re plus refractory metals (Mo, Hf, Zr, Ta, Nb, V) and W and / or Re plus refractory metals and Ti; the screened FCC high-entropy alloy systems include one category: W and Re plus Pt-based noble metals (Pt, Pd, Ru, Os, Ir, Rh).

[0050] As shown in Table 1, the experiment achieved W18.74 Re 22.31 Pt 31.55 Pd 15.30 Ru 12.10 ( w t%, code Pt-1), Re 14.62 Hf 28.05 Ta 28.42 Zr 14.33 Nb 14.58 ( w t%, code RE-3), W 28.63 Re 14.50 Ta 28.18 Zr 14.21 Nb 14.47 ( w The synthesis of three novel high-entropy alloys (t%, code RE-4) showed that their lattice constants and main crystalline phases formed were highly consistent with theoretical calculations.

[0051] Table 1. Theoretical calculations and experimental values ​​of the lattice constants of the main phases in three novel refractory high-entropy alloys. The thermal conductivity of the synthesized high-entropy alloy was tested using laser, the hardness of the alloy was tested using nanoindentation, and the tensile strength and elongation of the high-entropy alloy were tested using a small punch test. Figure 10 The curve data and micrographs of nanoindentation of the three components are shown. The hardness and thermal conductivity data of the three components are summarized in Table 2. The highest-performing refractory high-entropy alloy with W and Re is identified as RE-4, with a hardness of 801±9 HV and a thermal conductivity of 18.002±0.060 W / (m·K), which meets the standard of high-hardness and high-thermal-conductivity high-entropy alloys. The trends of the hardness and thermal conductivity experimental results of the three new refractory high-entropy alloys are highly consistent with the expectations of AI algorithm and theoretical calculation.

[0052] Table 2. Theoretical calculations and experimental values ​​of hardness, modulus, and thermal conductivity of three novel refractory high-entropy alloys. As can be seen from Table 2, the experimental results and theoretical calculation results are basically consistent in trend; Figure 11 This section compares the hardness and thermal conductivity of the alloy synthesized in this embodiment with those of other common high-entropy alloys to demonstrate the advantages of the alloy in this embodiment in terms of both thermal conductivity and hardness.

[0053] This embodiment also provides a high-entropy alloy composition design system covering the entire periodic table, including a memory and a processor. The memory stores a computer program, and the processor calls the computer program to execute the steps of the high-entropy alloy composition design method covering the entire periodic table as described above.

[0054] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for designing high-entropy alloy compositions covering the entire periodic table, characterized in that, include: Pre-collect general materials database resources, extract the core features and crystal diagrams of the materials, and use them to train the hybrid graph neural network to obtain a pre-trained large model; Collect alloy performance data from published papers and construct a probe test set; The probe test set is divided into multiple subspaces according to element type and phase structure; the pre-trained large model is used to predict alloy properties in each subspace, and high-performance subspaces are selected based on the prediction results. The alloy composition was optimized using an optimization algorithm for each high-performance subspace. Each high-performance subspace after composition optimization was screened by adding restrictions on raw material elements and target alloy properties to ensure alloy solid solution; The selected alloy composition is verified by both theoretical calculation and experimental methods. If the verification is successful, it is taken as the final high-entropy alloy composition.

2. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, The general materials database resources include the elemental composition, crystal structure information, and corresponding physical and chemical properties of crystalline materials; The extracted core features include space group, crystal density, and element-weighted electronegativity; In the process of extracting the crystal map, Pearson correlation coefficient is used for feature engineering to screen out sequence features that are strongly correlated with the target properties, and crystal map is constructed using lattice structure and interatomic topological relationships.

3. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, The hybrid graph neural network includes an MLP part and a CGCNN part. The output result is obtained after processing the output of the MLP part and the CGCNN part through a hybrid layer. During the training process of the hybrid graph neural network, the core sequence features are input into the MLP part, and the crystal graph is input into the CGCNN part.

4. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, The general materials dataset includes pure metals, semiconductors, metal halides, intermetallic compounds, and metal oxides; The probe test set data consists of high-entropy alloys or other multi-element alloys with properties similar to high-entropy alloys.

5. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, The process of selecting high-performance subspaces based on prediction results specifically includes: The correlation coefficient between the alloy property prediction results of the pre-trained large model in the subspace and the corresponding true values ​​is calculated, and the subspace with the correlation coefficient greater than the preset correlation threshold is selected as the high-performance subspace.

6. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, In the process of alloy composition optimization, if the prediction results of the pre-trained large model with multiple physical and chemical properties as targets are all high in the same subspace, then the non-dominated genetic algorithm is used to optimize the alloy composition of the subspace according to the physical and chemical properties with high performance. If, within the same subspace, only the pre-trained large model with a single physical property as the objective has high prediction performance, then the Bayesian optimization algorithm is used to optimize the alloy composition of the subspace according to the physical property with high performance.

7. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, The restrictions on the raw material elements include: the radius difference between solid solution atoms is within 15%, as defined by the Hume-Rothery rule; The limitations on the target alloy properties include: Various atomic radius differences in the synthesis of the target high-entropy alloy δ and Mixed enthalpy change ΔH mix Within a specific range that ensures the solid solution of the element; Average number of valence electrons in synthetic alloys VEC The specific range it occupies matches the component optimization subspace.

8. The method for designing high-entropy alloy compositions covering the entire periodic table according to claim 1, characterized in that, The theoretical calculation verification methods include density functional theory and molecular dynamics theory. The density functional theory is used to calculate the elastic constants, elastic modulus, micro Vickers hardness, and lattice thermal conductivity of the alloy. The molecular dynamics theory is used to simulate the metal melting and crystallization process in order to predict the solid solution phase; The experimental verification method includes melting high-entropy alloys using vacuum arc melting, testing the thermal conductivity of the synthesized high-entropy alloys using laser thermal conductivity, and testing the hardness of the alloys using nanoindentation.

9. A high-entropy alloy obtained using a high-entropy alloy composition design method covering the entire periodic table as described in any one of claims 1-8, characterized in that, The composition of the high-entropy alloy is as follows: W 18.74 Re 22.31 Pt 31.55 Pd 15.30 Ru 12.10 、Re 14.62 Hf 28.05 Ta 28.42 Zr 14.33 Nb 14.58 or W 28.63 Re 14.50 Ta 28.18 Zr 14.21 Nb 14.47 ; The W 18.74 Re 22.31 Pt 31.55 Pd 15.30 Ru 12.10 The high-entropy alloy achieves a single-target high hardness of 816±49 HV, the W 28.63 Re 14.50 Ta 28.18 Zr 14.21 Nb 14.47 The hardness of the refractory high-entropy alloy is 801±9 HV, and the thermal conductivity is 18.002±0.060 W / (m·K).

10. A high-entropy alloy composition design system covering the entire periodic table, characterized in that, It includes a memory and a processor, the memory storing a computer program, the processor invoking the computer program to perform the steps of the method as described in any one of claims 1 to 8.

Images