Atomic scale characterization method and system for ultralight elements in crystalline materials

Through machine learning models combined with high-resolution electron microscopy technology and atomic scale simulation, the difficulty of characterizing ultralight elements in crystal materials is solved, atomic scale characterization and interaction tracking is achieved, and the understanding and regulation of material performance is improved.

CN118800369BActive Publication Date: 2024-11-29ZHEJIANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411278822.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-12
Publication Date
2024-11-29
Estimated Expiration
2044-09-12

AI Technical Summary

Technical Problem

The prior art is difficult to accurately characterize the positions of ultralight elements such as hydrogen, oxygen, nitrogen, boron, etc. in crystal materials and their interactions with crystal defects on the atomic scale, and traditional high-resolution electron microscopy technology and atomic scale calculation simulation methods It is difficult to describe complex local structures and dynamic evolution processes.

Method used

Based on machine learning models, combined with high-resolution electron microscopy technology and atomic scale simulation, the atomic scale characterization of ultra-light elements can be realized through image calibration, local atomic environment vector calculation, atomic scale simulation and machine learning model training, and the atomic scale characterization of ultra-light elements can be tracked in situ. Interaction of crystal defects.

Benefits of technology

The accuracy of the atomic-scale characterization of ultralight elements in crystal materials is realized and the complex local structure is described, providing an effective tool to understand the impact of ultralight elements on material performance, and improving the understanding and regulation of phenomena such as hydrogen embrittlement.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118800369B_ABST
    Figure CN118800369B_ABST
Patent Text Reader

Abstract

The present invention discloses an atomic scale characterization method and system for ultralight elements in crystalline materials, belonging to the field of micro-nanomechanical characterization technology. Based on the influence of ultralight atoms on the local arrangement of metal atoms, the present invention first trains a machine learning model through atomic scale simulation data, so that it can predict the plane distance scalar between the metal atom and its nearest ultralight atom according to the local atomic environment vector of the metal atom; then, for the gridded microscopic image, the distance value between each grid node and multiple nearest neighbor metal atoms is calculated to form the first characterization vector of the grid node, and at the same time, the plane distance scalar predicted by the machine learning model for these nearest neighbor metal atoms is used to form the second characterization vector of the grid node, and the distance value between the two characterization vectors is used as the criterion scalar; through interpolation completion and threshold method, the area rich in ultralight atoms is determined. The present invention can realize the atomic scale characterization of ultralight atoms and has broad application prospects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of micro-nanomechanical characterization technology, and specifically relates to a method and system for characterizing ultralight elements in crystalline materials based on machine learning, atomic scale simulation, and high-resolution electron microscopy technology. Background Art

[0002] In crystalline materials, ultralight elements with low atomic numbers such as hydrogen (H), oxygen (O), nitrogen (N), and boron (B) tend to accumulate in typical crystal defects such as vacancies, dislocations, grain boundaries, and cracks, which have a significant impact on the physical, chemical, and mechanical properties of the material. For example, hydrogen often accumulates at crystal defects in metal materials, causing a decrease in the mechanical properties of the material, a decrease in plasticity, cracking or damage (hydrogen embrittlement), and leading to failure of the metal structure. In order to understand the root cause of the influence of ultralight elements on the properties of crystal defects, it is necessary to accurately characterize the positions of ultralight atoms at the atomic scale to reveal their interaction mechanism with typical crystal defects.

[0003] The impact of ultralight elements on material properties has attracted widespread attention in the fields of energy, catalysis, engineering structures, etc., but the atomic-scale characterization methods for ultralight elements are still imperfect. At present, the characterization and research methods of ultralight elements at the atomic scale mainly include modern electron microscopy characterization technology and atomic-scale computational simulation technology. In terms of experiments, modern electron microscopy characterization technology represented by high-resolution transmission electron microscopy (HRTEM) and scanning transmission electron microscopy (STEM) has realized atomic-scale in-situ observation of the evolution process of crystal defects in metal materials. Existing high-resolution electron microscopy imaging technology is based on the deflection of electron beams by atoms. Compared with metal atoms, the interaction between ultralight atoms and electron beams is extremely weak, making it difficult to effectively image them in electron microscopy images. Electron microscopic characterization combined with integrated differential phase contrast technology (iDPC-STEM) can capture the positions of ultralight atoms such as hydrogen and oxygen in a few structurally stable and well-periodic metal compounds. However, the crystal defect structures that are widely present in actual materials are more complex, and existing technologies cannot yet achieve atomic-scale characterization of ultralight elements during the dynamic evolution of complex structures. In terms of atomic-scale computational simulation, computational methods represented by density functional theory and molecular dynamics can provide various key physical and mechanical information for the microscopic mechanisms of the interaction between ultralight elements and defects. Limited by modeling accuracy, atomic-scale computational simulations are difficult to fully consider the complex local structures such as distortions, interfaces, and steps in actual material defects. The over-simplification of local structures affects the reliability and universality of computational simulation conclusions. For crystal defects containing different microstructures, atomic-scale simulations require a large amount of computing resources to model and calculate them one by one.

[0004] In summary, the problems existing in the existing technology can be summarized as follows: 1) The problem that traditional high-resolution electron microscopy characterization methods are difficult to achieve atomic-scale characterization of ultralight elements in crystal defects. 2) The problem that traditional high-resolution electron microscopy characterization methods are difficult to track the interaction between ultralight elements and crystal defects during the deformation of crystal defects. 3) The problem that atomic computational simulation methods are difficult to describe the complex local structures observed in actual crystal defects. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems existing in the prior art and provide a method and system for atomic scale characterization of ultralight elements in crystalline materials. The present invention establishes a machine learning model based on atomic scale simulation data, considers the influence of ultralight elements on the local atomic arrangement in crystalline materials and their defects, and realizes atomic scale characterization of ultralight elements in combination with electron microscopy.

[0006] The specific technical solutions adopted by the present invention are as follows:

[0007] In a first aspect, the present invention provides an atomic scale characterization method for ultralight elements in a crystalline material, comprising:

[0008] S1. Calibrate the image size of the crystal defect microscopic image of the metal material subjected to the external load so that the lattice spacing in the image coordinate system is equal to the actual theoretical calculated value, and use it as the calibrated image; establish an image coordinate system for the calibrated image, and identify the two-dimensional image coordinates of each metal atom in the image; grid the calibrated image to obtain uniformly distributed grid nodes;

[0009] S2, based on the two-dimensional image coordinates of all metal atoms in the calibrated image, calculate the distance values ​​of each metal atom and its m nearest neighboring metal atoms according to a first preset value m, to form a first local atomic environment vector corresponding to the metal atom;

[0010] S3. Based on the atomic scale model constructed according to the crystal defect information in the calibrated image, ultralight atoms of different contents are introduced at different positions near the crystal defect, thereby forming a series of simulation scenarios; in each simulation scenario, a simulated external load corresponding to the external load is applied, and a simulation based on density functional theory and / or molecular dynamics is performed to update the atomic arrangement and deformation state of the local area where the crystal defect is located in the atomic scale model;

