CN117727401A - Atomic scale high-dose irradiation damage prediction method and system based on deep learning - Google Patents
Atomic scale high-dose irradiation damage prediction method and system based on deep learning Download PDFInfo
- Publication number
- CN117727401A CN117727401A CN202311798562.0A CN202311798562A CN117727401A CN 117727401 A CN117727401 A CN 117727401A CN 202311798562 A CN202311798562 A CN 202311798562A CN 117727401 A CN117727401 A CN 117727401A
- Authority
- CN
- China
- Prior art keywords
- atomic
- deep learning
- model
- prediction
- simulation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 238000013135 deep learning Methods 0.000 title claims abstract description 48
- 238000004088 simulation Methods 0.000 claims abstract description 66
- 230000007547 defect Effects 0.000 claims abstract description 53
- 230000006870 function Effects 0.000 claims abstract description 30
- 238000012360 testing method Methods 0.000 claims abstract description 29
- 238000004458 analytical method Methods 0.000 claims abstract description 26
- 238000012800 visualization Methods 0.000 claims abstract description 7
- 238000012549 training Methods 0.000 claims description 24
- 238000006073 displacement reaction Methods 0.000 claims description 17
- 238000000329 molecular dynamics simulation Methods 0.000 claims description 17
- 239000000463 material Substances 0.000 claims description 16
- 230000008859 change Effects 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 14
- 230000003068 static effect Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 230000005855 radiation Effects 0.000 claims description 9
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000003780 insertion Methods 0.000 claims description 5
- 230000037431 insertion Effects 0.000 claims description 5
- 230000003993 interaction Effects 0.000 claims description 5
- 230000015572 biosynthetic process Effects 0.000 claims description 4
- 238000013136 deep learning model Methods 0.000 claims description 4
- 238000009830 intercalation Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 abstract description 9
- 238000004422 calculation algorithm Methods 0.000 description 5
- 238000009825 accumulation Methods 0.000 description 2
- 238000002939 conjugate gradient method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002040 relaxant effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000002687 intercalation Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012900 molecular simulation Methods 0.000 description 1
- 239000011824 nuclear material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 238000005204 segregation Methods 0.000 description 1
- 230000008961 swelling Effects 0.000 description 1
Landscapes
- Monitoring And Testing Of Nuclear Reactors (AREA)
Abstract
The invention discloses an atomic scale high-dose irradiation damage prediction method and system based on deep learning, and belongs to the technical field of simulation. Data are obtained and preprocessed; constructing a prediction model based on deep learning, and testing and adjusting; establishing relevant parameters and potential functions according to the system, and initializing acting force and relaxation operation on the atomic system; predicting the number and positions of franker defect pairs according to a prediction model, selecting a lattice atom and a gap position according to the prediction model, and placing the selected atom at a new position; performing energy minimization; iterating the interatomic force to update the atomic position; analyzing the result and performing a visualization operation on the result. The system comprises a learning prediction module, an analog simulation module and an analysis module. On the premise of saving the calculation cost, the method improves the scientificity of the high-dose irradiation result, reduces the time complexity of calculation, and optimizes the simulation step of high-dose irradiation damage.
Description
Technical Field
The invention relates to the technical field of simulation, in particular to an atomic scale high-dose irradiation damage prediction method and system based on deep learning.
Background
Nuclear reactors are widely favored because of their high energy density, low environmental pollution, and capability of large-scale grid-connected power generation. Today nuclear reactor technology is mature, but there are still problems that remain unsolved, wherein the choice of nuclear structural materials is a key factor in the service life of the nuclear reactor; irradiation environments are common in nuclear reactors, and ensuring that nuclear structural materials are still stable in the irradiation environment is a basis for addressing the service life and safety of nuclear reactors. Under the extreme conditions of long-term high temperature, high radiation dose and the like of a nuclear reactor, the service life of a nuclear structural material can be greatly reduced, and a microstructure can be seriously damaged; neutrons and energetic particles generated in nuclear reactors act on structural materials such that the structural materials microscopically exhibit a number of atomic scale defects; irradiation effects such as irradiation hardening, irradiation embrittlement, irradiation fatigue, swelling, element segregation, etc. are macroscopically generated.
