High-entropy ldh morphology prediction method based on machine learning potential, storage medium and equipment

Abstract

The application discloses a high-entropy LDH morphology prediction method based on a machine learning potential, a storage medium and equipment, and the method comprises the following steps: high-throughput ensemble sampling is performed on a layered double metal hydroxide system composed of different metal cations, and an initial configuration library of random cation arrangement samples is constructed; a pre-trained graph neural network machine learning potential function is used to perform structure relaxation on the initial configuration library, and a reserved sample set is obtained by screening according to an atomic force threshold value; atomic scale descriptors are extracted from the reserved sample set; a cross-scale constitutive mapping relationship is constructed, and the atomic scale descriptors are converted into mesoscopic growth parameters required by a mesoscopic diffusion interface growth model; and the mesoscopic growth parameters are input into the three-dimensional diffusion interface growth model to obtain a mesoscopic morphology evolution result. The application can characterize the influence of atomic scale structure disturbance on the layer-by-layer nucleation, lapping and interlocking morphology of the layered double metal hydroxide, and can be used for morphology prediction, component screening and structure regulation of the high-entropy layered double metal hydroxide.

Description

Technical Field

[0001] This invention relates to the fields of materials computational science and microstructure simulation technology, specifically to a high-entropy LDH morphology prediction method, storage medium, and device based on machine learning potential. Background Technology

[0002] Layered bimetallic hydroxides are a class of two-dimensional layered materials characterized by tunable metal sites on the layers, exchangeable anions between layers, and abundant surface hydroxyl groups and edge active sites. They can be used in electrocatalysis, energy storage, adsorption separation, environmental remediation, and related interfacial reaction systems. For these materials, performance depends not only on the metal elemental composition, valence state distribution, and local electronic structure, but also on the size, thickness, orientation, stacking method of the nanosheets, and the pore network formed between the layers. Especially in electrochemical reactions and interfacial mass transfer processes, the degree of exposure of edge sites, pore connectivity, and electrolyte accessibility directly affect the effective active area and apparent reaction response.

[0003] Morphology control of existing layered bimetallic hydroxides typically relies on adjusting experimental factors such as the type of metal salt, the molar ratio of metal ions, precipitants, complexing agents, interlayer anions, hydrothermal temperature, hydrothermal time, or substrate conditions. These methods can yield various mesoscopic morphologies, including sheet-like, flower-like, nanograss-like, layered array-like, or three-dimensional house-of-cards interlocking structures. Compared to dense, face-to-face stacked structures, open three-dimensional interlocking networks formed by cross-overlapping nanosheets generally possess more accessible edges, larger interface exposure areas, and more unobstructed mass transport channels, thus holding significant importance in applications such as catalysis and energy storage.

[0004] With the introduction of multi-principal metal cations, high-entropy or multi-principal layered bimetallic hydroxides have attracted attention. Introducing multiple metal cations into the same layer of these materials can generate richer local coordination environments, stronger local lattice distortions, higher degrees of configurational freedom, and more complex interlayer coupling states. Compared with traditional binary or low-component layered bimetallic hydroxides, multi-principal or high-entropy layered bimetallic hydroxides often exhibit morphological characteristics such as thinner lamellars, increased boundary wrinkles, and enhanced lamellar intersections and spatial interlocking. These morphological changes indicate that the increasing complexity of the metal composition not only alters the local structure and energy state but may also progressively affect mesoscopic assembly through processes such as interlayer slip, exfoliation, edge diffusion, and nucleation behavior.

[0005] However, current understanding of the morphological evolution of high-entropy layered bimetallic hydroxides still relies primarily on experimental observation and empirical induction. Scanning electron microscopy, X-ray diffraction, energy dispersive spectroscopy, and electrochemical testing can obtain information on morphology, phase composition, elemental distribution, and macroscopic response, but they cannot directly reveal how local structural perturbations caused by the random occupancy of different metal cations are transmitted to interlayer mechanics, edge dynamics, and mesoscopic nucleation processes. Judging the high-entropy effect solely based on the final experimental morphology makes it difficult to distinguish the relative contributions of factors such as mixing thermodynamics, local distortion, interlayer interactions, edge migration, and substrate nucleation to morphological evolution.

[0006] In theoretical calculations, density functional theory can accurately calculate local electronic structure, surface energy, adsorption energy, interlayer interactions, and reaction barriers. However, this method is usually limited by computational cost and spatiotemporal scale. For high-entropy layered bimetallic hydroxides containing multiple randomly occupied metal cations, a large number of configuration samples are required to obtain statistically representative atomic-scale information. Traditional first-principles calculations cannot simultaneously meet the requirements of configuration number, system size, and computational efficiency, and are even less able to directly simulate the dynamic process from atomic-scale perturbation to the formation of micrometer-scale layered networks.