[0011] S4. For each atomic scale model updated in each simulation scenario, recalculate the distance value between each metal atom and its m nearest neighboring metal atoms to form the second local atomic environment vector corresponding to the metal atom, and calculate the plane distance scalar between the metal atom and its nearest ultralight neighboring atom; use the second local atomic environment vector and plane distance scalar of each metal atom in each simulation scenario as the model input and output labels, respectively, construct training samples to add to the training data set, and train the machine learning model;

[0012] S5, inputting the first local atomic environment vector corresponding to each metal atom in the calibrated image into the trained machine learning model to predict the plane distance scalar corresponding to the metal atom;

[0013] S6, based on the two-dimensional image coordinates of all metal atoms in the calibrated image, according to the second preset value n, calculate the distance value between each grid node and its n nearest neighboring metal atoms to form a first characterization vector corresponding to the grid node, and at the same time use the plane distance scalar predicted by the machine learning model for the n nearest neighboring metal atoms to form a second characterization vector corresponding to the grid node, and use the distance value between the first characterization vector and the second characterization vector as a criterion scalar;

[0014] S7. Based on the criterion scalars of all grid nodes in the calibrated image, the criterion scalars of all positions in the image domain are obtained by interpolation and completion, and the positions where the criterion scalars do not exceed the judgment threshold are marked as regions rich in ultralight atoms.

[0015] As a preferred embodiment of the first aspect, the crystal defect microscopic image is an image of a metal material nanosample containing crystal defects captured by a high-resolution transmission electron microscope HRTEM, a scanning transmission electron microscope STEM or an atomic force microscope AFM.

[0016] As a preferred embodiment of the above-mentioned first aspect, the machine learning model is a feedforward neural network, a support vector machine, a convolutional neural network or a linear regression model.

[0017] As a preferred embodiment of the above-mentioned first aspect, the crystal defect information required for constructing the atomic scale model includes the element type, crystal structure, defect type and crystal defect geometry information extracted from the calibrated image.

[0018] As a preferred embodiment of the first aspect, the first preset value m is 10-30.

[0019] As a preferred embodiment of the first aspect, the second preset value n is 2-8.

[0020] As a preferred embodiment of the first aspect, the distance value between the first characterization vector and the second characterization vector is: calculating the difference vector between the first characterization vector and the second characterization vector, and then dividing the difference vector by its dimension value to obtain the modulus length.

[0021] In a second aspect, the present invention provides an atomic scale characterization system for ultralight elements in crystalline materials, comprising:

[0022] An image calibration module is used to calibrate the image size of a crystal defect microscopic image of a metal material subjected to an external load so that the lattice spacing in the image coordinate system is equal to the actual theoretical calculated value, which is used as a calibrated image; an image coordinate system is established for the calibrated image to identify the two-dimensional image coordinates of each metal atom in the image; and the calibrated image is gridded to obtain uniformly distributed grid nodes;

[0023] A local atomic environment vector calculation module is used to calculate the distance values ​​of each metal atom and its m nearest neighboring metal atoms based on the two-dimensional image coordinates of all metal atoms in the calibrated image according to a first preset value m, so as to form a first local atomic environment vector corresponding to the metal atom;

[0024] an atomic scale simulation module, for introducing ultralight atoms of different contents at different positions near the crystal defect based on the atomic scale model constructed according to the crystal defect information in the calibrated image, thereby forming a series of simulation scenarios; in each simulation scenario, applying a simulated external load corresponding to the external load, and performing a simulation based on density functional theory and / or molecular dynamics to update the atomic arrangement and deformation state of the local area where the crystal defect is located in the atomic scale model;

[0025] A machine learning module is used to update the atomic scale model for each simulation scenario, recalculate the distance value between each metal atom and its m nearest neighboring metal atoms, construct the second local atomic environment vector corresponding to the metal atom, and calculate the plane distance scalar between the metal atom and its nearest ultralight neighboring atom; the second local atomic environment vector and plane distance scalar of each metal atom in each simulation scenario are used as model input and output labels, respectively, to construct training samples to be added to the training data set, and train the machine learning model;

[0026] A plane distance scalar prediction module, used for inputting the first local atomic environment vector corresponding to each metal atom in the calibrated image into the trained machine learning model to predict the plane distance scalar corresponding to the metal atom;

[0027] a criterion scalar calculation module, for calculating the distance values ​​of each of the grid nodes and its n nearest neighboring metal atoms based on the two-dimensional image coordinates of all metal atoms in the calibrated image and according to a second preset value n, to form a first characterization vector corresponding to the grid node, and at the same time, using the plane distance scalar predicted by the machine learning model for the n nearest neighboring metal atoms to form a second characterization vector corresponding to the grid node, and using the distance value between the first characterization vector and the second characterization vector as the criterion scalar;

[0028] The ultralight atom region marking module is used to obtain the criterion scalars of all positions in the image domain through interpolation and completion based on the criterion scalars of all grid nodes in the calibrated image, and mark the positions where the criterion scalars do not exceed the judgment threshold as regions rich in ultralight atoms.

[0029] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the atomic scale characterization method of ultralight elements in crystalline materials as described in any one of the first aspects above can be implemented.

[0030] In a fourth aspect, the present invention provides a computer electronic device comprising a memory and a processor;

[0031] The memory is used to store computer programs;

[0032] The processor is used to implement the atomic scale characterization method of ultralight elements in crystalline materials as described in any one of the first aspects above when executing the computer program.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1) Traditional high-resolution electron microscopy characterization methods can only effectively image atoms with larger atomic numbers, such as metal atoms, but cannot achieve stable imaging of ultralight atoms. The method of the present invention is based on the influence of ultralight atoms on the local arrangement of metal atoms, and uses machine learning methods to predict their positions, thus achieving atomic-scale characterization of ultralight atoms.

[0035] 2) In order to understand the microscopic mechanism of the interaction between ultralight atoms and crystal defects, traditional atomic simulation or theoretical modeling methods need to greatly simplify the crystal defect model, which is difficult to fully describe the complex microstructure of crystal defects in actual materials. The method of the present invention directly uses high-resolution electron microscopy images as input, and the generalization performance of the machine learning model is good. It can take into account various local structures in the deformation process of crystal defects and realize the direct characterization of ultralight atoms in real structures.

[0036] 3) The method of the present invention can be used to characterize the position of ultralight atoms in real time during in-situ high-resolution characterization to track the microscopic mechanism of interaction between ultralight atoms and crystal defects, providing an efficient and accurate tool for understanding and regulating typical physical, chemical, and mechanical phenomena caused by ultralight atoms such as hydrogen embrittlement. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the basic steps of the atomic-scale characterization method for ultralight elements in crystalline materials.

[0038] Figure 2 Diagram of the modular composition of the atomic-scale characterization system for ultralight elements in crystalline materials.

[0039] Figure 3 A schematic diagram of the components of computer electronic equipment.

[0040] Figure 4 This is a flow chart of the atomic scale characterization of ultralight elements in crystalline materials in an embodiment of the present invention.

