A method for obtaining detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics

Abstract

This application discloses a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, relating to the field of materials science and technology. The method includes: acquiring an initial configuration file of the energetic material to be tested, and performing dynamic processing on the initial configuration file using a materials simulation model to generate a first training dataset; iteratively training the potential function using the first training dataset to obtain a structure file corresponding to each generation of training; sampling and screening the structure files, and calculating the accuracy adjustment parameters corresponding to the sampled and screened structure files using first-principles calculation software; adjusting the parameters of the materials simulation model using the accuracy adjustment parameters, and using the adjusted materials simulation model to generate a second training dataset from the structure files obtained from each generation of training and the first training dataset; and predicting the performance of the energetic material to be tested based on the potential function trained on the second training dataset. A general method for predicting material properties that balances accuracy and efficiency is proposed.

Description

Technical Field

[0001] This application relates to the field of materials science and technology, specifically to a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. Background Technology

[0002] Energetic materials (EMs) are widely used in civilian and military fields due to their ability to release high-density energy in extremely short periods of time, such as in mining blasting, aerospace propulsion, and munitions systems. Their core advantages lie in their high energy density and strong destructive power, making them indispensable in applications requiring concentrated energy output. Key performance indicators for energetic materials include detonation pressure and detonation velocity (Do). v The parameters, including the total energy released per unit mass of material during detonation, directly determine the destructive effectiveness of the material in practical applications.

[0003] In related technologies, methods for predicting the properties of energetic materials mainly include numerical prediction based on traditional models and prediction based on kinetic methods. However, traditional model-based methods rely on empirical data, are difficult to apply to novel energetic materials, and have low prediction accuracy; while kinetic methods are almost impossible to implement due to their large computational load and low efficiency.

[0004] Therefore, how to propose a method for predicting the performance of energetic materials that balances accuracy and efficiency has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, this application provides a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. The main purpose is to solve the problem that current methods for predicting the properties of energetic materials cannot simultaneously achieve both accuracy and efficiency.

[0006] In a first aspect, this application provides a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, including:

[0007] Obtain the initial configuration file of the energetic material to be tested, and use the material simulation model to perform dynamic processing on the initial configuration file to generate the first training dataset;

[0008] The potential function is iteratively trained using the first training dataset to obtain the structure file corresponding to each generation of training.

[0009] The structural files are sampled and screened, and the precision adjustment parameters corresponding to the sampled and screened structural files are calculated using first-principles calculation software.

[0010] The material simulation model is adjusted by the precision adjustment parameters, and a second training dataset is generated from the structure file obtained from each generation of training and the first training dataset using the adjusted material simulation model.

[0011] The performance of the energetic material under test is predicted based on the potential function trained on the second training dataset.

[0012] Optionally, the step of iteratively training the potential function using the first training dataset to obtain the structure file corresponding to each generation of training includes: creating a potential function training task and labeling the training results of each generation according to the iteration number; running molecular dynamics simulation training using the potential function in a discontinuous Hugoniot type ensemble; plotting a trend graph of thermodynamic quantities changing with the number of kinetic steps; the trend graph is used to determine the stability of the dynamics while determining the number of kinetic steps; detecting the number of kinetic steps and the number of iterations, and stopping the iterative training when the stopping condition is met.

[0013] Optionally, the discontinuous Hugoniot type ensemble includes: NPHugoniotniot or NVHugoniotniot ensemble, and the initial parameters and initial structure of the ensemble are set non-uniformly.

[0014] Optionally, the material simulation model includes the MatterSim model based on a deep neural network, whose inputs include molecular formulas or unit cell structures containing atomic coordinates and lattice parameters, and whose outputs are energy, force, and stress with quantum mechanical precision.

[0015] Optionally, the initial configuration file includes the chemical structural features of the energetic material to be tested, and the first training dataset includes the initial structure and the first evaluation parameters corresponding to the initial structure.

[0016] Optionally, the step of adjusting the parameters of the material simulation model using the accuracy adjustment parameters, and using the adjusted material simulation model to generate a second training dataset from the structure file obtained from each generation of training and the first training dataset, includes:

[0017] Using the material simulation model with adjusted parameters, a second training dataset is generated from the structure files obtained from each generation of training and the initial structure in the first training dataset.

[0018] Secondly, this application provides a device for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, comprising:

[0019] The acquisition unit is configured to acquire the initial configuration file of the energetic material to be tested, and to perform dynamic processing on the initial configuration file using a material simulation model to generate a first training dataset;

[0020] The training unit is configured to iteratively train the potential function using the first training dataset to obtain the structure file corresponding to each generation of training.