[0007] Mesoscopic simulation methods such as phase-field models and diffusion-interface models are suitable for describing processes such as lamellar growth, interface advancement, nucleation expansion, and network evolution, and can reproduce morphological evolution trends on a large spatial scale. However, parameters such as growth driving forces, mobility, nucleation rate, roughness, aspect ratio, and interface anisotropy in existing mesoscopic models are mostly derived from empirical settings, artificial fitting, or inverse deduction from single experiments, lacking explicit correspondences with real atomic structures, local disorder, interlayer slip, stripping barriers, and edge diffusion behavior. Therefore, although traditional mesoscopic models can simulate morphology, they often fail to explain why that morphology is generated by specific high-entropy components.

[0008] In recent years, machine learning potential functions based on graph neural networks have enabled structural relaxation, energy assessment, and high-throughput configuration screening of complex inorganic materials at relatively low computational cost, providing a new technical means for extracting atomic-scale descriptors of multi-principal layered materials. However, using machine learning potentials alone can only obtain atomic-scale information such as energy, force, and bond length distortion, and cannot directly output mesoscopic morphologies such as lamellar nucleation density, boundary roughness, aspect ratio variation, and three-dimensional interlocking networks.

[0009] Therefore, how to develop a cross-scale method that can transform the atomic-scale descriptor system obtained by machine learning potential into mesoscopic growth parameters and further drive diffusion interface or phase field model for morphology prediction, in order to solve the problem of lack of quantitative correlation between atomic structure perturbation and mesoscopic morphology evolution of high-entropy layered bimetallic hydroxides, is an urgent problem to be solved. Summary of the Invention

[0010] To address the technical problems existing in the prior art, the first objective of this invention is to provide a high-entropy LDH morphology prediction method based on machine learning potential. This method converts atomic-scale descriptors into mesoscopic growth parameters through cross-scale constitutive mapping, drives a three-dimensional diffusion interface growth model to obtain mesoscopic morphology evolution results, and achieves cross-scale prediction from atomic structure perturbation to mesoscopic morphology.

[0011] The second objective of this invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for predicting the morphology of high-entropy layered bimetallic hydroxides based on machine learning potential.

[0012] A third objective of this invention is to provide an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method for predicting the morphology of high-entropy layered bimetallic hydroxides based on machine learning potential.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] The high-entropy LDH topography prediction method based on machine learning potential includes the following steps:

[0015] High-throughput ensemble sampling was performed on layered bimetallic hydroxide systems with different metal cation compositions to construct an initial configuration library containing multiple random cation arrangement samples;

[0016] The initial configuration library is structurally relaxed using a pre-trained graph neural network machine learning potential function, and a retained sample set is obtained by screening based on the atomic force threshold.

[0017] Atomic-scale descriptors are extracted from the retained sample set. These atomic-scale descriptors include basic descriptors that reflect the thermodynamic stability of the mixture and the degree of local structural distortion, as well as high-level physical descriptors that reflect the thermodynamics of crystal interfaces, interlayer coupling, and edge migration dynamics.

[0018] Construct cross-scale constitutive mapping relationships to convert the atomic-scale descriptors into mesoscopic growth parameters required for the mesoscopic diffusion interface growth model;

[0019] The mesoscopic growth parameters are input into a three-dimensional diffusion interface growth model for numerical solution to obtain the mesoscopic morphology evolution results of each component system.

[0020] According to one example, the atomic force threshold screening includes: removing a sample when the maximum atomic force of the sample exceeds a preset hard threshold, and determining a strict convergence threshold for the remaining samples based on the force distribution within the group to form a retained sample set;

[0021] The degree of local structural distortion is characterized by the variance of metal-oxygen local bond lengths. The method also includes selecting representative configurations from the retained sample set based on a comprehensive score of group-average energy and average bond length variance, and calculating the high-level physical descriptor on the representative configuration.

[0022] According to one example, the cross-scale constitutive mapping relationship includes: determining an effective growth driving force based on the excess mixing energy in the atomic-scale descriptor, wherein when the excess mixing energy does not reach the disorder initiation threshold, the effective growth driving force is a lower limit of the basic driving force, and when the excess mixing energy reaches or exceeds the disorder initiation threshold, the effective growth driving force is jointly determined by the sum of the lower limit of the basic driving force and the disorder promotion coefficient multiplied by the difference between the excess mixing energy and the disorder initiation threshold.