In the research of novel structural materials, because irradiation damage is researched by experimental means, the formation and evolution of internal defects of the materials are difficult to be deeply researched, and the feasibility of the materials is reduced due to the high cost of the experimental means. With the development of computer science and technology, the research of the irradiation damage process of a structural material by molecular simulation becomes the main stream direction of researching the structural material, but usually the nuclear structural material is exposed to high-flux irradiation damage, for better comparison with experiments, the molecular dynamics is usually to simulate the irradiation damage process by introducing primary dislocation atoms, and if a better structure is obtained, the primary dislocation atoms are often required to be introduced for many times to realize the high-flux simulation of the irradiation damage. This requires molecular dynamics to simulate atomic systems on the order of tens or even hundreds of millions, which is a great challenge for molecular dynamics using simulation techniques to calculate tens of thousands of times. Therefore, optimizing the molecular dynamics method of high-throughput simulation is particularly important.
Disclosure of Invention
The invention aims to provide a deep learning-based atomic scale high-dose irradiation damage prediction method and a deep learning-based atomic scale high-dose irradiation damage prediction system, which improve the scientificity of high-dose irradiation results, reduce the time complexity of calculation and optimize the high-dose irradiation damage simulation step on the premise of saving the calculation cost.
In order to achieve the above object, the present invention provides a deep learning-based atomic scale high dose irradiation damage prediction method, comprising the steps of:
s1, acquiring data, acquiring a plurality of irradiation damage simulation result samples of various system lattice types and different system irradiation doses, preprocessing the acquired data, and dividing the preprocessed data into a training data set and a testing data set according to the ratio of 4:1;
s2, constructing a prediction model based on deep learning, wherein the prediction model comprises a franker defect pair prediction model based on the deep learning, a high-dose irradiation damage per atom displacement trend prediction model based on the deep learning and a high-dose irradiation damage simulation result effectiveness prediction model based on the deep learning;
s3, testing the prediction model, and adjusting the model in a continuous iteration mode according to the test result until the model prediction precision reaches the requirement;
s4, establishing and initializing an atomic system according to related parameters of the system, performing relaxation operation on the atomic system, calculating the total atomic number N, and setting N FI =0, where FI is franker defect pair insertion, N FI Number of insertions for franker defect pairs;
s5, predicting the number and the positions of the franker defect pairs according to a prediction model, selecting an atom positioned at a lattice site according to the prediction model, and selecting a gap position in a simulation unit according to the prediction model; placing the selected atom in a new location;
s6, iterating the interatomic acting force according to the potential function, updating the atomic position, and iterating N FI Numerical value, irradiation damage dose;
s7, calculating the energy of the atomic system, analyzing thermodynamic property information, energy information and defect information of the atomic system, and performing visual operation;
s8, judging the result validity and the accuracy of the current operation result according to the high-dose irradiation damage simulation result validity prediction model, and automatically adjusting the atomic system model to a proper structure when the current simulation result is abnormal;
s9, repeating the steps S5-S8 until the simulation result reaches the required irradiation damage dose.
Preferably, in step S2, a prediction model based on deep learning is constructed, which specifically includes:
s21, defining a model structure, wherein an LSTM model is formed by stacking five LSTM layers;
s22, utilizing Xavier initialization data, adjusting the initial range of parameters according to the constructed model, and selecting a scaling factor according to an activation function;
s23, transmitting the input data into an LSTM model through forward propagation, receiving the input data, the output of the previous time step and the internal state of the LSTM unit at each time step, and calculating the output of the current time step and the updated internal state;
s24, calculating a value of a loss function according to the output of the model and the real label, and adopting an absolute percentage error as the loss function;
s25, updating parameters of the LSTM model by using random gradient descent so as to minimize a loss function;
s26, repeating the steps S21-S25 until a preset stopping condition is reached.
Preferably, in the step S3, in the process of testing the prediction model, the bisection absolute percentage error is used as a prediction precision evaluation index;
the training set is as follows: d (D) train ＝{(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x n ,y n )}
The test set is: d (D) test ＝{(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x m ,y m )}
Wherein n is the total number of training samples, m is the total number of test samples, and when the training model function is f (x), the output value of the model is
The mean absolute percentage error is:
preferably, step S5 specifically includes:
s51, collecting atomic system parameters and atomic data of simulation results of irradiation damage, and processing the collected data through centralized and standardized operations, wherein the atomic system parameters comprise element types and lattice types;
s52, predicting the number and the positions of the pairs of the franker defects by using a franker defect pair prediction model and a high-dose irradiation damage per-atom displacement trend prediction model, and outputting the number and the positions of vacancies in the pairs of the franker defects and the number and the positions of interstitial atoms;
s53, selecting corresponding atoms at the vacancy positions and placing the atoms in interstitial atom positions.
