A method, system and device for predicting hydrogen embrittlement performance of a Ti-Al-V alloy

Abstract

The application provides a Ti-Al-V alloy hydrogen embrittlement performance prediction method, system and device, and belongs to the cross field of material calculation and artificial intelligence. The method comprises the following steps: covering initial data sets of key configurations in the whole hydrogen embrittlement process through a system, and adopting an active learning iteration framework to perform potential function training and optimization, so as to finally obtain a special potential function with the accuracy of first principles and the efficiency of molecular dynamics, which is used for large-scale and long-time hydrogen embrittlement process simulation. The method solves the limitation problems of existing atomic simulation methods in accuracy, efficiency and system applicability.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of materials computation and artificial intelligence, and specifically relates to a method, system and device for predicting the hydrogen embrittlement properties of Ti-Al-V alloys. Background Technology

[0002] Titanium alloys are widely used in aerospace, biomedical, and other fields due to their high strength, low density, and good corrosion resistance. However, titanium alloys are prone to hydrogen embrittlement in hydrogen-containing environments, leading to sudden brittle fracture under stresses far below their yield strength, which seriously threatens the structural safety. Therefore, understanding the diffusion and aggregation behavior of hydrogen in titanium alloys and its interaction with microstructures (such as grain boundaries, dislocations, and second phases) is crucial for predicting and mitigating hydrogen embrittlement.

[0003] Currently, atomic-scale simulations of the Ti-Al-VH system mainly rely on two types of potential functions: one is empirical potential functions (such as EAM and MEAM), which are computationally efficient but have limited accuracy and cannot accurately describe complex multi-element systems and the chemical interactions between hydrogen and defects. The other type is first-principles calculations (such as methods based on density functional theory), which provide high-precision data but are extremely expensive to compute, limiting them to simulations at the atomic scale and picosecond timescale, making it difficult to directly study the long-range diffusion behavior of hydrogen on macroscopic timescales and its impact on mechanical properties.

[0004] In summary, existing technologies suffer from the following technical problems: empirical potential functions lack sufficient accuracy in describing key interactions in the Ti-Al-VH system (such as hydrogen-metal bonding, hydrogen segregation energy at defects, and hydride phase transformation behavior); while first-principles calculations, although highly accurate, are limited by spatiotemporal scales and cannot be used for simulating hydrogen embrittlement processes on a practical engineering scale. Therefore, the lack of a suitable potential function for atomic-scale simulation of the Ti-Al-VH system in engineering applications makes it difficult to predict the hydrogen-induced failure process of hydrogen embrittlement in this alloy system, thus hindering the process optimization of high-resistance hydrogen-embrittlement titanium alloys. Summary of the Invention

[0005] To address the limitations of existing atomic simulation methods in terms of accuracy, efficiency, and system applicability in describing the hydrogen embrittlement behavior of titanium alloys, this invention provides a method, system, and apparatus for predicting the hydrogen embrittlement properties of Ti-Al-V alloys.

[0006] A first aspect of this invention provides a method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys, comprising the following steps: Construct an initial dataset, which contains multiple preset atomic configurations of four elements: Ti, Al, V, and H, and their corresponding first-principles calculation data; The potential function model is initially trained using the initial dataset to obtain the initial potential function model; Molecular dynamics simulations are performed using the initial potential function model to generate molecular dynamic trajectories containing new atomic configurations; the prediction uncertainty of each atomic configuration in the molecular dynamic trajectory is calculated, and atomic configurations with prediction uncertainties higher than a preset threshold are selected as new atomic configurations. First-principles calculations are performed on the selected new atomic configurations to obtain their corresponding energy and stress data; the new atomic configurations and their corresponding energy and stress data are added to the initial dataset to form an expanded dataset; The potential function model is retrained and updated using the augmented dataset; Repeat the above training steps until the prediction error of the updated potential function model converges and the prediction uncertainty is lower than the set threshold, thus obtaining the final machine learning potential function. The atomic configuration of the Ti-Al-VH system to be tested is input into the final machine learning potential function for molecular dynamics simulation, which predicts the index that determines the hydrogen embrittlement sensitivity.