[0021] The sampling unit is configured to sample and screen the structure files, and calculate the precision adjustment parameters corresponding to the sampled and screened structure files using first-principles calculation software.

[0022] The adjustment unit is configured to adjust the parameters of the material simulation model using the precision adjustment parameters, and to generate a second training dataset using the adjusted material simulation model, the structure file obtained from each generation of training, and the first training dataset.

[0023] The prediction unit is configured to predict the performance of the energetic material under test based on the potential function trained on the second training dataset.

[0024] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics as described in the first aspect.

[0025] Fourthly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics as described in the first aspect.

[0026] Fifthly, this application provides a computer program product having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics as described in the first aspect.

[0027] Using the above technical solution, this application provides a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. First, an initial configuration file of the energetic material to be tested is obtained, and the initial configuration file is dynamically processed using a material simulation model to generate a first training dataset. The potential function is iteratively trained using the first training dataset to obtain a structure file corresponding to each generation of training. The structure files are sampled and screened, and the accuracy adjustment parameters corresponding to the sampled and screened structure files are calculated using first-principles calculation software. The material simulation model is then adjusted using the accuracy adjustment parameters, and a second training dataset is generated using the adjusted material simulation model for each generation of training structure files and the first training dataset. Based on the potential function trained using the second training dataset, the performance of the energetic material to be tested is predicted. Compared with related technologies, this method first obtains a large dataset of a certain quality based on the initial configuration file using a large model (e.g., the first training set consisting of the initial structure and the corresponding first evaluation parameters is used as the first generation training samples). The purpose is mainly to obtain the structure file of each iteration result. Model processing is more time-saving and efficient than other methods. Then, first-principles calculations are used to select some of the training iterative structure files for calculation, obtaining more accurate data to adjust the model's accuracy. After adjusting the model, the model performance is improved, and a second training dataset is generated from the obtained structure files. The second training dataset has better accuracy than the first training dataset and is used to train the potential function to predict the performance of the energetic material under test.

[0028] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0029] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This paper illustrates a flowchart of a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, as provided in an embodiment of this application.

[0031] Figure 2 This paper illustrates a schematic diagram of the logic flow of a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, according to an embodiment of this application.

[0032] Figure 3 This illustration shows a schematic diagram of how a thermodynamic quantity changes with the number of kinetic steps, according to an embodiment of this application.

[0033] Figure 4 This illustration shows a schematic diagram of a device for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, according to an embodiment of this application. Detailed Implementation

[0034] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0035] As mentioned in the background, energetic materials (EMs) are widely used in civilian and military fields, such as mining blasting, aerospace propulsion, and munitions systems, due to their ability to release high-density energy in a very short time. Their core advantages lie in their high energy density and strong destructive power, making them indispensable in applications requiring concentrated energy output. Key indicators for evaluating the performance of energetic materials include detonation pressure, detonation velocity (Dv), and the total energy released per unit mass of material during detonation. These parameters directly determine the material's destructive effectiveness in practical applications. However, the detonation process is accompanied by extreme high-temperature and high-pressure conditions. Traditional experimental methods require kilogram-scale samples for testing, resulting in high experimental costs, complex processes, and certain dangers. Without effective prediction methods, significant resources must be invested in materials with poor performance, severely restricting the screening and design of new high-performance energetic materials.

[0036] Currently, theoretical methods for predicting the detonation performance of energetic materials are developing in two main directions. One is based on numerical models developed from classical detonation theories (such as Chapman–Jouguet, CJ theory), with representative software including CHEETAH and EXPLO5. These methods typically estimate detonation performance quickly by solving partial differential equations under the principle of minimum free energy and combining empirical parameters. Their advantage lies in their fast computation speed, providing prediction results within minutes. However, their accuracy depends on sufficient experimental data for parameterization, resulting in lower accuracy when dealing with novel materials. On the other hand, with improved computational capabilities, atomic-scale molecular dynamics (MD) simulations have become a research hotspot. For example, multi-scale shock technique (MSST) is used to describe detonation paths along Rayleigh lines. However, the MSST method requires orthorhombic unit cell structures, making it unsuitable for most non-orthorhombic crystal systems (such as common energetic crystals like TATB), and the extreme temperature and pressure conditions of the simulation environment may deviate from real detonation conditions. The use of the Hugoniot method in the past was limited by the extremely high energy barrier, and the simulation time required often exceeded hundreds of picoseconds, making it computationally unsustainable for traditional classical potential functions and ab initio methods, and difficult to apply in practice to the prediction of detonation performance.