[0023] According to one example, the cross-scale constitutive mapping relationship further includes: determining thermodynamic anisotropy parameters based on the basal surface energy, edge surface energy, basal water interface energy, and edge water interface energy in the atomic-scale descriptor, wherein the thermodynamic anisotropy parameters are used to characterize the growth competition relationship between the layered bimetallic hydroxide in the basal direction and the edge direction.

[0024] According to one example, the cross-scale constitutive mapping further includes: determining the in-plane mobility based on the edge diffusion kinetics surrogate in the atomic-scale descriptor, and determining the thickness-direction mobility based on the generalized stacking fault energy and the stripping energy in the atomic-scale descriptor; wherein the in-plane mobility is negatively correlated with the edge diffusion kinetics surrogate, and the thickness-direction mobility is negatively correlated with both the generalized stacking fault energy and the stripping energy.

[0025] According to one example, the cross-scale constitutive mapping relationship further includes: determining the heterogeneous nucleation rate based on the basal effective interface energy and the disorder-induced nucleation enhancement coefficient in the atomic-scale descriptor, wherein the heterogeneous nucleation rate is negatively correlated with the basal effective interface energy and positively correlated with the disorder-induced nucleation enhancement coefficient.

[0026] According to one example, the cross-scale constitutive mapping relationship further includes: determining the topographic roughness response function based on the interlayer stiffness surrogate and the degree of local structural distortion in the atomic-scale descriptor; and determining the effective aspect ratio based on the excess mixing energy and the edge diffusion dynamics surrogate in the atomic-scale descriptor.

[0027] According to one example, the three-dimensional diffusion interface growth model characterizes the solid-phase lamellar and solution regions with continuous phase field variables; each component system uses the same computational domain and basic numerical parameters, and the morphological differences between components are modulated by the nucleation tendency, anisotropic growth kinetic rate ratio, boundary roughness, and aspect ratio parameters input by the cross-scale constitutive mapping relationship; the mesoscopic morphological evolution results include the evolution process from initial random nucleation to the formation of lamellar cross-interlocking networks.

[0028] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the above-described high-entropy LDH topography prediction method based on machine learning potential.

[0029] An electronic device includes one or more processors and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the above-described high-entropy LDH topography prediction method based on machine learning potential.

[0030] The present invention has the following advantages:

[0031] This invention overcomes the shortcomings of existing methods for predicting the mesoscopic morphology of high-entropy layered bimetallic hydroxides, such as the difficulty in quantitatively transferring atomic-scale structural perturbations to the mesoscopic model, the reliance on empirically set parameters for mesoscopic phase field or diffusion interface models, and the lack of interpretable prediction methods for the morphological evolution trends of different component systems. It provides a cross-scale prediction method for the mesoscopic morphology of high-entropy layered bimetallic hydroxides based on machine learning potentials. This method achieves the prediction of the evolution trend of mesoscopic morphology from local structural disorder to layered interlocking by coupling high-throughput configuration sampling of machine learning potentials, extraction of atomic-scale physical descriptors, cross-scale constitutive mapping, and three-dimensional diffusion interface growth simulation.

[0032] This invention introduces a pre-trained machine learning potential function into the random configuration sampling and structural relaxation process of high-entropy layered bimetallic hydroxides, which can obtain statistical atomic-scale structural information of multi-component systems at a lower computational cost, alleviating the problem that traditional first-principles methods are unable to cover a large number of random cation configurations and mesoscale morphological evolution processes.

[0033] This invention extracts atomic-scale descriptors such as mixing energy, local bond length distortion, crystal interface energy, generalized stacking fault energy, exfoliation energy, interlayer stiffness, and edge diffusion barrier, and maps them to driving force, mobility, nucleation rate, roughness, and aspect ratio parameters in the mesoscopic growth model, thereby reducing the dependence of traditional phase-field models on empirically set mesoscopic parameters.

[0034] This invention can predict the relative trend of the evolution of layered bimetallic hydroxides from relatively regular two-dimensional sheet stacking to open three-dimensional interlocking network after the introduction of high-entropy components, and can quantitatively compare the nucleation tendency, boundary roughness, sheet cross-over and network openness of different component systems.

[0035] This invention enables trend correlation between simulated morphology and experimental scanning electron microscope images through a unified skeletal index, providing a quantitative screening basis for the component design, morphology control, effective interface exposure optimization, and electrochemical performance improvement of high-entropy layered bimetallic hydroxides. Attached Figure Description

[0036] Figure 1 This is a flowchart of the high-entropy layered bimetallic hydroxide morphology prediction method based on machine learning potential of the present invention.

