High-throughput sample preparation method for research on damage of ion irradiation microstructure

Abstract

The invention discloses a high-throughput sample preparation method for ion irradiation microstructure damage research. The high-throughput sample preparation method comprises the following steps: pretreating a nuclear structure material; according to a pre-trained prediction model, parameters of machining equipment for machining through the femtosecond laser machining technology are obtained; carrying out rough machining on the nuclear structure material by adopting a femtosecond machining technology to obtain a rough sample; the rough samples are screened, fine machining is conducted on the rough samples meeting the screening standard through the FIB technology, the samples meeting the high-throughput requirement are obtained, and machining equipment parameters of FIB fine machining are obtained according to a pre-trained prediction model; enabling the to-be-detected surface to be perpendicular to the irradiation direction, and then carrying out an irradiation damage experiment; and the irradiated sample is thinned to be less than 80 nm by using an FIB technology, and is well preserved for subsequent TEM testing.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ion irradiation sample preparation-characterization, in particular to a high-throughput sample preparation method for ion irradiation microstructure damage research. BACKGROUND

[0002] In the process of microstructure analysis of nuclear structure materials after ion irradiation, the sample with a depth greater than 3 μm needs to be extracted from the surface layer of the sample after irradiation, because the depth of the irradiation effect in the material is less than 2 μm. At the same time, the microstructure defects caused by ion irradiation are nanoscale, which can only be analyzed and characterized by transmission electron microscopy and atom probe, so the thickness of the prepared sample is less than 80 nm. The conventional sample preparation method suitable for the above two analysis and characterization methods and suitable for all materials is only focused ion beam cutting technology (FIB), and the observable area of the sample prepared by FIB is greater than 4 μm 2 , and the processing precision is less than 10 nm. However, this method has high sample preparation cost and long processing time, and the processing time for each sample is 1.5-2.5 hours, and the price is 3000-5000 yuan.

[0003] In addition, for polycrystalline metal materials, the sensitivity of different grain orientations to irradiation is significantly different, and the characteristics of the same microstructure defects projected in different directions are greatly different. Some microstructure defects caused by irradiation need to be obtained under a specific band axis when observed by transmission electron microscopy. Whether the microstructure defect can be observed is related to the band axis (vector b) and the diffraction orientation (vector g). At the same time, the orientation of the transmission sample is fixed and the thickness is thin, so it cannot be guaranteed that the specific band axis or diffraction vector of a sample is observed. Therefore, the irradiated sample shows the current situation of multiple FIB sample preparation, resulting in high sample preparation cost, long processing time, and inability to meet the high-throughput testing requirements. In the sample preparation process, if multiple irradiated samples are welded on the same copper grid, the sample processed later will contaminate the sample processed before. In addition, the sample prepared by this method will fall off and deform when subjected to external force, and the situation of knocking, colliding, falling, etc. should be prevented.

[0004] Currently, when using FIB technology alone to prepare multiple samples on the same copper grid, welding can easily contaminate the sample, and when the sample is subjected to external force, the sample will fall off and deform. In addition, the femtosecond laser processing technology has a heat-affected zone, which causes the microstructure of the sample in the heat-affected zone to change, and should be removed before irradiation testing. SUMMARY

[0005] To achieve the above and other related purposes, the present application discloses a high-throughput sample preparation method for ion irradiation microstructure damage research, comprising: preprocessing the nuclear structure material; According to the pre-trained prediction model, the machining equipment parameters for machining by the femtosecond laser machining technology are obtained. The nuclear structure material is coarsely machined by the femtosecond machining technology to obtain a coarse sample; The coarse sample is screened, and the screened compliant coarse sample is finely machined by the FIB technology to obtain a sample meeting the high-throughput requirement, wherein the machining equipment parameters for the FIB fine machining are obtained according to the pre-trained prediction model. The surface to be tested is perpendicular to the irradiation direction, and then the irradiation damage experiment is performed; The irradiated sample is thinned to less than 80 nm by the FIB technology, and is well preserved for subsequent TEM testing; The training of the prediction model includes: Historical data of sample machining are collected, material characteristic parameters, machining equipment parameters and environmental condition parameters of each case are recorded as input variables, and sample machining success rate and sample quality indicators are recorded as output variables; The historical data are preprocessed; According to the preprocessed data, a multilayer perception neural network is trained to obtain a prediction model.