[0041] Figure 5 Schematic diagram of the structure of the feedforward neural network used in the embodiment of the present invention;

[0042] Figure 6 Schematic diagram of some key parameter calculation methods in the embodiments of the present invention. (a) The calculation principle of vector; (b) is Scalar calculation principle; (c) Principles of vector calculation.

[0043] Figure 7 In the embodiment of the present invention, based on The Ni-H system hydrogen atom recognition effect and its verification based on the criterion; (a) is the recognition effect of hydrogen atoms in the system gap, (b) is the true position of hydrogen atoms corresponding to (a), Contour lines mark areas rich in ultralight atoms. The contour lines mark the specific positions of the hydrogen atoms. (c) is the identification effect of interstitial hydrogen atoms near the polycrystalline grain boundary, and (d) is a comparison of the actual positions of the hydrogen atoms corresponding to (c).

[0044] Figure 8 3 is a characterization of the segregated hydrogen atoms at the grain boundaries in the face-centered cubic metal Ni in the embodiment of the present invention; wherein (a) is a HRTEM image of a 24°

[110] asymmetric tilted grain boundary at a hydrogen atom concentration of 6.1%, (b) is a hydrogen atom characterization result corresponding to (a), (c) is a HRTEM image of the grain boundary at a hydrogen atom concentration of 29.9%, and (d) is a hydrogen atom characterization result corresponding to (c).

[0045] Fig. 9 In the embodiment of the present invention, based on The identification effect of hydrogen atoms in the Fe-Cr-H tilted grain boundary based on the criterion and its verification; the hydrogen atoms are distributed at the grain boundary and the tetrahedral voids inside the grain, (a1) and (a2) are Fe 80 Cr 20 Materials and Fe 90 Cr 10 HRTEM images of the Fe-Cr-H tilted grain boundaries in the material, (b1) and (b2) are Fe 80 Cr 20 Materials and Fe 90 Cr 10 The predicted values ​​of the scalar in the material, (c1) and (c2) are Fe 80 Cr 20 Materials and Fe 90 Cr 10 The recognition effect of hydrogen atoms in the material, (d1) and (d2) are Fe 80 Cr 20 Materials and Fe 90 Cr 10 This is the actual position of hydrogen atoms in the material.

[0046] Fig.10 In the embodiment of the present invention, based on The identification effect of Ti-TiH phase interface based on the judgment criteria and its verification; (a) is the HRTEM image of Ti-TiH phase interface, (b) is The predicted value of the scalar, (c) is the recognition effect of the phase interface position, and (d) is the actual position of the hydrogen atom. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0048] In the description of the present invention, it is to be understood that when an element is considered to be "connected" to another element, it may be directly connected to the other element or indirectly connected, that is, there are intermediate elements. On the contrary, when an element is said to be "directly" connected to another element, there are no intermediate elements.

[0049] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0050] In a preferred embodiment of the present invention, a method for atomic-scale characterization of ultralight elements in crystalline materials is provided. The method establishes a machine learning model based on atomic-scale simulation data, considers the influence of ultralight elements such as hydrogen, nitrogen, oxygen, and boron on the local atomic arrangement in crystalline materials and their defects, and realizes atomic-scale characterization of ultralight elements in combination with high-resolution electron microscopy. Figure 1 As shown, the specific steps of the atomic scale characterization method of ultralight elements in the crystal material include S1 to S7, and the specific implementation of each step is described in detail below.

[0051] S1. Calibrate the image size of the crystal defect microscopic image of the metal material under external load so that the lattice spacing in the image coordinate system is equal to the actual theoretical calculated value, which is used as the calibrated image. Then establish the image coordinate system for the calibrated image and identify the two-dimensional image coordinates of each metal atom in the image; at the same time, grid the calibrated image to obtain evenly distributed grid nodes.

[0052] It should be noted that the crystal defect microscopic image in the present invention is a microscopic image of a metal material with crystal defects taken by electron microscopic characterization technology. In an embodiment of the present invention, the metal material containing crystal defects can be made into a nano-sample, and then the corresponding image is obtained by photographing it by means of a high-resolution transmission electron microscope HRTEM, a scanning transmission electron microscope STEM or an atomic force microscope AFM, preferably a high-resolution transmission electron microscopic image. In addition, when taking a microscopic image of the nano-sample, the external load required for the test can be applied first and then photographed according to the needs of the test, or the external load required for the test can be applied during the photographing process. The external load information (including load form, load direction and maximum strain) when taking the microscopic image can be recorded in advance to facilitate the application of the corresponding simulated external load during the subsequent simulation. In an embodiment of the present invention, the applied external load can be uniaxial tension. Compression strain and other forms.

[0053] It should also be noted that the calibration of the image size in the present invention needs to be based on the lattice spacing in the image coordinate system being equal to the theoretical value or calculated value of the lattice spacing in the actual metal material. Since factors such as resolution, camera internal and external parameters will affect the size corresponding to a single pixel when the image is captured, it is necessary to measure the lattice spacing in the image coordinate system after obtaining the crystal defect microscopic image, and then determine the image scaling ratio based on the lattice spacing measurement value in the image coordinate system and the corresponding lattice spacing theoretical calculation value, and finally scale the crystal defect microscopic image according to the ratio to obtain the calibrated image, so that the lattice spacing directly measured in the image coordinate system can meet its theoretical calculation value. Therefore, the remaining distance values ​​measured based on the calibrated image can also be used to reflect its actual physical size in the metal material.

[0054] It should also be noted that each metal atom is a two-dimensional circular area in the calibrated image, and its two-dimensional image coordinates can be replaced by the image coordinates of the point where the center of the circle is located.

[0055] It should also be noted that when the calibrated image is gridded, the specific grid size can be optimized according to the actual image size and the accuracy requirements of the subsequent simulation. The entire image can be evenly divided into grids in the horizontal and vertical directions, and a grid intersection in the horizontal and vertical directions is a grid node.

[0056] S2. Based on the two-dimensional image coordinates of all metal atoms in the calibrated image, the distance values ​​between each metal atom and its m nearest neighboring metal atoms are calculated according to a first preset value m to form a first local atomic environment vector corresponding to the metal atom.

[0057] It should be noted that the first preset value m in the present invention is a parameter that can be optimized according to the actual characterization effect. In the embodiment of the present invention, after parameter optimization, the first preset value m is preferably 10-30.

[0058] In the above step S2, it is necessary to calculate the distance between each pair of metal atoms in the calibrated image, and then traverse each metal atom in the calibrated image. For any metal atom currently traversed, the m metal atoms closest to itself can be determined, and these m metal atoms are the nearest neighbor metal atoms. The m distance values ​​between the currently traversed metal atom and the m nearest neighbor metal atoms constitute the first local atomic environment vector corresponding to the currently traversed metal atom, and the dimension of the first local atomic environment vector is m. In the above calibrated image, the distance value between any two metal atoms can be calculated by the two-dimensional image coordinates of the two metal atoms in the calibrated image to obtain the corresponding spatial distance. It should be noted that in order to ensure the uniformity of subsequent model inputs, when the m distance values ​​constitute the first local atomic environment vector, they can be uniformly sorted according to fixed rules, for example, they can be sorted from small to large or from large to small according to the distance value.