[0007] Furthermore, the atomic configurations in the initial dataset include at least one of the following: perfect unit cells of α-Ti and β-Ti; Ti-Al-V ternary solid solution supercells; hydrogen-containing Ti-Al-V ternary solid solution models; titanium hydride crystal models; grain boundary models of HCP or BCC systems; wherein the titanium hydride crystal models include γ-TiH and δ-TiH. 1.5 Three hydrides: ε-TiH2 and ε-TiH2.

[0008] Furthermore, in the hydrogen-containing Ti-Al-VH solid solution model, hydrogen atoms occupy tetrahedral or octahedral interstitial sites, and the hydrogen concentration is set to increase incrementally.

[0009] Furthermore, the initial training includes configuring the network structure parameters of the potential function model and training the model using the initial dataset.

[0010] Furthermore, the molecular dynamics simulation using the initial potential function model includes at least one of the following simulation scenarios: NVT or NPT simulations were performed on solid solutions or hydrides in the temperature range of 300K to 1500K. High-temperature annealing simulation was performed on systems containing grain boundaries, phase boundaries, or dislocations; Simulate shear or tensile deformation by applying it to a model containing cracks.

[0011] Furthermore, the method includes a step of performing multi-scale verification on the potential function model after the prediction error of the updated potential function model converges and the prediction uncertainty is below a set threshold. The multi-scale verification includes at least one of the following: calculating the energy equation of state curve and comparing it with first-principles or experimental data; calculating the diffusion barrier of hydrogen in α-Ti or β-Ti; calculating the elastic constant and comparing it with the first-principles results; simulating the interaction between hydrogen and moving dislocations; and simulating the initiation and propagation of hydrogen-induced cracks.

[0012] Furthermore, the step of loading the target atomic model of the Ti-Al-VH system under test into the final machine learning potential function for prediction to obtain the hydrogen embrittlement performance prediction result includes the following steps: Construct a target atomic model for the Ti-Al-VH system whose hydrogen embrittlement properties are to be predicted; The target atom model and the final machine learning potential function are loaded into the molecular dynamics simulation program; Run molecular dynamics simulations and, based on the simulation outputs, calculate one or more indices for quantitatively assessing hydrogen embrittlement sensitivity, including at least one of the following: hydrogen diffusion activation energy in α-Ti or β-Ti phases, hydrogen segregation energy at grain boundaries or phase boundaries, and fracture energy at hydrogen-containing grain boundaries or phase boundaries.

[0013] A second aspect of this invention provides a system for predicting the hydrogen embrittlement properties of Ti-Al-V alloys, comprising: The construction module is used to construct an initial dataset, which contains multiple preset atomic configurations of four elements: Ti, Al, V, and H, and their corresponding first-principles calculation data. The initialization module uses the initial dataset to perform preliminary training on the potential function model to obtain the initial potential function model. The screening module is used to perform molecular dynamics simulation using the initial potential function model to generate molecular dynamics trajectories containing new atomic configurations; calculate the prediction uncertainty of each atomic configuration in the molecular dynamics trajectory, and screen out atomic configurations with prediction uncertainty higher than a preset threshold as new atomic configurations. An expansion module is used to perform first-principles calculations on the selected new atomic configurations to obtain their corresponding energy and stress data; and to add the new atomic configurations and their corresponding energy and stress data to the initial dataset to form an expanded dataset; The update module is used to retrain the potential function model using the expanded dataset and update the potential function model. The iterative module is used to repeatedly execute the above training steps until the prediction error of the updated potential function model converges and the prediction uncertainty is lower than a set threshold, thus obtaining the final machine learning potential function. The prediction module is used to load the target atomic model of the Ti-Al-VH system under test into the final machine learning potential function for prediction, and obtain the prediction results of hydrogen embrittlement performance.

