Intelligent prediction method and system for different incident particle energy effects

Abstract

The application provides a different incident particle energy effect intelligent prediction method and system, the prediction method comprises the following steps: step one, using a machine learning method to predict the information of primary displacement atoms generated by a device irradiated by incident particles with different energies; step two, according to the information of the primary displacement atoms, using a region growing algorithm to simulate the dynamic evolution of displacement cascade of the primary displacement atoms in the device. The application intelligently and dynamically predicts and simulates the displacement cascade caused by the irradiation damage of space semiconductor devices by incident particles with different energy information, provides a basis for quickly and accurately calculating the interaction between incident particles and displacement cascade, and thus helps to reveal the physical action phenomenon of different energy particles on materials.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of simulation of space irradiation, in particular to a method and system for intelligent prediction of effects of different incident particle energies. BACKGROUND

[0002] During space travel, the outer shell and internal devices of a spacecraft are subjected to strong particle flow radiation, which generates primary knock-on atoms (PKA) and a cascade collision triggered by PKA in the device, thereby causing irradiation defects in the material, which in turn leads to degradation of the material's macroscopic physical and mechanical properties, causing material failure and triggering major safety accidents.

[0003] The primary damage state obtained by displacement cascade has a significant impact on the microstructure evolution of the material and is the basis for the model of subsequent damage accumulation and evolution. Therefore, the defects generated by displacement cascade are very important for the study of material irradiation damage, and the identification of displacement cascade in simulation data is necessary for understanding the properties of the material. However, the displacement cascade process occurs on a nanometer and picosecond length and time scale, making it difficult to study directly through experimental methods. Although some existing technologies have explored the effects of various factors on displacement cascade, there is a lack of research on the results of displacement cascade, i.e., how to identify the defects obtained by displacement cascade. SUMMARY

[0004] The problem solved by the present application is to provide a method for predicting displacement cascade caused by irradiation of aerospace devices by incident particles of different energies.

[0005] To solve the above problems, the present application provides a method for intelligent prediction of effects of different incident particle energies, comprising:

[0006] Step one, using a machine learning method to predict the information of primary knock-on atoms generated by irradiation of devices by incident particles of different energies;

[0007] Step two, using a region growing algorithm to simulate the dynamic evolution of the primary knock-on atoms in the device according to the information of the primary knock-on atoms.

[0008] Preferably, the method of using machine learning to predict the information of primary knock-on atoms generated by irradiation of devices by incident particles of different energies comprises:

[0009] Obtain sample data from a database and divide the sample data into a training set and a test set;

[0010] Define the independent variables and dependent variables in the training set and the test set, respectively, wherein the independent variables are the energy information of the incident particles, and the dependent variables are the information of the primary knock-on atoms generated by irradiation of the device by the incident particles.

[0011] The nonlinear regression algorithm is used to establish a nonlinear regression equation according to the independent variables and the dependent variables, and the information of the primary knock-on atom generated by the device irradiated by the incident particles with different energies is predicted by the nonlinear regression equation.

[0012] Preferably, the information of the primary knock-on atom includes the energy, position, acceleration of the primary knock-on atom, and the conditions required for the primary knock-on atom to collide with its surrounding particles.

[0013] Preferably, the dynamic evolution simulation of the displacement cascade of the primary knock-on atom in the device according to the information of the primary knock-on atom by using the region growing algorithm includes:

[0014] The device is divided into different region blocks according to the position information of the primary knock-on atom;

[0015] At time t, it is judged whether the primary knock-on atom and its surrounding ions in each region block satisfy the first collision condition, and when the first collision condition is satisfied, the primary knock-on atom collides with its surrounding particles, changes the state of the particles around the primary knock-on atom, and generates a secondary knock-on atom;

[0016] At time t+1, it is judged whether the secondary knock-on atom and its surrounding particles satisfy the second collision condition, and when the second collision condition is satisfied, the secondary knock-on atom collides with its surrounding particles, changes the state of the particles around the secondary knock-on atom, and generates a new knock-on atom;

[0017] With the increase of time, iteration is repeatedly performed to make the collision spread and grow in the device until a specified time node is reached, and the process of the dynamic evolution simulation of the displacement cascade of the primary knock-on atom in the device is completed.

