Efficient prediction method for mechanical properties of water-containing montmorillonite

Abstract

The invention provides a method for efficiently predicting mechanical properties of water-containing montmorillonite. The method comprises the following steps: establishing a water-containing montmorillonite model through molecular dynamics modeling software; carrying out first principle calculation on the water-containing montmorillonite model to obtain a first data set trained by an atomic potential function model; training an atomic potential function model through the first data set, and checking whether the atomic potential function model is good or not after training is completed; performing molecular dynamics simulation by using the trained atomic potential function model, comparing empirical force fields at the same time, and obtaining a force field file based on result differences; and carrying out uniaxial stretching and compression simulation on the water-containing montmorillonite model based on the force field file by adopting molecular dynamics software to obtain the mechanical property of the water-containing montmorillonite. According to the method, the mechanical behaviors of the water-containing montmorillonite under different temperature, pressure and loading conditions can be simulated for the water-containing montmorillonite model, and compared with a classical molecular dynamics method, the simulation precision is improved, and higher calculation efficiency is also obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the mechanical properties of water-containing montmorillonite, and particularly relates to a method for efficiently predicting the mechanical properties of water-containing montmorillonite. Background Art

[0002] As a kind of silicate mineral widely distributed in the earth's crust, clay minerals have relatively slight water sensitivity. The mechanical properties of clay minerals are affected by interlayer water and ions, and will exhibit swelling, rheology, and strength anisotropy. In the field of geotechnical engineering, the mechanical properties of water-containing clay minerals will significantly affect the stability of foundations and slopes and the deformation characteristics of corresponding geotechnical engineering, which are important factors affecting the performance of geotechnical engineering. Therefore, in-depth research on the mechanical properties of water-containing clay minerals not only provides important theoretical support for engineering practice and technological innovation in related fields, but also provides a scientific basis for geological disaster prevention and control and geotechnical engineering treatment.

[0003] The development of numerical simulation technology has made it possible to study and predict the mechanical properties of clay minerals such as montmorillonite. Among them, classical molecular dynamics and first-principles have become effective tools for studying the mechanical properties of montmorillonite. The molecular dynamics method relies on Newton's classical mechanics method to study the motion of molecular systems. The motion trajectories of atoms and molecules obey Newton's classical force field, so that the positions and velocities of atoms and molecules at any moment can be obtained, and then the motion of the system within a certain period of time can be obtained. However, the calculation accuracy of the molecular dynamics method depends on the selected force field, and the force field file is programmed based on experience, which will significantly restrict the calculation accuracy of molecular dynamics. On the other hand, a machine learning force field file based on mathematical analytical expressions without physical meaning can use the accurate potential energy surface reference data obtained by first-principles calculations to train the model, so as to obtain a potential energy surface far more accurate than the empirical force field. However, because the machine learning force field has no physical assumptions, its portability is poor and it can only be applied to a certain specific system; while the empirical force field of classical molecular dynamics is based on the physical description of the real system, reflecting the bonding effects and Coulomb electrostatic forces between atoms, etc., so it has better portability for the system. Summary of the Invention

[0004] The object of the present invention is: aiming at the deficiencies in the above background art, to provide a solution capable of efficiently predicting the mechanical properties of water-containing montmorillonite, so as to significantly improve both accuracy and efficiency.

[0005] To achieve the above object, the present invention provides a method for efficiently predicting the mechanical properties of water-containing montmorillonite, including the following steps:

[0006] S1, establishing a water-containing montmorillonite model through molecular dynamics modeling software;

[0007] S2. Perform first-principles calculations on the water-containing montmorillonite model to obtain the first data set for training the atomic potential function model;

[0008] S3. Train the atomic potential function model with the first data set, and check the quality of the atomic potential function model after training;

[0009] S4. Use the trained atomic potential function model for molecular dynamics simulations, and at the same time adopt an empirical force field, set the same parameters as the atomic potential function model for classical molecular dynamics simulations, to compare and calculate the result differences between the atomic potential function model and the empirical force field, and obtain the force field file;

[0010] S5. Use molecular dynamics software to perform uniaxial tensile and compression simulations on the water-containing montmorillonite model based on the force field file, and obtain the mechanical properties of the water-containing montmorillonite according to the simulation results.

[0011] Furthermore, the water-containing montmorillonite model takes atoms as the basic unit. The montmorillonite model consists of two Si-O tetrahedrons and one Al-O octahedron to form a crystal layer, and at the same time, a water molecule model is inserted in the middle of the crystal layer to form the water-containing montmorillonite model.