[0037] To address at least one of the aforementioned problems, this embodiment proposes a method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. This method can be applied to a device or electronic device for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. This device or electronic device can be installed or integrated into a material prediction system and, during operation, can execute any of the methods mentioned below for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. Figure 1 As shown, the method includes:

[0038] S101, Obtain the initial configuration file of the energetic material to be tested, and use the material simulation model to perform dynamic processing on the initial configuration file to generate the first training dataset;

[0039] The initial configuration file refers to the chemical structural features of the energetic material to be tested, such as a molecular formula or unit cell structure containing atomic coordinates and lattice parameters. The specific format can be .cif or POSCAR. The initial configuration file is then subjected to dynamic processing using a materials simulation model to generate the first training dataset. This first training dataset includes a batch of initial structures obtained by running dynamics on the initial configuration file, along with corresponding first evaluation parameters. These first evaluation parameters include at least energy, force, and stress, which are the main parameters in the process of evaluating the explosive performance of the material.

[0040] S102, use the first training dataset to iteratively train the potential function to obtain the structure file corresponding to each generation of training;

[0041] A potential function refers to a neural network potential function used to calculate interatomic interactions, thereby simulating the dynamic behavior of materials under extreme conditions. Its core principle is to simulate interatomic interactions (potential energy) based on Machine Learning Interaction Potential (MLIP) technology, replacing or enhancing traditional empirical potential functions or quantum mechanical calculations (such as density functional theory, DFT). In one feasible approach, the NEP potential function can serve as a concrete application example. The potential function is iteratively trained using a material training model, with each training result being a batch of structure files, referred to as the current generation's structure files. These structure files, along with their corresponding first evaluation parameters, are then used as input for the next iteration, iterating until a preset stopping condition is met.

[0042] S103, the structural files are sampled and screened, and the precision adjustment parameters corresponding to the sampled and screened structural files are calculated using first-principles calculation software.

[0043] After training, a large number of refined structure files are obtained in batches. These files are then sampled and selected, and a number of structures are used in first-principles calculation software to calculate the precision adjustment parameters corresponding to the selected structure files. This mainly involves calculating a small subset of structures from the large number of refined structure files using first-principles calculation software to obtain parameters corresponding to the selected structures. These parameters, like the selected structures, include at least energy, force, and stress. Because the calculations are performed using first-principles calculations, their precision is higher than that of parameters obtained solely from the model (such as the first evaluation parameters).

[0044] The first-principles calculation software can be software that performs calculations based on first principles. In this embodiment, VASP (Vienna Ab-initio Simulation Package) quantum mechanics simulation software is used as an example. It is mainly used for atomic-scale simulations in the fields of materials science, chemistry, and physics. By solving density functional theory equations, it calculates the electronic structure, energy, force, stress, and other physical properties of materials with extremely high accuracy and is regarded as the "gold standard" of computational materials science.

[0045] S104 adjusts the parameters of the material simulation model by adjusting the precision parameters, and uses the adjusted material simulation model to generate a second training dataset from the structure file obtained from each generation of training and the first training dataset.

[0046] In S103, the parameters corresponding to the structure obtained through first-principles calculations are used to adjust the parameters of the material simulation model, thereby improving the accuracy of the adjusted model. Then, the structural files obtained from each generation of training and the initial structures in the first training dataset are re-processed dynamically, i.e., recalibrated to generate the second training dataset. The second training dataset includes structural files and their corresponding second evaluation parameters. These second evaluation parameters also include at least energy, force, and stress, but due to the improved training accuracy of the model, the accuracy of the second evaluation parameters is better than that of parameters obtained solely through the model (such as the first evaluation parameters).

[0047] S105 predicts the performance of the energetic material under test based on the potential function trained on the second training dataset.

[0048] The potential function is then trained using a second training dataset, and its performance is higher than that trained using the first training dataset. This allows for the testing of the performance of the energetic material under test.

[0049] In this embodiment, the initial configuration file of the energetic material to be tested is first obtained, and the initial configuration file is dynamically processed using a material simulation model to generate a first training dataset. The potential function is iteratively trained using the first training dataset to obtain the structure file corresponding to each generation of training. The structure files are sampled and screened, and the accuracy adjustment parameters corresponding to the sampled and screened structure files are calculated using first-principles calculation software. The parameters of the material simulation model are adjusted using the accuracy adjustment parameters, and a second training dataset is generated using the adjusted material simulation model for the structure files obtained from each generation of training and the first training dataset. Based on the potential function trained on the second training dataset, the performance of the energetic material to be tested is predicted. Compared with related technologies, this method first obtains a large dataset of a certain quality based on the initial configuration file through a large model, which is needed to generate the training potential function. Then, it combines first-principles calculations to select some training iterative structure files for calculation, and obtains more accurate data to adjust the model's accuracy. After adjusting the model, the model performance is improved, and a second training dataset is generated from the obtained structure files. The second training dataset has better accuracy than the first training dataset, and is then used to train the potential function to predict the performance of the energetic material under test.