[0059] S3. Based on the atomic scale model constructed according to the crystal defect information in the above-mentioned calibrated image, ultralight atoms of different contents are introduced at different positions near the crystal defect to form a series of simulation scenarios. In each simulation scenario, a simulated external load corresponding to the external load when the aforementioned microscopic image is taken is applied, and a simulation based on density functional theory and / or molecular dynamics is performed to update the atomic arrangement and deformation state of the local area where the crystal defect is located in the atomic scale model.

[0060] It should be noted that the above atomic scale model is constructed based on the crystal defect information obtained from the calibrated image. The atomic scale model construction method required for studying crystal defects belongs to the prior art and can be implemented by referring to any method in the prior art.

[0061] Before building an atomic-scale model, it is necessary to analyze the basic information of crystal defects from the calibrated image, including element type, crystal structure, main defect types and their corresponding geometric information. Defect types include point defects such as vacancies, interstitial atoms, and substitutional atoms, line defects such as screw dislocations, edge dislocations, and mixed dislocations, interface defects such as stacking faults, grain boundaries, twin boundaries, and phase boundaries, as well as complex defects such as voids, precipitate phases, cracks, and shear bands. Geometric information includes macroscopic geometric parameters necessary for building atomic models, such as crystal orientation, interface inclination, dislocation type, defect morphology, and geometric dimensions, but does not require the precise coordinates of each atom.

[0062] The construction method of the atomic scale model can be summarized as follows: in software environments such as MATLAB, Ovito, and Atomsk, set the element type, defect geometry and model size information to build an atomic scale model. Defect geometry includes macroscopic geometric parameters necessary for building an atomic model, such as crystal orientation, interface inclination, dislocation type, defect morphology, and geometric size. The atomic scale model is essentially a model that stores the type and position coordinates of each metal atom.

[0063] Once the atomic scale model is constructed, atomic scale simulation can be performed in software environments such as LAMMPS and VASP based on density functional theory, molecular dynamics, or a combination of the two to determine the atomic arrangement and deformation state of the crystal defect region in the atomic scale model under the external loads applied when taking the microscopic image of the metal material nano-sample. When atoms are rearranged or deformed, the two-dimensional image coordinates of some atoms in the atomic scale model will change, so each simulation scenario will result in an atomic scale model with updated atomic coordinates after the simulation.

[0064] S4. For each simulation scenario, the atomic-scale model is updated, and the distance values ​​between each metal atom and its m nearest neighboring metal atoms are recalculated to form the second local atomic environment vector corresponding to this metal atom. At the same time, the plane distance scalar between this metal atom and its nearest ultralight neighboring atoms is calculated. The second local atomic environment vector and plane distance scalar of each metal atom in each simulation scenario are used as model input and output labels, respectively, to construct training samples and add them to the training data set, and train the machine learning model.

[0065] It should be noted that in the above-mentioned S4 step, the calculation method of the second local atomic environment vector is the same as the aforementioned first local atomic environment vector, and the difference between the two is that the calculated data is different. For any kind of atomic scale model updated for a simulation scenario. When calculating the second local atomic environment vector of each metal atom, it is necessary to calculate the distance between the metal atoms in the atomic scale model updated, and then traverse each metal atom in the atomic scale model updated, for any metal atom currently traversed, the m metal atoms closest to itself can be determined, and these m metal atoms are the nearest neighbor metal atoms. The m distance values ​​between the metal atom currently traversed and the m nearest neighbor metal atoms in the atomic scale model updated constitute the second local atomic environment vector corresponding to the metal atom currently traversed, and the dimension of the second local atomic environment vector is also m. Similarly, when the m distance values ​​constitute the second local atomic environment vector, it is also necessary to uniformly sort them in a way from small to large or from large to small according to the distance value.

[0066] In addition, in the above S4 step, for each metal atom, in addition to calculating the second local atomic environment vector corresponding to the metal atom, it is also necessary to find the ultralight atom that is the nearest neighbor of the metal atom in the simulation results of the corresponding simulation scenario, and calculate the plane distance scalar between the metal atom and its nearest neighbor ultralight atom based on the two-dimensional image coordinates of the two. The nearest neighbor ultralight atom of a metal atom refers to the ultralight atom that is closest to the metal atom after a certain content (the value range is 0% - the theoretical saturation concentration that can fill all gaps) of the ultralight atom to be studied (hydrogen, nitrogen, oxygen, boron, etc.) is introduced at a random position during the simulation.

[0067] Therefore, each simulation scenario corresponds to a specific introduction content and introduction position when ultralight atoms are introduced. An atomic-scale model is obtained through simulation update, and each metal atom in this atomic-scale model can correspond to the second local atomic environment vector and the plane distance scalar. The second local atomic environment vector and the plane distance scalar of each metal atom in each simulation scenario are used as model input and output labels, respectively, to construct a training sample. The training samples of all metal atoms in all simulation scenarios constitute a training data set. This training data set can be used to train a machine learning model so that it can input the second local atomic environment vector of metal atoms, quickly predict the plane distance scalar of metal atoms, and reduce the amount of calculation required for simulation.

[0068] In the present invention, the machine learning model can adopt any model that can predict the plane distance scalar according to the second local atomic environment vector, such as a feedforward neural network, a support vector machine, a convolutional neural network or a linear regression model can realize this function. Of course, the machine learning model can also be realized by other neural networks such as deep learning, and there is no restriction on this. The training of the machine learning model belongs to the prior art. By setting a loss function, based on the training data set, the loss is calculated forward, and the network parameters are updated in reverse, and the optimization of the learnable parameters of the model is iteratively realized. When the convergence condition is reached, the iterative learning can be terminated. Of course, considering the computing resources required for atomic scale simulation and the training requirements of machine learning, the scale of the training data set should not be too large or too small. In actual implementation, a small-scale atomic scale simulation can be carried out first, and a preliminary training data set can be constructed. And according to the performance of the trained model on the test set, it is decided whether it is necessary to increase the scale of atomic scale simulation.

[0069] S5. Input the first local atomic environment vector corresponding to each metal atom in the calibrated image into the trained machine learning model to predict the plane distance scalar corresponding to the metal atom.

[0070] It should be noted that the trained machine learning model needs to achieve the corresponding prediction accuracy on the test set before it can be used for the actual prediction in this step. Generally speaking, when training the machine learning model on the aforementioned training data set, its input data needs to be normalized. If the training data has been normalized, the first local atomic environment vector in this step also needs to be normalized in the same way to ensure that accurate prediction output can be obtained.

[0071] S6. Based on the two-dimensional image coordinates of all metal atoms in the above-mentioned calibrated image, according to the second preset value n, calculate the distance values ​​of each grid node and its n nearest neighbor metal atoms to form a first characterization vector corresponding to this grid node. At the same time, the plane distance scalar predicted by the machine learning model for these n nearest neighbor metal atoms constitutes the second characterization vector corresponding to this grid node, and the distance value between the first characterization vector and the second characterization vector is used as the criterion scalar.

[0072] It should be noted that the second preset value n is a parameter that can be optimized according to the actual characterization effect. In the embodiment of the present invention, after parameter optimization, the second preset value n is preferably 2-8.