[0012] Furthermore, S2 includes the following sub-steps:

[0013] S21. Convert the water-containing montmorillonite model into an input text file and import it into the first-principles calculation software;

[0014] S22. Prepare all the atomic species pseudopotential files and the first input parameter file required for the first-principles calculation;

[0015] S23. Perform a structural optimization operation on the water-containing montmorillonite model to obtain the energy-minimized configuration;

[0016] S24. Use the result file of the energy-minimized configuration as the input structure file for the first-principles calculation, set a new first input parameter file, and set the expected temperature, pressure, and number of calculation steps for uniaxial tensile and compression simulations;

[0017] S25. Obtain the configuration obtained from the first-principles calculation, and use it to obtain multiple snapshots of the water-containing montmorillonite model as the first data set.

[0018] Furthermore, S3 specifically includes the following sub-steps:

[0019] S31. Divide the first data set into a training set and a test set according to a preset ratio;

[0020] S32. Write a second input parameter file, including the parameters related to the graph neural network model, as well as the AGAT layer, the number of neurons, the learning rate, the decay rate, and the number of training steps;

[0021] S33. Submit a task to train an atomic potential function model;

[0022] S34. Examine the quality of the atomic potential function model. Compare the calculation results of this model with the calculation results obtained only by first principles, analyze the error between the two, and determine whether the accuracy of the trained model meets the requirements of subsequent calculations. If the requirements are met, proceed to the next calculation; otherwise, continue training.

[0023] Further, S4 includes the following sub-steps:

[0024] S41. Perform molecular dynamics simulations using the trained atomic potential function model; at the same time, also use an empirical force field and set the same parameters as the atomic potential function model to perform classical molecular dynamics simulations;

[0025] S42. Obtain the results of the energy and force of the hydrated montmorillonite calculated by the two methods respectively, take the difference between the two results to form a second data set;

[0026] S43. Train the second data set using the deep potential energy method. The result obtained from the training is the force field correction term. Combine the correction term with the classical force field to obtain a force field file.

[0027] Further, S5 includes the following sub-steps:

[0028] S51. Establish an input parameter file for molecular dynamics software calculations, set periodic boundary conditions, select the parameters of the trained atomic potential function model and periodic boundary conditions, perform energy minimization optimization on the hydrated montmorillonite model, and assign initial velocities to atoms through the force field file for equilibrium relaxation;

[0029] S52. Perform uniaxial tensile and compression simulations on the hydrated montmorillonite model. The uniaxial tensile and compression simulations use the NVT ensemble. Achieve uniaxial tensile and compression simulations in the corresponding directions by defining the tensile and compression rates in the preset directions. At the same time, set the simulation parameters and obtain the mechanical properties of the hydrated montmorillonite according to the simulation results.

[0030] Further, the molecular dynamics software is LAMMPS molecular dynamics software.

[0031] Further, the results of the uniaxial tensile and compression simulations are the strain and stress information of the hydrated montmorillonite model in the tensile and compression directions. Plot the stress-strain curve and evaluate the elastic modulus and / or tensile strength of the hydrated montmorillonite according to the stress-strain curve.

[0032] The above solution of the present invention has the following beneficial effects:

[0033] An efficient prediction method for the mechanical properties of water-containing montmorillonite provided by the present invention uses a graph attention neural network. Under various environmental conditions (such as temperature, pressure, and stress state), an atomic potential function model is constructed and the empirical force field is corrected to accurately simulate the response behavior of its mechanical properties. At the same time, through an efficient numerical simulation technology, high-precision prediction and efficient analysis of the mechanical properties of clay minerals are realized. Compared with the traditional classical molecular dynamics method, the accuracy of this method is significantly improved, and it can more accurately capture the internal connection between the microscopic structure and macroscopic mechanical behavior of clay minerals. At the same time, compared with the computationally expensive first-principles calculation method, the present invention has greatly optimized the computational efficiency and can complete the simulation tasks of large-scale systems in a shorter time, providing strong technical support for the research and application in related fields.

[0034] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. Brief Description of the Drawings

[0035] Figure 1 is the step flow chart of the present invention;

[0036] Figure 2 is the specific step flow chart of S3 of the present invention;

[0037] Figure 3 is the stress-strain curve change diagram of water-containing Na-MMT under uniaxial tension conditions under different water contents of the present invention, where (a), (b), (c), and (d) represent water contents of 5%, 10%, 15%, and 20% respectively. Specific Embodiments

[0038] The following specific examples illustrate the embodiments of the present disclosure. Those skilled in the art can easily understand other advantages and effects of the present disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts belong to the scope of protection of the present disclosure.