Preferably, in step S6, two methods are used to update the atomic position, specifically:
s61, when a molecular statics method is adopted for simulation, calculating the energy and gradient initialization search direction of the initial state of the atomic system, traversing to obtain the minimum energy value of the atomic system in the conjugated direction, the displacement direction and the numerical value of each atom, and updating the atomic position;
s62, checking the change conditions of energy and gradient, when the energy change is small and the gradient is close to zero, reaching the minimum energy and ending iteration, and recording the change quantity of the atomic position in each iteration; ending the iterative steps S61-S62 when the energy minimization is satisfied;
s63, when a molecular dynamics method is adopted for simulation, firstly, energy minimization is carried out, the atomic position is updated, then, an NPT ensemble is utilized, the interatomic acting force is calculated through a potential function, the atomic displacement is calculated, and the atomic position is updated according to the atomic displacement;
s64, iterating the atomic positions and calculating the interatomic acting force, and repeating the steps S63-S64 until the set number of steps is reached.
Preferably, in step S6:
the EAM potential function is adopted, and the formula is as follows:
wherein F is i Is the embedding energy ρ h,i Is in position R when the intercalating atom is absent i Host charge density at Φ i,j Is the opposite potential between atom i and atom j, R i,j For the distance between atoms i and j, the electron density ρ at host lattice i is h,i Setting the atomic electron density as the current superposition of atomic electron densities;
iteration N FI Numerical value, calculate N FI ＝N FI +1；
Updating system irradiation dose cDPA value to beWherein N is the total number of system atoms.
The invention also provides a deep learning-based atomic scale high-dose irradiation damage prediction system, which comprises a learning prediction module consisting of a data processing unit, a training unit and a testing unit group, an analog simulation module consisting of a system building unit, a relaxation simulation unit, a molecular statics simulation unit, a molecular dynamics simulation unit and a visualization unit, and an analysis module consisting of a thermodynamic property information analysis unit, an energy information analysis unit and a defect information analysis unit.
Preferably, the data processing unit: the method comprises the steps of collecting a plurality of irradiation damage simulation result samples of different atomic systems, different initial conditions and different boundary conditions, and carrying out data preprocessing and data division;
training unit: the method is used for performing deep learning training and constructing a deep learning model;
test unit: for quantitatively evaluating the predictive model.
Preferably, the system establishment unit: establishing an atomic system according to the element types and the atomic system lattice parameters obtained through interaction with a user, and initializing initial acting force and initial energy parameters of atoms in the atomic system;
molecular statics simulation unit: performing high-dose damage simulation on an atomic system by a molecular statics method;
molecular dynamics simulation unit: performing high-dose damage simulation on an atomic system by a molecular dynamics method;
and a visualization unit: and the system is used for outputting a 3D atomic system graph according to the position coordinate information of the atomic system and visualizing the information obtained by the analysis module.
Preferably, the thermodynamic property information analysis unit is used for analyzing thermodynamic property information of an atomic system, including system temperature, system enthalpy change and entropy change;
an energy analysis unit: the method is used for analyzing energy information of an atomic system, including system energy, vacancy forming energy and surface energy;
defect analysis unit: the method is used for analyzing defects of an atomic system to obtain interstitial atoms, vacancy quantity and position parameters of the system and obtain structural evolution, defect formation and motion information of the material.
Therefore, the atomic scale high dose irradiation damage prediction method and system based on deep learning have the beneficial effects that:
1. the invention simulates the high-dose irradiation damage process by using the franker defect pair accumulation method, and reduces the simulation calculation amount of the high-dose irradiation damage to the maximum extent.
2. According to the method, the cascade collision process generated by irradiation damage is not considered, the franker defect pair is directly introduced, the requirement of carrying out detailed simulation calculation on each irradiation is avoided, and high-dose irradiation damage simulation can be carried out even for a larger atomic system.
3. The deep learning algorithm is used for constructing the franker defect pair prediction model, so that the accuracy of high-dose irradiation damage simulation is improved, and the complexity and nonlinear characteristics in the high-dose irradiation damage process can be better captured; not only reduces the calculated amount, but also improves the accuracy and reliability of the simulation by the study and optimization of the deep learning algorithm; this provides an efficient and viable method for studying the high dose radiation damage process and the related changes in the structural properties of the nuclear material.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a system architecture diagram of the present invention.
Detailed Description
The technical scheme of the invention is further described below through the attached drawings and the embodiments.