[0014] A third aspect of the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0015] A fourth aspect of the present invention provides a readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0016] The method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys provided by this invention has the following beneficial effects: By constructing a dedicated initial dataset covering four elements—Ti, Al, V, and H—the model's learning foundation is ensured to have a high degree of system specificity. This fundamentally overcomes the inaccuracy of general potential functions in describing the chemical bonding of complex multi-element systems, enabling the final potential function to achieve near-first-principles accuracy in predicting key parameters such as hydrogen dissolution energy and hydrogen segregation energy at phase / grain boundaries. Secondly, the constructed active learning iterative loop continuously explores new configurations through molecular dynamics simulations using the current model, intelligently selecting areas with high model prediction uncertainty for first-principles calculation calibration, followed by dataset expansion and retraining. This allows the potential function to actively learn non-equilibrium states and extreme configurations that may occur during hydrogen embrittlement and are difficult to enumerate in advance (such as high stress concentration regions at crack tips and dynamic hydrogen atom aggregation processes). This significantly enhances the model's numerical stability and extrapolation prediction capabilities when simulating complex dynamic failure processes, avoiding distortion or failure caused by configurations exceeding the training range in traditional methods when simulating such processes. Ultimately, by using the potential function obtained through closed-loop iterative training to convergence, the precision of first-principles calculations and the efficiency of molecular dynamics were successfully integrated. This enabled it to be directly used for large-scale, long-term atomic simulations, systematically studying the long-range diffusion of hydrogen, the interaction between hydrogen and moving dislocations, and the entire process of hydrogen-induced crack initiation and propagation. This provides a previously scarce computational tool with both high confidence and high practicality for revealing the hydrogen embrittlement mechanism of titanium alloys at the atomic scale, predicting material failure behavior, and guiding the design of hydrogen-resistant alloys. Attached Figure Description

[0017] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys according to an exemplary embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0020] Currently available potential function libraries lack a high-precision potential function that can accurately describe the four elements Ti, Al, V, and H and their complex interactions (especially the behavior of hydrogen at key defects such as α / β phase interfaces and grain boundaries). This limits research into the microscopic mechanisms of hydrogen embrittlement: due to the lack of reliable and efficient atomic simulation tools, it is difficult to systematically study the long-range diffusion kinetics of hydrogen in complex polycrystalline Ti-Al-V alloys, the dynamic interactions between hydrogen and propagating defects (dislocations, cracks), and the atomic mechanisms of hydrogen-induced microcrack initiation and propagation.

[0021] To overcome the aforementioned technical problems, this invention develops a high-precision machine learning potential function specifically for the Ti-Al-VH system. This machine learning potential function achieves near-first-principles calculation accuracy and computational efficiency comparable to empirical potential functions. It provides a computational tool for in-depth exploration of the microscopic mechanism of hydrogen embrittlement in titanium alloys at the atomic level, prediction of hydrogen-induced failure behavior of materials, and guidance for the design of hydrogen-resistant titanium alloys, filling the gap in high-precision and efficient simulation technology in this field.

[0022] Specifically, the present invention aims to solve the following specific technical problems: (1) Solve the contradiction between precision and efficiency: Provide a computational tool that can maintain the precision of first-principles calculations while having the computational efficiency of classical molecular dynamics when simulating the hydrogen-related behavior of the Ti-Al-VH system, thereby bridging the time and space gap from the electronic scale to the mesoscopic scale.

[0023] (2) Fill the gap in high-precision potential function for titanium alloy quaternary system: Develop a machine learning potential function specifically optimized for Ti-Al-VH four-component system to ensure that it can accurately describe all possible interatomic interactions in the system, especially the effect of the addition of vanadium (V) on phase stability, hydrogen dissolution and diffusion behavior, and hydrogen segregation energy at complex microstructures such as α / β phase interface.

[0024] (3) Overcome the shortcomings of insufficient reliability in simulating key hydrogen embrittlement configurations: ensure that the developed potential function fully covers the key and non-equilibrium atomic configurations involved in the hydrogen embrittlement process (such as the state of various grain boundaries, phase boundaries, dislocation cores, and crack tips under high local hydrogen concentration) in its training data, so as to ensure that the prediction results of the potential function remain stable and reliable when simulating extreme events such as hydrogen-induced crack initiation and propagation.