[0018] Preferably, the nonlinear regression algorithm is used to establish a nonlinear regression equation according to the independent variables and the dependent variables includes:

[0019] Obtain a scatter plot between the independent variables and the dependent variables;

[0020] According to the trend of the scatter plot, the correlation between the variables, and the number of strong influence points, it is judged whether to establish the nonlinear regression equation, wherein the strong influence points refer to the points deviating from the trend in the scatter plot;

[0021] When the scatter plot meets the set requirements, the nonlinear regression equation is established.

[0022] Preferably, the nonlinear regression equation is:

[0023] y = p0 + p1x + p2x 2 +p3x 3 +...+p n x n ;

[0024] Where x is the independent variable, y is the dependent variable, and p0, p1, p2, p3, and p n These are the coefficients of the polynomial factors, and n is the power of the independent variable.

[0025] Preferably, before establishing the regression model based on the independent variable and the dependent variable, the method further includes: preprocessing the sample data, wherein the preprocessing includes data normalization and data differencing.

[0026] Preferably, the preprocessing of the sample data includes:

[0027] Based on the balance of the sample data distribution, the extreme degree of the independent variable values, and the normality of the sample data, determine whether to preprocess the sample data;

[0028] If any one of the following conditions is not met: the distribution balance of the sample data, the extreme degree of the independent variable value, or the normality of the sample data, the sample is preprocessed.

[0029] Preferably, the method further includes: validating the regression model using cross-validation.

[0030] The advantages of the intelligent prediction method for the energy effects of different incident particles in this invention compared to existing technologies are as follows:

[0031] This invention provides intelligent and dynamic prediction and simulation of displacement cascades caused by irradiation damage to space semiconductor devices from incident particles with different energy information. Specifically, this invention first uses machine learning methods to predict the information of primary displaced atoms generated by incident particles of different energies. The predicted results are used as initial conditions for subsequent dynamic evolution. A region growth algorithm is then used for dynamic evolution to calculate the relationship between the displacement cascade collision effect after the generation of primary displaced atoms and the time growth, realizing the dynamic evolution effect of the collision effect spreading from primary displaced atoms to the entire macroscopic device over time. This invention provides a foundation for rapidly and accurately calculating the interaction between incident particles and displacement cascades, thereby helping to reveal the physical phenomena of the effects of particles of different energies on materials.

[0032] This invention also provides a system for predicting the effects of different types of space irradiation incident particles, comprising:

[0033] The acquisition module is used to acquire information on primary ex-situ atoms generated by incident particles irradiating the device at different energies using machine learning methods;

[0034] The shift cascade dynamic evolution module is used to simulate the shift cascade dynamic evolution of the primary deviated atoms in the device using a region growth algorithm based on the information of the primary deviated atoms.

[0035] The advantages of the system for predicting the effects of different types of space irradiation incident particles in this invention compared to the prior art are the same as those of the intelligent prediction method for the energy effects of different incident particles, and will not be repeated here. Attached Figure Description

[0036] Figure 1 This is a flowchart of the intelligent prediction method for the energy effects of different incident particles in an embodiment of the present invention;

[0037] Figure 2 This is a flowchart illustrating how machine learning methods are used in embodiments of the present invention to predict information about primary ex-situ atoms generated by incident particles irradiating devices at different energies.