[0039] Note that the following describes various aspects of embodiments within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on this disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement an apparatus and / or practice a method. Additionally, this apparatus can be implemented and this method can be practiced using other structures and / or functionality in addition to one or more of the aspects set forth herein.

[0040] It should also be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present disclosure schematically. The diagrams only show the components related to the present disclosure, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in its actual implementation can be arbitrarily changed, and the component layout type may also be more complex. Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0041] As Figure 1 shown, an embodiment of the present invention provides a method for efficiently predicting the mechanical properties of hydrated montmorillonite, specifically including the following steps:

[0042] S1. Use molecular dynamics modeling software to establish a hydrated montmorillonite model. Among them, the hydrated montmorillonite model is a visual three-dimensional model. This model takes atoms as basic units, and a crystal layer is composed of two Si-O tetrahedrons and one Al-O octahedron. This is the montmorillonite model. At the same time, a water molecule model is inserted in the middle of the crystal layer to form a hydrated montmorillonite model.

[0043] S2. Perform first-principles calculations on the hydrated montmorillonite model to obtain the first data set for training the atomic potential function model. This step specifically includes the following sub-steps:

[0044] S21. Convert the hydrated montmorillonite model into an input text file and import it into the first-principles calculation software. Among them, the input text file includes information such as atomic species, atomic coordinates, and atomic velocities.

[0045] S22. Prepare all the atomic species pseudopotential files and the first input parameter file required for the first-principles calculation. Among them, the pseudopotential file contains a virtual potential introduced during the simulation calculation to achieve an approximate calculation of a complex system. The parameters in the first input parameter file include temperature, pressure, number of calculation steps, etc.

[0046] S23. Perform a structural optimization operation on the water-containing montmorillonite model to obtain the energy-minimized configuration. At this time, set corresponding parameters such as the cutoff energy, calculation method, and number of calculation steps in the first input parameter file to control the structural optimization operation.

[0047] S24. Use the result file of the energy-minimized configuration as the input structure file for first-principles calculation. Set a new first input parameter file, and set the expected temperature, pressure, and number of calculation steps to perform uniaxial tensile and compression simulations.

[0048] S25. Obtain the configuration obtained from the first-principles calculation, and use it to obtain multiple snapshots of the water-containing montmorillonite model as the first data set.

[0049] S3. Train the atomic potential function model with the first data set, and check the quality of the atomic potential function model after training. Please also refer to Figure 2 , and this step specifically includes the following sub-steps:

[0050] S31. Use the GraNNField software to divide the first data set into a training set and a test set according to a specific ratio, such as 4:1.

[0051] S32. Write the second input parameter file, including parameters related to the graph neural network model such as atomic types and atomic interactions, as well as parameters such as the AGAT layer, number of neurons, learning rate, decay rate, and number of training steps.

[0052] S33. Submit the task and call the GATGNN model to start training the atomic potential function model.

[0053] S34. Check the quality of the atomic potential function model after training. Compare the calculation results of this model with the calculation results using only first principles, analyze the error between the two, so as to judge whether the accuracy of the trained model can meet the subsequent calculation requirements. If the requirements are met, the next calculation can be performed; otherwise, continue training.

[0054] S4. Perform molecular dynamics simulations using the trained atomic potential function model. At the same time, also use the empirical force field and set the same parameters as the atomic potential function model to perform classical molecular dynamics simulations to compare the result differences between the atomic potential function model and the empirical force field. This step specifically includes the following sub-steps:

[0055] S41. Use the GraNNField-LAMMPS interface to perform molecular dynamics simulations using the trained atomic potential function model; at the same time, also use the empirical force field and set the same parameters as the atomic potential function model to perform classical molecular dynamics simulations.

[0056] S42. Obtain the results of the energy and force of hydrated montmorillonite calculated by the two methods respectively, take the difference between the two results to form a second data set.

[0057] S43. Use the deep potential energy method to train the second data set. The result obtained from the training is the force field correction term. Combine the correction term with the classical force field to obtain a new force field file. Among them, the core of the deep potential energy method is to construct the mapping relationship of atomic energy, force, and potential energy surface through a deep neural network. By minimizing the prediction errors (loss functions) of energy, force, and virial tensor, the model is approximated to the quantum chemistry calculation results. Through this step, a force field that retains the accuracy of the first principle and the physical description of the clay mineral system can be obtained.

