Fuel microstructure prediction method based on generative deep learning model

Abstract

The invention relates to the technical field of computational material science, in particular to a fuel microstructure prediction method based on a generative deep learning model, and the method comprises the following steps: obtaining a data set; the data set comprises SEM photos under different process conditions; preprocessing the data set, and obtaining a training set based on the preprocessed data set; constructing a generative deep learning model based on the conditional variation auto-encoder model; inputting the data of the training set into a generative deep learning model, and carrying out model training to obtain a trained generative deep learning model; and inputting the process conditions into the trained generative deep learning model to generate a specific fuel microstructure. According to the method, the microstructure image is directly generated from the chemical components and the process preparation parameters by means of the generative deep learning model, professional limitation is avoided, the image consistent with the microscopic detection result can be directly output, and the problems that traditional experiments and simulation calculation are high in cost, long in consumed time, complex in process and the like are solved.

Description

Technical Field

[0001] The present application relates to the field of computational materials science and technology, and in particular to a method for predicting fuel microstructure based on a generative deep learning model. Background Art

[0002] Nuclear fuel is the core of nuclear reactors, and its performance is related to the performance and safety of the reactor. Previous studies have shown that adding Ti to UO2 can effectively improve the comprehensive performance of the fuel. Different Ti doping amounts, sintering temperatures, and sintering holding times will affect the microstructure of UO2 fuel. In order to obtain the microstructure of fuel under different conditions, traditional methods require a large number of experimental studies to obtain it. However, the experimental conditions of UO2 are relatively harsh, costly, and long, and the fuel microstructure prediction method is difficult to carry out.

[0003] At present, the prediction of microstructure generally adopts the modeling-simulation calculation process. The microstructure parameters, such as main phases and grain size, are obtained through simulation calculation methods such as first principles and thermodynamics. The method of using microstructure parameters to build a microstructure model has the following disadvantages: (1) The simulation calculation process is complicated and requires professional personnel and software to perform a large number of calculations. (2) There is a significant difference between the simulation calculation results and the actual results. Therefore, a fuel microstructure prediction method is urgently needed to reduce experimental work and get rid of professional limitations. Summary of the invention

[0004] The present application provides a fuel microstructure prediction method based on a generative deep learning model. With the help of a generative deep learning model, a microstructure image is directly generated from chemical composition and process preparation parameters, freeing itself from professional limitations and being able to directly output images that are consistent with microscopic detection results, thereby solving the problems of high cost, long time consumption, and complex process of traditional experiments and simulation calculations.

[0005] The present application provides a fuel microstructure prediction method based on a generative deep learning model, comprising the following steps: first, obtaining a data set; the data set includes SEM photos under different process conditions, and the different process conditions are composed of different chemical compositions, sintering temperatures, and holding time combinations; then, preprocessing the data set to obtain a preprocessed data set, and based on the preprocessed data set, obtaining a training set; next, constructing a generative deep learning model based on a conditional variational autoencoder model; and inputting the data of the training set into the generative deep learning model to perform model training to obtain a trained generative deep learning model; finally, inputting the process conditions into the trained generative deep learning model to generate a specific fuel microstructure.

[0006] In some exemplary embodiments, obtaining a data set includes: first, collecting SEM images of fuels for different chemical compositions and preparation process conditions; then, based on the SEM images of the fuels, establishing SEM photo data sets corresponding to different chemical compositions-sintering temperatures-holding times.

[0007] In some exemplary embodiments, the data set is preprocessed, including: preprocessing the images and experimental data in the data set respectively; wherein, the image is preprocessed, including: measuring the length of the ruler on the image, calculating the scale of the image, scaling the image based on the scale, converting the image into a grayscale image, and performing histogram equalization on the image to enhance the image contrast and image visual effect; then converting the image into an RGB image, and finally completing the image preprocessing through random cropping and horizontal and vertical flipping operations; the experimental data is preprocessed, including: using the mean variance normalization method to process the sintering temperature, Ti content and holding time variables, calculating the mean and standard deviation of each column and performing data conversion.