[0050] The above explanation focused on the specific steps; the following analysis will examine the logic of these steps. In S101 and S102 of this embodiment, a large amount of high-quality datasets (e.g., the first training set consisting of the initial structure and corresponding first evaluation parameters, used as the first generation training samples) are obtained through the model. The purpose is primarily to obtain the structure files of each iteration result. Processing through the model is more time-efficient and effective than other methods. Furthermore, a selection process is performed on the large number of structure files and corresponding evaluation parameters. Directly performing first-principles calculations on a large number of structure files would be extremely time-consuming and resource-intensive, which is why related technologies cannot effectively integrate first-principles methods with practical applications. By selecting a portion for first-principles calculations, highly accurate energy, force, and stress parameters are obtained. However, these parameters are not used for performance prediction but rather to further improve the model's performance; therefore, they are called accuracy adjustment parameters. After improving the model's training accuracy, the energy, force, and stress are recalculated on the structure files obtained from each previous generation of training, serving as the second training dataset for the potential function. It can be seen that the second training dataset has higher accuracy than the previous one. The potential function is then trained to analyze and predict the performance of the energetic material under test. This method significantly saves CPU computing resources and time. Furthermore, when recalculating energy, force, and stress on the structure files obtained from each previous training iteration, resources are saved and efficiency is improved because the structure files are from previous iterations. Therefore, the energetic material performance prediction method proposed in this embodiment can simultaneously achieve both accuracy and efficiency, greatly accelerating the training of the machine learning potential function. Combined with relevant experimental data, it shows a speedup of nearly four orders of magnitude compared to the first-principles method, making high-speed prediction with first-principles accuracy possible.

[0051] It is important to clarify that although the first evaluation parameter, the accuracy adjustment parameter, and the second evaluation parameter are essentially energy, force, and stress parameters, the difference between the first evaluation parameter and the second evaluation parameter lies in the fact that the parameters of the model are adjusted during the calculation, so they are not the same; while the accuracy adjustment parameter is calculated based on first principles, and its purpose is to adjust the accuracy of the model, so it also needs to be distinguished.

[0052] Optionally, the potential function is iteratively trained using the first training dataset to obtain a structure file corresponding to each generation of training, including: creating a potential function training task and labeling the training results of each generation according to the iteration number; running molecular dynamics simulation training using the potential function in a discontinuous Hugoniot type ensemble; plotting the trend of thermodynamic quantities changing with the number of kinetic steps; the trend plot is used to determine the stability of the dynamics while determining the number of kinetic steps; detecting the number of kinetic steps and the number of iterations, and stopping the iterative training when the stopping condition is met.

[0053] Among them, the non-continuous Hugoniot type ensembles include: NPHugoniotniot or NVHugoniotniot ensembles, and the initial parameters and initial structure of the ensembles are set non-uniformly.

[0054] In this embodiment, the ensemble is a core concept in statistical mechanics and molecular dynamics, referring to a group of systems with the same macroscopic conditions (such as temperature, pressure, volume, etc.) but different microscopic states, playing a crucial role in predicting detonation properties. In the simulation, the ensemble determines how the system evolves over time and affects the macroscopic properties of the material (such as detonation parameters).

[0055] Another creative aspect of this embodiment is the proposal to employ discontinuous Hugoniot-type ensembles, including NPHugoniotniot or NVHugoniotniot ensembles, for the first time. Firstly, NPHugoniotniot or NVHugoniotniot is more efficient than other ensembles in obtaining points on the Hugoniot curve. For example, the MSST method may fail to converge to the Hugoniot curve when the initial detonation velocity is lower than the actual material detonation velocity Dv. Furthermore, the ReaxCJ method, which uses NVT for extrapolation, requires at least three data points for each Hugoniot point to fit the curve.

[0056] Second, NPHugoniotniot or NVHugoniotniot are more versatile. MSST ensembles require the initial structural file to be orthogonal, while NPHugoniot ensembles can be applied to structural files of any shape and are suitable for a wider range of energetic materials.