[0025] (4) To achieve systematic simulation and verification of hydrogen embrittlement mechanism: Provide a potential function that has been verified not only by traditional physical property testing, but also by simulation of specific hydrogen embrittlement processes, so that it can be directly used to systematically study the long-range diffusion dynamics of hydrogen in Ti-Al-V alloy, the interaction between hydrogen and moving dislocations, the atomic mechanism of hydrogen-induced grain boundary embrittlement and microcrack propagation, and provide a quantitative and microscopic tool basis for understanding and predicting the hydrogen embrittlement behavior of this system. The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] First, this invention provides a method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys, specifically as follows: Figure 1 As shown, it includes the following steps: Step S1: Systematically construct the initial dataset.

[0027] An initial dataset for constructing the Ti-Al-VH quaternary system is provided, which includes various preset atomic configurations and their corresponding first-principles calculation data.

[0028] (1) α / β-Ti matrix: Perfect unit cells of α-Ti with hexagonal close-packed (HCP) structure and β-Ti with body-centered cubic (BCC) structure were constructed respectively. Full-degree-of-freedom structure optimization was performed using first-principles parameters (such as PAW-PBE functional, 520eV cutoff energy) that have undergone rigorous convergence testing to ensure that the error between the calculated lattice constants (such as α-Ti: a≈2.95Å, c≈4.68Å) and the experimental values ​​is controlled within ±1%, thereby verifying the reliability of the calculation method.

[0029] (2) Alloying model: In the α-Ti and β-Ti supercells, based on the nominal composition of engineering alloys such as TA17, Al and V atoms with different proportions and random distributions were introduced to replace atoms, and a Ti-Al-V ternary solid solution model was constructed. Through structural optimization, the effects of Al and V added individually and together on the lattice constant, volume and stability of the two Ti matrices were quantitatively analyzed.

[0030] (3) Hydrogen solid solution model: In the optimized α / β-Ti and Ti-Al-V alloy supercells, a single H model in which hydrogen atoms occupy tetrahedral / octahedral interstitials was systematically constructed; and a solid solution model with increasing hydrogen concentration (e.g., Ti:H atomic ratio from 2:1, 1:1 to 1:2) was constructed. All models were fully relaxed, and the variation of lattice parameters and supercell volume with hydrogen content was accurately statistically analyzed.

[0031] (4) Sampling of hydride systems: Constructing stoichiometric γ-TiH (FCT) and δ-TiH 1.5 Crystal models of titanium hydrides such as FCC and ϵ-TiH2 (BCT) were developed. The structure of each hydride was optimized, and its equation of state (EOS) was calculated by applying a series of volumetric strains. The equilibrium volume, bulk modulus, and other parameters were obtained by fitting the Birch-Murnaghan equation, and the relative stability and mechanical properties of hydrides with different hydrogen contents were evaluated.

[0032] (5) Sampling of grain boundary models. Construct grain boundary models for the HCP system and the BCC system. For example, the {10-11}<10-12> and {10-12}<10-11> twin boundary models for the HCP structure; and the {110}<1-11>, {112}<11-1>, and {332}<11-3> grain boundary models for the BCC grain boundary.

[0033] (6) Dataset integration: Summarize all the above optimized configurations and their corresponding first-principles calculation results (total system energy, atomic forces, stress tensor), process them according to the standardized format (such as the format required by DeepMD-kit), and generate the initial training data file.

[0034] Step S2: Iterative training and optimization of the potential function based on active learning.

[0035] Based on this initial dataset, a machine learning potential function model is trained using an active learning iterative framework, which includes the following sub-steps: S2.1. Train the potential function model using the current dataset; S2.2. Perform molecular dynamics simulations using the current potential function model to generate new atomic configurations; S2.3. Select configurations from the new atomic configurations whose model prediction uncertainty is higher than a set threshold; S2.4. Perform first-principles calculations on the selected configurations and add the results to the dataset; S2.5. Repeat S2.1 to S2.4 until the prediction error of the potential function model converges and the uncertainty is below the set threshold.