[0037] Figure 2 This is a prediction diagram of the mesoscopic morphology of the binary control system LDH.

[0038] Figure 3 This is a predicted mesoscopic morphology diagram of the ternary HE3 system LDH.

[0039] Figure 4 This is a predicted mesoscopic morphology diagram of the quaternary HE4 system LDH.

[0040] Figure 5 This is a predicted mesoscopic morphology diagram of the pentagonal HE5 system LDH.

[0041] Figure 6 Scanning electron microscope images of different LDH components. Detailed Implementation

[0042] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0043] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0044] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0045] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all promotional information and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0046] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily essential for implementing this application, and therefore cannot be used to limit the scope of protection of this application.

[0047] The methods provided in this application relate to the fields of materials computational science and microstructure simulation technology, especially the cross-scale prediction technology of mesoscopic morphology of high-entropy layered bimetallic hydroxides, which are specifically described in the embodiments below.

[0048] Random cation arrangement configurations were constructed for layered bimetallic hydroxide systems composed of different metal cations. The structural relaxation and convergence screening were performed using a pre-trained graph neural network machine learning potential. Atomic-scale descriptors reflecting mixing thermodynamics, local bond length distortion, crystal interface thermodynamics, interlayer slip and exfoliation, interlayer stiffness, and edge migration dynamics were extracted from the relaxed atomic structures. These atomic-scale descriptors were mapped to parameters in a mesoscopic growth model, including effective growth driving force, thermodynamic anisotropy, in-plane mobility, thickness-direction mobility, heterogeneous nucleation rate, morphological roughness, and effective aspect ratio. These mesoscopic parameters were input into a three-dimensional diffusion interface growth model to obtain the relative morphological evolution results of layered bimetallic hydroxides with different compositions from initial nucleation to lamellar cross-overlapping, spatial interlocking, and the formation of an open three-dimensional network.

[0049] Reference Figure 1 First, initial atomic structures of layered bimetallic hydroxide systems with different components were constructed. These different component systems included a binary control system and a multi-principal or high-entropy layered bimetallic hydroxide system containing three or more metal cations. Each system used a uniform layered bimetallic hydroxide structural template and a consistent interlayer anion environment, and the metal cation sites were randomly occupied to form multiple initial configuration samples, thereby reducing the random influence of a single random configuration on subsequent calculation results.

[0050] Secondly, the initial configuration samples are subjected to full-degree-of-freedom structural relaxation using a pre-trained machine learning potential function. The machine learning potential function is preferably a graph neural network machine learning potential function, more preferably the CHGNet general graph neural network potential function. During structural relaxation, atomic coordinates and cell parameters are allowed to be optimized simultaneously, and the relaxed samples are screened based on the maximum residual atomic force. In this embodiment, CHGNet is used as a unified high-throughput relative trend screening potential function. Its usability is jointly verified through a unified configuration template, a unified interlayer environment, periodic boundary conditions, the same maximum relaxation steps, the same residual force screening rules, and experimental morphological trends. The energy output by CHGNet is used for relative descriptors and inter-group trend comparisons under the same computational process, without directly interpreting a single energy value as an absolute thermodynamic quantity. In the embodiments, the Control, HE3, HE4, and HE5 mainline systems retained 22, 24, 24, and 21 effective configurations respectively after CHGNet relaxation from 30 initial random configurations. The number of strictly convergent samples were 2, 3, 3, and 2, respectively, and the maximum residual force ranges of the retained samples were 0.076606-0.099956 eV / Å, 0.068063-0.100010 eV / Å, 0.070587-0.099947 eV / Å, and 0.074396-0.100090 eV / Å, respectively. The obtained relative morphological evolution trends were further verified by XRD, TEM / STEM-EDS, SEM, and skeletonization statistics. When the maximum atomic force of a sample exceeds a preset hard threshold, the sample is removed. For the remaining samples, an adaptive convergence threshold is further determined based on the force distribution within the group to obtain the retained sample set. Specifically, the atomic force screening adopts a two-level threshold: a preset hard threshold F. hard =0.12 eV / Å, configurations with maximum residual atomic forces higher than this value are directly discarded; the default strict convergence threshold F strict =0.08 eV / Å. If the number of samples for a certain component under the default strict convergence threshold is less than 2, then the maximum residual force of the retained samples in that group is sorted in ascending order, and the second smallest value F(2) is added to 10. -6 eV / Å was used as a candidate threshold, and the range was limited to 0.08–0.10 eV / Å, i.e., F * strict =min[max(F(2)+10 -6 [0.08), 0.10] eV / Å; maximum residual force not higher than F * strict The configurations that are strictly converged are used as samples, and the other configurations that are not higher than the hard threshold are used as samples that are loosely retained.