[0073] In the above step S6, it is also necessary to first calculate the distance from each grid node to all metal atoms in the calibrated image, and then traverse each grid node in the calibrated image. For any currently traversed grid node, the n metal atoms closest to itself can be determined, and these n metal atoms are the nearest neighbor metal atoms. The n distance values ​​between the currently traversed grid node and the n nearest neighbor metal atoms constitute the first characterization vector corresponding to the currently traversed grid node, and the dimension of the first characterization vector is n. In the above calibrated image, the distance value between any grid node and the metal atom can be calculated from the two-dimensional image coordinates of the grid node to the center of the metal atom in the calibrated image to obtain the corresponding spatial distance. It should be noted that in order to ensure the uniformity of subsequent model inputs, when the n distance values ​​constitute the first characterization vector, they can be uniformly sorted according to fixed rules, for example, they can be sorted from small to large or from large to small according to the distance value. In addition, when constructing the second characterization vector corresponding to the grid node, the plane distance scalars of the n nearest neighbor metal atoms closest to the grid node (previously predicted by the aforementioned machine learning model in S5) can be used to form the second characterization vector in the same order as the first characterization vector, that is, if each metal atom in the calibrated image has a unique index, then the index sequence of the n metal atoms corresponding to the n distance values ​​in the first characterization vector is exactly the same as the index sequence of the n metal atoms corresponding to the n plane distance scalars in the second characterization vector. This ensures that the first characterization vector and the second characterization vector are comparable.

[0074] It should also be noted that the scalar used as the criterion is the distance value calculated based on the first characterization vector and the second characterization vector, but the distance value between two vectors can be calculated in a variety of ways, such as Euclidean distance, cosine distance, dot product, etc., which can be optimized and selected based on actual conditions.

[0075] S7. Based on the criterion scalars of all grid nodes in the calibrated image, the criterion scalars of all positions in the image domain are obtained by interpolation and completion, and the positions where the criterion scalars do not exceed the judgment threshold are marked as areas rich in ultralight atoms.

[0076] It should be noted that the purpose of the above interpolation completion is to encrypt the sparsely sampled data in the image space, because the above calculation process only obtains the criterion scalar of the grid node, and there is no criterion scalar at the non-grid node position. Of course, in theory, if the grid is dense enough, there is no need for interpolation completion, but this will consume a lot of computing resources and take too long. Therefore, in general, in this step, linear interpolation or other interpolation methods can be used to complete the criterion scalars at other non-grid nodes on the image based on the criterion scalars of the existing grid nodes, so as to facilitate the subsequent determination of the position of the ultralight atoms.

[0077] It should also be noted that in the present invention, the criterion scalar at each position on the image is a landmark value used to characterize whether there is an ultralight atom at this position. The smaller the value, the more likely it is that there is an ultralight atom at this position. Therefore, the present invention can be based on the criterion scalar and combined with the threshold method to realize the judgment of the ultralight atom-rich area. The specific method is to optimize an optimal threshold value based on the measured data. When the criterion scalars of all positions in a certain area are less than the judgment threshold, the area can be regarded as an ultralight atom-rich area. In actual implementation, the area rich in ultralight atoms can be divided by the contour method. Moreover, in addition to marking the area rich in ultralight atoms, in the embodiment of the present invention, the marked area can be reduced to the scale of a single atom by further reducing the judgment threshold, thereby determining the specific position of the superhydrogen atom.

[0078] It should be noted that the above-mentioned judgment threshold is a relative value. When the distance value calculation method between the first characterization vector and the second characterization vector is different, the judgment threshold to be selected is also different. Therefore, the specific value of the judgment threshold needs to be adjusted and optimized on the measured data according to the calculation method of the criterion scalar, and finally the best value is determined. In an embodiment of the present invention, the distance value between the first characterization vector and the second characterization vector can be obtained by first calculating the difference vector between the first characterization vector and the second characterization vector, and then dividing the difference vector by its dimension value (i.e., the second preset value n) to obtain the modulus length, which is recorded as the distance value of the two characterization vectors. Based on this distance value calculation method, in an embodiment of the present invention, the judgment threshold in the calibrated image not exceeding The area is determined to be rich in ultralight atoms, and the threshold value of the calibrated image is not more than The area is determined as the specific location of the superhydrogen atom.

[0079] It can be seen that the existing traditional high-resolution electron microscopy methods can usually only achieve stable and clear imaging of atoms with larger atomic numbers such as metal atoms, while it is difficult to reliably characterize ultra-light atoms such as hydrogen atoms that are gathered in large quantities at crystal defects. The method described in S1 to S7 of the present invention derives the position of ultra-light atoms from the influence of ultra-light atoms on the local atomic arrangement, and can achieve ultra-light atomic characterization of complex crystal defects during deformation. This method does not require hardware improvements to existing electron microscopy characterization equipment, so it is easy to apply and promote. Moreover, compared with traditional computational simulation methods that require simplified modeling or one-to-one modeling of the crystal defects to be studied, when the range of sample coverage in the training data set is wide enough, the machine learning prediction model obtained by the method of the present invention will have good generalization ability, and a single machine learning model can achieve the characterization of multiple crystal defects under different ultra-light atomic distributions and different deformation states.

[0080] In addition, it should be noted that in the atomic scale characterization of ultralight elements in the crystalline material described in the above embodiment, each step S1 to S7 can be implemented by means of program code.

[0081] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides an atomic scale characterization system for ultralight elements in crystalline materials corresponding to the atomic scale characterization method for ultralight elements in crystalline materials provided in the above embodiment, such as Figure 2 As shown, the system includes:

[0082] An image calibration module is used to calibrate the image size of a crystal defect microscopic image of a metal material subjected to an external load so that the lattice spacing in the image coordinate system is equal to the actual theoretical calculated value, which is used as a calibrated image; an image coordinate system is established for the calibrated image to identify the two-dimensional image coordinates of each metal atom in the image; and the calibrated image is gridded to obtain uniformly distributed grid nodes;

[0083] A local atomic environment vector calculation module is used to calculate the distance values ​​of each metal atom and its m nearest neighboring metal atoms based on the two-dimensional image coordinates of all metal atoms in the calibrated image according to a first preset value m, so as to form a first local atomic environment vector corresponding to the metal atom;

[0084] an atomic scale simulation module, for introducing ultralight atoms of different contents at different positions near the crystal defect based on the atomic scale model constructed according to the crystal defect information in the calibrated image, thereby forming a series of simulation scenarios; in each simulation scenario, applying a simulated external load corresponding to the external load, and performing a simulation based on density functional theory and / or molecular dynamics to update the atomic arrangement and deformation state of the local area where the crystal defect is located in the atomic scale model;

[0085] A machine learning module is used to update the atomic scale model for each simulation scenario, recalculate the distance value between each metal atom and its m nearest neighboring metal atoms, construct the second local atomic environment vector corresponding to the metal atom, and calculate the plane distance scalar between the metal atom and its nearest ultralight neighboring atom; the second local atomic environment vector and plane distance scalar of each metal atom in each simulation scenario are used as model input and output labels, respectively, to construct training samples to be added to the training data set, and train the machine learning model;

[0086] A plane distance scalar prediction module, used for inputting the first local atomic environment vector corresponding to each metal atom in the calibrated image into the trained machine learning model to predict the plane distance scalar corresponding to the metal atom;

[0087] a criterion scalar calculation module, for calculating the distance values ​​of each of the grid nodes and its n nearest neighboring metal atoms based on the two-dimensional image coordinates of all metal atoms in the calibrated image and according to a second preset value n, to form a first characterization vector corresponding to the grid node, and at the same time, using the plane distance scalar predicted by the machine learning model for the n nearest neighboring metal atoms to form a second characterization vector corresponding to the grid node, and using the distance value between the first characterization vector and the second characterization vector as the criterion scalar;

[0088] The ultralight atom region marking module is used to obtain the criterion scalars of all positions in the image domain through interpolation and completion based on the criterion scalars of all grid nodes in the calibrated image, and mark the positions where the criterion scalars do not exceed the judgment threshold as regions rich in ultralight atoms.