[0036] Specifically, a deep learning potential function generator (DP-GEN) framework is adopted to implement an active learning iteration of "exploration-selection-learning" in order to efficiently expand the configuration space and improve the reliability of the potential function.

[0037] (1) Initial model training: Starting with the above initial dataset, configure the network parameters of the potential function model (e.g., 4 hidden layers, 128 neurons per layer) and perform initial training.

[0038] (2) Configuration space exploration: Using the current potential function, drive molecular dynamics simulations covering a wide range of phase spaces to explore new configurations. This includes NVT / NPT simulations of solid solutions and hydrides in the temperature range of 300K to 1500K, sampling vibrational and phase transition information at finite temperatures.

[0039] High-temperature annealing simulations were performed on systems containing grain boundaries, phase boundaries, and dislocations to sample the evolution and recombination of defect structures.

[0040] Apply shear or tensile deformation to the interface or cracked model and sample the atomic configuration under non-equilibrium, high-stress conditions.

[0041] (3) Configuration screening and data supplementation: By automatically analyzing the molecular dynamics trajectories generated by exploratory simulations, and based on the uncertainty quantification index of the machine learning potential function prediction results (e.g., the prediction variance of the force on each atom), novel atomic configurations with low confidence and "insufficient representativeness" in the current potential function model are systematically identified and extracted.

[0042] (4) First-principles calculation and iterative update: The selected new configurations are submitted for first-principles calculation to obtain accurate "energy-force-stress" data, which is then added to the training set. Subsequently, the potential function model is retrained using the expanded dataset to update the network parameters.

[0043] (5) Iterative convergence: Repeat steps 2 to 4 until the model’s prediction error on the independent validation set (e.g., root mean square error of atomic force < 0.05 eV / Å, root mean square error of energy < 5 meV / atom) converges and the uncertainty in the newly explored hydrogen embrittlement-related phase space is below the set threshold.

[0044] Step S3: Rigorous verification and application at multiple scales.

[0045] The hydrogen embrittlement process of the Ti-Al-VH system was simulated using the trained machine learning potential function.

[0046] After training, the potential function is validated across scales to ensure its predictive power. Once validated, the potential function is used to conduct large-scale molecular dynamics simulations.

[0047] (1) Energy state equation (EOS verification): EOS curves for structures with different hydrogen contents and alloying elements were calculated and compared with first-principles calculations and experimental data.

[0048] (2) Kinetic properties: The diffusion barrier of hydrogen in α / β-Ti and alloys was calculated and is consistent with the DFT-NEB calculation results and experimental values ​​(e.g., ~0.54eV in α-Ti).

[0049] (3) Mechanical properties: Calculate the elastic constants of α / β-Ti, Ti-Al-V alloys and hydrides, compare them with the first-principles results, and control the relative error within a reasonable range (e.g., within 10%).

[0050] (4) Hydrogen-defect interaction: study the interaction between hydrogen and moving dislocations, and the effect of hydrogen on the cohesive strength of grain boundaries / phase boundaries.

[0051] (5) Hydrogen-induced failure: Simulate the entire process of hydrogen promoting microcrack initiation and intergranular or transgranular propagation under external load, revealing the hydrogen embrittlement mechanism at the atomic scale.

[0052] Through the complete process of "systematic initial data construction, active learning and iterative optimization, and rigorous multi-scale verification", the shortcomings of existing potential functions in terms of specificity, configuration coverage and reliability of hydrogen embrittlement prediction in the Ti-Al-VH system are solved, providing a high-precision and high-efficiency calculation tool for the study of the microscopic mechanism of hydrogen embrittlement in titanium alloys and performance optimization.

[0053] Compared with the prior art, the present invention has the following advantages: 1. Addressing the issue of inaccurate predictions: A dedicated initial dataset containing all relevant elemental combinations (solid solutions, hydrides) and defects is constructed. An active learning method is employed to specifically collect extreme and non-equilibrium configuration data closely related to the hydrogen embrittlement process, continuously expanding the training sample. This strategy ensures, from the fundamental level of model training, that the potential function can faithfully characterize the complex interactions of the Ti-Al-VH system, thereby significantly improving the accuracy and reliability of its predictions of key physicochemical parameters such as hydrogen dissolution energy, hydrogen segregation energy, and grain boundary fracture energy.