[0057] Furthermore, this embodiment proposes that the initial parameter settings and initial structure of the ensemble are non-uniform, since CJ theory indicates that the final state after impact is the most stable state. However, MD methods such as MSST and NPHugoniot cannot realistically simulate the detonation reaction path. Therefore, it proposes ignoring the reaction path and only requiring the final state to be the most stable state. Improvements are made to the initial parameters and initial structure, including: the initial parameters are obtained through kinetic calculations after the experimental structure or molecules are stacked into a crystal structure; the initial velocity is calculated from atomic mass and artificially specified density. The initial structure only needs to be consistent with the target structure in terms of the number of atoms and element types. Previous researchers using this method for detonation performance prediction used initial structures and initial parameters (initial energy, initial volume, initial pressure) automatically obtained by a program without changing the initial parameters. This embodiment is the first to use a non-uniform initial structure and initial parameters in detonation performance prediction, thus enabling the prediction of the detonation velocity and detonation pressure of energetic materials based on their molecular composition. This method greatly accelerates the speed of predicting detonation performance using dynamics, making it possible to predict detonation performance using first-principles methods, and expanding the application scope of molecular dynamics in predicting detonation performance, thus serving as a general-purpose method.

[0058] This study employs an innovative method for constructing the initial kinetic structure in detonation velocity simulations. This method uses a gas molecule packing structure as the initial configuration while strictly maintaining the types and numbers of atoms in the system consistent with the target product. This structural design strategy effectively reduces the potential barrier height that needs to be overcome during kinetic simulations, thereby significantly improving the efficiency of the system's evolution to the final stable state. From a theoretical and methodological perspective, this technique represents a significant breakthrough in the principle of detonation velocity kinetic prediction. This choice of initial structure provides a new computational paradigm for studying the equation of state of matter under extreme conditions. The establishment of this method provides a new theoretical foundation and technical approach for multi-scale simulation of the detonation performance of high-energy materials.

[0059] Furthermore, combined Figure 2 This diagram illustrates the logical flow of a method for obtaining detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics. Using a synthesized energetic material as the test system and combining it with a real-world scenario, the steps for training the potential function in the above embodiments are explained. The crystal structure POSCAR file and the in.yaml configuration parameter file are placed in the main working directory folder. The machine learning potential function large-scale model MatterSim and the currently most efficient machine learning potential function NEP are used for efficient and high-precision detonation velocity prediction. First:

[0060] Initial dataset construction: The main purpose is to run dynamics on the initial configuration files of energetic materials to obtain the first training dataset. An `init` directory is created in the main working directory, and the `POSCAR` file from the main working directory is copied to the `init` directory. In this directory, the NPHugoniot ensemble dynamics task, which creates dynamics at different pressures based on the `POSCAR`, will be run, using MatterSim as the potential function, for 40,000 steps. After completion, the module will automatically extract the structures from all task folders (at this point, the initial structures in the first training dataset and the corresponding first evaluation parameters are obtained), and the number of frames specified by the `ini_frames` parameter will be extracted as the initial structures for active learning.

[0061] Training employs an active learning approach, including tasks for training the potential function, model bias, and structure labeling. Each task comprises three sub-steps: creation, execution, and post-processing. One iteration constitutes one generation, specifically including:

[0062] 1. Establish the training task for the NEP potential function. This step will create a folder corresponding to the algebra in the working directory, such as iter.000000 representing the first generation task. Within this algebra folder, a directory named 00.NEP will be created as the training folder for the NEP potential function. Within this folder, a dataset folder will be created, merging all structure files tagged with MatterSim into this folder. Structure files of a certain proportion will be set as the training set, and the remaining structures as the test set. After this part is completed, 0 (the algebra) will be recorded in the record.NEP file in the working directory.

[0063] 2. Run the NEP training task. The created NEP training task is run using GPU resources. The NEP program updates the parameters of the NEP potential function using the backpropagation algorithm on the training set until the specified number of steps is reached. After the task completes, a file named `task_finished` will be generated in the corresponding task folder. If this step is executed again and the `task_finished` file is detected in that folder, it will be skipped, and the next task will be executed. After this part is completed, 1 (algebraic number) will be recorded in the `record.NEP` file in the working directory.

[0064] 3. Run the NEP training post-processing task. After this part is completed, the number 2 (algebra) will be recorded in the record.NEP file in the working directory.

[0065] 4. Create the gpumd task. This step will generate a 01.gpumd folder in the folder corresponding to the algebra. The task folder is generated according to the settings in the model_devi_general general parameter. It is recommended to set the ensemble to NPHugoniot. This ensemble requires additional settings for pressure (equilibrium pressure) and pperiod (pressure bath parameter). An example code is as follows:

[0066] Ensembles:[nphugo] # Active learning for sampling dynamic systems; "NPHugoniot" is recommended for burst velocity simulations.