[0038] Figure 3 This is a flowchart of a prediction model for the effects of different types of incident particles based on multinomial regression in an embodiment of the present invention. Detailed Implementation

[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0040] Please see Figure 1 As shown, an embodiment of the present invention provides an intelligent prediction method for the energy effects of different incident particles, comprising:

[0041] Step 1: Use machine learning methods to predict information about primary ex-situ atoms produced by irradiating the device with incident particles of different energies;

[0042] Step 2: Based on the information of the primary deviated atoms, a region growth algorithm is used to simulate the dynamic evolution of the displacement cascade of the primary deviated atoms in the device.

[0043] During space irradiation, high-energy incident particles (neutrons, electrons, and heavy ions, etc.) collide with the lattice atoms of a material, transferring some of their energy to the lattice atoms. If the energy gained by the collided atom exceeds its displacement threshold energy, it breaks free from the constraints of surrounding atoms and leaves its original lattice site, becoming a primary knock-on atom (PKA). The atom that leaves its lattice site becomes an interstitial atom, and the original lattice site becomes vacant due to the lack of atoms. If the PKA energy is high enough, it will continue to interact with another lattice atom in the material (secondary knock-on atom). If the energy gained by the secondary knock-on atom exceeds its displacement threshold energy, it will also detach from the lattice site and collide with other atoms, continuing this cycle until the energy gained by the last collided atom is insufficient to overcome its displacement threshold energy. This series of collision processes is called cascade collisions. The interaction between PKA and the lattice atoms in the material produces displacement cascades, thereby causing irradiation damage to the material. The primary damage state obtained by displacement cascade is the basis for damage accumulation and evolution. Therefore, the simulation of damage caused by displacement cascade plays an important role in studying the microstructural evolution of materials by incident particles.

[0044] In the cascade collision process described above, the information of the PKA generated after incident particles of different energies irradiate the device, such as energy and type, are different. Therefore, the displacement defects caused by the cascade collisions of PKAs are also different. Therefore, in order to examine the relationship between the displacement cascade effect generated after the interaction between incident particles with different energies and the device, this embodiment provides the above-mentioned prediction method. First, a machine learning method is used to predict the information of the primary displaced atoms generated by incident particles of different energies. The predicted results are used as the initial conditions for subsequent dynamic evolution. The region growth algorithm is used for dynamic evolution to calculate the relationship between the displacement cascade collision effect after the generation of primary displaced atoms and the growth of time, realizing the dynamic evolution effect of the collision effect spreading from primary displaced atoms to the entire macroscopic device over time.

[0045] This enables intelligent and dynamic prediction and simulation of the displacement cascade caused by irradiation damage to space semiconductor devices from incident particles with different energy information. It provides a foundation for rapidly and accurately calculating the interaction between incident particles and the displacement cascade, thereby helping to reveal the physical phenomena of different energy particles affecting materials. The information of the primary displaced atom includes its energy, position, acceleration, and the conditions required for its collision with surrounding particles, such as trajectory, thermal peaks, and quenching conditions.

[0046] In some implementations, combined with Figure 2 As shown, the information on primary ex-situ atoms produced by irradiating a device with incident particles of different energies using machine learning methods includes:

[0047] Obtain sample data from the database and divide the sample data into a training set and a test set;

[0048] Independent and dependent variables are defined in the training set and the test set, respectively, wherein the independent variable is the energy information of the incident particle, and the dependent variable is the information of the shift cascade dynamic evolution effect generated by the incident particle irradiation device;

[0049] A regression model is established based on the independent variable and the dependent variable, and the correlation analysis between the independent variable and the dependent variable is performed through the regression model.

[0050] In some implementations, establishing a regression model based on the independent variable and the dependent variable includes:

[0051] Obtain a scatter plot between the independent variable and the dependent variable;

[0052] Based on the clues and trends in the scatter plot, the correlation between variables, and the number of points with strong influence, determine whether to establish a nonlinear regression equation, where points with strong influence refer to points in the scatter plot that deviate from the trend.

[0053] When the scatter plot meets the set requirements, the nonlinear regression equation is established.