[0006] Preferably, the preprocessing of the nuclear structure material includes: The nuclear structure material is cut into a metal sheet of 4.5*4.5*0.5 mm, the metal sheet is polished to 100-200 mu m thick by sandpaper, and then the sample is electrolytic etched for 15 seconds to eliminate the stress on the surface of the sample, and the sample is taken out and cleaned with anhydrous ethanol, and the sample is dried and stored in vacuum.

[0007] Preferably, the preprocessing of the historical data includes: The data is cleaned to remove abnormal values, missing values and repeated values in the data; Feature encoding, the feature variables in the data are converted into numerical features by using one-hot encoding; The encoded features are standardized.

[0008] Preferably, the training of the prediction model according to the preprocessed data by using the multilayer perception neural network includes: The number of neurons is determined according to the number of input variables, and the number of neurons is the same as the number of input variables; A plurality of hidden layers are set, and the activation function uses the ReLU function; An output layer is set, for the sample machining success rate, the output layer is set with one neuron, and for the sample quality indicators, the number of neurons is set according to the number of indicators in the sample quality indicators, and the activation function is a linear function; The loss function of the sample preparation success rate adopts a binary cross-entropy function; and the loss function of the sample quality index adopts a mean square error function. The processed historical data is divided into a training set, a validation set and a test set; The training set is used for model training to obtain a preliminary prediction model; The validation set is used for evaluating the preliminary prediction model after training; The hyperparameters of the neural network are adjusted according to the evaluation results, including searching for optimal parameters in a predefined hyperparameter space by using a grid search or a random search; After the optimal hyperparameters are determined, the model is retrained using the combined data of the training set and the validation set to improve the performance of the model by making full use of more data; The retrained model is finally evaluated on the test set to obtain the evaluation results of the model on the test set. If the evaluation results meet the preset standard, the model is used as the prediction model; otherwise, the training is repeated.

[0009] Preferably, the material property parameters include material type, grain size, elastic modulus and thermal conductivity; the machining equipment parameters include pulse energy, pulse frequency, scanning speed, spot diameter and repetition frequency of femtosecond laser, ion source type, ion source brightness, ion beam energy, ion beam current and beam spot size of FIB; the environmental condition parameters include temperature, humidity and cleanliness level of the machining environment; and the sample quality index includes thickness uniformity, surface roughness and heat-affected zone size.

[0010] Preferably, the machining equipment parameters for femtosecond laser machining technology obtained according to the pre-trained prediction model include: The range of the machining equipment parameters for femtosecond laser machining technology is set to define the search space of the parameters; Multiple machining equipment parameter combinations are randomly initialized in the search space; Each parameter combination is input into the trained machine learning model together with given material and environmental parameters to predict the sample preparation success rate and the sample quality index and calculate the fitness; The particle swarm algorithm is used to find the optimal solution of the machining equipment parameter combinations until a termination condition is met, i.e., the maximum number of iterations is reached or the fitness converges, and the optimal parameter combination obtained is the recommended machining equipment parameter.

[0011] Preferably, the machining equipment parameters for FIB finishing obtained according to the pre-trained prediction model include: The range of the machining equipment parameters for FIB finishing is set to define the search space of the parameters; Multiple machining equipment parameter combinations are randomly initialized in the search space; inputting each parameter combination into the trained machine learning model with given material and environmental parameters, predicting the sample preparation success rate and sample quality indicators, and calculating the fitness; adopting the particle swarm optimization algorithm to find the optimal solution of the parameter combination of the processing equipment until the termination condition is met, i.e., the maximum number of iterations is reached or the fitness converges, and the final optimal parameter combination is the recommended processing equipment parameter.