Unless defined otherwise, technical or scientific terms used herein should be given the ordinary meaning as understood by one of ordinary skill in the art to which this invention belongs. The terms "first," "second," and the like, as used herein, do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises", and the like, means that elements or items preceding the word are included in the element or item listed after the word and equivalents thereof, but does not exclude other elements or items. The terms "disposed," "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. "upper", "lower", "left", "right", etc. are used merely to indicate relative positional relationships, which may also be changed when the absolute position of the object to be described is changed.
Example 1
As shown in fig. 1, the invention provides a deep learning-based atomic scale high dose irradiation damage prediction method, which comprises the following steps:
firstly (step S1) acquiring data, collecting a plurality of irradiation damage simulation result samples of various system lattice types and different system irradiation doses, preprocessing the acquired data, and dividing the preprocessed data into a training data set and a testing data set according to the ratio of 4:1.
And secondly (step S2) constructing a prediction model based on deep learning, wherein the prediction model comprises a franker defect pair prediction model based on the deep learning, a high-dose irradiation damage per atom displacement trend prediction model based on the deep learning and a high-dose irradiation damage simulation result effectiveness prediction model based on the deep learning.
In the process of constructing the model, the method further comprises the following steps:
s21, defining a model structure, wherein an LSTM model is formed by stacking five LSTM layers;
s22, utilizing Xavier initialization data, adjusting the initial range of parameters according to the constructed model, and selecting a scaling factor according to an activation function;
s23, transmitting the input data into an LSTM model through forward propagation, receiving the input data, the output of the previous time step and the internal state of the LSTM unit at each time step, and calculating the output of the current time step and the updated internal state;
s24, calculating a value of a loss function according to the output of the model and the real label, and adopting an absolute percentage error as the loss function;
s25, updating parameters of the LSTM model by using random gradient descent so as to minimize a loss function;
s26, repeating the steps S21-S25 until a preset stop condition is reached
And thirdly (S3) testing the prediction model, and adjusting the model in a continuous iteration mode according to the test result until the model prediction precision reaches the requirement.
In the process of testing the prediction model, taking the bisection absolute percentage error as a prediction precision evaluation index;
the training set is as follows: d (D) train ＝{(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x n ,y n )}
The test set is: d (D) test ＝{(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x m ,y m )}
Wherein n is the total number of training samples, m is the total number of test samples, and when the training model function is f (x), the output value of the model is
The mean absolute percentage error is:
fourth (step S4) according to the system establishment related parameters, establishing and initializing an atomic system, performing relaxation operation on the atomic system, calculating the total atomic number N, and setting N FI =0, where FI is franker defect pair insertion, N FI Is the number of franker defect pairs inserted.
Fifthly (step S5) predicting the number and positions of franker defect pairs according to the prediction model, selecting an atom located at a lattice site according to the prediction model, and selecting a gap position in the simulation unit according to the prediction model. The selected atom is placed in a new location.
The method comprises the following specific steps:
s51, collecting atomic system parameters (including element types, lattice types and the like) and atomic data of simulation results of irradiation damage simulation, and processing the collected data through centralized and standardized operations;
s52, predicting the number and the positions of the pairs of the franker defects by using the franker defect pair prediction model and the high-dose irradiation damage per-atom displacement trend prediction model, and outputting the number and the positions of vacancies and the number and the positions of interstitial atoms in the pairs of the franker defects.
S53, selecting corresponding atoms at the vacancy positions and placing the atoms in interstitial atom positions.
Step S6, iterating the action force among atoms according to the potential function, calculating the energy of an atomic system and updating the atomic position; then iterate N FI Numerical value, irradiation damage dose. The method specifically comprises the following steps:
s61, when a molecular statics method is adopted for simulation, calculating the energy and gradient initialization search direction of the initial state of the atomic system, traversing to obtain the minimum energy value of the atomic system in the conjugated direction, the displacement direction and the numerical value of each atom, and updating the atomic position;
s62, checking the change conditions of energy and gradient, when the energy change is small and the gradient is close to zero, reaching the minimum energy and ending iteration, and recording the change quantity of the atomic position in each iteration; ending the iterative steps S61-S62 when the energy minimization is satisfied;
s63, when a molecular dynamics method is adopted for simulation, firstly, energy minimization is carried out, the atomic position is updated, then, an NPT ensemble is utilized, the interatomic acting force is calculated through a potential function, the atomic displacement is calculated, and the atomic position is updated according to the atomic displacement;
s64, iterating the atomic positions and calculating the interatomic acting force, and repeating the steps S63-S64 until the set number of steps is reached.