[0067] Structure_id: [2] # The number of the structure file selected for dynamics

[0068] Pressure[20,25,30,35,40,45,50,60] # List of simulated pressures, in GPa

[0069] Pperiod: 2000 # Simulation pressure bath parameter settings, in steps

[0070] For example, the configuration parameters for other parts are as follows:

[0071] ini_run_steps: 100000 # Number of run steps for generation 0 dynamics

[0072] max candidate: 1000 # The number of candidate structures needed in each generation during active learning.

[0073] model_devi time step: 0.2 #Time step size for each generation during active learning

[0074] gpu_available: [0,1,2,3] # Available GPU numbers

[0075] task per gpu: 1 # Number of tasks running per GPU

[0076] time step: 0.2 #Time step of dynamics

[0077] max run steps: 1000000 # The program will automatically stop when the number of steps exceeds this maximum running step count and the number of generations is greater than or equal to 20.

[0078] rho: 1.62 # Density, unit is g / cm³ 3

[0079] The rho parameter will be used to calculate the initial volume during the NPHugoniot run. After this part finishes running, the number 3 (algebraic) will be recorded in the record.NEP file in the working directory.

[0080] In this embodiment, the discontinuous Hugoniot is used: it can be directly based on molecular configuration without periodic boundary conditions, and is applicable to any energetic material (such as molecular crystals, polymer-bonded explosives, etc.), thus solving the problem of unpredictability of some new materials. In addition, it is close to the detonation physics process and has higher computational efficiency.

[0081] 5. Run the gpumd task. Collect all tasks in 01.gpumd and group them into multiple task queues based on available GPUs. Each task will continuously update the atomic positions based on the input files model.xyz, run.in, and nep.txt in its corresponding folder until the required number of steps is reached. After completion, a task_finished file will be generated in the folder. If a task_finished file is detected before running, that folder will be skipped. The rho parameter will be used to calculate the initial volume during the NPHugoniot run. After this part is completed, 4 (algebra) will be recorded in the record.NEP file in the working directory.

[0082] 6. Perform post-processing for the gpumd task. The program will plot the thermodynamic quantities as a function of the number of kinetic steps in each task folder, based on the thermo.out file, such as... Figure 3 The diagram illustrates how thermodynamic quantities such as temperature, potential-to-kinetic energy ratio, pressure, and volume change with the number of kinetic steps (with time as a reference). Users can use this diagram to determine if the kinetics are stable. During post-processing, if multiple generations of thermodynamic data are unstable, the initial dataset construction should be checked. The program checks the trajectory files in each folder. When the shortest atomic distance in a trajectory is less than 0.5, or the highest temperature is higher than 10000K, if the number of frames at this point is less than 70% of the total number of frames, the first two frames of that trajectory will be used for MatterSim kinetics. Frames will then be extracted from this sequence as candidate structures. After this part finishes running, 5 (generation number) will be recorded in the record.NEP file in the working directory.

[0083] 7. Create a MatterSim labeling task. Label the candidate structures collected in steps 4, 5, and 6 of the previous section using MatterSim to determine their energy, force, and stress levels for use in the next round of active learning. After this part is completed, the number 6 (algebra) will be recorded in the record.NEP file in the working directory.

[0084] 8. Run MatterSim's tagging task. After this part is completed, 7 (algebraic number) will be recorded in the record.NEP file in the working directory.

[0085] 9. Perform post-processing of MatterSim tasks and run the speed-up task. Here, the number of steps in the gpumd task in the second major step is checked. When the number of steps in steps 4, 5, and 6 of the gpumd task exceeds 400,000, it is considered to be stable, and a speed-up task test will be performed. The structure of the speed-up task is extracted by default from the last frame of the last task folder of the current generation of gpumd tasks as the initial structure. When stable: original is set to true, the POSCAR or .cif file in the working directory will also be used as the initial structure, creating a new speed-up task in the 03.shock folder. The number 8 (generation) will be recorded in the record.NEP file in the working directory.

[0086] The detonation velocity test creates NPHugoniot dynamics task folders with different pressures based on a specified structure file. After the run is complete, the thermodynamic average of the last 80% of the data for each trajectory is calculated. The average pressure and average temperature in each task folder are considered as a point on the PV plane. A quadratic function is fitted to these points, and a tangent line is drawn through the initial data points. The slope of the tangent line is converted using a fixed relationship to obtain the detonation velocity. After one generation ends, the next generation begins. By analyzing the detonation velocity test results obtained in step 9, the program will automatically terminate when the detonation velocity change is within 1%.

