A method for targeted design of high-calorific-value energetic alloy materials using coupled element information

Abstract

This application relates to the field of computational materials science and technology, and particularly to a method for targeted design of high-calorific-value energetic alloy materials by coupling elemental information. The method includes the following steps: First, establishing a dedicated energetic alloy material dataset; then, constructing a deep reinforcement learning model for high-calorific-value energetic alloy materials, training the model based on the dedicated energetic alloy material dataset, directionally generating potential energetic alloy material components, and assigning crystal templates to obtain energetic alloy materials with crystal structures; subsequently, training a calorific value and formation energy prediction model based on the dedicated energetic alloy material dataset, predicting newly generated energetic alloy materials, and selecting high-calorific-value energetic alloy materials; finally, using VASP software for computational verification of candidate materials. This application, based on a smart paradigm for materials research and development, combines a deep reinforcement learning model, a neural network prediction model, and quantum mechanical computational verification to achieve targeted design of high-calorific-value energetic alloy materials.

Description

Technical Field

[0001] This application relates to the field of computational materials science and technology, and in particular to a method for directional design of high-calorific-value energetic alloy materials by coupling elemental information. Background Technology

[0002] Energetic alloy materials refer to a class of alloy materials with energetic properties. They belong to metallic fuels with energetic formulations and energy density, playing an important role in the field of energetic materials. Metallic fuels are a special type of fuel that differs from conventional fossil fuels. Due to their advantages such as high energy density and high calorific value, abundant raw materials, zero carbon emissions, long-term storage, and recyclability, they have been widely used in explosives (solid propellants, explosives, and other energetic formulations) and play an indispensable role.

[0003] In existing research and engineering applications, aluminum, magnesium, and boron are the most typical and well-studied metallic fuel elements. Among them, aluminum powder, due to its high energy release and cost advantages, has long served as the performance benchmark for solid propellant systems. However, its combustion process is generally plagued by problems such as particle agglomeration and oxide coating, limiting its actual combustion efficiency. In contrast, while elemental metallic fuels such as magnesium and boron have advantages in theoretical energy density or ignition performance, they are constrained by inherent defects such as oxide film inhibition, ignition delay, and insufficient combustion efficiency, making it difficult for them to surpass aluminum powder in combustion performance in engineering applications. Alloys, on the other hand, can have their properties adjusted according to different requirements, and the overall performance of alloy fuels is generally superior to that of elemental fuels, making alloy fuels have a more promising application prospect. Alloying to form energetic alloy materials can improve combustion behavior to some extent, but the design and optimization of existing energetic alloy materials still mainly rely on experimentally driven trial-and-error methods, which are not only costly and time-consuming to develop, but also pose significant safety risks under high-energy, high-combustion, and potentially toxic conditions. More importantly, this type of method relies heavily on the accumulated experience of existing fuel systems, making it difficult to conduct systematic exploration in the vast space of multi-element alloy compositions, which fundamentally limits the efficiency of discovering new high-performance metallic fuels.

[0004] With the rapid growth of materials data and the maturation of artificial intelligence technology, materials research is gradually shifting from an experience-driven to a data-driven paradigm. In recent years, deep learning models trained on first-principles data have been able to efficiently generate and screen large numbers of stable inorganic crystal structures, demonstrating significant advantages in crystal structure prediction and discovery. Simultaneously, autoregressive and diffusion models trained on large-scale crystal structure data have also achieved the automatic generation of reasonable crystal structures. Furthermore, researchers have utilized deep reinforcement learning for reverse design of inorganic materials. The literature incorporates machine learning prediction models based on component characteristics to predict material properties, while also drawing on template-matching-based crystal structure prediction (CSP) to achieve reverse design of inorganic materials. Some researchers have also applied graph convolution to the crystal field, forming a crystal graph convolutional neural network (CGCNN), which can achieve high-throughput prediction of crystal structures and is very suitable for preliminary pre-screening of large numbers of generated crystals. However, the aforementioned data-driven crystal generation technologies are mainly concentrated in structural materials and functional materials. Their application in the crystal structure design and performance-oriented generation of metallic fuels, especially high-calorific-value energetic alloys, has not yet been systematically implemented, leaving a significant technological gap. Summary of the Invention

[0005] This application provides a method for targeted design of high-calorific-value energetic alloy materials by coupling element information. Based on the intelligent paradigm of materials research and development, the method couples element information related to high-calorific-value energetic alloy materials, and combines deep reinforcement learning models, neural network prediction models, and quantum mechanical calculation verification to achieve targeted design of high-calorific-value energetic alloy materials.

[0006] To address the aforementioned technical problems, this application provides a method for targeted design of high-calorific-value energetic alloy materials by coupling elemental information. This method includes the following steps: First, establishing a dedicated dataset of energetic alloy materials containing the formation energy and calorific value properties; then, constructing a deep reinforcement learning model of the high-calorific-value energetic alloy material by coupling elemental information, and training the model based on the dedicated energetic alloy material dataset to generate potential energetic alloy material components and assigning them with synthetic potential crystal templates to obtain energetic alloy materials with crystal structures; next, training two crystal graph convolutional neural network prediction models, one for calorific value and the other for formation energy, based on the dedicated energetic alloy material dataset; subsequently, predicting the performance of the energetic alloy materials with crystal structures to virtually select candidate high-calorific-value energetic alloy materials; finally, performing quantum mechanical calculations on the selected candidate high-calorific-value energetic alloy materials to verify their calorific value and stability.

[0007] In some exemplary embodiments, a dedicated dataset of energetic alloy materials containing the formation energy and calorific value of energetic alloy materials is established, including: First, extracting the material composition, crystallographic files, and formation energy information of energetic alloy materials from public databases such as Material Project, and performing deduplication processing, using the same material composition and crystal structure as the data deduplication standard; then, based on Hess's Law and combined with the theoretical combustion reaction of energetic alloy materials, the calorific value of combustion is calculated by the difference between the formation energy of the energetic alloy material and the combustion products, and an automated processing script is written to perform high-throughput calculation of its calorific value to obtain the calorific value information; finally, storing the extracted material composition, crystallographic files, and formation energy information, as well as the calorific value information calculated by Hess's Law, together to construct a dedicated dataset of energetic alloy materials.

[0008] In some exemplary embodiments, constructing a deep reinforcement learning model for high-calorific-value energetic alloy materials that couples element information includes: firstly, constructing a composition generation model and an accuracy R model that couples element information, respectively. 2 The model is a "composition-calorific value" prediction model with a value of 0.986 and a calorific value reward function constrained by physical experience. Then, the composition generation model and the "composition-calorific value" prediction model are placed in the same deep reinforcement learning framework, and the calorific value reward function based on physical experience is used for reward value feedback to form a deep reinforcement learning model for high calorific value energetic alloy materials that couples element information.

[0009] In some exemplary embodiments, constructing a component generation model that couples element information includes: First, based on a dedicated energetic alloy material dataset, summarizing physical experience from data on high-calorific-value regions of energetic alloy materials to derive element information affecting the calorific value of energetic alloy materials; then, based on the element information affecting the calorific value of energetic alloy materials, performing data statistics and probability calculations on high-energy elements in high-calorific-value energetic alloy materials, and applying the calculated probabilities to the first step of element constraint in the component generation model, transmitting physical experience to the model in digital form, assisting the model in recognizing the high-calorific-value characteristics of energetic alloy materials, and accelerating model iteration efficiency.

[0010] In some exemplary embodiments, the accuracy R is constructed. 2 The "composition-calorific value" prediction model with an accuracy of 0.986 includes: based on a dedicated energetic alloy material dataset, using the elements and stoichiometry in the chemical formula as inputs to the "composition-calorific value" prediction model, a "composition-calorific value" prediction model with a prediction accuracy of 0.986 is constructed, realizing efficient prediction of the composition of newly generated energetic alloy materials without depending on the specific crystal structure, and providing calorific value information of the newly generated high-energy alloy composition for the calorific value reward function.