[0058] S5. Use the LAMMPS molecular dynamics software to perform uniaxial tensile and compression simulations on the hydrated montmorillonite model based on the force field file, and obtain the mechanical properties of hydrated montmorillonite according to the simulation results. This step specifically includes the following sub-steps:

[0059] S51. Establish an input parameter file for the LAMMPS molecular dynamics software calculation. In this file, set the periodic boundary conditions, select the trained atomic potential function model parameters and periodic boundary conditions, perform energy minimization optimization on the hydrated montmorillonite model, and assign initial velocities to the atoms through the force field file for equilibrium relaxation.

[0060] S52. Perform uniaxial tensile and compression simulations on the hydrated montmorillonite model. The uniaxial tensile and compression simulations adopt the NVT ensemble, and realize the uniaxial tensile and compression simulations in the corresponding directions by defining the tensile and compression rates in the preset directions, and set the simulation parameters at the same time.

[0061] The results of the uniaxial tensile and compression simulations are the strain and stress information of the hydrated montmorillonite model in the tensile and compression directions; plot the stress-strain curve, and evaluate the elastic modulus and / or tensile strength of the hydrated montmorillonite according to the stress-strain curve.

[0062] As described above, the method provided in this embodiment can, for clay mineral models with different water content characteristics, use the graph attention neural network to construct an atomic potential function model and correct the empirical force field under various environmental conditions (such as temperature, pressure, and stress state), accurately simulate the response behavior of its mechanical properties, and at the same time, through efficient numerical simulation techniques, achieve high-precision prediction and efficient analysis of the mechanical properties of clay minerals.

[0063] Compared with the traditional classical molecular dynamics method, this method has significantly improved simulation accuracy and can more accurately capture the internal connection between the microstructure and macroscopic mechanical behavior of clay minerals. At the same time, compared with the computationally expensive first-principles calculation method, this invention has greatly optimized the computational efficiency and can complete the simulation tasks of large-scale systems in a shorter time, providing strong technical support for research and applications in related fields.

[0064] The following further illustrates the effect of this method through specific cases. In S1, an aqueous montmorillonite model with the chemical formula Na 0.75 nH2OSi4[Al 3.5 Mg 0.5 O 20 (OH)4 is established using the amorphous cell module of Materials Studio software, and the water content of each montmorillonite unit cell is controlled to be 5%, 10%, 15% and 20%.

[0065] In S2, the aqueous montmorillonite model is imported into the VASP software, and the aqueous montmorillonite model is converted into a POSCAR text file. Prepare all the atomic species pseudopotential files POTCAR, K-point parameter files KPOINTS and the first input parameter file INCAR required for first-principles calculations. Perform a structural optimization operation on the aqueous montmorillonite model to obtain the energy-minimized configuration. In the INCAR file, set the cut-off energy to 650 eV, ISIF to 3, and IBRION to 2, that is, select the generalized gradient approximation method, choose 1000 calculation steps, and select the default values for the remaining parameters and then perform the structural optimization operation. After obtaining the energy-minimized configuration, use the CONTCAR file as the input structure file POSCAR for first-principles calculations and the lmp file for classical molecular dynamics calculations. In the first-principles calculations, set a new first input parameter file INCAR, set the expected temperature to 300 K, the pressure gradients to 0, 2, 4, 6, 8, 10 GPa respectively, and the calculation steps to 6000 for uniaxial tensile and compression simulations to obtain the output result files. Each output result file of the aqueous montmorillonite contains 6000 groups of data. Each group of calculated data contains parameters such as the coordinates, forces, and energies of the atoms. Using the last 4000 groups of stable calculation data, a total of 24000 steps of calculation data can be obtained, generating 24000 groups of snapshots for the first dataset of the graph neural network.

[0066] In S3, select the GATGNN model, and divide the first dataset into two parts, the training set and the test set, in a ratio of 4:1. Among them, there are 19200 groups of training set data and 4800 groups of test set data. Set the parameters related to the graph attention neural network model and the training hyperparameters. The cut-off radius is set to The size of the neural network is set to [25, 50, 100], the learning rate starts from 0.001, the termination learning rate is 0.000015, 5 AGAT layers are set, and there are 64 neurons. After submitting the task to start the training of the atomic potential function model, after the training is completed, using the GraNNField-LAMMPS interface, the trained atomic potential function model is used for molecular dynamics simulation of hydrated clay minerals. The simulation temperature is set to 300K, and the pressures of 5GPa, 10GPa, 15GPa, and 20GPa are selected to perform tensile and compression simulations on hydrated montmorillonite respectively. The number of calculation steps is set to 6000 steps, and two result values of energy and force of hydrated montmorillonite are obtained respectively. Then, with the same parameters set, classical molecular dynamics simulation is carried out using the empirical force field ClayFF and two result values are obtained. The difference between the result values of the two methods is formed into a second data set. Using the last 4000 groups of stable calculation data, a total of 16000 groups of data can be obtained. The deep potential energy method is used to train these 16000 groups of data, and the trained model is the correction term. The correction term is combined with the empirical force field to obtain a force field file that retains the accuracy of the first principles and the physical description of the clay mineral system.

[0067] In S4, the hydrated montmorillonite model is converted into a text file that can be recognized by the LAMMPS molecular dynamics software, and an input parameter file in.lammps of the LAMMPS molecular dynamics software is established. In this file, periodic boundary conditions are set, and the replicate instruction is used to perform unit cell expansion operation on the hydrated montmorillonite model to establish a montmorillonite supercell of 4×4×2 to achieve larger atomic-scale calculations. Using the LAMMPS molecular dynamics software, the trained potential function model parameters and periodic boundary conditions are selected to perform energy minimization optimization on the hydrated montmorillonite model. The pair_style hybrid / overlay instruction is set to use the trained force field file, initial velocities are assigned to the atoms, and equilibrium relaxation is performed. The equilibrium relaxation operation uses the npt ensemble, the relaxation temperature is 300K, and the relaxation time is 200ps. After the relaxation is completed, uniaxial tensile and compression simulations are carried out on the hydrated montmorillonite model. The uniaxial tensile simulation uses the nvt ensemble, and uniaxial tensile and compression simulations in the x, y, and z directions are achieved by defining the tensile and compression rates in a certain direction. The simulation temperature is 300K, and the strain rate of the simulation tensile and compression is 0.00001, and the maximum strain is 25%. The result of the uniaxial tensile simulation is the strain and stress information of the hydrated montmorillonite model after uniaxial tensile simulation in the tensile and compression directions; the stress-strain curve is plotted within the small strain range (0% - 3%), and the elastic modulus and / or tensile strength of the hydrated montmorillonite are evaluated according to the stress-strain curve.

[0068] Figure 3It is a graph showing the variation of the stress-strain curve of water-containing Na-MMT under uniaxial tension conditions for four water content conditions. From Figure 3 it can be seen that when the strain is small, the water-containing montmorillonite is in the elastic stage; after reaching the tensile limit, the water-containing montmorillonite undergoes brittle fracture. Moreover, the greater the water content, the smaller the elastic modulus and tensile strength of the water-containing montmorillonite. Therefore, the trend of the mechanical properties changing with the molecular weight is consistent with the actual trend that the greater the water content, the lower the tensile strength and elastic modulus.

[0069] Table 1 presents the comparison results of the calculation efficiency between the deep potential molecular dynamics and the first principles. Therefore, this method has a significantly improved calculation efficiency compared to the first principles. Under the same system and calculation conditions, the efficiency of this method has increased by nearly a hundred times.

[0070] Table 1 Comparison of the calculation efficiency between this method and the first principles for the montmorillonite system

[0071]

[0072] Table 2 shows the comparison of the Young's modulus results obtained by this method through the corrected force field, the empirical force field, and the first principles calculation. It can be seen that compared with the classical molecular dynamics simulation using the empirical force field, the machine learning force field shows higher accuracy in predicting the mechanical behavior of the system, and its calculation results have better consistency with the first principles calculation values.

[0073] Table 2 Comparison of the Young's modulus results obtained by using the corrected force field for the montmorillonite system

[0074]

[0075]

[0076] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0077] The above embodiments only represent several implementation manners of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application should be subject to the appended claims.

Claims

1. An efficient prediction method for the mechanical properties of hydrated montmorillonite, characterized in that, It includes the following steps: S1. Establish a hydrated montmorillonite model using molecular dynamics modeling software; S2. Perform first-principles calculations on the hydrated montmorillonite model to obtain the first data set for training the atomic potential function model; S3. Train the atomic potential function model with the first data set, and check the quality of the atomic potential function model after training; S4. Use the trained atomic potential function model for molecular dynamics simulation, and at the same time adopt the empirical force field, set the same parameters as the atomic potential function model for classical molecular dynamics simulation, so as to compare and calculate the result differences between the atomic potential function model and the empirical force field, and obtain the force field file; S5. Use molecular dynamics software to perform uniaxial tension and compression simulations on the hydrated montmorillonite model based on the force field file, and obtain the mechanical properties of the hydrated montmorillonite according to the simulation results.

2. The high-efficiency prediction method for the mechanical properties of water-containing montmorillonite according to claim 1, wherein The hydrated montmorillonite model takes atoms as the basic unit. The montmorillonite model consists of two Si-O tetrahedrons and one Al-O octahedron to form a crystal layer, and at the same time, a water molecule model is inserted in the middle of the crystal layer to form the hydrated montmorillonite model.

3. The high-efficiency prediction method for the mechanical properties of a water-containing montmorillonite according to claim 1, wherein S2 includes the following sub-steps: S21. Convert the hydrated montmorillonite model into an input text file and import it into the first-principles calculation software; S22. Prepare all atomic species pseudopotential files and the first input parameter file required for first-principles calculations; S23. Perform a structural optimization operation on the hydrated montmorillonite model to obtain the energy-minimized configuration; S24. Use the result file of the energy-minimized configuration as the input structure file for first-principles calculations, set a new first input parameter file, and set the expected temperature, pressure, and number of calculation steps for uniaxial tension and compression simulations; S25. Obtain the configurations obtained from first-principles calculations, and use them to obtain multiple snapshots of the hydrated montmorillonite model as the first data set.

4. The high-efficiency prediction method for the mechanical properties of water-containing montmorillonite according to claim 3, wherein, S3 specifically includes the following sub-steps: S31. Divide the first data set into a training set and a test set according to a preset ratio; S32. Write the second input parameter file, including parameters related to the graph neural network model, as well as the AGAT layer, the number of neurons, the learning rate, the decay rate, and the number of training steps; S33. Submit the task to train the atomic potential function model; S34. Check the quality of the atomic potential function model, compare the calculation results of this model with the calculation results using only first-principles, analyze the error between the two, and judge whether the accuracy of the trained model meets the subsequent calculation requirements. If it meets the requirements, proceed to the next calculation; otherwise, continue training.

5. The high-efficiency prediction method for the mechanical properties of hydrated montmorillonite according to claim 4, characterized in that, S4 includes the following sub-steps: S41. Use the trained atomic potential function model for molecular dynamics simulation; at the same time, also adopt the empirical force field, and set the same parameters as the atomic potential function model for classical molecular dynamics simulation; S42. Obtain the results of the energy and force of the hydrated montmorillonite calculated by the two methods respectively, take the difference between the two results to form the second data set; S43. Train the second data set using the deep potential energy method, and the result obtained from training is the force field correction term. Combine the correction term with the classical force field to obtain the force field file.

6. A method for efficiently predicting the mechanical properties of water-containing montmorillonite according to claim 5, characterized in that, S5 includes the following sub-steps: S51. Establish an input parameter file for molecular dynamics software calculations, set periodic boundary conditions, select the trained atomic potential function model parameters and periodic boundary conditions, perform energy minimization optimization on the water-containing montmorillonite model, assign initial velocities to atoms through a force field file, and perform equilibrium relaxation; S52. Conduct uniaxial tensile and compression simulations on the water-containing montmorillonite model. The uniaxial tensile and compression simulations use the NVT ensemble. Uniaxial tensile and compression simulations in the corresponding directions are achieved by defining the tensile and compression rates in the preset directions. At the same time, set the simulation parameters, and obtain the mechanical properties of the water-containing montmorillonite according to the simulation results.

7. A method for efficiently predicting the mechanical properties of a hydrated montmorillonite according to any one of claims 1-6, characterized in that, The molecular dynamics software is LAMMPS molecular dynamics software.

8. A method for efficiently predicting the mechanical properties of water-containing montmorillonite according to any one of claims 1-6, characterized in that, The results of the uniaxial tensile and compression simulations are the strain and stress information of the water-containing montmorillonite model in the tensile and compression directions. Plot the stress-strain curve, and evaluate the elastic modulus and / or tensile strength of the water-containing montmorillonite according to the stress-strain curve.

Images