The potential function adopts an EAM potential function (it should be noted that, in order to accurately describe the interaction between atoms of the material, different materials adopt a plurality of different potential functions, and if the invention mainly uses the EAM potential function for calculation), the formula is as follows:
wherein F is i Is the embedding energy ρ h,i Is in position R when the intercalating atom is absent i Host charge density at Φ i,j Is the opposite potential between atom i and atom j, R i,j For the distance between atoms i and j, the electron density ρ at host lattice i is h,i Setting the atomic electron density as the current superposition of atomic electron densities;
iteration N FI Numerical value, calculate N FI ＝N FI +1；
Updating system irradiation dose cDPA value to beWherein N is the total number of system atoms.
Seventhly (step S7) analyzing thermodynamic property information, energy information and defect information of the atomic system, and performing visual operation.
And eight (S8) judging the result validity and the accuracy of the current operation result according to the high-dose irradiation damage simulation result validity prediction model, and automatically adjusting the atomic system model to a proper structure if the current simulation result is abnormal.
Nine (step S9) repeating steps S5-S8 until the simulation result reaches the required irradiation damage dose.
Example 2
As shown in fig. 2, the present invention further provides a deep learning-based atomic scale high dose radiation damage prediction system, comprising:
1. the learning prediction module comprises a data processing unit, a training unit and a testing unit.
A data processing unit: the method is used for collecting a plurality of irradiation damage simulation result samples of different atomic systems, different initial conditions and different boundary conditions, preprocessing the collected data, and dividing the preprocessed data into a training data set and a testing data set according to a ratio of 4:1.
Training unit: for performing a deep learning training using the training data set generated by the data processing unit, a deep learning model is constructed, the deep learning model comprising:
1) Franker defect pair prediction model based on deep learning: the franker defect pair accumulation method applied to high-dose irradiation damage is trained through cascade collision results based on a deep learning algorithm, and the number of franker defect pairs introduced into an atomic system each time is predicted;
2) Deep learning-based high-dose irradiation damage per atom displacement trend prediction model: based on a deep learning algorithm, according to the number training of the franker defects obtained by introducing primary dislocation atoms each time under different irradiation doses in high-dose irradiation damage, predicting the number of the franker defects introduced each time in an atomic system;
3) High-dose irradiation damage simulation result effectiveness prediction model based on deep learning: the method is used for judging the result validity and the accuracy of the current operation result, and if the current simulation result is abnormal, the atomic system model is automatically adjusted to a proper structure.
Test unit: for quantitative evaluation of the model.
2. The simulation module comprises a system establishment unit, a relaxation simulation unit, a molecular statics simulation unit, a molecular dynamics simulation unit and a visualization unit.
A system establishment unit: and constructing an atomic system according to system establishment information such as element types, atomic system lattice parameters and the like obtained through interaction with a user, and initializing the atomic system, wherein the parameters comprise initial acting force, initial energy and the like of atoms are determined.
Molecular statics simulation unit: the method is used for simulating high-dose irradiation damage to an atomic system by a molecular statics method. Based on a molecular statics method, the model is fully relaxed through energy minimization, and the balance structure with the lowest energy is obtained. The position coordinates and force information of the atomic system are iterated by introducing franker defect pairs and minimizing the energy of the system after introduction. The above operation is repeated until the set irradiation damage dose is reached. In the process of simulating high-dose irradiation damage, a high-dose irradiation damage simulation result validity prediction model obtained by a learning prediction module is used for judging the result validity and the accuracy of the current operation result, and if the current simulation result is abnormal, the atomic system model is automatically adjusted to a proper structure.
Molecular dynamics simulation unit: the method is used for simulating the high-dose irradiation damage of the atomic system by a molecular dynamics method. Performing conjugate gradient descent (CG) energy minimization treatment on the initial atomic system based on a statics relaxation means, fully relaxing the model to obtain a balance structure with the lowest energy, performing thermodynamic relaxation on the initial atomic system based on a molecular dynamics method, and fully relaxing the model by using an isothermal and isobaric ensemble (NPT) to obtain a relatively stable structure; on the basis, the behavior of the simulated irradiation damage is generated by introducing a certain number of franker defects, and the system is quickly brought into a stable structure by using energy minimization for an atomic system. The thermodynamic relaxation of systems containing defective atoms is performed by molecular dynamics methods, using for example NPT ensembles for atomic systems, by calculating interatomic interactions such as intercalation atomic potential (EAM) to bring the system into equilibrium. The above process is repeated continuously to simulate the high dose irradiation damage process. In the process, a high-dose irradiation damage simulation result validity prediction model is used for judging the result validity and accuracy of the current operation result, and if an abnormal condition occurs in the current simulation result, the atomic system model is automatically adjusted to a proper structure.