[0054] The setting requirements include that the scatter plot has a clue trend, the correlation between the variables is linear, and the number of strong influence points is less than a set value.

[0055] The nonlinear regression equation includes:

[0056] y = p0 + p1x + p2x 2 +p3x 3 +...+p n x n ;

[0057] Where p0, p1, p2, p3 and pn are the coefficients of the polynomial factors; x is the independent variable, representing the energy of the incident particle itself; y is the dependent variable, representing the energy, acceleration, position, and conditions required for collision with surrounding particles generated by incident particles with different energies on the semiconductor device material; and n is the power of the independent variable (x).

[0058] Regression algorithms are highly accurate at determining model fit, and can accurately regress data even with minimal input. When studying a single variable or the relationship between two variables, regression algorithms can more directly yield the desired model. As a form of regression algorithm, multinomial regression is divided into univariate m-degree multinomial regression and multivariate linear regression. Univariate m-degree multinomial regression, broadly speaking, belongs to nonlinear regression and is categorized as a univariate multinomial regression model. Multivariate linear regression is a type of linear regression. Univariate m-degree multinomial regression is the most common type of multinomial regression. It solves nonlinear models involving a single output variable and a single input variable, where the relationship between the output and input variables is a superposition of powers. Therefore, this embodiment uses the aforementioned multinomial regression equation.

[0059] In multinomial regression algorithms, n represents the regression order, and different n values ​​result in different regression effects. This embodiment considers not only the fit between the actual and predicted curves during the actual modeling process, but also takes into account the actual situation during the fitting process, avoiding the neglect of extreme values ​​in the model due to excessively reducing error during modeling.

[0060] In some implementations, before establishing the regression model, the process further includes preprocessing the sample data, including: determining whether to preprocess the sample data based on the balance of the sample data distribution, the extreme values ​​of the independent variable, and the normality of the sample data; and preprocessing the sample data when any one of the following conditions is not met: the balance of the sample data distribution, the extreme values ​​of the independent variable, or the normality of the sample data. Preprocessing includes data normalization and data differencing.

[0061] Specifically, in combination Figure 3 As shown, parameters such as element type and energy corresponding to the incident particle type are obtained from the constructed particle parameter database and input into the simulation system. Then, by examining whether the data distribution is balanced, observing whether the values ​​of the independent variables are too extreme, and preliminarily observing the normality of the variables, a decision is made on whether the data needs to be preprocessed. Figure 3In this model, Y indicates that the data needs to be preprocessed, and N indicates that regression analysis should be performed directly. Based on this, a scatter plot is created for each independent and dependent variable to observe trends, whether the scatter plot shows any telltale signs, whether the correlation is linear or curvilinear, and whether there are any strong points deviating from the trend. Regression curves are established using the scatter plots of the variables for experimental analysis, leading to the establishment of a regression equation. Then, the regression equation is used to analyze the correlation between the dependent variable (the dynamic evolution effect of the displacement cascade caused by particle action) and the independent variable (particle energy information), for example, the relationship between the incident particle energy and the number of defects generated after irradiating the device. The calculated analysis results are stored in a database for backup and can be accessed in a data visualization format for users to easily adjust and optimize the prediction model parameters.

[0062] In some implementations, after the incident particles act on the device to generate primary displaced atoms, these primary displaced atoms then collide with other lattice atoms as new incident particles, thus forming a series of cascaded collisions. Therefore, in order to calculate a series of cascaded shift dynamic evolution processes, this embodiment uses a region-growing algorithm to realize the dynamic evolution process of the shifted atoms. The idea is to use the initial pka as the center point and perform dynamic evolution simulations in all directions at different times.

[0063] Specifically, the step of performing dynamic evolution simulation of the shift cascade of the primary deviated atoms in the device using a region growth algorithm based on the information of the primary deviated atoms includes:

[0064] To prevent seed points of different structures in the device from diffusing to other structures and causing erroneous growth during the evolution process, the device is divided into different regions at the initial moment using the predicted position information of the primary displaced atoms.