[0054] 2. Enhanced Simulation Stability: This invention introduces an active learning closed loop of "simulation-screening-learning." By automatically identifying and screening configurations with high model prediction uncertainty in molecular dynamics exploration simulations (such as crack tips and high-hydrogen segregation regions), these configurations are added to the training set, enabling the potential function to adaptively cover the real atomic environments that may occur throughout the hydrogen embrittlement process. This mechanism significantly improves the numerical stability and result confidence of the model when simulating complex dynamic events such as crack initiation and propagation, and hydrogen-induced interface failure.

[0055] 3. Addressing the "Impracticality" Issue: This invention addresses multi-scale verification of hydrogen embrittlement, requiring the potential function to accurately predict key indicators directly determining hydrogen embrittlement sensitivity, such as the hydrogen diffusion barrier, hydrogen segregation, and grain boundary fracture energy. This makes the final model a specialized tool calibrated for hydrogen embrittlement, directly usable for evaluating material resistance to hydrogen embrittlement, optimizing composition and processes, and significantly enhancing its practicality and engineering value.

[0056] Based on the above inventive concept, the present invention also provides a system for predicting the hydrogen embrittlement properties of Ti-Al-V alloys, comprising: The building module is used to construct the initial dataset, which contains multiple preset atomic configurations of the four elements Ti, Al, V, and H, along with their corresponding first-principles calculation data.

[0057] The initialization module uses the initial dataset to perform preliminary training on the potential function model, thereby obtaining the initial potential function model.

[0058] The filtering module is used to perform molecular dynamics simulations using an initial potential function model to generate molecular dynamic trajectories containing new atomic configurations; calculate the prediction uncertainty of each atomic configuration in the molecular dynamic trajectory, and filter out atomic configurations with prediction uncertainties higher than a preset threshold as new atomic configurations.

[0059] The expansion module is used to perform first-principles calculations on the selected new atomic configurations to obtain their corresponding energy and stress data; the new atomic configurations and their corresponding energy and stress data are added to the initial dataset to form an expanded dataset.

[0060] The update module is used to retrain the potential function model using the expanded dataset and update the potential function model.

[0061] The iterative module is used to repeatedly execute the above training steps until the prediction error of the updated potential function model converges and the prediction uncertainty is lower than a set threshold, thus obtaining the final machine learning potential function.

[0062] The prediction module is used to load the target atomic model of the Ti-Al-VH system under test into the final machine learning potential function for prediction, and obtain the prediction results of hydrogen embrittlement performance.

[0063] The iterative module is used to repeatedly execute the above training steps until the prediction error of the updated potential function model converges and the prediction uncertainty is lower than a set threshold, thus obtaining the final machine learning potential function.

[0064] The present invention also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The steps of the provided method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys.

[0065] This invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for various operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above-mentioned functions. Figure 1 The steps of the provided method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the patent of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for predicting the hydrogen embrittlement properties of Ti-Al-V alloys, characterized in that, The method includes the following steps: Construct an initial dataset, which contains multiple preset atomic configurations of four elements: Ti, Al, V, and H, and their corresponding first-principles calculation data; The potential function model is initially trained using the initial dataset to obtain the initial potential function model; Molecular dynamics simulations are performed using the initial potential function model to generate molecular dynamic trajectories containing new atomic configurations; the prediction uncertainty of each atomic configuration in the molecular dynamic trajectory is calculated, and atomic configurations with prediction uncertainties higher than a preset threshold are selected as new atomic configurations. First-principles calculations are performed on the selected new atomic configurations to obtain their corresponding energy and stress data; the new atomic configurations and their corresponding energy and stress data are added to the initial dataset to form an expanded dataset; The potential function model is retrained using the augmented dataset to update the potential function model; the prediction error of the updated potential function model converges and the prediction uncertainty is lower than a set threshold to obtain the final machine learning potential function. The target atomic model of the Ti-Al-VH system to be tested is loaded into the final machine learning potential function for prediction, and the hydrogen embrittlement performance prediction results are obtained.