[0051] Next, atomic-scale descriptors are extracted from the retained sample set. The atomic-scale descriptors include at least one of the following: system average energy, metal site average energy, excess mixing energy, metal-oxygen local bond length variance, basal surface energy, edge surface energy, basal-water interface energy, edge-water interface energy, generalized stacking fault energy, peeling energy, interlayer stiffness surrogate quantity, edge attachment kinetic surrogate quantity, and edge diffusion kinetic surrogate quantity.

[0052] Among them, the variance of the local bond length of the metal-oxygen region is used to characterize the degree of local structural distortion, and its expression is:

[0053] (1)

[0054] In the formula, For local bond length variance, The number of metal-oxygen bonds included in the statistics. For the first Metal-oxygen bond length The average bond length of the metal-oxygen bond is given.

[0055] Excess mixing energy is used to characterize the thermodynamic stability of a multi-component system relative to a linear combination of endmember reference states, and its expression is:

[0056]

[0057] In the formula, For excess mixing energy, The average metal site energy of the target system is given. The reference energy is obtained by linear combination of the average energies of the endmember reference system.

[0058] To transfer atomic-scale information to mesoscopic morphology models, this invention constructs a cross-scale constitutive mapping relationship.

[0059] Effective growth drivers can be expressed as:

[0060]

[0061] In the formula, For effective growth driving force, The lower limit of the basic driving force The disorder promotion coefficient, This represents the disorder initiation threshold. This relationship is used to describe the combined effect of hybrid stabilization and local structural distortion on the crystal growth driving force.

[0062] Thermodynamic anisotropy parameters can be expressed as:

[0063]

[0064] In the formula, For thermodynamic anisotropy parameters, For the surface energy of the base plane, For the surface energy of the edge surface, For the base surface water interface energy, This represents the interfacial energy between the edge and water surfaces. This parameter is used to characterize the thermodynamic competition between basal broadening and edge-direction growth in layered bimetallic hydroxides.

[0065] The in-plane mobility can be modulated by the edge diffusion barrier, and its expression is as follows:

[0066]

[0067] In the formula, In-plane mobility The basic in-plane mobility. The effective edge diffusion barrier after reduction by the liquid phase environment. Boltzmann's constant, Where is the absolute temperature. The thickness-direction mobility can be jointly controlled by interlayer slip and peel resistance, and its expression is:

[0068]

[0069] In the formula, For the thickness direction mobility, The basic mobility in the thickness direction, The unit event energy barrier is obtained by converting the generalized stacking fault energy. The energy barrier per unit event is calculated from the stripping energy. It is the generalized stacking fault energy weighting factor.

[0070] Heterogeneous nucleation rate can be expressed as:

[0071]

[0072] In the formula, For heterogeneous nucleation rate, Based on prenucleation factors, These are the coefficients of the classical nucleation term. For the effective interface energy of the base plane, To prevent tiny constants with a denominator of zero, This is the disorder-induced nucleation enhancement coefficient.

[0073] The surface roughness response function can be expressed as:

[0074]

[0075] In the formula, Let be the surface roughness response function. Based on roughness, This is the local disorder roughness amplification factor. This is the interlayer stiffness modulation factor. This is a proxy for inter-layer stiffness. and These are the lower and upper limits of roughness, respectively. This represents the interval clipping function.

[0076] The effective length-to-diameter ratio can be expressed as:

[0077] (9)

[0078] In the formula, For an effective aspect ratio, The aspect ratio is the scaling factor. The dynamic compressibility index is... and These represent the lower and upper limits of the aspect ratio, respectively. The effective aspect ratio is determined by the thermodynamic anisotropy parameter and the ratio of in-plane mobility to thickness-direction mobility. The excess mixing energy influences the lamellar expansion process through the effective growth driving force, while the edge diffusion kinetic proxy participates in regulating the effective aspect ratio through the in-plane mobility.

[0079] Through the above constitutive mapping relationship, the atomic-scale descriptors obtained by machine learning potential calculation can be transformed into mesoscopic growth model parameters, so that the phase field or diffusion interface simulation has a clear atomic structure source and reduces the dependence on empirical parameters.

[0080] The coefficients in formulas (3) to (9) are determined using a standardized anchoring method applicable to all components, without performing component-specific empirical parameter tuning for HE3, HE4, or HE5. Typical values ​​for each parameter in this embodiment are shown in Table 1.