[0087] In combination Figure 2 In addition to explaining the steps described above in this embodiment, conditions for ending training can be set according to requirements. Users can view the speedup test results in the 03.shock folder of each generation of tasks to observe convergence. When the user believes the current speedup has reached the required level, the task can be terminated using `kill -TERM -(pid)`. Running this program will generate a large number of batch-refined structure files. Approximately one hundred frames can be extracted from each initial structure for stress, force, and energy calculations using first-principles software such as VASP, for fine-tuning in MatterSim. After fine-tuning in MatterSim, it will be used to label all structures. This saves significant CPU computing resources and time, and the entire process can be completed within one day.

[0088] Optionally, the material simulation model includes the MatterSim model based on deep neural networks, whose inputs include molecular formulas or unit cell structures containing atomic coordinates and lattice parameters, and whose outputs are energy, force, and stress with quantum mechanical precision.

[0089] In this embodiment, the MatterSim model is a large-scale atomic-scale simulation model based on machine learning and deep learning. It is primarily used for efficiently predicting the physicochemical properties of materials (such as energy, force, and stress), and serves as the core tool for data generation and distillation in this embodiment. Data is generated using the large model and then transferred to a smaller model (NEP) for efficient computation, significantly improving the efficiency and accuracy of detonation performance prediction. The main steps of this invention involve extracting the structure through active learning combined with a large model, and then obtaining the corresponding energy, force, and stress parameters through first-principles calculations. The large model is then fine-tuned and subjected to model distillation, ultimately leading to detonation property simulation testing. Compared to related technologies, this embodiment is the first to predict detonation performance using large-scale model distillation technology. This method significantly accelerates the training of the machine learning potential function, speeding up the process by nearly four orders of magnitude compared to first-principles methods, making first-principles accuracy in detonation velocity prediction possible. Furthermore, discontinuous Hugoniot-type ensembles are employed, including NPHugoniotniot and NVHugoniotniot ensembles. Previous researchers using this method to predict detonation performance used initial structures and parameters (initial energy, initial volume, initial pressure) obtained automatically by a program, without altering the initial parameters; that is, the structure and parameters were uniformly matched. This embodiment, for the first time, uses non-uniform initial structures and parameters in detonation performance prediction. This method significantly accelerates the speed of dynamic prediction of detonation performance, making first-principles methods for predicting detonation performance possible.

[0090] Furthermore, as Figures 1 to 3 The specific implementation of the method shown in this embodiment provides a device for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, such as... Figure 4 As shown, the device includes: an acquisition unit 401, a training unit 402, a sampling unit 403, an adjustment unit 404, and a prediction unit 405.

[0091] The acquisition unit 401 is configured to acquire the initial configuration file of the energetic material to be tested, and to perform dynamic processing on the initial configuration file using a material simulation model to generate a first training dataset.

[0092] Training unit 402 is configured to iteratively train the potential function using the first training dataset to obtain the structure file corresponding to each generation of training.

[0093] Sampling unit 403 is configured to perform sampling and screening of the structure file, and calculate the precision adjustment parameters corresponding to the structure file after sampling and screening using first-principles calculation software.

[0094] The adjustment unit 404 is configured to adjust the parameters of the material simulation model through the precision adjustment parameters, and use the adjusted material simulation model to generate a second training dataset from the structure file obtained from each generation of training and the first training dataset.

[0095] The prediction unit 405 is configured to predict the performance of the energetic material under test based on the potential function trained on the second training dataset.

[0096] In specific application scenarios, training unit 402 is further configured to create a potential function training task and label the training results of each generation according to the iteration number; run molecular dynamics simulation training using the potential function in a discontinuous Hugoniot type ensemble; plot a trend graph of thermodynamic quantities changing with the number of kinetic steps; the trend graph is used to determine the stability of the kinetics while determining the number of kinetic steps; detect the number of kinetic steps and the number of iterations, and stop iterative training when the stopping condition is met.

[0097] In a specific application scenario, the adjustment unit 404 is further configured to use the material simulation model with adjusted parameters to generate the second training dataset from the structure file obtained from each generation of training and the initial structure in the first training dataset.

[0098] It should be noted that other corresponding descriptions of the functional units involved in the device for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics provided in this embodiment can be found in [reference needed]. Figures 1 to 3 The corresponding descriptions in [the document] will not be repeated here.