2. The method according to claim 1, characterized in that, The atomic configurations in the initial dataset include at least one of the following: perfect unit cells of α-Ti and β-Ti; Ti-Al-V ternary solid solution supercells; hydrogen-containing Ti-Al-V ternary solid solution models; titanium hydride crystal models; grain boundary models of HCP or BCC systems; the titanium hydride crystal models include γ-TiH and δ-TiH. 1.5 Three hydrides: ε-TiH2 and ε-TiH2.

3. The method according to claim 2, characterized in that, In the hydrogen-containing Ti-Al-VH solid solution model, hydrogen atoms occupy tetrahedral or octahedral interstitial sites, and the hydrogen concentration is set to increase incrementally.

4. The method according to claim 1, characterized in that, The initial training of the potential function model using the initial dataset includes configuring the network structure parameters of the potential function model and training the model using the initial dataset.

5. The method according to claim 1, characterized in that, The molecular dynamics simulation is performed using the initial potential function model, and the molecular dynamics simulation includes at least one of the following simulation scenarios: NVT or NPT simulations were performed on solid solutions or hydrides in the temperature range of 300K to 1500K. High-temperature annealing simulation was performed on systems containing grain boundaries, phase boundaries, or dislocations; Simulate shear or tensile deformation by applying it to a model containing cracks.

6. The method according to claim 1, characterized in that, It also includes a step of performing multi-scale verification of the potential function model after the prediction error of the updated potential function model converges and the prediction uncertainty is below a set threshold. The multi-scale verification includes at least one of the following: calculating the energy state equation curve and comparing it with first-principles or experimental data. Calculate the diffusion energy barrier of hydrogen in α-Ti or β-Ti; calculate the elastic constants and compare them with first-principles results; Simulate the interaction between hydrogen and moving dislocations; simulate the initiation and propagation of hydrogen-induced cracks.

7. The method according to claim 1, characterized in that, The process of loading the target atomic model of the Ti-Al-VH system under test into the final machine learning potential function for prediction, and obtaining the hydrogen embrittlement performance prediction result, includes the following steps: Construct a target atomic model for the Ti-Al-VH system whose hydrogen embrittlement properties are to be predicted; The target atom model and the final machine learning potential function are loaded into the molecular dynamics simulation program; Run molecular dynamics simulations and, based on the simulation outputs, calculate one or more indices for quantitatively assessing hydrogen embrittlement sensitivity, including at least one of the following: hydrogen diffusion activation energy in α-Ti or β-Ti phases, hydrogen segregation energy at grain boundaries or phase boundaries, and fracture energy at hydrogen-containing grain boundaries or phase boundaries.

8. A system for predicting the hydrogen embrittlement properties of Ti-Al-V alloys, comprising: The construction module is used to construct an initial dataset, which contains multiple preset atomic configurations of four elements: Ti, Al, V, and H, and their corresponding first-principles calculation data. The initialization module uses the initial dataset to perform preliminary training on the potential function model to obtain the initial potential function model. The screening module is used to perform molecular dynamics simulation using the initial potential function model to generate molecular dynamics trajectories containing new atomic configurations; calculate the prediction uncertainty of each atomic configuration in the molecular dynamics trajectory, and screen out atomic configurations with prediction uncertainty higher than a preset threshold as new atomic configurations. An expansion module is used to perform first-principles calculations on the selected new atomic configurations to obtain their corresponding energy and stress data. The new atomic configuration and its corresponding energy and stress data are added to the initial dataset to form an expanded dataset. The update module is used to retrain the potential function model using the expanded dataset and update the potential function model. The iterative module is used to repeatedly execute the above training steps until the prediction error of the updated potential function model converges and the prediction uncertainty is lower than a set threshold, thus obtaining the final machine learning potential function. The prediction module is used to load the target atomic model of the Ti-Al-VH system under test into the final machine learning potential function for prediction, and obtain the prediction results of hydrogen embrittlement performance.

9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 7.

Images