[0089] The specific implementation method of each functional module in the above system is the same as the corresponding method steps in the above embodiment, and will not be repeated here.

[0090] Similarly, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer electronic device corresponding to the atomic scale characterization method of ultralight elements in crystalline materials provided in the above embodiment, such as Figure 3 As shown, the computer electronic device includes a memory and a processor;

[0091] The memory is used to store computer programs;

[0092] The processor is used to implement the atomic scale characterization method of ultralight elements in crystalline materials as described above when executing the computer program.

[0093] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0094] Therefore, based on the same inventive concept, another preferred embodiment of the present invention also provides a computer-readable storage medium corresponding to the atomic scale characterization method of ultralight elements in crystal materials provided in the above embodiment, and a computer program is stored on the storage medium. When the computer program is executed by the processor, it can implement the atomic scale characterization method of ultralight elements in crystal materials as described above.

[0095] It is understandable that the above storage medium and memory may be random access memory (RAM) or non-volatile memory (NVM), such as at least one disk memory. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0096] It can be understood that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0097] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In each embodiment provided in this application, the division of steps or modules in the device and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0098] The present invention will further demonstrate the detailed implementation process and technical effects of the atomic scale characterization method of ultralight elements in crystalline materials shown in the above steps S1 to S7 in a specific example through a specific embodiment, so as to facilitate understanding of the essence of the present invention.

[0099] Example

[0100] In an embodiment, the atomic scale characterization method of ultralight elements in a crystalline material is specifically implemented as follows: Figure 4 As shown, the specific steps are as follows:

[0101] Step 1: For the metal material nano-specimen during or after the external load is applied, the high-resolution electron microscopic image of crystal defects obtained by the HRTEM method is calibrated, and the lattice spacing in the image coordinate system is equal to the theoretical value or calculated value by proportionally enlarging or reducing the image size. The calibrated image is called the calibrated image. In addition, the basic information of the external load corresponding to the external load applied to the nano-specimen is recorded, including the load form, load direction, and maximum strain.

[0102] Step 2: Analyze the calibrated image to obtain basic information about crystal defects, including element type, crystal structure, main defect types and their corresponding geometric information. Defect types include point defects such as vacancies, interstitial atoms, and substitutional atoms, line defects such as screw dislocations, edge dislocations, and mixed dislocations, interface defects such as stacking faults, grain boundaries, twin boundaries, and phase boundaries, as well as complex defects such as voids, precipitate phases, cracks, and shear bands. Geometric information includes macroscopic geometric parameters necessary for constructing atomic models, such as crystal orientation, interface inclination, dislocation type, defect morphology, and geometric dimensions, but does not require the precise coordinates of each atom.

[0103] Step 3: Identify the coordinates of metal atoms in the calibrated image, establish a rectangular coordinate system in the image plane, identify the coordinates of each metal atom based on the center of the atom, and obtain the two-dimensional coordinates of each metal atom in the image plane. In addition, the horizontal and vertical grids are evenly divided at intervals of 0.5 angstroms in the horizontal and vertical directions, so that evenly distributed grid nodes are obtained at the intersection of the horizontal and vertical grids. A single grid node is denoted as N ij , where i and j are the horizontal and vertical serial numbers of the node respectively.

[0104] Step 4: Calculate the local atomic environment (LAE) vector of each metal atom , the LAE matrix is ​​formed by the LAE vectors of all metal atoms in the calibrated image. Each metal atom in the calibrated image is traversed as the central atom, and the central atom and its nearest 1, 2, ..., The plane distances of the metal atoms (sorted from small to large in distance value, the spatial distances of the m nearest neighbor metal atoms to the central atom in the two-dimensional image plane are ), and sort them from small to large according to the distance, and get Vector. Positive integers is the cutoff coefficient, which means the number of neighboring atoms that need to be included in the distance calculation. The reasonable value range of is [10, 30]. The value selected in this embodiment is , that is, for each central atom, the plane distances of the 1st to 20th neighboring atoms are considered.

[0105] Step 5: Based on MATLAB software, an atomic scale model that conforms to the basic information of the crystal defect obtained in the second step is constructed. At random positions near the defect, a specified amount of ultralight atoms to be studied (the value range is 0% ~ the theoretical saturation concentration that can fill all gaps) is introduced. A series of simulation scenarios can be formed by introducing ultralight atoms of different contents at different positions. For each simulation scenario, the corresponding simulated external load is applied according to the basic information of the external load recorded in the first step. Under the simulated external load, a simulation based on density functional theory and molecular dynamics is performed, including static structural energy minimization, relaxation at finite temperature, and 10 6 ~10 11 / s strain rate range of tensile, compressive, shear, torsion or mixed load dynamic simulation. After atomic-scale simulation of each simulation scenario, the atomic arrangement and deformation in the local defect area of ​​the atomic-scale model can be determined, thereby obtaining the corresponding ultra-light atom introduction content and introduction position of this simulation scenario, and finally updating the atomic-scale model with the corresponding defect local atomic arrangement and deformation state.

[0106] Step 6: Update the atomic-scale model obtained in step 5 for each simulation scenario and recalculate the atomic mass of each metal atom using the method shown in step 4. vector, and based on the simulation results, the plane distance scalar between each metal atom and its nearest ultralight atom is calculated, denoted as . The corresponding values ​​of each metal atom in each simulation scenario are Vector and The scalar is constructed as a training sample, and all training samples form a training data set. The input of the training data set is Vector, each piece of data is stored in the form of a 1×20 vector; the output of the database is Scalar.

[0107] Step 7: Based on the training set obtained in step 6, The vector is the input, The scalar is the output, and the machine learning model is trained. The machine learning model in this embodiment is a multi-layer feedforward neural network, and its network structure is as follows Figure 5 As shown in the figure, the topology is 20-20-20-20-1, which includes 1 input layer with 20 nodes, 3 hidden layers with 20 nodes each, and 1 output layer with a single node. During training, the loss function is mean-square error (MSE), the initial learning rate is 0.001, and the network parameters are updated using Levenberg-Marquardt. The input data is normalized before training, and the normalization method is Z-score normalization, that is, , in Each piece of data The normalized value, is the mean of all data, is the standard deviation of all data. During training, 70% of the training data is randomly divided into a training set, 15% into a validation set, and 15% into a test set.