[0012] Preferably, it comprises: When rough machining is performed using femtosecond laser processing technology, the optical imaging system of the femtosecond laser is used to screen the nuclear structure material, and the nuclear structure material is processed into a predetermined size and shape; When fine machining is performed using FIB technology, the FIB technology is used to remove the heat-affected zone of the femtosecond laser on the irradiation surface of the rough machining sample, and multiple microcolumns are processed to the same height, with an error of less than 50 nm.

[0013] Preferably, the fitness is: Fitness = w1 x sample preparation success rate + w2 x (1-thickness uniformity tolerance / thickness uniformity error) + w3 x (1-surface roughness tolerance / surface roughness error) + w4 x (1-heat-affected zone size tolerance / heat-affected zone size error) w 1 x sample preparation success rate w 2 x (1-thickness uniformity tolerance / thickness uniformity error) w 3 x (1-surface roughness tolerance / surface roughness error) w 4 x (1-heat-affected zone size tolerance / heat-affected zone size error) wherein: w1, w2, w3, and w4 are weight coefficients for adjusting the relative importance of each indicator in the fitness calculation; the sample preparation success rate is a binary indicator indicating whether the sample is successfully prepared using the current parameter combination; the thickness uniformity error is the deviation between the actual thickness uniformity and the target thickness uniformity; the thickness uniformity tolerance is the allowable range of thickness uniformity error; the surface roughness error is the deviation between the actual surface roughness and the target surface roughness; the surface roughness tolerance is the allowable range of surface roughness error; the heat-affected zone size error is the deviation between the actual heat-affected zone size and the target heat-affected zone size; and the heat-affected zone size tolerance is the allowable range of heat-affected zone size error.

[0014] In a second aspect, the present application provides a high-throughput sample preparation system for ion irradiation microstructure damage research, comprising: a pretreatment device for pretreating the nuclear structure material; a model training module for collecting historical data of sample preparation, recording material characteristic parameters, processing equipment parameters, and environmental condition parameters of each case as input variables, and recording sample preparation success rate and sample quality indicators as output variables; preprocessing the historical data; According to the pretreated data, a multilayer perception neural network is trained to obtain a prediction model; The femtosecond laser processing equipment is used to coarsely process the nuclear structure material by using the femtosecond processing technology to obtain a coarse sample; wherein the processing equipment parameters of the femtosecond laser processing technology are obtained according to the prediction model trained in advance; The FIB equipment is used to finely process the coarse sample by using the FIB technology to obtain a sample meeting the high-throughput requirement; wherein the processing equipment parameters of the FIB fine processing are obtained according to the prediction model trained in advance.

[0015] By using the above technical solution, multiple samples can be prepared on the same carrier without welding, the sample preparation efficiency is high, the cost is low, high-throughput sample preparation can be realized, the problems of high cost, long time consumption and non-batch in the current sample preparation method are effectively solved, the sample preparation efficiency is improved, the different grain orientations can be fully statistically sized and counted in the TEM test process after irradiation, and the qualitative and quantitative correlation between the grain orientation and the microstructure damage of the irradiation dose can be established. BRIEF DESCRIPTION OF DRAWINGS

[0016] The above and other features, advantages, and aspects of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings that illustrate the present disclosure by way of example. The drawings provided herein are for illustrative purposes only and constitute part of the detailed description. In the drawings, the same or similar reference signs refer to the same or similar elements, wherein: Figure 1 The flowchart of the embodiments of the present application is shown. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] With reference to Figure 1 The embodiments of the present application provide a high-throughput sample preparation method for ion irradiation microstructure damage research, comprising: The nuclear structure material is pretreated; According to the prediction model trained in advance, the processing equipment parameters of the femtosecond laser processing technology are obtained; The nuclear structure material is coarsely processed by using the femtosecond laser processing technology to obtain a coarse sample; The crude samples are screened, and the qualified crude samples are refined using FIB technology to obtain samples that meet the high-throughput requirements. The processing equipment parameters for FIB refining are obtained based on a pre-trained prediction model.