[0065] At time t, it is determined whether the initial seed point (i.e. the primary despot atom) in each region block and its surrounding ions satisfy the first collision condition. When the first collision condition is satisfied, the primary despot atom collides with its surrounding particles, changing the state (energy, position, acceleration, etc.) of the particles around the primary despot atom, and generating secondary despot atoms.

[0066] At time t+1, it is determined whether the newly generated particle (the secondary de-positioned atom) and its surrounding particles satisfy the second collision condition. When the second collision condition is satisfied, the secondary de-positioned atom collides with its surrounding particles, changing the state of the particles surrounding the secondary de-positioned atom and generating a new de-positioned atom.

[0067] As time progresses, repeated iterations are performed to allow the collisions to diffuse and grow within the device until a specified time point is reached, completing the dynamic evolution simulation process of the primary displaced atoms' shift cascade in the device.

[0068] In some preferred methods, to obtain a model with strong generalization ability, the data is divided into test data and training data. When the model also has a good fit on the test data, it can be said that the model has strong generalization ability. However, there is still a problem: if the model derived from the training data does not perform well on the test data, the parameters of the test data need to be adjusted. In other words, this is to make the model fit the test dataset better. The problem is that the model may be overfitting the test dataset. To address this, the dataset is divided into three parts, including a validation dataset. The training data is used for training, the validation data is used to verify the model's performance and adjust hyperparameters, and the test data is used as the final evaluation criterion. Therefore, preferably, the intelligent prediction method for different incident particle energy effects further includes: using cross-validation to validate the regression model.

[0069] Cross-validation is a model validation technique used to evaluate the results of statistical analysis. It is mainly applied to predicting and estimating the accuracy of predictive models in practice. The purpose of cross-validation is to test the model's ability to predict new data, thereby avoiding problems such as overfitting. Cross-validation divides the sample data into complementary subsets, performs analysis on one subset (called the training set), and validates the analysis on the other subset (called the validation set or test set). To reduce the dependence of experimental results on the data segmentation, and to ensure the stability and balance of the experimental results and increase the persuasiveness of the experiment, this embodiment uses the K-fold cross-validation method. The dataset is divided into ten equal segments, each segment having one and only one chance to serve as the test set. Each training iteration uses only one segment of data as the test set. Each time, the model is predicted using the remaining segments of the dataset, and the model's accuracy is tested on the next segment of the dataset. This process is repeated N times, and the average prediction accuracy obtained by the model in N iterations is recorded. This average represents the average performance with unchanged model parameters, and is also a stable performance, thus reducing randomness to some extent.

[0070] Another embodiment of the present invention provides a system for predicting the effects of different types of space irradiation incident particles, comprising:

[0071] The acquisition module is used to acquire information on primary ex-situ atoms generated by incident particles irradiating the device at different energies using machine learning methods;

[0072] The shift cascade dynamic evolution module is used to simulate the shift cascade dynamic evolution of the primary deviated atoms in the device using a region growth algorithm based on the information of the primary deviated atoms.

[0073] While the disclosure is as stated above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this disclosure, and all such changes and modifications will fall within the protection scope of this invention.

Claims

1. A method for intelligent prediction of the energy effects of different incident particles, characterized in that, include: Step one involves using machine learning methods to predict information about primary ex-situ atoms produced by irradiating the device with incident particles of different energies, including: Obtain sample data from the database and divide the sample data into a training set and a test set; Independent and dependent variables are defined in the training set and the test set, respectively, wherein the independent variable is the energy information of the incident particle, and the dependent variable is the information of the primary desitu atoms generated by the incident particle irradiating the device; A nonlinear regression algorithm is used to establish a nonlinear regression equation based on the independent and dependent variables, and the nonlinear regression equation is used to predict the information of the primary ex-situ atoms generated by the incident particles of different energies irradiating the device. Step 2: Based on the information of the primary deviated atoms, a region growth algorithm is used to simulate the dynamic evolution of the displacement cascade of the primary deviated atoms in the device.