[0108] Step 8: The metal atoms in the calibrated image obtained in step 4 are The vector is input into the machine learning model obtained in step 7 to obtain the Scalar predicted value.

[0109] Step 9: For each node N in the calibrated image obtained in step 3 ij Perform distance calculation. The specific calculation method is: for each grid node N in the calibrated image ij , determine the n metal atoms closest to it, and calculate node N ij The plane distances of the 1st, 2nd, …, nth nearest neighbor metal atoms (sorted in ascending order by distance value, the n nearest neighbor metal atoms and node N ij The two-dimensional image plane space distances are ,…, ), denoted as ,…, ; Get the node N ij The 1st, 2nd, …, nth nearest neighbor metal atom of The scalar predicted value is denoted by ,…, Where n is the number of neighboring atoms taken into consideration, and its reasonable value range is [2, 8] after testing. The value selected in this embodiment is , that is, for a node, the 1st to 4th nearest neighbor atoms of the node are considered for calculation.

[0110] In this embodiment, the above The vector describes the local atomic environment of the metal atom. The calculation principle is shown in Figure 6 As shown in (a); The scalar describes the distance between a metal atom and its nearest ultralight neighbor. The calculation principle is given in Figure 6 As shown in (b); The vectors describe each grid node N ij The distances to their nearest metal atoms and ultralight atoms are calculated by referring to Figure 6 As shown in (c) in .

[0111] Step 10: For each grid node N in the calibrated image ij , calculate the criterion scalar ,in Represents the magnitude of a vector.

[0112] Step 11: Calculate the criterion scalar for each grid node in the calibrated image The calculated results are linearly interpolated to obtain Contour plot of values ​​in the entire image coordinate system. The region with the criterion is marked as a region rich in ultralight atoms, and vice versa, it is marked as a region without ultralight atoms. is the ultralight atom determination threshold. In this embodiment, two types of Values. The region is determined to be rich in ultralight atoms, satisfying The area is determined as the specific location of the superhydrogen atom.

[0113] Based on the steps described in the first to eleventh steps above, the characterization performance tests and verifications are performed on different metal materials.

[0114] Figure 7 The characterization effect and verification of the characterization method of this embodiment on light atoms (hydrogen atoms) in face-centered cubic metal nickel (Ni) are given. Hydrogen atoms are distributed in the octahedral voids of the Ni lattice and the atomic gaps at the grain boundaries. Figure 7 (a) shows the characterization results of hydrogen atoms distributed in Ni in the form of circles. This method can accurately determine the boundary between the hydrogen-rich region and the hydrogen-free region (in ), and the distribution of hydrogen atoms (in red contour lines The above results are highly consistent with the actual distribution pattern of hydrogen atoms shown in Figure (b), and accurately capture the outline of the hydrogen-rich area. Figure 7 Figure (c) shows the characterization effect of hydrogen atoms in a more complex Ni polycrystal. The model includes multiple grains with different crystal orientations and their interfaces (called grain boundaries). Hydrogen atoms tend to aggregate at grain boundaries. The method of this embodiment accurately characterizes the aggregation of hydrogen atoms at grain boundaries, which is consistent with the reference value of Figure (d).

[0115] It should be noted that in order to train Figure 7 The machine learning model of hydrogen atom distribution shown in (a) above established a Ni-interstitial hydrogen atom model with circular distribution of different radii and irregular distribution in the fifth step, and performed atomic-scale simulation to obtain training data; for training prediction Figure 7 (c) Machine learning model of hydrogen atom distribution shown. In the fifth step, a series of randomly oriented polycrystalline Ni models were established and different concentrations of hydrogen atoms were introduced at the grain boundaries, followed by atomic-scale simulations to obtain training data.

[0116] Figure 8 The ultralight element (hydrogen) characterization results of 24°

[110] asymmetric tilted grain boundaries in face-centered cubic Ni during compression deformation are given. Figure 8 (a) and (c) show HRTEM images of grain boundaries with different hydrogen concentrations. Both grain boundaries are structures deformed under 5.8% uniaxial compressive strain in the vertical direction. Due to the limitations of the electron microscopy characterization principle, only the information of metal atoms can be obtained. Figure 8 (b) and (d) in the figure respectively characterize the hydrogen element for (a) and (c). It can be seen that the area of ​​the hydrogen-rich region and the hydrogen atom concentration in the characterization result in (d) are significantly higher than those in (b), which can distinguish the difference in hydrogen concentration between the two grain boundaries of (a) and (c).

[0117] It should be noted that Figure 8 (b) and (d) are predicted by the same machine learning model. The training data is constructed by establishing a 24°

[110] asymmetric tilted grain boundary and introducing different concentrations of hydrogen into it to obtain multiple initial models; subjecting the initial model to uniaxial compression of 0% to 6%, and outputting an atomic model for every 1% compression strain; and extracting the input and output parameters of the metal atoms in all atomic models to form a training database.

[0118] Fig. 9 The characterization results of hydrogen atoms accumulated in the symmetric tilted grain boundary system of body-centered cubic iron-chromium (Fe-Cr) alloy are given. In body-centered cubic metals, hydrogen atoms are distributed at the grain boundaries and in the tetrahedral voids inside the grains. Fig. 9 For the images shown in (a1) and (a2), we first predict The distribution of parameters is Fig. 9 As shown in (b1) and (b2), it represents the distance between each metal atom and its nearest hydrogen atom. Then the hydrogen atom distribution is determined. Fig. 9 (c1) and (c2) divide the hydrogen-rich region (red contour) and the hydrogen atom distribution (orange contour), which is consistent with Fig. 9 The reference values ​​in (d1) and (d2) are consistent.

[0119] It should be noted that Fig. 9 (c1) and (c2) in the figure are predicted by the same machine learning model. The training data is constructed by randomly establishing several Fe-Cr grain boundary models, where the molar fraction of Fe is between 75% and 95%, and the molar fraction of Cr is between 25% and 5%, respectively, to cover the elemental composition of Fe80Cr20 and Fe90Cr10 alloys. Hydrogen atoms with different distributions and concentrations are introduced at the grain boundaries and tetrahedral voids, and atomic-scale simulations are performed. The atomic-scale models are output and extracted to obtain training data.

[0120] Fig.10 The characterization results of hydrogen atoms in hexagonal closest packed metal titanium (Ti) are given. The hydrogen atoms aggregated in Ti may form a face-centered tetragonal lattice type TiH phase with Ti, and the interface between Ti and TiH phase is identified using the method of this embodiment. Fig.10 The image shown in Figure (a) is performed in sequence The prediction of parameters and hydrogen characterization are shown in Fig.10 As shown in (b) and (c), the interface between Ti and TiH phases is divided by red contour lines. Fig.10 Figure (d) in the middle gives the reference value of the two-phase interface, indicating that the characterization results based on the method of this embodiment are consistent with the actual results.