And a visualization unit: and the system is used for outputting a 3D atomic system graph according to the position coordinate information of the atomic system and outputting information obtained by a visual analysis module.
Introducing a franker defect pair, inputting atomic system parameters such as an atomic system lattice type, an element type and the like to obtain a result according to the franker defect pair prediction model obtained by the learning prediction module, correspondingly selecting a plurality of atoms in the atomic system according to the prediction model result, and then moving the selected atoms to a gap position predicted by the prediction model to realize the introduction of the plurality of pairs of franker defect pairs. And adjusting the number of the franker defect pairs introduced each time according to the deep learning-based high-dose irradiation damage per-atom displacement trend prediction model obtained by the learning prediction module.
The energy minimization uses a conjugate gradient method, the conjugate gradient method adopts a PRP algorithm, and the formula is as follows:
x k+1 ＝x k +α k d k
wherein x is an objective function argument, alpha k Optimal solution step length d for one-dimensional search k G is conjugate gradient k Is the gradient of the objective function.
3. The analysis module comprises a thermodynamic property information analysis unit, an energy information analysis unit and a defect information analysis unit.
Thermodynamic property information analysis unit: the method is used for analyzing thermodynamic property information of an atomic system, wherein the thermodynamic property information comprises system temperature, system enthalpy change, entropy change and the like.
Energy information analysis unit: for analyzing atomic system energy information including system energy, vacancy forming energy, surface energy, and the like.
Defect information analysis unit: for analyzing defects of an atomic system; and analyzing the atomic system by using a Wigner-Seitz analysis method to obtain the number and position parameters of interstitial atoms and vacancies of the system and obtain the information of structural evolution, defect formation, movement and the like of the material.
Therefore, the atomic scale high-dose irradiation damage prediction method and the atomic scale high-dose irradiation damage prediction system based on deep learning are adopted, the scientificity of high-dose irradiation results is improved on the premise of saving the calculation cost, the calculation time complexity is reduced, and the high-dose irradiation damage simulation step is optimized.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention and not for limiting it, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that: the technical scheme of the invention can be modified or replaced by the same, and the modified technical scheme cannot deviate from the spirit and scope of the technical scheme of the invention.
Claims (10)
1. The atomic scale high dose irradiation damage prediction method based on deep learning is characterized by comprising the following steps of:
s1, acquiring data, acquiring a plurality of irradiation damage simulation result samples of various system lattice types and different system irradiation doses, preprocessing the acquired data, and dividing the preprocessed data into a training data set and a testing data set according to the ratio of 4:1;
s2, constructing a prediction model based on deep learning, wherein the prediction model comprises a franker defect pair prediction model based on the deep learning, a high-dose irradiation damage per atom displacement trend prediction model based on the deep learning and a high-dose irradiation damage simulation result effectiveness prediction model based on the deep learning;
s3, testing the prediction model, and adjusting the model in a continuous iteration mode according to the test result until the model prediction precision reaches the requirement;
s4, establishing and initializing an atomic system according to related parameters of the system, performing relaxation operation on the atomic system, calculating the total atomic number N, and setting N FI =0, where FI is franker defect pair insertion, N FI Number of insertions for franker defect pairs;
s5, predicting the number and the positions of the franker defect pairs according to a prediction model, selecting an atom positioned at a lattice site according to the prediction model, and selecting a gap position in a simulation unit according to the prediction model; placing the selected atom in a new location;
s6, iterating the interatomic acting force according to the potential function, updating the atomic position, and iterating N FI Numerical value, irradiation damage dose;
s7, calculating the energy of the atomic system, analyzing thermodynamic property information, energy information and defect information of the atomic system, and performing visual operation;
s8, judging the result validity and the accuracy of the current operation result according to the high-dose irradiation damage simulation result validity prediction model, and automatically adjusting the atomic system model to a proper structure when the current simulation result is abnormal;
s9, repeating the steps S5-S8 until the simulation result reaches the required irradiation damage dose.