2. The intelligent prediction method for the energy effects of different incident particles according to claim 1, characterized in that, The information about the primary despot atom includes its energy, position, acceleration, and the conditions required for it to collide with surrounding particles.

3. The intelligent prediction method for the effects of different incident particle energies according to claim 2, characterized in that, The step of performing dynamic evolution simulation of the displacement cascade of the primary deviated atoms in the device using a region growth algorithm based on the information of the primary deviated atoms includes: Based on the position information of the primary displaced atoms, the device is divided into different regions; At time t, it is determined whether the primary desitu atom in each region block and its surrounding ions satisfy the first collision condition. When the first collision condition is satisfied, the primary desitu atom collides with its surrounding particles, changing the state of the particles around the primary desitu atom and generating a secondary desitu atom. At time t+1, it is determined whether the secondary de-positioned atom and its surrounding particles satisfy the second collision condition. When the second collision condition is satisfied, the secondary de-positioned atom collides with its surrounding particles, changing the state of the particles surrounding the secondary de-positioned atom and generating a new de-positioned atom. As time progresses, repeated iterations are performed to allow the collisions to diffuse and grow within the device until a specified time point is reached, completing the dynamic evolution simulation process of the primary displaced atoms' shift cascade in the device.

4. The intelligent prediction method for the energy effects of different incident particles according to claim 1, characterized in that, The step of employing a nonlinear regression algorithm to establish a nonlinear regression equation based on the independent and dependent variables includes: Obtain a scatter plot between the independent variable and the dependent variable; Based on the clues and trends in the scatter plot, the correlation between variables, and the number of points with strong influence, determine whether to establish the nonlinear regression equation, where points with strong influence refer to points in the scatter plot that deviate from the trend. When the scatter plot meets the set requirements, the nonlinear regression equation is established.

5. The intelligent prediction method for different incident particle energy effects according to claim 4, characterized in that, The nonlinear regression equation is: ； Where x is the independent variable, y is the dependent variable, and p0, p1, p2, p3, and p n These are the coefficients of the polynomial factors, and n is the power of the independent variable.

6. The intelligent prediction method for the energy effects of different incident particles according to claim 1, characterized in that, Before establishing the regression model based on the independent variable and the dependent variable, the method further includes: preprocessing the sample data, wherein the preprocessing includes data normalization and data differencing.

7. The intelligent prediction method for different incident particle energy effects according to claim 6, characterized in that, The preprocessing of the sample data includes: Based on the balance of the sample data distribution, the extreme degree of the independent variable values, and the normality of the sample data, determine whether to preprocess the sample data; If any one of the following conditions is not met: the distribution balance of the sample data, the extreme degree of the independent variable value, or the normality of the sample data, the sample is preprocessed.

8. The intelligent prediction method for different incident particle energy effects according to claim 6, characterized in that, Also includes: The regression model was validated using cross-validation.

9. A system for predicting the effects of different types of incident particles from space irradiation, characterized in that, include: The acquisition module is used to acquire information on primary desitu atoms produced by incident particles of different energies irradiating a device using machine learning methods. This includes: acquiring sample data from a database and dividing the sample data into a training set and a test set; defining independent and dependent variables in the training set and the test set respectively, wherein the independent variable is the energy information of the incident particles, and the dependent variable is the information on the primary desitu atoms produced by the incident particles irradiating the device; and using a nonlinear regression algorithm to establish a nonlinear regression equation based on the independent and dependent variables, and using the nonlinear regression equation to predict the information on the primary desitu atoms produced by the incident particles of different energies irradiating the device. The shift cascade dynamic evolution module is used to simulate the shift cascade dynamic evolution of the primary deviated atoms in the device using a region growth algorithm based on the information of the primary deviated atoms.

Images