[0019] The surface to be tested is perpendicular to the irradiation direction before the irradiation damage experiment is performed. The irradiated sample was thinned to less than 80 nm using FIB technology and preserved intact for subsequent TEM testing.

[0020] Preferably, the pretreatment of the nuclear structure material includes: The core structure material was cut into metal sheets of 4.5*4.5*0.5 mm. The metal sheets were sanded to a thickness of 100~200μm. The sample was then electrolytically etched for 15 seconds to eliminate the stress on the sample surface. The sample was then removed and cleaned with anhydrous ethanol. After drying, the sample was vacuum stored.

[0021] The training of the prediction model includes: Collect historical data on sample preparation, record the material property parameters, processing equipment parameters, and environmental condition parameters for each case as input variables, and record the sample preparation success rate and sample quality indicators as output variables. Preprocess historical data; Based on the preprocessed data, a multilayer perceptron neural network is used for training to obtain a prediction model.

[0022] Preferably, the material property parameters include material type, grain size, elastic modulus, and thermal conductivity; the processing equipment parameters include the pulse energy, pulse frequency, scanning speed, spot diameter, and repetition frequency of the femtosecond laser, and the ion source type, ion beam energy, and ion beam current of the FIB; the environmental condition parameters include the temperature, humidity, and cleanliness level of the processing environment; and the sample quality indicators include thickness uniformity, surface roughness, and heat-affected zone size.

[0023] The material type is identified by its material number or name. The grain size is measured using an optical microscope or electron backscatter diffraction technique, with the unit being micrometers. The elastic modulus is determined through mechanical experiments, with the unit being gigapascals. The thermal conductivity is obtained using thermal analysis techniques such as laser flare, with the unit being watts per meter Kelvin.

[0024] The unit of laser pulse energy in femtosecond laser processing equipment is microjoule, the unit of pulse frequency is Hertz, and the unit of scanning speed is millimeters per second; the spot diameter is the diameter of the laser beam after focusing, and the unit is micrometer; the repetition frequency is the repetition frequency of the laser pulse, and the unit is Hertz; the unit of ion source in FIB equipment is ampere per square centimeter, the unit of ion beam energy is kilovolt, the unit of ion beam current is nanoampere, and the unit of beam spot size is nanometer.

[0025] The temperature of the processing environment is recorded in degrees Celsius using an environmental temperature sensor, the humidity is measured in percentage using a humidity sensor, and the cleanliness level is rated according to the standard of the laboratory environment, such as ISO 5, ISO 6, etc.

[0026] The thickness uniformity is calculated by measuring the thickness of the sample at multiple positions, and the ratio of the maximum deviation of the thickness to the average thickness is calculated and expressed in percentage, the surface roughness is measured using an atomic force microscope and expressed in nanometer root mean square value, and the heat-affected zone size is observed and measured using a scanning electron microscope and expressed in microns.

[0027] Preferably, the preprocessing of the historical data comprises: cleaning the data to remove outliers, missing values and duplicate values in the data; feature encoding, converting the feature variables in the data into numerical features using one-hot encoding; standardizing the encoded features.

[0028] Preferably, the training of the preprocessed data using a multilayer perception neural network to obtain a prediction model comprises: determining the number of neurons according to the number of input variables, the number of neurons being the same as the number of input variables; setting multiple hidden layers, the activation function being ReLU function, and the number of hidden layers being set by a person skilled in the art according to actual needs; setting an output layer, for the sample preparation success rate, setting 1 neuron in the output layer; for the sample quality indicators, setting neurons according to the number of indicators in the sample quality indicators, and the activation function being a linear function; wherein the loss function of the sample preparation success rate adopts a binary cross-entropy function, and the loss function of the sample quality indicators adopts a mean square error function; dividing the processed historical data into a training set, a validation set and a test set, the ratio being 7:1:2; using the training set to train the model to obtain a preliminary prediction model; using the validation set to evaluate the trained preliminary prediction model; adjusting the hyperparameters of the neural network according to the evaluation results, including using grid search or random search to find the optimal parameters in the predefined hyperparameter space; after determining the optimal hyperparameters, retraining the model using the combined data of the training set and the validation set to improve the performance of the model.

[0029] The retrained model is then evaluated on the test set to obtain the evaluation result. If the evaluation result meets the preset standard, the model is used as the prediction model; otherwise, the training is repeated.

[0030] Preferably, the parameters of the processing equipment obtained based on the pre-trained prediction model for processing using femtosecond laser technology include: Set the range of processing equipment parameters when performing femtosecond laser processing technology, and define the parameter search space; Randomly initialize multiple combinations of processing equipment parameters within the search space; Each parameter combination, along with given material and environmental parameters, is input into a trained machine learning model to predict sample preparation success rate and sample quality indicators, and to calculate fitness. The particle swarm optimization algorithm is used to find the optimal solution for the combination of processing equipment parameters until the termination condition is met, that is, the maximum number of iterations or fitness convergence is reached. The final optimal parameter combination is the recommended processing equipment parameters.

[0031] Preferably, the processing equipment parameters for the FIB finishing are obtained based on a pre-trained prediction model, including: Set the range of machining equipment parameters for FIB finishing and define the parameter search space; Randomly initialize multiple combinations of processing equipment parameters within the search space; Each parameter combination, along with given material and environmental parameters, is input into a trained machine learning model to predict sample preparation success rate and sample quality indicators, and to calculate fitness. The particle swarm optimization algorithm is used to find the optimal solution for the combination of processing equipment parameters until the termination condition is met, that is, the maximum number of iterations or fitness convergence is reached. The final optimal parameter combination is the recommended processing equipment parameters.

[0032] Preferably, the fitness is: fitness = w 1×sample preparation success rate+ w 2×(1-thickness uniformity tolerance / thickness uniformity error)+ w 3×(1-Surface roughness tolerance / Surface roughness error)+ w 4 × (1 - tolerance for heat-affected zone size / error for heat-affected zone size); Wherein: w1, w2, w3, w4 are weight coefficients, used to adjust the relative importance of each index in fitness calculation;The sample preparation success rate is a binary index, indicating whether the sample is successfully prepared using the current parameter combination;The thickness uniformity error is the deviation between the actual thickness uniformity and the target thickness uniformity;The thickness uniformity tolerance is the allowable thickness uniformity error range;The surface roughness error is the deviation between the actual surface roughness and the target surface roughness;The surface roughness tolerance is the allowable surface roughness error range;The heat affected zone size error is the deviation between the actual heat affected zone size and the target heat affected zone size;The heat affected zone size tolerance is the allowable heat affected zone size error range.

[0033] When rough machining is performed using femtosecond laser processing technology, the optical imaging system of the femtosecond laser is used to screen the nuclear structure material, and the nuclear structure material is processed into a predetermined size and shape; When fine machining is performed using FIB technology, the FIB technology is used to remove the heat affected zone of the femtosecond laser on the irradiation surface of the rough machining sample, and multiple microcolumns are processed to the same height, with an error of less than 50 nm.

[0034] When the target size and shape of the femtosecond laser are machined, microstructure damage occurs in the heat affected zone of the sample, and the heat affected zone of the irradiation surface needs to be removed by FIB, about 2-3 μm, and the size of the removed part needs to be reserved during size design, and the heights of multiple microcolumns need to be ensured to be the same to ensure that the irradiation dose of each microcolumn is the same during ion irradiation. In addition, the irradiation surface of multiple microcolumns needs to be flat, with an error of less than 50 nm. In the process of machining the sample, welding is not required, and multiple samples do not interfere with each other during the thinning process after irradiation, which is more conducive to TEM observation after the sample preparation is completed. The purpose of the present application is to provide a method for realizing high-throughput sample preparation, and the sample prepared according to the method can solve the problems of high cost, long time consumption and non-batch production in the current sample preparation method, and improve the sample preparation efficiency.

[0035] Embodiments of the present application also provide a high-throughput sample preparation system for ion irradiation microstructure damage research, comprising: A pretreatment device for pretreating the nuclear structure material; A model training module for collecting historical data of sample preparation, recording material characteristic parameters, processing equipment parameters and environmental condition parameters of each case as input variables, and recording sample preparation success rate and sample quality indicators as output variables; Pretreating the historical data; According to the pretreated data, a multilayer perception neural network is trained to obtain a prediction model; The femtosecond laser processing device is used for rough machining of a nuclear structure material by using a femtosecond processing technology to obtain a rough sample; wherein the processing device parameters of the femtosecond laser processing technology are obtained according to a pre-trained prediction model. The FIB device is used for fine machining of the rough sample by using a FIB technology to obtain a sample meeting a high-throughput requirement; wherein the processing device parameters of the FIB fine machining are obtained according to a pre-trained prediction model.

[0036] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless specifically so defined.

[0037] For the method embodiments, for the sake of simple description, they are all described as a combination of a series of actions, but those skilled in the art should know that the embodiments of the present application are not limited by the order of the described actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential for the embodiments of the present application.

[0038] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and the necessary general hardware platform. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments of the present application.

[0039] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A high-throughput sample preparation method for studying microstructure damage under ion irradiation, characterized in that, include: Pretreatment of nuclear structural materials; Based on the pre-trained prediction model, the parameters of the processing equipment for processing using femtosecond laser technology are obtained; Femtosecond laser processing technology was used to rough process the nuclear structure material to obtain a rough sample; The crude samples are screened, and the qualified crude samples are refined using FIB technology to obtain samples that meet the high-throughput requirements. The processing equipment parameters for refining using FIB technology are obtained based on a pre-trained prediction model. The surface to be tested is perpendicular to the irradiation direction before the irradiation damage experiment is performed. The irradiated sample was thinned to less than 80 nm using FIB technology and preserved intact for subsequent TEM testing. The training of the prediction model includes: Collect historical data on sample preparation, record the material property parameters, processing equipment parameters, and environmental condition parameters for each case as input variables, and record the sample preparation success rate and sample quality indicators as output variables. Preprocess historical data; Based on the preprocessed data, a multilayer perceptron neural network is used for training to obtain a prediction model.

2. The method according to claim 1, characterized in that, The pretreatment of the nuclear structure material includes: The core structure material was cut into metal sheet samples of 4.5*4.5*0.5 mm. The samples were sanded to a thickness of 100~200 μm, and then electrolytically etched for 15 seconds to eliminate the stress on the sample surface. The samples were then removed, cleaned with anhydrous ethanol, dried, and vacuum stored.

3. The method according to claim 1, characterized in that, The preprocessing of historical data includes: Clean the data to remove outliers, missing values, and duplicate values. Feature encoding uses one-hot encoding to convert feature variables in the data into numerical features. The encoded features are then standardized.

4. The method according to claim 1, characterized in that, The step of training a prediction model using a multilayer perceptron neural network based on the preprocessed data includes: The number of neurons is determined based on the number of input variables, and the number of neurons is the same as the number of input variables. Multiple hidden layers are set, and the ReLU function is used as the activation function; The output layer is configured such that, for sample preparation success rate, one neuron is set in the output layer; for sample quality indicators, the number of neurons is set according to the number of indicators in the sample quality indicators, and the activation function is a linear function. Among them, the loss function for sample preparation success rate adopts the binary cross-entropy function; the loss function for sample quality index adopts the mean square error function. The processed historical data is divided into training set, validation set and test set; The model is trained using the training set to obtain a preliminary prediction model; The initial prediction model after training is evaluated using a validation set; The hyperparameters of the neural network are adjusted based on the evaluation results, including searching for the optimal parameters in a predefined hyperparameter space using grid search or random search. After determining the optimal hyperparameters, the model is retrained using a combination of training and validation data to make full use of more data and improve the model's performance. The retrained model is then evaluated on the test set to obtain the evaluation result. If the evaluation result meets the preset standard, the model is used as the prediction model; otherwise, the training is repeated until the evaluation result meets the preset standard.

5. The method according to claim 1, characterized in that, The material property parameters include material type, grain size, elastic modulus, and thermal conductivity; the processing equipment parameters include the pulse energy, pulse frequency, scanning speed, spot diameter, and repetition frequency of the femtosecond laser, and the ion source type, ion beam energy, and ion beam current of the FIB; the environmental condition parameters include the temperature, humidity, and cleanliness level of the processing environment; and the sample quality indicators include thickness uniformity, surface roughness, and heat-affected zone size.

6. The method according to claim 1, characterized in that, The parameters of the processing equipment obtained based on the pre-trained prediction model using femtosecond laser processing technology include: Set the range of processing equipment parameters when performing femtosecond laser processing technology, and define the parameter search space; Randomly initialize multiple combinations of processing equipment parameters within the search space; Each parameter combination, along with given material and environmental parameters, is input into a trained machine learning model to predict sample preparation success rate and sample quality indicators, and to calculate fitness. The particle swarm optimization algorithm is used to find the optimal solution for the combination of processing equipment parameters until the termination condition is met, that is, the maximum number of iterations or fitness convergence is reached. The final optimal parameter combination is the recommended processing equipment parameters.

7. The method according to claim 1, characterized in that, The processing equipment parameters for FIB finishing are obtained based on a pre-trained prediction model and include: Set the range of machining equipment parameters for FIB finishing and define the parameter search space; Randomly initialize multiple combinations of processing equipment parameters within the search space; Each parameter combination, along with given material and environmental parameters, is input into a trained machine learning model to predict sample preparation success rate and sample quality indicators, and to calculate fitness. The particle swarm optimization algorithm is used to find the optimal solution for the combination of processing equipment parameters until the termination condition is met, that is, the maximum number of iterations or fitness convergence is reached. The final optimal parameter combination is the recommended processing equipment parameters.

8. The method according to claim 1, characterized in that, include: When using femtosecond laser processing technology for rough processing, the optical imaging system of the femtosecond laser is used to screen the core structure material and process the core structure material into a preset size and shape; When using FIB technology for finishing, the heat-affected zone of the femtosecond laser on the surface of the rough-processed sample is removed by FIB technology, and multiple micropillars are processed to the same height so that the error is less than 50 nm.

9. The method according to claim 6 or 7, characterized in that, The fitness is: fitness = w 1×sample preparation success rate+ w 2×(1-thickness uniformity tolerance / thickness uniformity error)+ w 3×(1-Surface roughness tolerance / Surface roughness error)+ w 4 × (1 - tolerance for heat-affected zone size / error for heat-affected zone size); in: w1, w2, w3, and w4 are weighting coefficients used to adjust the relative importance of each indicator in the fitness calculation; sample preparation success rate is a binary indicator that indicates whether sample preparation was successful using the current parameter combination; thickness uniformity error is the deviation between the actual thickness uniformity and the target thickness uniformity; thickness uniformity tolerance is the allowable range of thickness uniformity error; surface roughness error is the deviation between the actual surface roughness and the target surface roughness; surface roughness tolerance is the allowable range of surface roughness error; heat-affected zone size error is the deviation between the actual heat-affected zone size and the target heat-affected zone size; heat-affected zone size tolerance is the allowable range of heat-affected zone size error.

10. A high-throughput sample preparation system for studying microstructure damage under ion irradiation, characterized in that, include: A pretreatment device for pretreating nuclear structural materials; The model training module is used to: collect historical data on sample preparation, record material property parameters, processing equipment parameters, and environmental condition parameters for each case as input variables, and record sample preparation success rate and sample quality indicators as output variables. Preprocess historical data; Based on the preprocessed data, a multilayer perceptron neural network is used for training to obtain a prediction model; Femtosecond laser processing equipment is used to rough process nuclear structure materials using femtosecond laser processing technology to obtain rough samples; the processing equipment parameters for the femtosecond laser processing technology are obtained based on a pre-trained prediction model; The FIB device is used to refine crude samples using FIB technology to obtain samples that meet high-throughput requirements. The processing parameters of the FIB technology for refining are obtained based on a pre-trained prediction model.

Images