[0121] It should be noted that in order to train Fig.10 In the corresponding example, the machine learning model required for hydrogen atom characterization uses the following training data construction method: an atomic scale model of the Ti-TiH two-phase is established, and 0-4% uniaxial tensile / compressive strain is applied in different directions. An atomic scale simulation is performed every 0.5% strain to obtain a stable structure under the strain, and the atomic scale model is output and the training data is extracted.

[0122] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for atomic scale characterization of ultralight elements in crystalline materials, characterized in that: include: S1. Calibrate the image size of the crystal defect microscopic image of the metal material subjected to the external load so that the lattice spacing in the image coordinate system is equal to the actual theoretical calculated value, which is used as the calibrated image; Establishing an image coordinate system for the calibrated image and identifying the two-dimensional image coordinates of each metal atom in the image; gridding the calibrated image to obtain evenly distributed grid nodes; S2, based on the two-dimensional image coordinates of all metal atoms in the calibrated image, calculate the distance values ​​of each metal atom and its m nearest neighboring metal atoms according to a first preset value m, to form a first local atomic environment vector corresponding to the metal atom; S3. Based on the atomic scale model constructed according to the crystal defect information in the calibrated image, ultralight atoms of different contents are introduced at different positions near the crystal defect, thereby forming a series of simulation scenarios; in each simulation scenario, a simulated external load corresponding to the external load is applied, and a simulation based on density functional theory and / or molecular dynamics is performed to update the atomic arrangement and deformation state of the local area where the crystal defect is located in the atomic scale model; S4. For each simulation scenario, the atomic scale model is updated to recalculate the distance values ​​between each metal atom and its m nearest neighbor metal atoms to form the second local atomic environment vector corresponding to the metal atom, and the plane distance scalar between the metal atom and its nearest neighbor ultralight atom is calculated at the same time; The second local atomic environment vector and plane distance scalar of each metal atom in each simulation scenario are used as model input and output labels respectively, training samples are constructed and added to the training data set, and the machine learning model is trained; S5, inputting the first local atomic environment vector corresponding to each metal atom in the calibrated image into the trained machine learning model to predict the plane distance scalar corresponding to the metal atom; S6, based on the two-dimensional image coordinates of all metal atoms in the calibrated image, according to the second preset value n, calculate the distance value between each grid node and its n nearest neighboring metal atoms to form a first characterization vector corresponding to the grid node, and at the same time use the plane distance scalar predicted by the machine learning model for the n nearest neighboring metal atoms to form a second characterization vector corresponding to the grid node, and use the distance value between the first characterization vector and the second characterization vector as a criterion scalar; S7. Based on the criterion scalars of all grid nodes in the calibrated image, the criterion scalars of all positions in the image domain are obtained by interpolation and completion, and the positions where the criterion scalars do not exceed the judgment threshold are marked as regions rich in ultralight atoms.

2. The atomic scale characterization method of ultralight elements in crystalline materials according to claim 1, characterized in that: The crystal defect microscopic image is an image of a metal material nano-sample containing crystal defects captured by a high-resolution transmission electron microscope HRTEM, a scanning transmission electron microscope STEM or an atomic force microscope AFM.

3. The atomic scale characterization method of ultralight elements in crystalline materials according to claim 1, characterized in that: The machine learning model is a feedforward neural network, a support vector machine, a convolutional neural network or a linear regression model.

4. The atomic scale characterization method of ultralight elements in crystalline materials according to claim 1, characterized in that: The crystal defect information required for constructing the atomic scale model includes the element type, crystal structure, defect type and crystal defect geometry information extracted from the calibrated image.

5. The atomic scale characterization method of ultralight elements in crystalline materials according to claim 1, characterized in that: The first preset value m is between 10 and 30.

6. The atomic scale characterization method of ultralight elements in crystalline materials according to claim 1, characterized in that: The second preset value n is between 2 and 8.

7. The atomic scale characterization method of ultralight elements in crystalline materials according to claim 1, characterized in that: The distance value between the first characterization vector and the second characterization vector is obtained by calculating the difference vector between the first characterization vector and the second characterization vector, and then dividing the difference vector by its dimension value to obtain the modulus length.

8. An atomic scale characterization system for ultralight elements in crystalline materials, characterized in that: include: An image calibration module is used to calibrate the image size of a crystal defect microscopic image of a metal material subjected to an external load so that the lattice spacing in the image coordinate system is equal to the actual theoretical calculated value as a calibrated image; Establishing an image coordinate system for the calibrated image and identifying the two-dimensional image coordinates of each metal atom in the image; gridding the calibrated image to obtain evenly distributed grid nodes; A local atomic environment vector calculation module is used to calculate the distance values ​​of each metal atom and its m nearest neighboring metal atoms based on the two-dimensional image coordinates of all metal atoms in the calibrated image according to a first preset value m, so as to form a first local atomic environment vector corresponding to the metal atom; an atomic scale simulation module, for introducing ultralight atoms of different contents at different positions near the crystal defect based on the atomic scale model constructed according to the crystal defect information in the calibrated image, thereby forming a series of simulation scenarios; in each simulation scenario, applying a simulated external load corresponding to the external load, and performing a simulation based on density functional theory and / or molecular dynamics to update the atomic arrangement and deformation state of the local area where the crystal defect is located in the atomic scale model; A machine learning module is used to update the atomic scale model for each simulation scenario, recalculate the distance value between each metal atom and its m nearest neighbor metal atoms, construct the second local atomic environment vector corresponding to the metal atom, and calculate the plane distance scalar between the metal atom and its nearest neighbor ultralight atom; The second local atomic environment vector and plane distance scalar of each metal atom in each simulation scenario are used as model input and output labels respectively, training samples are constructed and added to the training data set, and the machine learning model is trained; A plane distance scalar prediction module, used for inputting the first local atomic environment vector corresponding to each metal atom in the calibrated image into the trained machine learning model to predict the plane distance scalar corresponding to the metal atom; a criterion scalar calculation module, for calculating the distance values ​​of each of the grid nodes and its n nearest neighboring metal atoms based on the two-dimensional image coordinates of all metal atoms in the calibrated image and according to a second preset value n, to form a first characterization vector corresponding to the grid node, and at the same time, using the plane distance scalar predicted by the machine learning model for the n nearest neighboring metal atoms to form a second characterization vector corresponding to the grid node, and using the distance value between the first characterization vector and the second characterization vector as the criterion scalar; The ultralight atom region marking module is used to obtain the criterion scalars of all positions in the image domain through interpolation and completion based on the criterion scalars of all grid nodes in the calibrated image, and mark the positions where the criterion scalars do not exceed the judgment threshold as regions rich in ultralight atoms.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the atomic scale characterization method of ultralight elements in crystalline materials as described in any one of claims 1 to 7 can be implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the atomic scale characterization method of ultralight elements in crystalline materials as described in any one of claims 1 to 7 when executing the computer program.

Citation Information

Patent Citations

  • Solid-liquid two-phase abrasive flow precision machining molecular dynamics construction method

    CN110334462A

  • Method and apparatus for determination of an isotope concentration using a microscope with energetic particles

    EP3282468A1