2. The deep learning-based atomic scale high dose radiation damage prediction method according to claim 1, wherein in step S2, a deep learning-based prediction model is constructed, specifically comprising:
s21, defining a model structure, wherein an LSTM model is formed by stacking five LSTM layers;
s22, utilizing Xavier initialization data, adjusting the initial range of parameters according to the constructed model, and selecting a scaling factor according to an activation function;
s23, transmitting the input data into an LSTM model through forward propagation, receiving the input data, the output of the previous time step and the internal state of the LSTM unit at each time step, and calculating the output of the current time step and the updated internal state;
s24, calculating a value of a loss function according to the output of the model and the real label, and adopting an absolute percentage error as the loss function;
s25, updating parameters of the LSTM model by using random gradient descent so as to minimize a loss function;
s26, repeating the steps S21-S25 until a preset stopping condition is reached.
3. The deep learning-based atomic scale high dose irradiation damage prediction method and system according to claim 2, wherein step S3 uses the bisection absolute percentage error as a prediction accuracy evaluation index in the process of testing the prediction model;
the training set is as follows: d (D) train ＝{(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x n ,y n )}
The test set is: d (D) test ＝{(x 1 ,y 1 ),(x 2 ,y 2 ),...,(x m ,y m )}
Wherein n is the total number of training samples, m is the total number of test samples, and the training model function isf (x), the model has an output value of
The mean absolute percentage error is:
4. the deep learning-based atomic scale high dose radiation damage prediction method according to claim 3, wherein step S5 specifically comprises:
s51, collecting atomic system parameters and atomic data of simulation results of irradiation damage, and processing the collected data through centralized and standardized operations, wherein the atomic system parameters comprise element types and lattice types;
s52, predicting the number and the positions of the pairs of the franker defects by using a franker defect pair prediction model and a high-dose irradiation damage per-atom displacement trend prediction model, and outputting the number and the positions of vacancies in the pairs of the franker defects and the number and the positions of interstitial atoms;
s53, selecting corresponding atoms at the vacancy positions and placing the atoms in interstitial atom positions.
5. The deep learning-based atomic scale high dose irradiation damage prediction method according to claim 4, wherein in step S6, two methods are used to update the atomic position, specifically:
s61, when a molecular statics method is adopted for simulation, calculating the energy and gradient initialization search direction of the initial state of the atomic system, traversing to obtain the minimum energy value of the atomic system in the conjugated direction, the displacement direction and the numerical value of each atom, and updating the atomic position;
s62, checking the change conditions of energy and gradient, when the energy change is small and the gradient is close to zero, reaching the minimum energy and ending iteration, and recording the change quantity of the atomic position in each iteration; ending the iterative steps S61-S62 when the energy minimization is satisfied;
s63, when a molecular dynamics method is adopted for simulation, firstly, energy minimization is carried out, the atomic position is updated, then, an NPT ensemble is utilized, the interatomic acting force is calculated through a potential function, the atomic displacement is calculated, and the atomic position is updated according to the atomic displacement;
s64, iterating the atomic positions and calculating the interatomic acting force, and repeating the steps S63-S64 until the set number of steps is reached.
6. The deep learning-based atomic scale high dose radiation damage prediction method according to claim 5, wherein in step S6:
the EAM potential function is adopted, and the formula is as follows:
wherein F is i Is the embedding energy ρ h,i Is in position R when the intercalating atom is absent i Host charge density at Φ i,j Is the opposite potential between atom i and atom j, R i,j For the distance between atoms i and j, the electron density ρ at host lattice i is h,i Setting the atomic electron density as the current superposition of atomic electron densities;
iteration N FI Numerical value, calculate N FI ＝N FI +1；
Updating system irradiation dose cDPA value to beWherein N is the total number of system atoms.
7. Atomic scale high dose irradiation damage prediction system based on deep learning, its characterized in that: the system comprises a learning prediction module consisting of a data processing unit, a training unit and a testing unit group, an analog simulation module consisting of a system building unit, a relaxation simulation unit, a molecular statics simulation unit, a molecular dynamics simulation unit and a visualization unit, and an analysis module consisting of a thermodynamic property information analysis unit, an energy information analysis unit and a defect information analysis unit.
8. The deep learning based atomic scale high dose radiation damage prediction system of claim 7, wherein:
a data processing unit: the method comprises the steps of collecting a plurality of irradiation damage simulation result samples of different atomic systems, different initial conditions and different boundary conditions, and carrying out data preprocessing and data division;
training unit: the method is used for performing deep learning training and constructing a deep learning model;
test unit: for quantitatively evaluating the predictive model.