[0099] Based on the above, Figures 1 to 3 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figures 1 to 3 The method shown.

[0100] Based on the above, Figures 1 to 3 Accordingly, this embodiment also provides a computer program product on which a computer program is stored, which, when executed by a processor, implements the above-described method. Figures 1 to 3 The method shown.

[0101] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0102] Based on the above, Figures 1 to 3 The method shown, and Figure 4 To achieve the above objectives, this application also provides an electronic device, which can be configured on a computer side, etc. This device includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to achieve the above-described virtual device embodiments. Figures 1 to 3 The method shown.

[0103] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0104] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0105] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms, or it can be implemented by hardware. By applying the solution of this embodiment, compared with related technologies, firstly, based on the initial configuration file, a large model is used to obtain a large dataset of a certain quality required for generating the training potential function (for example, the first training set composed of the initial structure and the corresponding first evaluation parameters is used as the first generation training sample). The purpose is more to obtain the structure file of each iteration result. Model processing is more time-saving and efficient than other methods. Then, the first principle is combined to screen some of the trained iterative structure files for calculation to obtain more accurate data for adjusting the accuracy of the model. After adjusting the model, the model performance is improved, and a second training dataset is generated again from the obtained structure files. The second training dataset has better accuracy than the first training dataset, and is then used to train the potential function to predict the performance of the energetic material under test.

[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0108] Similar parts between the embodiments provided in this application can be referred to mutually. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. A method for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, characterized in that, include: Obtain the initial configuration file of the energetic material to be tested, and use the material simulation model to perform dynamic processing on the initial configuration file to generate the first training dataset; The potential function is iteratively trained using the first training dataset to obtain the structure file corresponding to each generation of training. The structural files are sampled and screened, and the precision adjustment parameters corresponding to the sampled and screened structural files are calculated using first-principles calculation software. The material simulation model is adjusted by the precision adjustment parameters, and a second training dataset is generated from the structure file obtained from each generation of training and the first training dataset using the adjusted material simulation model. The performance of the energetic material under test is predicted based on the potential function trained on the second training dataset.

2. The method according to claim 1, characterized in that, The step of iteratively training the potential function using the first training dataset to obtain the structure file corresponding to each generation of training includes: Create a potential function training task and label the training results of each generation according to the number of iterations; Molecular dynamics simulation training was conducted using the aforementioned potential function in a discontinuous Hugoniot-type ensemble. Plot a trend graph of thermodynamic quantities as a function of the number of kinetic steps; the trend graph is used to determine the stability of the kinetics given the number of kinetic steps. The number of dynamic steps and iterations are checked, and the iterative training is stopped when the stopping condition is met.

3. The method according to claim 2, characterized in that, The discontinuous Hugoniot type ensembles include: NPHugoniotniot or NVHugoniotniot ensembles, and the initial parameters and initial structure of the ensembles are set non-uniformly.

4. The method according to claim 1, characterized in that, The material simulation model includes the MatterSim model based on deep neural networks, whose inputs include molecular formulas or unit cell structures containing atomic coordinates and lattice parameters, and whose outputs are energy, force, and stress with quantum mechanical precision.

5. The method according to claim 1, characterized in that, The initial configuration file includes the chemical structural features of the energetic material to be tested, and the first training dataset includes the initial structure and the first evaluation parameters corresponding to the initial structure.

6. The method according to claim 5, characterized in that, The step of adjusting the parameters of the material simulation model using the precision adjustment parameters, and then using the adjusted material simulation model to generate a second training dataset from the structure file obtained from each generation of training and the first training dataset, includes: Using the material simulation model with adjusted parameters, a second training dataset is generated from the structure files obtained from each generation of training and the initial structure in the first training dataset.

7. A device for obtaining the detonation velocity and detonation pressure of energetic materials based on atomic-scale dynamics, characterized in that, include: The acquisition unit is configured to acquire the initial configuration file of the energetic material to be tested, and to perform dynamic processing on the initial configuration file using a material simulation model to generate a first training dataset; The training unit is configured to iteratively train the potential function using the first training dataset to obtain the structure file corresponding to each generation of training. The sampling unit is configured to sample and screen the structure files, and calculate the precision adjustment parameters corresponding to the sampled and screened structure files using first-principles calculation software. The adjustment unit is configured to adjust the parameters of the material simulation model using the precision adjustment parameters, and to generate a second training dataset using the adjusted material simulation model, the structure file obtained from each generation of training, and the first training dataset. The prediction unit is configured to predict the performance of the energetic material under test based on the potential function trained on the second training dataset.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

10. A computer program product having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images