[0011] In some exemplary embodiments, the calorific value reward function constrained by physical experience is constructed by adding auxiliary terms that affect the chemical rationality, alloy synthesability, and combustion performance of energetic alloy materials, in addition to considering the calorific value characteristics predicted by the "composition-calorific value" prediction model, as shown in the following formula:

[0012] in, It is the final state reward value. It is the main reward item based on the predicted calorific value calculation, that is, the normalized result of the calorific value predicted by the "component-calorific value" prediction model. It is an auxiliary modulation factor composed of multiple a priori physical characteristics, namely, the product of normalized coefficients of the material's physicochemical properties, synthesis properties, and combustion efficiency. yes The weighting factor; the main reward item is obtained by normalizing the calorific value predicted by the "component-calorific value" prediction model. The formula is shown below:

[0013] in, It is the calorific value of the energetic alloy material predicted by the "composition-calorific value" prediction model. The normalization function is used to map the predicted calorific value to a predetermined reward interval [0,1]; while considering the influence of material physicochemical properties, synthesis properties, and combustion efficiency, auxiliary terms are obtained. The resulting compounds are weighted positively for those with a balance of charge neutrality and electronegativity and an appropriate number of element types; those with a positive enthalpy of formation as estimated by the Miedema semi-empirical model are weighted negatively; and those with a positive combustion efficiency as determined by the Glassman combustion rule are weighted positively. The formula is shown below:

[0014] in, It is a physicochemical rationality modulation term. It is a modulatory term for alloy synthesability. It is a combustion efficiency modulation item.

[0015] In some exemplary embodiments, based on a dedicated dataset of energetic alloy materials, two crystal graph convolutional neural network prediction models are trained, one for calorific value and the other for formation energy, including: The dataset of the dedicated energetic alloy material was divided into a training set, a validation set, and a test set in a 6:2:2 ratio. Two independent crystal graph convolutional neural network prediction models were trained, using formation energy and calorific value as the neural network prediction targets, respectively. The prediction accuracy R for the material's calorific value and formation energy was [value missing]. 2 The values ​​are 0.996 and 0.98, respectively.

[0016] In some exemplary embodiments, the performance of the generated energetic alloy material is predicted, and candidate high-calorific-value energetic alloy materials are virtually screened. This includes: based on two crystal graph convolutional neural network prediction models trained, the newly generated energetic alloy material is used as input to predict the formation energy and calorific value of the newly generated energetic alloy material, and the calorific value is set at a dual screening threshold of 31.1 kJ / g higher than 1 kJ / g and the formation energy is set at a value less than 0 eV / atom, respectively, to obtain the virtually screened stable and high-calorific-value energetic alloy material.

[0017] In some exemplary embodiments, quantum mechanical calculations are performed on the selected high-calorific-value energetic alloy materials to verify their calorific value and stability, including: based on the calorific value and formation energy of the selected high-calorific-value energetic alloy materials, candidate materials with negative formation energy are selected and ranked by calorific value, and the 10 candidate high-calorific-value energetic alloy materials with the highest calorific value are selected for quantum mechanical calculation verification; Based on VASP software, energy minimization calculations were performed on the atomic crystal model structures of 10 selected high-calorific-value energetic alloy materials. The cell energy was extracted based on the final energy of the unit cell, and the formation energy was calculated. Based on Hess's Law, the combustion enthalpy was evaluated based on the difference between the theoretical combustion products and the formation energy of the high-calorific-value energetic alloy material. By normalizing by unit mass, the combustion calorific value of the corresponding high-calorific-value energetic alloy material was calculated, and it was verified whether the calorific value reached the preset threshold of 31.1 kJ / g. If the calorific value met the standard, stability verification was performed.

[0018] In some exemplary embodiments, if the calorific value meets the standard, stability verification is performed. The verification process includes: for candidate high calorific value energetic alloy materials that meet the standard, based on their crystal structure after energy minimization, determining the range of changes in crystal structure parameters, analyzing the relationship between formation energy and unit cell volume, calculating phonon spectra, and performing room temperature ab initio molecular dynamics simulations, to verify the crystal structure stability, energy stability, kinetic stability, and room temperature thermodynamic stability of the candidate high calorific value energetic alloy materials, respectively.

[0019] The technical solution provided in this application has at least the following advantages: This application provides a method for targeted design of high-calorific-value energetic alloy materials by coupling elemental information. The method includes the following steps: First, establishing a dedicated dataset of energetic alloy materials containing the formation energy and calorific value properties; then, constructing a deep reinforcement learning model of the high-calorific-value energetic alloy material by coupling elemental information, and training the model based on the dedicated energetic alloy material dataset to generate potential energetic alloy material components and assigning them with crystal templates of synthetic potential, thereby obtaining energetic alloy materials with crystal structures; next, training two crystal graph convolutional neural network prediction models, one for calorific value and the other for formation energy, based on the dedicated energetic alloy material dataset; subsequently, predicting the performance of the energetic alloy materials with crystal structures to virtually select candidate high-calorific-value energetic alloy materials; finally, performing quantum mechanical calculations on the selected candidate high-calorific-value energetic alloy materials to verify their calorific value and stability. This application achieves directional design guided by calorific value by coupling elemental information related to the calorific value of energetic alloy materials and based on a deep reinforcement learning model of energetic alloy materials. It combines a crystal graph convolutional neural network model to quickly predict the calorific value and formation energy properties of the material, and further uses first-principles calculations of quantum mechanics for multi-dimensional high-precision verification. This constitutes an integrated "directional generation-prediction-verification" workflow for the directional design of high-calorific-value energetic alloy materials.

[0020] The design route for high-calorific-value energetic alloy materials in this application consists of three steps: generation of high-calorific-value targets for energetic alloy materials, high-throughput virtual screening of energetic alloy materials, and computational verification of high-calorific-value energetic alloy materials. Based on a data-driven intelligent design method, this application effectively avoids the safety risks inherent in traditional experimental processes and couples calorific-value influence element information to achieve targeted design of novel high-calorific-value energetic alloy materials at low cost and high efficiency. On one hand, this application uses element frequency statistics from a self-built dedicated energetic alloy material dataset, combined with the calorific value of the corresponding element itself for dual verification. Physical experience-based element information is incorporated into the deep reinforcement learning model to extract key information from existing data, improving the iteration efficiency of generating novel energetic alloy materials. On the other hand, a calorific-value reward function constrained by physical experience is designed in the deep reinforcement learning targeted generation model for high-calorific-value energetic alloy materials. The calorific-value reward function embeds auxiliary functions for the physicochemical rationality, alloy synthesizability, and combustion performance of the energetic alloy material, reflecting the influence of synthesis potential and combustion efficiency during the generation design, avoiding blindly pursuing a single calorific-value performance. Furthermore, in the process of high-throughput performance prediction of newly generated energetic alloy materials, this application trained two high-precision crystal graph convolutional neural network models for calorific value prediction and formation energy prediction, respectively, to achieve accurate performance prediction of energetic alloy materials with crystal structures, rather than relying on empirical formulas. In the final stage, based on first-principles calculations of density functional theory (DFT), quantum mechanical calculations were performed on the directionally generated and screened high-calorific-value energetic alloy materials using VASP software, thereby verifying their calorific value characteristics and multidimensional stability. Attached Figure Description

[0021] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments, and unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0022] Figure 1 This is a flowchart illustrating a method for directional design of high-calorific-value energetic alloy materials using coupled element information, as provided in an embodiment of this application.

[0023] Figures 2A to 2D A flowchart illustrating the specific architecture of a method for directional design of high-calorific-value energetic alloy materials based on coupling element information, provided in an embodiment of this application.

[0024] Figure 3 This is a schematic diagram of elemental information that is highly correlated with calorific value in an energetic alloy material dataset provided in an embodiment of this application.

[0025] Figure 4This is a diagram illustrating the accuracy of the "component-calorific value" prediction model on the test set in a deep reinforcement learning-oriented design model provided in an embodiment of this application.

[0026] Figure 5 This is a diagram illustrating the accuracy of the test set of the high-throughput virtual screening model for predicting calorific value in an embodiment of this application.

[0027] Figure 6 This is a diagram illustrating the accuracy of the formation energy prediction model test set in a high-throughput virtual screening process provided in an embodiment of this application.

[0028] Figure 7 The figure shows four stability verification results using VASP provided in one embodiment of this application. Detailed Implementation

[0029] As can be seen from the background technology, existing data-driven crystal generation technology is mainly concentrated in the fields of structural materials and functional materials. It has not yet been systematically applied to the crystal structure design and performance-oriented generation of metal fuels, especially high-energy metal alloy fuels, and there is still a significant technological gap.

[0030] To this end, information on elements with high calorific value impact was collected, and a directional design workflow for high calorific value energetic alloy materials was constructed, providing an achievable technical path for the rational design and rapid evaluation of high calorific value energetic alloy materials.

[0031] Due to the inherent risks of high cost, low efficiency, and potential toxicity associated with high-energy, high-explosive, and high-toxicity experimental designs for energetic alloys, related research relies heavily on experimental experience, making it difficult to conduct systematic explorations across a broad compositional space. Furthermore, current experimental assessments of the theoretical energy density of energetic alloys typically employ a weighted estimation method based on the energy densities of constituent elements, failing to adequately consider energy loss effects during alloy formation. Simultaneously, existing crystal structure generation methods lack the capability for targeted design in the field of energetic alloys, making it difficult to meet the structural construction requirements of high-calorific-value energetic alloys.

[0032] To address the aforementioned technical problems, this application provides a method for the targeted design of high-calorific-value energetic alloy materials by coupling elemental information. This method includes the following steps: First, establishing a dedicated dataset of energetic alloy materials containing the formation energy and calorific value properties; then, constructing a deep reinforcement learning model for high-calorific-value energetic alloy materials by coupling elemental information, and training the model based on the dedicated dataset to generate potential energetic alloy material components and assigning them with synthetic potential crystal templates to obtain energetic alloy materials with crystal structures; next, training two crystal graph convolutional neural network prediction models, one for calorific value and the other for formation energy, based on the dedicated dataset; subsequently, predicting the performance of energetic alloy materials with crystal structures to virtually select candidate high-calorific-value energetic alloy materials; finally, performing quantum mechanical calculations on the selected candidate high-calorific-value energetic alloy materials to verify their calorific value and stability. Based on the intelligent paradigm of materials research and development, this method couples elemental information related to high-calorific-value energetic alloy materials, and combines deep reinforcement learning models, neural network prediction models, and quantum mechanical calculation verification to achieve targeted design of high-calorific-value energetic alloy materials.

[0033] The embodiments of this application will now be described in detail with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details have been provided in the embodiments of this application to facilitate a better understanding of the application. However, the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments.

[0034] See Figure 1 This application provides a method for targeted design of high-calorific-value energetic alloy materials by coupling element information, comprising the following steps: Step S1: Establish a dedicated dataset of energetic alloy materials, including the formation energy and calorific value properties of energetic alloy materials; Step S2: Construct a deep reinforcement learning model for high-calorific-value energetic alloy materials that couples element information, and train the model based on a dedicated energetic alloy material dataset to generate potential energetic alloy material components in a targeted manner, and assign crystal templates with synthesis potential to obtain energetic alloy materials with crystal structures.

[0035] Step S3: Based on a dedicated dataset of energetic alloy materials, two crystal diagram convolutional neural network prediction models are trained, one for calorific value and the other for formation energy. Subsequently, the performance of energetic alloy materials with crystal structures is predicted, and candidate high-calorific-value energetic alloy materials are virtually screened.

[0036] Step S4: Perform quantum mechanical calculations on the selected high-calorific-value energetic alloy materials to verify their calorific value and stability.

[0037] The method for targeted design of high-calorific-value energetic alloy materials provided in this application consists of four steps: constructing a dedicated energetic alloy material dataset, generating high-calorific-value targets for energetic alloy materials, high-throughput virtual screening of energetic alloy materials, and computational verification of high-calorific-value energetic alloy materials. Figures 2A to 2D A flowchart illustrating the specific architecture of the method for directional design of high-calorific-value energetic alloy materials based on the coupling element information provided in this application.

[0038] like Figure 2A As shown, a dedicated dataset for energetic alloy materials, including formation energy and calorific value properties, was established. This included extracting material composition, crystallographic files, and formation energy information from public databases such as the Material Project, and performing deduplication. The data was stored in a table, using identical material composition and crystal structure as the deduplication criterion. Based on Hess's Law and the theoretical combustion reaction of energetic alloy materials, the calorific value was calculated from the difference between the formation energy of the energetic alloy material and the combustion products. An automated processing script was written to perform high-throughput calculations of the calorific value. The extracted material composition, crystallographic files, formation energy information, and calorific value information calculated using Hess's Law were stored together to construct the dedicated dataset for energetic alloy materials.

[0039] In some embodiments, such as Figure 2B As shown, a deep reinforcement learning model for high-calorific-value energetic alloy materials coupled with elemental information is constructed, including: firstly, constructing a composition generation model and an accuracy R-value model that are coupled with elemental information. 2 The model is a "composition-calorific value" prediction model with a value of 0.986 and a calorific value reward function constrained by physical experience. Then, the composition generation model and the "composition-calorific value" prediction model are placed in the same deep reinforcement learning framework, and the calorific value reward function based on physical experience is used for reward value feedback to form a deep reinforcement learning model for high calorific value energetic alloy materials that couples element information.

[0040] In some embodiments, constructing a component generation model coupled with elemental information includes: based on a dedicated energetic alloy material dataset, summarizing physical experience from high-calorific-value region data of energetic alloy materials to deduce elemental information affecting the calorific value of energetic alloy materials; performing data statistics and probability calculations on high-energy elements in high-calorific-value energetic alloy materials, and applying this probability to the first step of the reinforcement learning-oriented generation model for energetic alloy materials—elemental constraint—to digitally transmit physical experience to the model input, assisting the model in recognizing the high-calorific-value characteristics of energetic alloy materials and accelerating model iteration efficiency.

[0041] The composition generation model generates high-energy alloy compositions with a reward-maximizing objective, and then feeds the newly generated compositions into a built-in composition-calorific value prediction model. In some embodiments, an accuracy R is constructed. 2 The "composition-calorific value" prediction model with an accuracy of 0.986 includes: based on a dedicated energetic alloy material dataset, using the elements and stoichiometry in the chemical formula as inputs to the "composition-calorific value" prediction model, a "composition-calorific value" prediction model with a prediction accuracy of 0.986 is constructed, realizing efficient prediction of the composition of newly generated energetic alloy materials without depending on the specific crystal structure, and providing calorific value information of the newly generated high-energy alloy composition for the calorific value reward function.

[0042] In some embodiments, constructing a calorific reward function with physical experience constraints includes: in order to construct a calorific reward function with physical experience constraints specifically for the generation of energetic alloy materials, in addition to considering the calorific value characteristics predicted by the prediction model, auxiliary terms that affect the chemical rationality, alloy synthesizability, and combustion performance of the energetic alloy materials are added.

[0043] The expected reward value is output through the calorific value reward function, and then the value is fed back to the component generation model to complete the cognitive update and iteration of the model, thereby optimizing the performance of the newly generated component. Finally, for the newly generated energetic alloy material component, a crystal structure template matching method is used to endow it with a crystal structure with synthesis potential, resulting in a novel energetic alloy material with a crystal structure.

[0044] like Figure 2C As shown, in the high-throughput virtual screening process of energetic alloy materials, the dataset of energetic alloy materials was divided into a training set, a validation set, and a test set in a 6:2:2 ratio. Two independent crystal graph convolutional neural network prediction models were trained, using formation energy and thermal energy as the neural network prediction performance targets, respectively. The accuracy R0 of the thermal energy prediction model and the formation energy prediction model were [missing information]. 2 The values ​​are 0.996 and 0.98, respectively. Based on the two crystal diagram convolutional neural network prediction models obtained through training, the newly generated energetic alloy material with crystal structure is used as input to predict the formation energy and calorific value of the newly generated energetic alloy material. The calorific value is higher than 31.1 kJ / g and the formation energy is less than 0 eV / atom, respectively, as dual screening thresholds to obtain the virtually screened stable and high-calorific-value energetic alloy material.

[0045] like Figure 2DAs shown, in the calculation and verification process of high-calorific-value energetic alloy materials, based on the calorific value and formation energy of the selected high-calorific-value energetic alloy materials, candidate materials with negative formation energy were selected and ranked by calorific value. The 10 candidate high-calorific-value energetic alloy materials with the highest calorific value were selected for quantum mechanical calculation verification. The quantum mechanical calculation software VASP was used to perform energy minimization calculations on the 10 selected high-calorific-value energetic alloy materials, extracting their relaxed energy and thus calculating the formation energy. Based on Hess's law, the calorific value of the 10 high-calorific-value energetic alloy materials was obtained. If the calculated calorific value reached 31.1 kJ / g, the following steps were performed sequentially: crystal structure image discrimination after relaxation, plotting the "formation energy-volume" curve (EV curve), phonon spectrum calculation, and AIMD room-temperature relaxation. Four stability verification methods were used to verify its structural stability, energy stability, kinetic stability, and room-temperature thermodynamic stability, respectively. If all verifications met the standards, it indicated that a new high-calorific-value energetic alloy material was successfully generated.

[0046] In some embodiments, if the calorific value meets the standard, four types of stability verification are performed. The verification process includes: if the calorific value meets the standard, stability verification is performed. The verification process includes: for the candidate high calorific value energetic alloy material that meets the standard, based on its crystal structure after energy minimization, the crystal structure parameter variation range is determined, the relationship between formation energy and unit cell volume is analyzed, phonon spectrum is calculated, and room temperature ab initio molecular dynamics simulation is performed in sequence to verify the crystal structure stability, energy stability, kinetic stability and room temperature thermodynamic stability of the candidate high calorific value energetic alloy material respectively.

[0047] Specifically, the structural stability process includes: based on the crystal structure image of the high-calorific-value energetic alloy material after energy minimization, judging whether crystal collapse has occurred according to the changes in unit cell parameters and atomic positions. If the changes are small, it indicates that the crystal structure of the high-calorific-value energetic alloy material has good stability.

[0048] The energy stability verification process includes: based on the crystal structure of the high-calorific-value energetic alloy material with energy minimization, and analyzing its unit cell volume. Volumetric strain was calculated, and the atomic model structure was relaxed using the NVT ensemble. The cell energy at each volumetric strain point was calculated, and the formation energy was also calculated. The relationship curve between the formation energy and the cell volume was plotted using the third-order Birch–Murnaghan equation. Since the formation energy corresponding to the equilibrium volume at the lowest point of the curve is negative, it indicates that the crystal structure of this high-calorific-value energetic alloy material has energy stability.

[0049] The dynamic stability verification process includes: based on the crystal structure of the high-calorific-value energetic alloy material with energy minimization, constructing a crystal supercell model of the high-calorific-value energetic alloy material, applying small displacements to the asymmetric equivalent atoms to obtain the second-order force constant matrix, and calculating the phonon dispersion relation therefrom. When the obtained phonon spectrum does not contain imaginary frequencies in all high-symmetric wave vector paths, it is determined that the crystal structure of the high-calorific-value energetic alloy material has dynamic stability.

[0050] The process of verifying the thermodynamic stability at room temperature includes: constructing a supercell model of the high-calorific-value energetic alloy material based on the crystal structure of the high-calorific-value energetic alloy material with energy minimization; performing time evolution at room temperature using ab initio molecular dynamics simulation; obtaining the changes in system energy and temperature over time by solving the electronic structure in real time; and determining that the crystal structure of the high-calorific-value energetic alloy material has thermodynamic stability at room temperature when the system's energy and temperature only exhibit stable fluctuations around the mean after thermal equilibrium without lattice disorder, structural collapse, or chemical bond breakage.

[0051] The following detailed description of the method for directional design of high-calorific-value energetic alloy materials based on coupled element information provided in this application is illustrated through specific embodiments.

[0052] This application primarily utilizes a smart paradigm for materials research and development, focusing on the targeted design of high-calorific-value energetic alloys. By combining a deep reinforcement learning model, a neural network prediction model, and quantum mechanical calculations for verification, the targeted design of high-calorific-value energetic alloys is achieved. It should be noted that, compared to experimental design, besides the innovation in technical methods, this application focuses on energetic alloys with crystalline structures as its design objective, rather than being limited to compositional design.

[0053] The method for targeted design of high-calorific-value energetic alloy materials provided in this application includes the following steps: First, extract the material composition, crystallographic files, and formation energy information of the energetic alloy materials from public databases such as Material Project, and perform deduplication processing, using identical material composition and crystal structure as the data deduplication standard. Based on Hess's Law and combined with the theoretical combustion reaction formula of energetic alloy materials, its calorific value is calculated from the difference between the formation energy of the energetic alloy material and the combustion products, and an automated processing script is written to perform high-throughput calculation of its calorific value. The extracted material composition, crystallographic files, formation energy information, and calorific value information calculated by Hess's Law are jointly stored to construct a dedicated energetic alloy material dataset.

[0054] Next, in order to construct a deep reinforcement learning-oriented generation model for high-calorific-value energetic alloy materials, the elemental information affecting the calorific value of energetic alloy materials is derived from the dataset information. A dedicated calorific value prediction model is embedded and the reward function is maximized as the reward guide to generate high-calorific-value energetic alloy components. Then, the crystal structure prediction model is used to match and obtain potential new energetic alloy materials with crystal structures.

[0055] Then, the dedicated energetic alloy material dataset was divided into training, validation, and test sets according to a ratio of 6:2:2. Using this dataset as input, two crystal graph neural network prediction models with accuracies of 0.980 and 0.996 were trained, respectively, with calorific value and formation energy as performance targets. These models performed high-throughput predictions of the calorific value and formation energy of directionally generated energetic alloy materials with crystalline structures. A dual threshold was set: a calorific value higher than 31.1 kJ / g and a formation energy lower than 0 eV / atom. Potential high-calorific-value energetic alloy materials were virtually screened. Based on the calorific value and formation energy of the screened high-calorific-value energetic alloy materials, candidate materials with negative formation energies were ranked by calorific value. The top 10 candidate high-calorific-value energetic alloy materials were then subjected to quantum mechanical calculations for verification. Finally, quantum mechanical calculations were performed using VASP software to verify the results. Energy minimization calculations were performed on the atomic crystal model structures of the selected 10 high-calorific-value energetic alloys. Based on the final unit cell energy, the unit cell energy was extracted and the formation energy was calculated. Based on Hess's Law, the enthalpy of combustion was evaluated according to the difference between the theoretical combustion products and the formation energy of the high-calorific-value energetic alloy. By normalizing by unit mass, the calorific value of the corresponding high-calorific-value energetic alloy was calculated, and it was verified whether the calorific value reached the preset threshold of 31.1 kJ / g. If the calorific value met the threshold, four stability calculations were performed: crystal image discrimination after structural relaxation, plotting the "formation energy-volume" curve (EV curve), phonon spectrum calculation, and room-temperature AIMD simulation. These multi-dimensional verifications validated the stability of the newly generated high-calorific-value energetic alloy. Finally, the rationality of the directionally designed high-calorific-value energetic alloy was verified. The flowcharts for each process are as follows: Figures 2A to 2D As shown.

[0056] The embodiments of this application will now be described in detail with reference to the accompanying drawings, so that those skilled in the art can better understand and implement this application.

[0057] The following are specific examples of the implementation process of the technical solution to be protected in this application. However, this application may also adopt other implementation methods different from those described herein. Those skilled in the art can adopt different technical solutions under the guidance of the ideas in this application to realize the scope of this application. Therefore, this application is not strictly limited to the specific implementation examples below.

[0058] The method for directional design of high-calorific-value energetic alloy materials provided in this application includes the following steps: Step 1: At the outset, this application establishes a dataset of energetic alloy materials presented in Crystallographic Information File (CIF) format. Material composition, crystallographic files, and formation energy information for energetic alloy materials are extracted from publicly available databases such as the Material Project, and deduplication is performed, using identical material composition and crystal structure as the deduplication criterion. For the key performance indicator of energetic crystalline materials, the theoretical calorific value of its combustion reaction is evaluated based on Hess's Law, i.e., by calculating the difference between the formation energy of the energetic alloy material and the formation energy of specific combustion products. Given that metallic fuels are solid phases, their enthalpy change changes very little with temperature; therefore, this application approximates the enthalpy change at 0 K for characterizing the enthalpy change at room temperature. The formation energy at 0 K is used to calculate the combustion enthalpy using Hess's Law, based on the difference between the formation energy of the combustion products and the formation energy of the energetic alloy material. Subsequently, mass normalization is performed to convert the combustion enthalpy into calorific value. The extracted material composition, crystallographic files, formation energy information, and calorific value information calculated using Hess's law were jointly stored to construct a dedicated energetic alloy materials dataset. This calorific value result not only reveals the stored level of chemical energy within the material but also provides a quantitative physical criterion for screening candidate materials with high energy density (calorific value higher than that of Al powder). The constructed energetic alloy materials dataset contains 66 energetic metallic elements, comprising 27,459 original data entries for energetic alloy candidate materials, with calorific value being the core target attribute in this study. This dataset lays the data foundation for the targeted design and high-throughput virtual screening of energetic alloy materials. Step 2: A deep reinforcement learning model is proposed that couples elemental information related to the calorific value of energetic alloy materials to achieve high-throughput targeted generation of energetic alloy materials. This model couples elemental information related to the calorific value of the energetic alloy material dataset and performs targeted design of energetic alloy materials based on a stacked recurrent neural network (Stack-RNN) generative model with an embedded calorific value reward function. The specific principle is as follows: Stack-RNN is a recurrent neural network (RNN) with a stacked structure. By combining multiple simple RNN layers and allowing information to flow non-linearly between these layers, it enhances the model's ability to understand sequential data. It can flexibly capture contextual information at different levels and has greater potential for understanding and generating sequence patterns. This helps generative models understand complex chemical concepts such as ionic charge, electronegativity, and combustion efficiency. The main generation process is triggered by a predefined starting state. At each time step, the generative model predicts the next element and its corresponding composition based on the currently generated material composition information, and gradually appends the prediction results to the generated sequence until a preset termination condition is met, thereby completing the generation of a single energetic alloy material candidate composition.

[0059] The overall deep reinforcement learning-based generative model is implemented in two steps. The first step involves basic training of the model on an energetic alloy materials dataset. This step is solely for the model to learn the effective composition of chemical formulas. The generative model is pre-trained using 27,459 previously collected energetic alloy materials to capture chemically effective combination patterns, ensuring the chemical validity of the generated materials. Subsequently, the component generation model and the "component-calorific value" prediction model are combined into a deep reinforcement learning framework. An auxiliary term for the reward function, incorporating physical characteristics related to the combustion of energetic alloy materials, is added to the calorific value reward function to generate higher-quality energetic alloy components and improve iteration efficiency. The "component-calorific value" prediction model predicts the calorific value of each generated energetic alloy component and assigns a reward value based on the reward function, ultimately feeding this information back to the component generation model. The component generation model is configured to maximize the expected reward, optimizing towards a higher expected value for the reward function. Simultaneously, this application incorporates domain-specific elemental information related to the combustion calorific value of energetic alloy materials into the model, assisting the model in recognizing the calorific value properties of energetic alloy materials and accelerating the model's understanding of these materials. Based on the calorific value-related elemental information appearing in the constructed energetic alloy material dataset, this application performs statistical analysis on the data in the high calorific value region (above 31.1 kJ / g for aluminum powder). For example... Figure 3 As shown, the elements highly correlated with the calorific value of energetic alloy materials are displayed. Blue represents the frequency of elements appearing in the high-calorific-value region of the dedicated energetic alloy material dataset (the top 10 elements statistically highest), and red represents the calorific value performance of the corresponding elemental form. Here, the top 5 elements with excellent frequency and corresponding elemental calorific value are selected as high-energy elements for energetic alloy fuels, including B, Li, Be, Mg, and Al. To improve the iterative efficiency of the generative model, this application sets these 5 elements as the first elements to appear in the sequence generation stage, and determines their random occurrence probability based on the frequency of the element. Let the number of times a certain element appears in the high-calorific-value data region be denoted as . The probability that it is the first element is defined as:

[0060] in, This represents the first element (random variable) when the sequence is generated. This represents the summation of the frequencies of these five elements.

[0061] The purpose of the "composition-calorific value" prediction model is to predict the calorific value of each component generated by the model and pass the predicted calorific value information to a calorific value reward function with physical experience constraints. Since this stage only generates the composition of energetic alloy materials, this application employs a machine learning prediction model that only accepts stoichiometry as input, and fine-tunes it using a collected dataset to obtain an accuracy R0. 2 The "component-calorific value" prediction model with a value of 0.986, such as... Figure 4 The test set accuracy results demonstrate that the machine learning prediction model can accurately predict the calorific value of energetic alloy components.

[0062] The reward value originates from the reward function in the model. In addition to considering the calorific value characteristics predicted by the "component-calorific value" prediction model, auxiliary terms affecting the combustion performance of energetic alloy materials are also included. The purpose of setting the reward function is to clarify the optimization objective of the entire strategy network, namely, maximizing the predicted reward value. However, the setting of the reward function, besides considering the calorific value characteristics of energetic alloy materials, also needs to consider factors affecting the preparation and combustion performance of energetic alloy materials, such as chemical rationality, alloy synthesability, and combustion efficiency. To strengthen these physical constraints, this application adds supplementary reward terms to the reward function, resulting in the main reward function:

[0063] in, It is the final state reward value. It is the main reward item based on the predicted calorific value calculation, that is, the normalized result of the calorific value predicted by the "component-calorific value" prediction model. It is an auxiliary modulation factor composed of multiple a priori physical characteristics, namely, the product of normalized coefficients of the material's physicochemical properties, synthesis properties, and combustion efficiency. yes The weighting factor.

[0064] Calorific value is the core optimization objective of the entire generative model, and a pre-trained embedded "component-calorific value" prediction model is used for prediction. Considering the inconsistency in units between reward items, normalization processing is required for attributes including calorific value; therefore, the calorific value function of the main item is as follows:

[0065] in, It is the calorific value of the energetic alloy material predicted by the "composition-calorific value" prediction model. represents the normalization function, used to map the predicted heat value to a predetermined reward interval [0,1].

[0066] The auxiliary items primarily evaluate the synthesis and combustion performance of the generated components to prevent the formation of large quantities of unreasonable energetic alloy components, thereby improving the quality of the generated data. Specifically, compounds with a balanced charge neutrality and electronegativity and an appropriate number of element types are positively weighted; compounds with a positive enthalpy of formation estimated by the Miedema semi-empirical model are negatively weighted; and compounds with positive combustion efficiency as determined by the Glassman combustion rule are positively weighted. These supplementary items are collectively included in an auxiliary reward factor.

[0067] in, It is a physicochemical rationality modulation term. It is a modulatory term for alloy synthesability. This refers to the combustion efficiency modulation term. The acquisition of each value is explained as follows: The auxiliary reward term is used to evaluate the synthesis and combustion potential of newly generated energetic alloy materials to avoid generating energetic alloy combinations that are clearly unreasonable at the chemical or engineering level. Specifically, for generated compositions that meet the requirements of charge neutrality, reasonable electronegativity distribution, and appropriate element types and quantities, a positive modulation weight is assigned; for generated compositions whose formation enthalpy estimated based on the Miedema semi-empirical alloy formation model is negative or close to zero, they are considered to have basic syntheticability at the alloy level, and no additional reward is added; however, when the estimated formation enthalpy is positive and deviates from a reasonable range, an incremental penalty is applied to the generation result only according to the magnitude of its positive value to suppress combinations that are clearly unfavorable to alloy formation. Furthermore, based on Glassman's empirical rules for metallic fuel combustion, the effective combustion efficiency of the generated compositions is evaluated, and positive modulation is applied to generated results with higher combustion efficiency. The above evaluation results are uniformly mapped to dimensionless modulation factors, which are used to correct the main reward corresponding to the predicted calorific value, thereby improving the overall feasibility of generating energetic alloy materials in terms of chemistry, synthesis and combustion while ensuring that the high calorific value optimization target remains unchanged.

[0068] After obtaining the target energetic alloy composition output by the model, a potential crystal structure is assigned to the composition using a crystal template matching method. This structure prediction method relies on metric learning and automatically selects template structures with high chemical substitutability from a crystal structure database to best match the unknown stable structure of the target chemical composition. For each target composition, the model predicts the closest matching crystal structure. This provides crystallographically plausible candidate schemes for subsequent high-throughput machine learning screening and first-principles verification. Another role of crystal structure matching is to establish connections with experimentally observed compounds with similar structures, which can provide researchers attempting to achieve new analogues in the laboratory with more useful information regarding synthetic routes and conditions.

[0069] Step three is primarily used to evaluate the calorific value and thermodynamic stability of the energetic alloy material with a crystalline structure generated in step two. This application uses calorific value and formation energy as dual screening criteria for verification and evaluation. Calorific value is used as the target performance descriptor, with the calorific value of aluminum powder (31.1 kJ / g) as the relevant engineering benchmark. Simultaneously, a negative formation energy is required to ensure thermodynamic stability. A high-throughput screening was performed using a Crystal Graph Convolutional Neural Network (CGCNN) model, which explicitly incorporates atomic configuration and bonding environment. The CGCNN transforms atoms (nodes), bonds (edges), and their features (such as interatomic distances) into features in vector space, representing the local features of the energetic alloy material's crystal structure. These local features can capture the atomic-level arrangement and interactions within the material's microstructure.

[0070] First, the energetic alloy material dataset collected in step one was divided into a training set, a validation set, and a test set in a ratio of 6:2:2. The CGCNN model was trained on this dataset, which contains labeled material formation energies and calorific values. After model fine-tuning, the accuracy R0 was obtained for each dataset. 2 The formation energy prediction model with a value of 0.98 (e.g.) Figure 5 (as shown) and accuracy R 2 A calorific value prediction model with a value of 0.996 (such as...) Figure 6 As shown in the figure, the high accuracy of both demonstrates that the model can accurately predict the newly designed energetic alloy materials. It was then applied to rapidly evaluate the energetic alloy materials generated in the second generation stage. By setting thresholds for calorific value and generation energy, candidate energetic alloy materials with low performance or instability can be eliminated in high throughput, while retaining the high-calorific-value energetic alloy materials virtually screened.

[0071] Step 4: Based on the calorific value and formation energy of the high-calorific-value energetic alloy materials selected in Step 3, the calorific value reward function in the deep reinforcement learning generative model is used for calculation and sorting. The 10 with the highest reward values ​​are selected for quantum mechanical calculation verification. The Vienna Ab initio Simulation software package (VASP) is used for this quantum mechanical calculation within the Projected Enhanced Wave (PAW) framework. The exchange-related functional is approximated using the Perdew-Burke-Ernzerhof (PBE) generalized gradient approximation (GGA). Energy minimization calculations are performed on the atomic crystal model structures of the 10 selected high-calorific-value energetic alloy materials. Based on the final unit cell energy, the unit cell energy is extracted and the formation energy is calculated. Based on Hess's Law, the combustion enthalpy is evaluated according to the difference between the theoretical combustion products and the formation energy of the high-calorific-value energetic alloy material. By normalizing by unit mass, the combustion calorific value of the corresponding high-calorific-value energetic alloy material is calculated, and it is verified whether the calorific value reaches the preset threshold of 31.1 kJ / g. If the calorific value meets the threshold, four stability verifications are performed.

[0072] Given the novelty of the generated energetic alloy material, four stability verification methods were implemented to further ensure the reliability of the structure. First, the crystal structure image of the energetic alloy material after relaxation was evaluated. If the atomic-level structure of the energetic alloy material remains stable, and the bond lengths and bond angles change only slightly, it indicates that the designed energetic alloy material possesses structural stability. Based on the crystal structure of the high-calorific-value energetic alloy material with minimal energy, its unit cell volume was... Volumetric strain was assessed, and the atomic model structure was relaxed using the NVT ensemble. The cell energy at each volumetric strain point was calculated, along with the formation energy. The relationship between the formation energy and cell volume (EV curve) was plotted using the third-order Birch–Murnaghan equation. The negative formation energy corresponding to the equilibrium volume at the lowest point of the curve indicates that the crystal structure of this high-calorific-value energetic alloy possesses energy stability. The phonon dispersion relation of the energetic alloy was calculated using the Phonopy software package. Based on the energy-minimizing crystal structure of the high-calorific-value energetic alloy, a supercell model of the high-calorific-value energetic alloy was constructed. Small displacements were applied to the asymmetric equivalent atoms to obtain the second-order force constant matrix, and the phonon dispersion relation was calculated accordingly. When the obtained phonon spectrum contains no imaginary frequencies along all high-symmetric wave vector paths, the crystal structure of the high-calorific-value energetic alloy is determined to possess dynamic stability. Based on VASP software, ab initio molecular dynamics simulations were used to study the time evolution of the crystal structure of a high-calorific-value energetic alloy at room temperature. By solving the electronic structure in real time, the changes in energy and temperature over time were obtained. When the system reached thermal equilibrium, its energy and temperature exhibited only stable fluctuations around the mean without lattice disorder, structural collapse, or chemical bond breakage, the crystal structure of the high-calorific-value energetic alloy was determined to possess thermodynamic stability and storability at room temperature. These four rigorous stability assessments demonstrate that the novel energetic alloy material meets stability requirements. Finally, by comprehensively considering the verified calorific value and stability, novel energetic alloy materials were precisely selected, effectively guiding experimental design.

[0073] Compared with existing technologies, the method for targeted design of high-calorific-value energetic alloy materials provided in this application has the following advantages: First, addressing the inefficiency in designing high-calorific-value energetic alloys, this application differs from traditional experience-dependent experimental designs. It proposes a complete intelligent design framework for high-calorific-value energetic alloys, realizing a complete workflow from energetic alloy data sets to new material design. For targeted design of energetic alloys, the overall approach employs a high-throughput targeted generation, neural network model performance prediction, and high-precision quantum mechanical multiple verification technique, combining deep reinforcement learning for coupled element information, neural network prediction, and quantum mechanical calculations. This framework enables targeted design of energetic alloys for high-calorific-value target performance.

[0074] Secondly, in the preliminary dataset construction work, this application collected data from multiple public databases to ensure the integrity of energetic alloy data. Furthermore, the process of calculating energetic alloy materials using Hess's Law was transformed into an automated calculation task. For newly designed materials, only the formation energy needs to be calculated, and then input into the automated calculation program to obtain the corresponding calorific value.

[0075] In addition, in the step of generating high calorific value targets for energetic alloy materials, this application performs statistical analysis of element frequencies based on high calorific value region data in a dedicated energetic alloy material dataset. The statistical results, combined with the combustion calorific value of the element, select five high-energy elements. This probability is applied to the first step of the component generation model, which restricts the elements, and the physical experience related to the combustion calorific value of energetic alloy materials is digitally transmitted to the model for input.

[0076] In addition, in the step of generating high calorific value targets for energetic alloy materials, a physical interpretability auxiliary function is set for the synthesis and combustion performance of energetic alloy materials to optimize the reward mechanism and improve the quality of data generation.

[0077] For compositions that satisfy the requirements of charge neutrality, reasonable electronegativity distribution, and appropriate element types and quantities, a positive modulation weight is assigned. For compositions where the formation enthalpy estimated based on the Miedema semi-empirical alloy formation model is positive but deviates from a reasonable range, an incremental penalty is applied to the formation result only based on the magnitude of the positive value to suppress combinations that are clearly unfavorable to alloy formation. Based on Glassman's empirical rules for metal combustion, the effective combustion efficiency of the generated compositions is evaluated, and positive modulation is applied to compositions with higher combustion efficiency.

[0078] In addition, in the high-throughput virtual screening of energetic alloy materials, this application trained a dedicated machine learning calorific value prediction model and a machine learning formation energy prediction model based on its own constructed energetic alloy material dataset, and fine-tuned them to high accuracy.

[0079] Moreover, in the calculation and verification of high-calorific-value energetic alloy materials, this application did not simply use the formation energy to determine the stability of the newly designed energetic alloy materials. Instead, it used crystal structure images after structural relaxation, the "formation energy-volume" curve (EV curve), phonon spectrum calculation, and room temperature AIMD simulation to demonstrate the stability of the designed high-calorific-value energetic alloy materials from multiple dimensions. At the same time, it also demonstrated the potential of the newly designed high-calorific-value energetic alloy materials in experimental synthesis and subsequent long-term storage.

[0080] In summary, the method for targeted design of high-calorific-value energetic alloy materials provided in this application employs a data-driven machine learning approach to intelligently design energetic alloy materials. Compared to traditional experimental design, this method overcomes the inefficiency of designing novel energetic alloy materials. It enables efficient design of high-calorific-value energetic alloy materials within a broad compositional space at low cost in a safe environment free from explosions and toxic hazards, and allows for simulation verification, thereby effectively improving design efficiency and accelerating experimental design.

[0081] In the target generation stage of energetic alloy materials in this application, multidimensional elemental information related to combustion calorific value is coupled to assist the generation model in recognizing and utilizing statistical laws. A specific reward function with physical empirical characteristics designed for high-calorific-value energetic alloy materials is embedded. This reward function takes the target calorific value as the main term and adds auxiliary modulation terms for chemical rationality, synthesis potential, and normalized transformation of combustion efficiency. This combines the intelligent design process with material synthesis and combustion performance, enabling the generation of higher-quality data and providing a reasonable explanation for the combustion performance of energetic alloy materials.

[0082] In the high-throughput virtual screening stage of energetic alloy materials in this application, a high-precision calorific value prediction model with data fine-tuning is proposed. This model can compensate for the energy loss when the theoretical energy density of energetic alloy materials is estimated by weighting the energy density ratio of the constituent elements in the experiment. Compared with expensive density functional theory calculations, it can predict the calorific value of newly designed energetic alloy materials more efficiently and at a lower cost.

[0083] In this application, the quantum mechanical calculation software VASP is used to verify the virtual screening of high calorific value energetic alloy materials. Compared with the long cycle and high cost of experimental synthesis, this method can more efficiently realize the calorific value and stability verification of novel high calorific value energetic alloy materials. Moreover, the four verification methods designed in this application can verify the synthesis potential and long-term storage of novel high calorific value energetic alloy materials from multiple dimensions.

[0084] To verify the method provided in this application, this application collected raw data and calculated calorific value performance, constructing a dedicated dataset of "energetic alloy materials - calorific value" with 27,459 data entries. Based on this dataset, the component generation model in deep reinforcement learning, the "component-calorific value" prediction model, and two crystal graph convolutional neural network prediction models in high-throughput virtual screening were trained. The training effect has been demonstrated in the specific scheme, achieving high-throughput virtual prediction of the generated energetic alloy materials. Then, the predicted calorific value and predicted formation energy of the selected stable and high-calorific-value energetic alloy materials were input into the calorific value reward function of the deep reinforcement learning model, and the top 10 high-calorific-value energetic alloy materials with the highest reward values ​​were selected. Finally, quantum mechanical calculations based on VASP software were used to screen out three new high-calorific-value energetic alloy materials that meet the standards for both calorific value and stability. Atomic simulation verification yielded the calorific value properties of the three new high-calorific-value energetic alloy materials, as shown in Table 1 below. It can be seen that the combustion performance of the three exceeds the energy limit of traditional aluminum powder. The four stability verification results are as follows... Figure 7As shown, the figures 1, 2, 3, and 4 in the lower right corner represent the crystal structure image after structural relaxation, the energy-volume (EV) curve, the calculated phonon spectrum, and the room-temperature AIMD simulation results, respectively. The verification results are as follows: After structural relaxation, the cell parameters and atomic positions of the three novel high-calorific-value energetic alloys remain essentially unchanged, indicating structural stability. The lowest points in the energy-volume (EV) curves are all less than 0 eV / atom, indicating energy stability. The phonon spectra do not exhibit imaginary frequencies (i.e., the ordinate is less than 0) at high symmetry points, indicating kinetic stability. AIMD simulation results show that the temperature and energy of the three novel high-calorific-value energetic alloys fluctuate around the equilibrium point, and no significant structural changes occur, indicating good thermodynamic stability at room temperature and the ability to maintain long-term storage stability. Based on the above calorific value and stability verifications, this application demonstrates that it can achieve the directional design of high-calorific-value energetic alloys.

[0085] Table 1. Three validated new high-calorific-value energetic alloy materials

[0086] Based on the above technical solutions, this application provides a method for targeted design of high-calorific-value energetic alloy materials by coupling element information. This method includes the following steps: First, establishing a dedicated dataset of energetic alloy materials containing the formation energy and calorific value properties; then, constructing a deep reinforcement learning model of high-calorific-value energetic alloy materials by coupling element information, and training the model based on the dedicated dataset to generate potential energetic alloy material components and assigning them with synthetic potential crystal templates to obtain energetic alloy materials with crystal structures; next, training two crystal graph convolutional neural network prediction models for calorific value and formation energy, respectively, based on the dedicated dataset; subsequently, predicting the performance of energetic alloy materials with crystal structures to virtually select candidate high-calorific-value energetic alloy materials; finally, performing quantum mechanical calculations on the selected candidate high-calorific-value energetic alloy materials to verify their calorific value and stability.

[0087] This application utilizes a data-driven intelligent design method combined with computational simulation to avoid safety risks in experiments from a technical perspective, and achieves cost-effective and efficient targeted design of novel high-calorific-value energetic alloy materials. On one hand, this application uses element frequency statistics from a collected dataset, combined with the calorific value of the corresponding element's elemental form, for dual verification. It incorporates elemental information with physical experience into a deep reinforcement learning model, applying existing experience and improving design efficiency. On the other hand, the reward function for targeted design embeds an auxiliary function for the combustion performance of the energetic alloy material, reflecting the influence of synthesis potential and combustion efficiency during the design generation process, avoiding blindly pursuing only calorific value performance. Furthermore, in high-throughput prediction, this application trains a high-precision calorific value prediction model, achieving accurate prediction of the calorific value performance of the energetic alloy material, rather than relying on error-laden approximate estimations. Based on the crystal structure of the newly generated high-calorific-value energetic alloy material, this application can use first-principles calculations based on density functional theory (DFT) to perform quantum mechanical calculations on the targeted high-calorific-value energetic alloy material, verifying its calorific value and multi-dimensional stability.

[0088] Those skilled in the art will understand that the above-described embodiments are specific examples of implementing this application, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of this application. Any person skilled in the art can make their own modifications and alterations without departing from the spirit and scope of this application; therefore, the scope of protection of this application should be determined by the scope defined in the claims.

Claims

1. A method for directional design of high-calorific-value energetic alloy materials using coupled element information, characterized in that, Includes the following steps: Establish a dedicated dataset for energetic alloy materials that includes the formation energy and calorific value properties of energetic alloy materials; A deep reinforcement learning model for high-calorific-value energetic alloy materials coupled with element information is constructed. The model is trained based on a dedicated dataset of energetic alloy materials to generate potential energetic alloy material components in a targeted manner. Crystal templates with synthetic potential are then assigned to obtain energetic alloy materials with crystal structures. Based on a dedicated dataset of energetic alloy materials, two crystal graph convolutional neural network prediction models were trained, one for calorific value and the other for formation energy. Subsequently, the performance of energetic alloy materials with crystal structures was predicted, and candidate high-calorific-value energetic alloy materials were virtually screened. Quantum mechanical calculations were performed on the selected candidate high-calorific-value energetic alloy materials to verify their calorific value and stability.

2. The method for directional design of high-calorific-value energetic alloy materials according to claim 1, characterized in that, The establishment of a dedicated dataset for energetic alloy materials, including the formation energy and calorific value properties of energetic alloy materials, includes: Material composition, crystallographic files and formation energy information of energetic alloy materials were extracted from public databases, and data cleaning was performed with the same material composition and crystal structure as the data cleaning standard. Based on Hess's Law and combined with the theoretical combustion reaction of energetic alloy materials, the combustion calorific value information of energetic alloy materials is obtained; The extracted material composition, crystallographic files, formation energy information, and obtained calorific value information are stored together to construct a dedicated dataset of energetic alloy materials.

3. The method for directional design of high-calorific-value energetic alloy materials according to claim 1, characterized in that, The construction of a deep reinforcement learning model for high-calorific-value energetic alloy materials coupled with elemental information includes: A component generation model coupled with element information, a "component-calorific value" prediction model, and a calorific value reward function constrained by physical experience were constructed respectively. By placing the component generation model and the "component-calorific value" prediction model in the same deep reinforcement learning framework and using the calorific value reward function based on physical experience for reward value feedback, a deep reinforcement learning model for high calorific value energetic alloy materials with coupled element information is formed.

4. The method for directional design of high-calorific-value energetic alloy materials according to claim 3, characterized in that, Constructing a component generation model that couples element information, including: Based on a dedicated dataset of energetic alloy materials, we statistically summarize the elemental information that affects the calorific value of energetic alloy materials and apply it to the element sequence generation constraints in the composition generation model.

5. The method for directional design of high-calorific-value energetic alloy materials according to claim 3, characterized in that, Constructing a "component-calorific value" prediction model includes: Based on the material composition and corresponding calorific value information in the dedicated energetic alloy material dataset, the elements and stoichiometry in the material composition are used as inputs to the "composition-calorific value" prediction model, and a "composition-calorific value" prediction model that can efficiently predict the composition of newly generated energetic alloy materials is constructed.

6. The method for directional design of high-calorific-value energetic alloy materials according to claim 3, characterized in that, Construct a heat value reward function with physical experience constraints, including: In addition to considering the calorific value characteristics predicted by the "composition-calorific value" prediction model, auxiliary terms affecting the chemical rationality, alloy synthesability, and combustion performance of energetic alloy materials are added, as shown in the following formula: in, It is the final state reward value. It is the main reward item based on the predicted calorific value calculation, that is, the normalized result of the calorific value predicted by the "component-calorific value" prediction model. It is an auxiliary modulation factor composed of multiple a priori physical characteristics, namely, the product of normalized coefficients of the material's physicochemical properties, synthesis properties, and combustion efficiency. yes The weighting factor; By normalizing the calorific value predicted by the "component-calorific value" prediction model, the main reward item is obtained. The formula is shown below: in, It is the calorific value of energetic alloy materials predicted by the "composition-calorific value" prediction model. The normalization function is used to map the predicted heat value to a predetermined reward interval [0,1]. Taking into account the influence of material physicochemical properties, synthesis properties, and combustion efficiency, auxiliary terms are obtained. The resulting compounds are weighted positively for those with a balance of charge neutrality and electronegativity and an appropriate number of element types; those with a positive enthalpy of formation as estimated by the Miedema semi-empirical model are weighted negatively; and those with a positive combustion efficiency as determined by the Glassman combustion rule are weighted positively. The formula is shown below: in, It is a physicochemical rationality modulation term. It is a modulatory term for alloy synthesability. It is a combustion efficiency modulation item.

7. The method for directional design of high-calorific-value energetic alloy materials according to claim 1, characterized in that, Based on a dedicated dataset of energetic alloy materials, two crystal graph convolutional neural network prediction models were trained, one for calorific value and the other for formation energy. The dataset of the dedicated energetic alloy material was divided into a training set, a validation set, and a test set in a ratio of 6:2:

2. The formation energy and heat value were used as the neural network prediction performance targets, and two independent crystal graph convolutional neural network prediction models were trained to achieve efficient prediction of the formation energy and heat value of energetic alloy materials.

8. The method for directional design of high-calorific-value energetic alloy materials according to claim 1, characterized in that, The properties of the generated energetic alloy materials were predicted, and candidate high-calorific-value energetic alloy materials were virtually screened, including: Based on two crystal diagram convolutional neural network prediction models trained, the newly generated energetic alloy material is used as input to predict the formation energy and calorific value of the newly generated energetic alloy material. The calorific value is higher than 31.1 kJ / g and the formation energy is less than 0 eV / atom, respectively, as dual screening thresholds to obtain the virtually screened stable and high-calorific-value energetic alloy material.

9. The method for directional design of high-calorific-value energetic alloy materials according to claim 1, characterized in that, Quantum mechanical calculations were performed on the selected high-calorific-value energetic alloy materials to verify their calorific value and stability, including: Based on the calorific value and formation energy of the selected high-calorific-value energetic alloy materials, candidate materials with negative formation energy were selected and ranked by calorific value. The top 10 candidate high-calorific-value energetic alloy materials with the highest calorific value were selected for quantum mechanical calculation verification. Based on VASP software, energy minimization calculations were performed on the atomic crystal model structures of 10 selected high-calorific-value energetic alloy materials. Based on the final energy of the unit cell, the unit cell energy was extracted and the formation energy was calculated. Based on Hess's Law, the enthalpy of combustion is evaluated according to the difference between the theoretical combustion products and the formation energy of high-calorific-value energetic alloy materials. By normalizing the unit mass, the calorific value of the corresponding high-calorific-value energetic alloy material is calculated, and it is verified whether the calorific value reaches the preset threshold of 31.1 kJ / g. If the calorific value meets the standard, stability verification is carried out.

10. The method for directional design of high-calorific-value energetic alloy materials according to claim 9, characterized in that, If the calorific value meets the standard, stability verification will be conducted, and the verification process includes: For candidate high-calorific-value energetic alloy materials that meet the calorific value requirements, based on their crystal structure after minimizing energy, the following steps are performed in sequence: determining the range of changes in crystal structure parameters, analyzing the relationship between formation energy and unit cell volume, calculating phonon spectra, and performing ab initio molecular dynamics simulations at room temperature. These steps are used to verify the crystal structure stability, energy stability, kinetic stability, and room temperature thermodynamic stability of the candidate high-calorific-value energetic alloy materials.

Images