[0081] Table 1 shows the typical values ​​of the relevant parameters in formulas (3) to (9).

[0082]

[0083] The calculation temperature is set to 300 K, and the Boltzmann constant is a conventional physical constant. When converting the generalized stacking fault energy and stripping energy from surface energy density to the unit event energy barrier, the area of ​​a preset characteristic event can be used for the conversion. The effective edge diffusion barrier after liquid phase reduction is preferred for the edge diffusion barrier; if a reduction value is unavailable, the proportional reduction value of the original edge diffusion barrier can be used. The numerical stability term is only used to avoid the denominator being zero in the nucleation rate expression. The above coefficients can be sensitively perturbed within the range of 0.8–1.2 times to check the robustness of the predicted trend, and are not used as empirical parameters for specific components.

[0084] In mesoscopic simulations, this invention uses continuous phase field variables to characterize the solid-phase lamellars and solution or porous regions of layered bimetallic hydroxides. Different component systems are subjected to the same computational domain, substrate conditions, time step, diffusion coefficient, and fundamental numerical parameters. Component differences are reflected by parameters such as nucleation tendency, anisotropic growth kinetics, boundary roughness, and effective aspect ratio obtained from cross-scale constitutive mapping. This allows for the simulation of the relative morphological evolution of different component systems from initial random nucleation to lamellar overlap, spatial interlocking, and the formation of open three-dimensional networks.

[0085] This invention can also perform two-dimensional projection, edge extraction, and skeletonization analysis on the simulated three-dimensional topographic field to obtain topographic statistical indicators such as intersection density, skeleton length density, edge length density, open region ratio, and specific surface area surrogate quantity. These topographic statistical indicators can be used to characterize the complexity of layered networks, the degree of boundary roughness, and the degree of three-dimensional interlocking, and can be used to verify trends with statistical results obtained from experimental scanning electron microscope images through the same image processing procedure.

[0086] Mesoscopic model implementation and effect verification

[0087] The three-dimensional diffusion interface growth model uses normalized phase field indicator variables. Characterizing the solid-phase lamellar regions of LDH, Indicates solid-phase lamellar structure. Represents a solution or porous region; a single sheet is a sign distance function of hexagonal lamellae. generate, Interface width Each grid, the overall topographic field is taken Normalized growth unit supply field Satisfying explicit difference form ,in Using three-dimensional six-neighbor central difference, , , To satisfy explicit diffusion stability, when Exceed The time is divided according to the sub-step size. The computation region is... Each grid, normalized time step The total number of steps is 300, and the VTK topography field is output every 20 steps. Field. Boundary conditions are approximated by zero flux at the sides and bottom, and zero flux at the top. Fixed at 1.0 to indicate the supply of external growth units. The bottom region of each grid is fixed as the base layer. Initial conditions are... , A nucleus count of 1 is assigned to the basal layer and 0 to the remaining regions. Hexagonal thin-plate nuclei are then generated on the basal layer, with a base nucleus number of 185. These are then corrected according to the composition coverage factors Control / HE3 / HE4 / HE5 = 1.00 / 1.05 / 1.10 / 1.16 and the nucleation density factor, ultimately limiting the nucleus number to the range of 170-320. Initial nucleus formation... The coordinates are located 0-5 grids above the base, and the orientation is determined by a random angular distribution consisting mainly of obliquely upright thin sheets with a small number of nearly vertical thin sheets; the in-plane radius and thickness vary locally. The in-plane / thickness migration rate and the contact crowding inhibition term gradually increase. When the contact ratio exceeds 0.06, in-plane growth is inhibited; when it exceeds 0.10, thickness growth stops; and when it exceeds 0.16, the growth of the lamellar layer stops.

[0088] This embodiment selects binary Control, ternary HE3, quaternary HE4, and pentagonal HE5 layered bimetallic hydroxide systems as the calculation objects. Specifically, the Control system is a Ni-Fe layered bimetallic hydroxide, the HE3 system is a Ni-Co-Fe layered bimetallic hydroxide, the HE4 system is a Ni-Co-Zn-Fe layered bimetallic hydroxide, and the HE5 system is a Ni-Co-Zn-Fe-Al layered bimetallic hydroxide. All systems use a unified layered bimetallic hydroxide structural template and a consistent interlayer anion environment to ensure the comparability of calculation results between different component systems.

[0089] First, initial atomic structures for different component systems are constructed by randomly occupying metal cation sites according to a predetermined composition. To reduce the statistical randomness caused by a single random structure, 30 random cation configuration samples are generated for each principal system; simultaneously, an endmember reference system is constructed for subsequent calculation of the excess mixing energy reference state.

[0090] Subsequently, a pre-trained graph neural network machine learning potential function is used to perform full-degree-of-freedom structural relaxation on the initial configuration. In this embodiment, the machine learning potential function is the CHGNet general graph neural network potential function. During the relaxation process, atomic coordinates and cell parameters are allowed to be optimized simultaneously. After relaxation, the samples are screened based on the maximum residual atomic force, and samples that have not converged sufficiently or have abnormal local structures are removed to obtain a retained sample set.

[0091] Next, basic atomic-scale descriptors are extracted from the retained samples, including the total system energy, average metal site energy, metal-oxygen bond length distribution, average bond length variance, and maximum bond length variance. Among these, the local metal-oxygen bond length variance characterizes the degree of local structural distortion, and the excess mixing energy characterizes the thermodynamic stability of the multi-component system relative to the end-member reference state. The retained samples are comprehensively scored based on the average energy and average bond length variance, and the sample with the lowest score is selected as the representative configuration of the system. The specific scoring function is as follows: for the same component... The first Let there be 1 retained sample, and its normalized energy at the metal site be . The average metal-oxygen bond variance is The mean values ​​of this sample group are respectively and The standard deviations are respectively and Then the overall score The energy term and the local distortion term are weighted equally, i.e. The equivalent normalized weights are all 0.5. Representative configurations are selected primarily from the strictly convergent sample pool. The smallest sample; if no strictly convergent sample exists in the group, then select from the loosely reserved samples not exceeding the hard threshold using the same scoring function. The smallest sample size. Further calculations are performed on representative configurations for basal surface energy, edge surface energy, basal water interface energy, edge water interface energy, generalized stacking fault energy, peeling energy, interlayer stiffness surrogate, and edge diffusion dynamics surrogate.

[0092] Then, according to the aforementioned cross-scale constitutive mapping relationship, the atomic-scale descriptors are converted into mesoscopic growth parameters, including effective growth driving force, thermodynamic anisotropy parameters, in-plane mobility, thickness direction mobility, heterogeneous nucleation rate, and morphological roughness response function.

[0093] Finally, the mapped mesoscopic growth parameters are input into a three-dimensional diffusion interface growth model. This model uses continuous phase field variables to characterize the LDH solid-phase lamellars and solution or pore regions. Under the same computational domain, substrate conditions, and basic numerical parameters, the mesoscopic morphological evolution process of different component systems from initial nucleation, lamellar expansion, boundary roughening to lamellar interlocking is simulated. In this embodiment, the total number of steps is set to 300, and the morphological differences between components are entirely modulated by the aforementioned mapped mesoscopic growth parameters.

[0094] Figures 2 to 5 The mesoscopic morphology prediction results for the Control, HE3, HE4, and HE5 systems are shown in the figures. As can be seen from the figures, with the increase of component complexity, the degree of cross-overlap and local interlocking between LDH sheets is enhanced, indicating that the method of the present invention can reflect the mesoscopic morphology evolution trend caused by high-entropy components.

[0095] Figure 6 Scanning electron microscope (SEM) images of different LDH components are shown, illustrating the actual experimental morphologies of the Control, HE3, HE4, and HE5 systems. Figures 2 to 5 The prediction results and Figure 6 A comparison of the experimental morphologies shows that the morphology evolution trend predicted by the present invention (the degree of lamellar cross-overlap and interlocking increases with the increase of component complexity) is consistent with the experimental observation results, verifying the reliability of the method of the present invention.

[0096] Through the above calculation process, the relative nucleation tendency, lamellar boundary roughening degree, aspect ratio variation and three-dimensional interlocking network formation trend of high-entropy LDH systems with different components can be obtained, thereby realizing cross-scale prediction from atomic-scale structural perturbation to mesoscopic morphological evolution.

[0097] Those skilled in the art will understand that all or part of the steps of the high-entropy layered bimetallic hydroxide morphology prediction method based on machine learning potential in the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed by a processor, it implements the steps of the high-entropy layered bimetallic hydroxide morphology prediction method based on machine learning potential. The computer-readable storage medium can be a read-only memory, random access memory, disk, optical disk, or any other form of storage medium.

[0098] Furthermore, this application also provides an electronic device including one or more processors and a storage device. The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the above-described method for predicting the morphology of high-entropy layered bimetallic hydroxides based on machine learning potentials. The electronic device can be a computer, server, workstation, or other device with computing capabilities.

Claims

1. A high-entropy LDH topography prediction method based on machine learning potential, characterized in that, Includes the following steps: High-throughput ensemble sampling was performed on layered bimetallic hydroxide systems with different metal cation compositions to construct an initial configuration library containing multiple random cation arrangement samples; The initial configuration library is structurally relaxed using a pre-trained graph neural network machine learning potential function, and a retained sample set is obtained by screening based on the atomic force threshold. Atomic-scale descriptors are extracted from the retained sample set. These atomic-scale descriptors include basic descriptors that reflect the thermodynamic stability of the mixture and the degree of local structural distortion, as well as high-level physical descriptors that reflect the thermodynamics of crystal interfaces, interlayer coupling, and edge migration dynamics. Construct cross-scale constitutive mapping relationships to convert the atomic-scale descriptors into mesoscopic growth parameters required for the mesoscopic diffusion interface growth model; The mesoscopic growth parameters are input into a three-dimensional diffusion interface growth model for numerical solution to obtain the mesoscopic morphology evolution results of each component system.

2. The method according to claim 1, characterized in that, The atomic force threshold screening includes: when the maximum atomic force of a sample exceeds a preset hard threshold, the sample is removed; and for the remaining samples, a strict convergence threshold is determined based on the force distribution within the group to form a retained sample set. The degree of local structural distortion is characterized by the variance of metal-oxygen local bond lengths. The method also includes selecting representative configurations from the retained sample set based on a comprehensive score of group-average energy and average bond length variance, and calculating the high-level physical descriptor on the representative configuration.

3. The method according to claim 1, characterized in that, The cross-scale constitutive mapping relationship includes: determining the effective growth driving force based on the excess mixing energy in the atomic scale descriptor, wherein when the excess mixing energy does not reach the disorder initiation threshold, the effective growth driving force is the lower limit of the basic driving force, and when the excess mixing energy reaches or exceeds the disorder initiation threshold, the effective growth driving force is jointly determined by the sum of the lower limit of the basic driving force and the disorder promotion coefficient multiplied by the difference between the excess mixing energy and the disorder initiation threshold.

4. The method according to claim 3, characterized in that, The cross-scale constitutive mapping relationship further includes: determining thermodynamic anisotropy parameters based on the basal surface energy, edge surface energy, basal water interface energy, and edge water interface energy in the atomic scale descriptor. The thermodynamic anisotropy parameters are used to characterize the growth competition relationship between the layered bimetallic hydroxide in the basal direction and the edge direction.

5. The method according to claim 3, characterized in that, The cross-scale constitutive mapping relationship further includes: determining the in-plane mobility based on the edge diffusion dynamics surrogate quantity in the atomic-scale descriptor, and determining the thickness-direction mobility based on the generalized stacking fault energy and the stripping energy in the atomic-scale descriptor; wherein the in-plane mobility is negatively correlated with the edge diffusion dynamics surrogate quantity, and the thickness-direction mobility is negatively correlated with both the generalized stacking fault energy and the stripping energy.

6. The method according to claim 3, characterized in that, The cross-scale constitutive mapping relationship further includes: determining the heterogeneous nucleation rate based on the effective interface energy of the basal plane and the disorder-induced nucleation enhancement coefficient in the atomic-scale descriptor, wherein the heterogeneous nucleation rate is negatively correlated with the effective interface energy of the basal plane and positively correlated with the disorder-induced nucleation enhancement coefficient.

7. The method according to claim 3, characterized in that, The cross-scale constitutive mapping relationship further includes: determining the topographic roughness response function based on the interlayer stiffness surrogate and the degree of local structural distortion in the atomic-scale descriptor; and determining the effective aspect ratio based on the excess mixing energy and the edge diffusion dynamics surrogate in the atomic-scale descriptor.

8. The method according to claim 1, characterized in that, The three-dimensional diffusion interface growth model characterizes the solid-phase lamellar and solution regions with continuous phase field variables; each component system uses the same computational domain and basic numerical parameters, and the morphological differences between components are modulated by the nucleation tendency, anisotropic growth kinetic rate ratio, boundary roughness and aspect ratio parameters input by the cross-scale constitutive mapping relationship; the mesoscopic morphological evolution results include the evolution process from initial random nucleation to the formation of lamellar cross-interlocking networks.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the high-entropy LDH topography prediction method based on machine learning potential as described in any one of claims 1 to 8.

10. An electronic device, characterized in that, The method includes one or more processors and a storage device, the storage device being used to store one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the high-entropy LDH topography prediction method based on machine learning potential as described in any one of claims 1 to 8.

Images