9. The deep learning based atomic scale high dose radiation damage prediction system of claim 8, wherein:
a system establishment unit: establishing an atomic system according to the element types and the atomic system lattice parameters obtained through interaction with a user, and initializing initial acting force and initial energy parameters of atoms in the atomic system;
a molecular statics simulation unit and a molecular dynamics simulation unit: the method is used for performing high-dose damage simulation on an atomic system;
and a visualization unit: and the system is used for outputting a 3D atomic system graph and visualizing the information obtained by the analysis module.
10. The deep learning based atomic scale high dose radiation damage prediction system of claim 9, wherein:
the thermodynamic property information analysis unit is used for analyzing thermodynamic property information of the atomic system, including system temperature, system enthalpy change and entropy change;
an energy analysis unit: the method is used for analyzing energy information of an atomic system, including system energy, vacancy forming energy and surface energy;
defect analysis unit: the method is used for analyzing defects of an atomic system to obtain interstitial atoms, vacancy quantity and position parameters of the system and obtain structural evolution, defect formation and motion information of the material.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311798562.0A CN117727401A (en) | 2023-12-25 | 2023-12-25 | Atomic scale high-dose irradiation damage prediction method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311798562.0A CN117727401A (en) | 2023-12-25 | 2023-12-25 | Atomic scale high-dose irradiation damage prediction method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117727401A true CN117727401A (en) | 2024-03-19 |
Family
ID=90201451
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311798562.0A Pending CN117727401A (en) | 2023-12-25 | 2023-12-25 | Atomic scale high-dose irradiation damage prediction method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117727401A (en) |
-
2023
- 2023-12-25 CN CN202311798562.0A patent/CN117727401A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Schoofs | Experimental design and structural optimization | |
| Ghannadi et al. | The application of PSO in structural damage detection: An analysis of the previously released publications (2005–2020) | |
| Jiang et al. | A holistic feature selection method for enhanced short-term load forecasting of power system | |
| CN114429078B (en) | Short-term wind power prediction method and system | |
| Aral et al. | Genetic algorithms in search of groundwater pollution sources | |
| Zhang et al. | A novel seepage behavior prediction and lag process identification method for concrete dams using HGWO-XGBoost model | |
| CN115034129B (en) | NOx emission concentration soft measurement method for thermal power plant denitration device | |
| CN106951995A (en) | A kind of EHV transmission electric field extreme learning machine predicts multiple-objection optimization screen method | |
| Kookalani et al. | Structural analysis of GFRP elastic gridshell structures by particle swarm optimization and least square support vector machine algorithms | |
| Rezaee et al. | A hybrid approach based on inverse neural network to determine optimal level of energy consumption in electrical power generation | |
| CN117786824B (en) | Tunnel environment parameter design method and system based on multi-objective optimization | |
| Pitike et al. | Accurate Fe–He machine learning potential for studying He effects in BCC-Fe | |
| Kim et al. | Efficient combinatorial optimization under uncertainty. 1. Algorithmic development | |
| Zhang et al. | Structural system identification and damage detection using adaptive hybrid Jaya and differential evolution algorithm with mutation pool strategy | |
| Guo et al. | Improved cat swarm optimization algorithm for assembly sequence planning | |
| CN117556713A (en) | Uncertainty quantization method for CFD multi-credibility high-dimensional correlation flow field | |
| CN117727401A (en) | Atomic scale high-dose irradiation damage prediction method and system based on deep learning | |
| Tung et al. | An application study of response surface method on axial gadolinium designs of BWR fuel assemblies | |
| Jones et al. | Adapting data-driven techniques to improve surrogate machine learning model performance | |
| Weerasinghe et al. | Molecular-Dynamics Analysis of the Mechanical Behavior of Plasma-Facing Tungsten | |
| Raza et al. | Towards explainable message passing networks for predicting carbon dioxide adsorption in metal-organic frameworks | |
| CN112199765A (en) | Partitioning method and device for concrete dam, electronic equipment and readable medium | |
| Li | The optimization method of CNC lathe performance based on Morris sensitivity analysis and improved GA algorithm | |
| Dheeshjith et al. | Transfer Learning for Emulating Ocean Climate Variability across $ CO_2 $ forcing | |
| Khutia et al. | Material parameter optimisation of Ohno-Wang kinematic hardening model using multi objective genetic algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |