Nuclear power steel grain size prediction method based on high-throughput characterization and machine learning

Abstract

The invention discloses a nuclear power steel grain size prediction method based on high-throughput characterization and machine learning, and relates to the technical field of material grain size analysis and prediction. Multivariable complex parameter combinations of a large amount of nuclear power austenitic stainless steel under different thermal deformation and thermal treatment conditions are obtained at a time through the high-throughput characterization technology; high-precision grain size information is obtained in combination with an EBSD large-area splicing technology, a reliable experimental database is established by using a WOA-XGBoost machine learning algorithm, and accurate prediction of the grain size of a large-size sample is realized. The method not only improves the data acquisition efficiency, but also remarkably reduces the research and production cost, optimizes the process parameters, and improves the material performance and quality. In addition, the development of an intelligent manufacturing technology is promoted, the intelligence and automation of material research and development and production processes are promoted, and the goal of sustainable development can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of material grain size analysis and prediction, and in particular to a method for predicting the grain size of nuclear power steel based on high-throughput characterization and machine learning. Background Art

[0002] In the production of nuclear power main pipelines, due to the difficulty in controlling the grain size, coarse grains and mixed grains are easily produced, resulting in the scrapping of forgings and causing huge losses. Therefore, it is of great significance to accurately evaluate the grain size of nuclear power materials.

[0003] The size and uniformity of grain size have an important influence on the service performance of products. Thermal deformation (such as forging, rolling) and heat treatment in hot manufacturing play a decisive role in grain configuration. Accurate prediction models for grain evolution are mostly material experiments combined with phenomenological models and physical models. However, general phenomenological models and physical models do not consider the coupling effects of thermal deformation and heat treatment processes, which are inconsistent with the actual production process, and the prediction results cannot reflect the actual production situation. Material experiments often use multiple small cylindrical specimens for compression experiments. Each hot compression specimen only corresponds to grain information under a certain fixed deformation temperature, constant strain rate, and strain amount. The experimental statistical samples are severely limited, which often leads to large model calculation errors and slow calculation speed. Especially for the prediction of grain size within the gradient strain range of large-size forgings, the amount of data involved in the generation process of large-size forgings is very large, which makes the traditional model too time-consuming, and the small cylindrical specimens used in the laboratory stage cannot obtain the continuous gradient strain results of the real forgings in the production process, resulting in inaccurate model predictions or even failure to produce results. Summary of the invention

[0004] In view of this, the present invention provides a method for predicting the grain size of nuclear power steel based on high-throughput characterization and machine learning, so as to realize rapid prediction of the grain size of large-size forgings under continuous gradient strain conditions.

[0005] To this end, the present invention provides the following technical solutions:

[0006] The present invention provides a method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning, comprising:

[0007] Prepare samples required for austenitic stainless steel experiments for nuclear power;

[0008] Conduct several groups of thermal deformation and heat treatment experiments on the samples, perform high-throughput characterization on the samples after thermal deformation and heat treatment, and obtain large-area spliced ​​electron backscatter diffraction EBSD images of the samples;

[0009] The large-area spliced ​​image of electron backscatter diffraction (EBSD) is divided into multiple sub-images, each of which contains at least one grain;

[0010] A plane rectangular coordinate system is established with the core of the sample as the central axis, and the coordinates of the center point of each sub-image and the average grain size in the sub-image are counted. The average grain size is assigned to the central coordinate to construct a grain size data set for each region.

[0011] A thermal compression test simulation is performed on the sample, and the strain values ​​corresponding to each center coordinate after the simulation are extracted to construct a strain data set;

[0012] Combine the grain size data set and strain data set of each region to jointly construct the process parameters and grain size data set of each point of austenitic stainless steel for nuclear power;

[0013] Constructing a machine learning-based austenitic stainless steel grain size prediction model, and training the prediction model using the process parameters of each point of austenitic stainless steel for nuclear power and the grain size data set after solid solution;

[0014] The trained prediction model is used to predict the grain size of austenitic stainless steel.

[0015] Furthermore, the machine learning model is WOA-XGBoost.

[0016] Furthermore, several groups of thermal deformation and heat treatment experiments were carried out on the samples, and high-throughput characterization was performed on the samples after thermal deformation and heat treatment to obtain large-area EBSD stitching images of the samples, including:

[0017] Austenitic stainless steel for nuclear power is processed into double wedge specimens;

[0018] Homogenize the sample;

[0019] The homogenized samples were subjected to thermal deformation test on a hydraulic press;

[0020] The thermally deformed sample is cut along the long axis direction to obtain a No. 1 sample and a No. 2 sample in a mirror image relationship; the No. 1 sample and the No. 2 sample are rectangular samples from the center to the edge of the sample;

[0021] Sample No. 2 was selected for solution treatment, and then sample No. 2 was air-cooled to room temperature;

[0022] After cutting the No. 1 and No. 2 specimens into multiple sections, the surfaces of the specimens were mechanically polished using sandpaper and various diamond suspensions;

[0023] An EBSD large-area mosaic image of the sample was obtained using an emission scanning electron microscope.

[0024] Furthermore, thermal deformation experiments were conducted on the samples, including deformation temperatures of 950°C, 1000°C, 1050°C, 1100°C, 1150°C and 1180°C, and a deformation speed of 5 mm / s.

[0025] Furthermore, the size of the No. 1 sample and the No. 2 sample obtained after cutting is 80 mm×5 mm.

[0026] Furthermore, sample No. 2 is subjected to a solution treatment, including: subjecting sample No. 2 to a solution treatment at 1060° C. for 4 hours.

[0027] Furthermore, after cutting sample No. 1 and sample No. 2 into multiple sections, the sample surfaces were mechanically polished using sandpaper and a variety of diamond suspensions, including: after cutting the samples into four sections of 17 mm, 19 mm, 21 mm, and 23 mm, the sample surfaces were mechanically polished using sandpaper and 3 μm, 1 μm, and 0.02 μm diamond suspensions.

[0028] Furthermore, the large-area EBSD stitching results were divided into multiple 800 × 800 μm subsets.

[0029] Advantages and positive effects of the present invention: The present invention uses high-throughput characterization technology to obtain a large number of multivariate complex parameter combinations of austenitic stainless steel for nuclear power under different thermal deformation and heat treatment conditions at one time, combines EBSD large-area stitching technology to obtain high-precision grain size information, and uses the WOA-XGBoost machine learning algorithm to establish a reliable experimental database to achieve accurate prediction of the grain size of large-size specimens; a large amount of data from high-throughput characterization can be used as training data for machine learning, which can provide application conditions for machine learning to achieve large-size specimen grain size prediction. This method not only improves data collection efficiency, but also significantly reduces research and production costs, while optimizing process parameters and improving material performance and quality. In addition, the invention also promotes the development of intelligent manufacturing technology, promotes the intelligence and automation of material research and development and production processes, and helps achieve the goal of sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1 It is a framework diagram of a nuclear power steel grain size prediction model based on high-throughput characterization and machine learning in an embodiment of the present invention;

[0032] Figure 2 This is a flow chart of a method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning in an embodiment of the present invention;

[0033] Figure 3 The figure is a specific dimension diagram of the double wedge specimen of X2 CrNiMo 18.12 (CN) austenitic stainless steel in the embodiment of the present invention;

[0034] Figure 4 This is a process diagram of a double wedge specimen thermal deformation experiment in an embodiment of the present invention;

[0035] Figure 5 Schematic diagram of double wedge specimen cutting in an embodiment of the present invention;

[0036] Figure 6 This is a schematic diagram of the sample characterization area in an embodiment of the present invention;

[0037] Figure 7 It is a strain cloud diagram of the deformation temperature of 950°C, 1000°C, 1050°C, 1100°C, 1150°C, and 1180°C of the deform simulation in the embodiment of the present invention;

[0038] Figure 8 It is a quantitative analysis diagram of the equivalent strain distribution on the center line of the cross section of the sample in the embodiment of the present invention;

[0039] Fig. 9 The IPF diagram of the sample with a heat deformation temperature of 1180°C after solid solution and the strain contour diagram of the sample with a heat deformation temperature of 1180°C after solid solution in the embodiment of the present invention;

[0040] Fig.10 This is a performance diagram of the WOA-XGBoost model in an embodiment of the present invention;

[0041] Fig.11 This is a graph of the prediction results of the WOA-XGBoost model test set in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0043] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0044] like Figure 1-2 As shown, a method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning in an embodiment of the present invention comprises the following steps:

[0045] Step 1: Prepare the samples required for the austenitic stainless steel experiment for nuclear power;

[0046] In this embodiment, X2 CrNiMo 18.12 (CN) austenitic stainless steel is used and processed into a double wedge-shaped specimen; the long axis of the specimen is 140 mm, the total height is 45 mm, the thickness is 50 mm, and the wedge side height is 20 mm. The specific sample dimensions are as follows: Figure 3 shown.

[0047] Step 2: Perform several groups of thermal deformation and heat treatment experiments on the sample, perform high-throughput characterization on the sample after thermal deformation and heat treatment, and obtain a large-area spliced ​​electron backscatter diffraction EBSD image of the sample;

[0048] In actual product production, the grains in each region of the billet evolve under the coupling of various inhomogeneous physical fields, and the combination of multivariable complex parameters determines the final grain size distribution of the billet. However, most existing studies only predict the final grain size from the perspective of a single process of thermal deformation or heat treatment, and few people consider the impact of the parameter coupling of thermal deformation and heat treatment processes on the grain size prediction model.

[0049] High-throughput experiments can obtain a large amount of characterization data. With the development of high-throughput characterization technology and big data analysis methods for material micro-region composition and performance, the research and development of new materials based on the idea of ​​material genome engineering has been greatly promoted, becoming an important method for modern material design and process design.

[0050] In this embodiment, high-throughput characterization technology is used to obtain a large number of multivariate complex parameter combinations of austenitic stainless steel for nuclear power under different thermal deformation and heat treatment conditions at one time, which greatly improves the data collection efficiency.

[0051] In specific implementation, the steps of high-throughput characterization specifically include:

[0052] Step 201, heat the sample at 1200°C for 8 hours to perform homogenization treatment.

[0053] Step 202, subjecting the homogenized X2 CrNiMo 18.12 (CN) austenitic stainless steel double wedge specimen to a hot deformation test on a 20MN hydraulic press;

[0054] The deformation temperatures are 950°C, 1000°C, 1050°C, 1100°C, 1150°C, and 1180°C, the deformation is 25 mm, and the pressing rate is 5 mm / s. Figure 4 shown.

[0055] Step 203, cutting the thermally deformed sample along the long axis direction to obtain a rectangular sample from the center to the edge of the sample;

[0056] like Figure 5 As shown, the cut samples include sample No. 1 and sample No. 2, with a size of 80 mm × 5 mm, and sample No. 1 and sample No. 2 are in a mirror image relationship.

[0057] Step 204, select sample No. 2 and carry out solution treatment at 1060°C for 4 hours, and then air-cool the sample to room temperature.

[0058] Step 205: Cutting and EBSD characterization of sample No. 1 and sample No. 2. Figure 6 As shown, the sample was cut into four sections of 17 mm, 19 mm, 21 mm, and 23 mm, and then the surface of the sample was mechanically polished using sandpaper and 3 μm, 1 μm, and 0.02 μm diamond suspensions.

[0059] Step 206: Use a Zeiss Sigma 300 field emission scanning electron microscope to obtain an electron backscatter diffraction (EBSD) image with a voltage of 20 kV and a working distance of 17 mm.

[0060] Step 3: Divide the large-area spliced ​​image of electron backscatter diffraction (EBSD) into multiple sub-images, each of which contains at least one grain; establish a plane rectangular coordinate system with the core of the sample as the central axis, count the coordinates of the center point of each sub-image and the average grain size in the sub-image, assign the average grain size to the central coordinates, and construct a grain size data set for each region;

[0061] In the specific implementation, the EBSD large-area stitching results are divided into multiple 800×800μm subsets (one sub-image is a subset, and each sub-image contains at least one grain), and a plane rectangular coordinate system is established with the core of the sample as the central axis. The coordinates of the center point of each subset and the average grain size in the subset are counted, and the average grain size is assigned to the central coordinate to construct the grain size data set of each region. The EBSD large-area stitching technology is used to obtain high-precision grain size information, providing an accurate data basis for the model. In another embodiment, the EBSD large-area stitching image can also be segmented in other forms, such as segmenting the EBSD large-area stitching image into multiple subsets of different sizes, each subset is a sub-image, and each sub-image contains a fixed number (such as 40) of grains.

[0062] Step 4: Perform a thermal compression test simulation on the sample, extract the strain values ​​corresponding to each center coordinate after the simulation, and construct a strain data set;

[0063] In the specific implementation, finite element analysis software such as deform is used to simulate the hot compression experiment. The deformation temperatures are 950°C, 1000°C, 1050°C, 1100°C, 1150°C, and 1180°C, and the deformation speed is 5 mm / s. The strain values ​​corresponding to the center coordinates after the simulation are extracted to construct a strain data set, such as Figure 7 shown.

[0064] Step 5: Combine the grain size datasets and strain datasets of each region to jointly construct the process parameters of each point and the grain size dataset after solid solution for austenitic stainless steel for nuclear power;

[0065] Quantitative analysis of the equivalent strain distribution on the center line of the specimen cross section is as follows: Figure 8 As shown; the IPF diagram of the sample with heat deformation temperature of 1180℃ after solid solution is as follows Fig. 9 (a) is shown; the strain contour diagram of the sample after solution treatment at a heat deformation temperature of 1180°C is shown in Fig. 9 (b) as shown.

[0066] Step 6: Construct a prediction model for austenitic stainless steel grain size based on machine learning, and train the prediction model using the process parameters of each point of austenitic stainless steel for nuclear power and the grain size data set after solid solution;

[0067] Preferably, the machine learning model can use WOA-XGBoost to construct an austenitic stainless steel grain size prediction model based on WOA-XGBoost, and the prediction model is trained using a large amount of hot deformation-solid solution grain size data obtained from high-throughput characterization experiments.

[0068] WOA-XGBoost is a machine learning model that combines the Whale Optimization Algorithm (WOA) and the XGBoost (Extreme GradientBoosting Model) algorithm to improve the prediction accuracy of regression and classification tasks.

[0069] XGBoost is an algorithm based on Gradient Boosting Decision Tree (GBDT), which is often used to solve classification and regression problems. Both XGBoost and GBDT are models based on the idea of ​​boosting integration. Compared with traditional GBDT, XGboost uses the second-order Taylor expansion to optimize the loss function and improve the calculation accuracy; uses regularization to simplify the model to avoid model overfitting; and uses the Blocks storage structure to enable the model to perform parallel calculations to improve the calculation speed. XGBoost consists of multiple weak learners, and the result of the previous weak learner will affect the generation of the next weak learner, and its influencing variable is called deviation. Therefore, the objective function E of XGBoost consists of two parts: the loss function L and the model complexity function Ω. The mathematical expression of the objective function E is as follows:

[0070] E=L+Ω (1)

[0071]

[0072] where y i is the true value of the i-th sample, is the predicted value of the i-th sample, T is the total number of nodes, and γ and λ are both penalty terms. The smaller the objective function, the higher the model accuracy. In order to minimize the objective function, a new function g is introduced. i With h i , and Taylor expansion is performed on equation (2) to simplify the objective function as follows:

[0073]

[0074] XGBoost is an improved GBDT algorithm that can train models faster and more efficiently.

[0075] WOA is a new type of swarm intelligence optimization search method, which originates from the simulation of the hunting behavior of humpback whale groups in nature. The algorithm simulates the whale population blowing bubbles while hunting, gradually restricting the range of activity of prey along a spiral path, and finally completing the foraging behavior. The WOA algorithm divides whale hunting behavior into three strategies, namely search and foraging strategy, contraction and encirclement strategy, and spiral update strategy. The target prey to be captured by the whale is assumed to be the optimal solution, and the position of each whale is assumed to be a potential solution. After knowing the position of the prey, each whale continuously updates its position according to the distance from the prey position, gradually encircling the prey, and uses this behavior of updating the position and gradually encircling the prey to act as an iterative algorithm in the computer. This iterative algorithm is expressed by the following formula:

[0076]

[0077] Where k represents the current number of iterations, represents the position vector of the optimal solution before the iteration, and Represent the solution vectors of the kth and k+1th iterations respectively. and is a constant vector, and its calculation formula is as follows:

[0078]

[0079] in, and Represents two random vectors ranging between 0 and 1. is a vector that decreases linearly from 2 to 0 as the number of iterations increases. Its calculation formula is:

[0080]

[0081] Where I represents the total number of iterations. When each whale shrinks its position according to equation (2), it will spit out bubbles along a spiral path, gradually narrowing the activity range of its prey. The formula for simulating its spiral motion is as follows:

[0082]

[0083] The shape of its spiral path is controlled by b and l, where l is a random number between -1 and 1. In order to achieve better hunting results, a random number p between 0 and 1 is generated to determine whether the algorithm chooses the bubble attack strategy. If p>0.5, the bubble attack strategy is entered, otherwise the vector The absolute value of determines whether to search for prey or surround prey. When , adopt the strategy of surrounding the prey; when When , a random search strategy is adopted. The mathematical expression of the random search strategy is as follows:

[0084]

[0085]

[0086] In the formula, represents a randomly selected whale position, i.e., a random solution.

[0087] Through research, it is found that A is of great significance to the prediction of grain evolution. According to these research contents, This largely regulates the ability of the WOA algorithm to predict grain evolution.

[0088] The WOA-XGBoost model uses the WOA algorithm to optimize the hyperparameters of XGBoost to improve the model's predictive performance. This model performs well when processing high-dimensional data and large-scale data sets. The model optimizes the objective function by simulating the behavior of whales looking for food, and gradually iteratively updates the position of candidate solutions to find the optimal hyperparameter combination. This method not only improves the performance of traditional XGBoost, but also demonstrates the application potential of swarm intelligence optimization algorithms in the field of machine learning.

[0089] The model training process is as follows:

[0090] Divide the data set: Divide the preprocessed data into a training set and a validation set, usually with a ratio of 80% training set and 20% validation set, for model training and performance verification.

[0091] Training model: The WOA-XGBoost model is trained using the training set data. During the training process, the model parameters are iteratively updated to minimize the difference between the predicted grain size and the actual grain size.

[0092] Verification and adjustment: Use the validation set to evaluate the prediction performance of the model and adjust the model settings based on the validation results, such as increasing the number of training rounds or further adjusting hyperparameters until a satisfactory prediction effect is obtained.

[0093] The model was evaluated using mean square error (MSE), coefficient of determination (R 2 ) and other statistical indicators to evaluate the prediction performance of the model to ensure that the model can accurately capture the relationship between process parameters and grain size. Fig.10 shown.

[0094] Step 7: Use the trained prediction model to predict the grain size of austenitic stainless steel.

[0095] The prediction results of the WOA-XGBoost model test set are as follows Fig.11As shown, the prediction model can accurately predict the grain size, reduce a large number of tests and repeated experiments in the industrial production process, and significantly reduce research and production costs.

[0096] In the above embodiment, high-throughput characterization is used to obtain several groups of multivariate complex parameter combinations of austenitic stainless steel for nuclear power at one time after thermal deformation and heat treatment, and a "process parameter-grain size" sample data set corresponding to the grain size of each area of ​​the sample is established. The EBSD large-area stitching technology is used to obtain the 80mm×5mm grain size information after solid solution. A reliable experimental database is established in combination with a machine learning algorithm, and the WOA-XGBoost model in the commonly used regression algorithm is used to train the experimental data set. The model can accurately predict the grain size of large-size samples corresponding to different thermal deformation parameters and solid solution parameters, greatly reducing the cost of testing and industrial production.

[0097] The method in the above embodiment is not only applicable to the prediction of grain size of steel materials for nuclear power, but can also be extended to the prediction and analysis of microstructure of other materials, and has wide applicability.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning, characterized in that: include: Prepare samples required for austenitic stainless steel experiments for nuclear power; Conduct several groups of thermal deformation and heat treatment experiments on the samples, perform high-throughput characterization on the samples after thermal deformation and heat treatment, and obtain large-area spliced ​​electron backscatter diffraction EBSD images of the samples; The large-area spliced ​​electron backscatter diffraction (EBSD) image is divided into multiple sub-images, each of which contains at least one grain; A plane rectangular coordinate system is established with the core of the sample as the central axis, and the coordinates of the center point of each sub-image and the average grain size in the sub-image are counted. The average grain size is assigned to the central coordinate to construct a grain size data set for each region. A thermal compression test simulation is performed on the sample, and the strain values ​​corresponding to each center coordinate after the simulation are extracted to construct a strain data set; Combine the grain size data set and strain data set of each region to jointly construct the process parameters and grain size data set of each point of austenitic stainless steel for nuclear power; Constructing a machine learning-based austenitic stainless steel grain size prediction model, and training the prediction model using the process parameters of each point of austenitic stainless steel for nuclear power and the grain size data set after solid solution; The trained prediction model is used to predict the grain size of austenitic stainless steel.

2. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 1, characterized in that: The machine learning model is WOA-XGBoost.

3. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 1, characterized in that: Conduct several sets of thermal deformation and heat treatment experiments on the sample, perform high-throughput characterization on the sample after thermal deformation and heat treatment, and obtain a large-area spliced ​​electron backscatter diffraction EBSD image of the sample, including: Austenitic stainless steel for nuclear power is processed into double wedge specimens; Homogenize the sample; The homogenized samples were subjected to thermal deformation test on a hydraulic press; The thermally deformed sample is cut along the long axis direction to obtain a No. 1 sample and a No. 2 sample in a mirror image relationship; the No. 1 sample and the No. 2 sample are rectangular samples from the center to the edge of the sample; Sample No. 2 was selected for solution treatment, and then sample No. 2 was air-cooled to room temperature; After cutting the No. 1 and No. 2 specimens into multiple sections, the surfaces of the specimens were mechanically polished using sandpaper and various diamond suspensions; An EBSD large-area mosaic image of the sample was obtained using an emission scanning electron microscope.

4. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 3 is characterized in that: Thermal deformation experiments were carried out on the samples, including deformation temperatures of 950°C, 1000°C, 1050°C, 1100°C, 1150°C and 1180°C, and a deformation speed of 5 mm / s.

5. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 3, characterized in that: The size of the No. 1 and No. 2 samples obtained after cutting is 80 mm×5 mm.

6. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 3, characterized in that: The No. 2 sample is subjected to a solution treatment, including: the No. 2 sample is subjected to a solution treatment at 1060° C. for 4 hours.

7. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 3, characterized in that: After cutting Sample No. 1 and Sample No. 2 into multiple sections, the surfaces of the samples were mechanically polished using sandpaper and a variety of diamond suspensions, including: after cutting the samples into four sections of 17 mm, 19 mm, 21 mm, and 23 mm, the surfaces of the samples were mechanically polished using sandpaper and 3 μm, 1 μm, and 0.02 μm diamond suspensions.

8. The method for predicting grain size of nuclear power steel based on high-throughput characterization and machine learning according to claim 1, characterized in that: The large-area EBSD stitching results were divided into multiple 800 × 800 μm subsets.

Images