[0008] In some exemplary embodiments, a generative deep learning model is constructed based on a conditional variational autoencoder model; and the data of the training set is input into the generative deep learning model to perform model training to obtain a trained generative deep learning model, including: first, based on the conditional variational autoencoder model, the depth of the encoder and the decoder is increased, the conditional variational autoencoder model is optimized, and a generative deep learning model is constructed; then, training parameters are set, and the generative deep learning model is trained and optimized based on the training set to obtain a trained generative deep learning model.

[0009] In some exemplary embodiments, the depth of the encoder and decoder is increased, including: using an attention mechanism to embed different process conditions to increase the depth of the encoder and decoder; expanding the encoder and decoder to 6 units, each unit including a convolutional layer, a pooling layer, and a normalization layer, and the last two units are connected using a residual structure.

[0010] In some exemplary embodiments, when training a generative deep learning model, an encoder is used separately for each process condition image, and all images share a decoder; the input process conditions are combined into an embedding vector with a length of 10 through a linear layer, and the embedding vector and the latent vector of the encoder are concatenated and then input into the decoder; the loss function during training is composed of reconstruction loss and KL divergence loss.

[0011] In some exemplary embodiments, the reconstruction loss is expressed using a cross entropy loss, and the corresponding procedure is: F.BCE_loss(image, generated_image) The corresponding procedure for KL divergence loss is: kl_loss = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp()) Among them, mu is the mean of the latent variable and logvar is the standard deviation of the latent variable.

[0012] In some exemplary embodiments, process conditions are input into a trained generative deep learning model to generate a specific fuel microstructure, including: first, converting the input component process conditions into an input embedding vector through a linear layer, combining the embedding vector and latent variables generated by training, and using an attention mechanism to generate new latent variables; then, inputting the new latent variables into a trained decoder to generate a corresponding image.

[0013] In some exemplary embodiments, the input component process conditions are converted into input embedding vectors through a linear layer, and the embedding vectors and latent variables generated by training are combined to generate new latent variables using an attention mechanism, including: recording the input variables as a i , the latent variable is recorded as z i , where i is the sequence number, z i The latent variable generated for training consists of 64 numbers; the calculation formula of the latent variable is:

[0014] Among them, z4 is the latent variable of the test set data with sequence number 4, and a4 is the input variable.

[0015] In some exemplary embodiments, new latent variables are input into a trained decoder to generate corresponding images, which are used to show the grain shape and size of the microstructure.

[0016] The technical solution provided by the embodiments of the present application has at least the following advantages: The embodiment of the present application provides a fuel microstructure prediction method based on a generative deep learning model, comprising the following steps: first, obtaining a data set; the data set includes SEM photos under different process conditions, and the different process conditions are composed of different chemical compositions, sintering temperatures, and holding time combinations; then, preprocessing the data set to obtain a preprocessed data set, and based on the preprocessed data set, obtaining a training set; next, constructing a generative deep learning model based on a conditional variational autoencoder model; and inputting the data of the training set into the generative deep learning model to perform model training to obtain a trained generative deep learning model; finally, inputting the process conditions into the trained generative deep learning model to generate a specific fuel microstructure.

[0017] The fuel microstructure prediction method based on the generative deep learning model provided in this application can quickly realize the prediction of the fuel microstructure under different chemical compositions and preparation process conditions by using the generative deep learning method. The generated images can be directly compared with the SEM images to truly reflect the microstructure of the fuel, greatly reducing the workload of experiments and traditional simulation calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplifications do not constitute limitations on the embodiments. Unless otherwise stated, the pictures in the drawings do not constitute proportional limitations.

[0019] Figure 1 A schematic flow chart of a fuel microstructure prediction method based on a generative deep learning model provided in an embodiment of the present application.

[0020] Figure 2 A schematic diagram for comparing the original image and the preprocessed image provided in an embodiment of the present application.

[0021] Figure 3 The loss function provided in the embodiment of the present application changes with the training round curve.

[0022] Figure 4 A schematic diagram for comparing the original image and the generated image provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] As can be seen from the background technology, the traditional method requires a lot of experimental research to obtain the fuel microstructure under different conditions, while the experimental conditions of UO2 are relatively harsh, the cost is high, the cycle is long, and the fuel microstructure prediction method is difficult to carry out.

[0024] In order to solve the above technical problems, the embodiment of the present application provides a fuel microstructure prediction method based on a generative deep learning model, the method comprising the following steps: first, obtaining a data set; the data set includes SEM photos under different process conditions, and the different process conditions are composed of different chemical compositions, sintering temperatures and holding time combinations; then, preprocessing the data set to obtain a preprocessed data set, and based on the preprocessed data set, obtaining a training set; next, based on a conditional variational autoencoder model, constructing a generative deep learning model; and inputting the data of the training set into the generative deep learning model, performing model training, and obtaining a trained generative deep learning model; finally, inputting the process conditions into the trained generative deep learning model to generate a specific fuel microstructure. The fuel microstructure prediction method based on a generative deep learning model provided in the present application directly generates microstructure images from chemical composition and process preparation parameters with the help of a generative deep learning model, gets rid of professional restrictions, and can directly output images consistent with microscopic detection results, solving the problems of high cost, long time consumption, and complex process of traditional experiments and simulation calculations.

[0025] The various embodiments of the present application will be described in detail below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present application and implement it.

[0026] The following contents are all for the purpose of describing in detail the specific examples of the specific implementation process provided by the technical solutions to be protected in this application. However, this application may also adopt other implementation methods different from this description. Technical personnel in this field may, under the guidance of the ideas of this application, adopt technical solutions of different viewpoints to implement the contents within the scope of this application. Therefore, this application is not strictly limited to the specific implementation cases below.

[0027] See also Figure 1 The present application embodiment provides a fuel microstructure prediction method based on a generative deep learning model, comprising the following steps: Step S1, obtaining a data set; the data set includes SEM photos under different process conditions, and the different process conditions are composed of different combinations of chemical composition, sintering temperature and holding time.

[0028] Step S2: preprocess the data set to obtain a preprocessed data set, and obtain a training set based on the preprocessed data set.

[0029] Step S3: Based on the conditional variational autoencoder model, a generative deep learning model is constructed; and the data of the training set is input into the generative deep learning model to perform model training to obtain a trained generative deep learning model.

[0030] Step S4: input the process conditions into the trained generative deep learning model to generate a specific fuel microstructure.

[0031] In some embodiments, obtaining a data set in step S1 includes: collecting SEM images of the fuel for different chemical compositions and preparation process conditions; and establishing a SEM photo data set corresponding to different chemical compositions-sintering temperatures-holding times based on the SEM images of the fuel.

[0032] Specifically, in step S1, the data set required for model training is obtained through experiments and literature collection. The data type of the training data set is Ti content-sintering temperature-holding time-corresponding SEM photos.

[0033] In some embodiments, the data set is preprocessed in step S2, including: preprocessing the images and experimental data in the data set respectively; wherein, the image is preprocessed, including: measuring the length of the ruler on the image, calculating the scale of the image, scaling the image based on the scale, converting it into a grayscale image, and performing histogram equalization on the image to enhance the image contrast and image visual effect; then converting the image into an RGB image, and finally completing the image preprocessing by random cropping and horizontal and vertical flipping operations; the experimental data is preprocessed, including: using the mean variance normalization method to process the sintering temperature, Ti content and holding time variables, calculating the mean and standard deviation of each column and performing data conversion.

[0034] In some embodiments, in step S3, a generative deep learning model is constructed based on the conditional variational autoencoder model; and the data of the training set is input into the generative deep learning model to perform model training to obtain a trained generative deep learning model, including: first, based on the conditional variational autoencoder model, the depth of the encoder and the decoder is increased, the conditional variational autoencoder model is optimized, and a generative deep learning model is constructed; then, training parameters are set, and model training and optimization of the generative deep learning model are performed based on the training set to obtain a trained generative deep learning model.

[0035] Specifically, this application optimizes and improves the conditional variational autoencoder (CVAE) model for the scenario of predicting the microstructure of fuel, embeds the component process conditions with the attention mechanism, increases the depth of the encoder-decoder, sets training parameters, and performs model training and optimization based on the enhanced data set.

[0036] Finally, the specified Ti doping content, sintering temperature and sintering holding time data are input into the model, and the model can generate a specific fuel microstructure.

[0037] By using a generative deep learning method, this application can quickly predict the microstructure of fuels under different chemical compositions and preparation process conditions. The generated images can be directly compared with SEM images to truly reflect the microstructure of the fuel, greatly reducing the workload of experiments and traditional simulation calculations.

[0038] Specifically, in the process of establishing the data set, SEM images of the fuel are collected for different chemical compositions and preparation process conditions, and a chemical composition-sintering temperature-holding time-corresponding SEM photo data set is established. The data sources include actual experimental results and fuel microstructure images in the paper. In this embodiment, SEM images corresponding to 12 process conditions are collected. The Ti content, sintering temperature, and sintering holding time in the process conditions are specifically shown in Table 1. Among them, serial numbers 1, 4, 5, and 11 are from the literature, and the rest are experimental supplements.

[0039] Table 1 Temperature, Ti content and time data sets

[0040] In the process of data preprocessing, the image was preprocessed first. The ImageJ software was used to measure the length of the ruler on the image and calculate the scale of the image. The scale of sequence number 4 was the smallest and was not scaled. The other images were scaled according to the 4-bit standard. The specific measurement results and reduction ratios are shown in Table 2.

[0041] Table 2 Actual ratio and reduction ratio of images

[0042] Use opencv to read the reduced image, convert the read image into a grayscale image, use the cv2.equalizeHist function to perform histogram equalization on the image to increase the image contrast and visual effect, and then convert the image into an RGB image. Perform image enhancement before training. First use the RandomCrop method to randomly crop the image to a size of 256×256, and then randomly flip it horizontally and vertically (with a probability of 0.2). This operation can increase the amount of data. After preprocessing, the image is as follows Figure 2 As shown, Figure 2 The left side is the original image and the right side is the preprocessed image.

[0043] Then the experimental data were preprocessed. Since the range of the three variables of temperature, Ti content, and time is large, in order to eliminate the influence of their range, the three variables of temperature, Ti content, and time were processed using mean-variance normalization, that is, the mean and standard deviation of each column were calculated, and each data was subtracted from the mean of the column and divided by the standard deviation. Finally, the data were obtained as shown in Table 3.

[0044] Table 3 Data normalization results

[0045] We selected sequence numbers 4 and 11 from the data as the test set, and the rest of the data as the training set for training.

[0046] In some embodiments, the depth of the encoder and decoder is increased, including: using an attention mechanism to embed different process conditions to increase the depth of the encoder and decoder; expanding the encoder and decoder to 6 units, each unit including a convolutional layer, a pooling layer, and a normalization layer, and the last two units are connected using a residual structure.

[0047] In some embodiments, when training a generative deep learning model, an encoder is used separately for each image of a process condition, and all images share a decoder; the input process conditions are combined into an embedding vector of length 10 through a linear layer, and the embedding vector and the latent vector of the encoder are concatenated and then input into the decoder.

[0048] Specifically, in the process of building and training a generative deep learning model, the embodiment of the present application selected the CVAE model as the basic model and optimized it for the fuel application scenario. The specific optimization contents are: (1) Increase the depth of the encoder-decoder to 6 units, each unit includes a convolutional layer, a pooling layer and a normalization layer, and the last two units are connected with a residual structure.

[0049] (2) Each condition image uses a separate encoder and shares a decoder.

[0050] (3) The input conditions of chemical composition, sintering temperature, and holding time are combined into an embedding vector of length 10 through a linear layer.

[0051] (4) The embedding vector and the encoder’s latent vector are concatenated and then input into the decoder.

[0052] The loss function during training consists of two parts: reconstruction loss and KL divergence loss. The reconstruction loss uses cross entropy loss (Cross - EntropyLoss), and the corresponding program is: F.BCE_loss(image, generated_image) The corresponding procedure for KL divergence loss is: kl_loss = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp()) Where mu is the mean of the latent variable and logvar is the standard deviation of the latent variable.

[0053] For other parameter settings, the total number of epochs is 1000, the batch size is set to 4, the optimizer is SGD, BCE+KL is used as the loss function, the training learning rate decreases according to the cosine function, and the training is 1000 rounds. The parameters of the decoder, input conditional linear layer, latent variables, and the generated images are saved every 10 rounds. The loss function during training is as follows Figure 3 shown.

[0054] In some embodiments, in step S4, the process conditions are input into a trained generative deep learning model to generate a specific fuel microstructure, including: first, the input component process conditions are converted into an input embedding vector through a linear layer, and the embedding vector and latent variables generated by the training are combined to generate new latent variables using an attention mechanism; then, the new latent variables are input into a trained decoder to generate a corresponding image.

[0055] The specific process is as follows: (1) Let the input variable be a i , the latent variable is recorded as z i , i is the serial number, a i For the data in Table 3, z i The latent variable generated for training consists of 64 numbers. Taking number 4 in the test set as an example, the calculation formula of the latent variable is used for calculation:

[0056] (2) Input the new latent variable into the trained decoder to generate the corresponding image.

[0057] Figure 4 The original and generated images are shown in Figure 2, showing the main features of the microstructure - grain shape and size.

[0058] Based on the above technical scheme, an embodiment of the present application provides a fuel microstructure prediction method based on a generative deep learning model, including the following steps: first, obtaining a data set; the data set includes SEM photos under different process conditions, and the different process conditions are composed of different chemical compositions, sintering temperatures and holding time combinations; then, preprocessing the data set to obtain a preprocessed data set, and based on the preprocessed data set, obtaining a training set; next, based on a conditional variational autoencoder model, constructing a generative deep learning model; and inputting the data of the training set into the generative deep learning model to perform model training to obtain a trained generative deep learning model; finally, inputting the process conditions into the trained generative deep learning model to generate a specific fuel microstructure.

[0059] The fuel microstructure prediction method based on the generative deep learning model provided in this application can quickly realize the prediction of the fuel microstructure under different chemical compositions and preparation process conditions by using the generative deep learning method. The generated images can be directly compared with the SEM images to truly reflect the microstructure of the fuel, greatly reducing the workload of experiments and traditional simulation calculations.

[0060] Those skilled in the art can understand that the above-mentioned embodiments are specific examples for implementing the present application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present application. Any person skilled in the art can make their own changes and modifications without departing from the spirit and scope of the present application, so the scope of protection of the present application shall be based on the scope defined in the claims.

Claims

1. A fuel microstructure prediction method based on a generative deep learning model, characterized in that: The following steps are involved: Acquire a data set; the data set includes SEM photos under different process conditions, where the different process conditions are composed of different combinations of chemical composition, sintering temperature and holding time; Preprocessing the data set to obtain a preprocessed data set, and obtaining a training set based on the preprocessed data set; Build a generative deep learning model based on the conditional variational autoencoder model; and inputting the data of the training set into the generative deep learning model to perform model training to obtain a trained generative deep learning model; The process conditions are input into the trained generative deep learning model to generate the fuel-specific microstructure.

2. The fuel microstructure prediction method based on a generative deep learning model according to claim 1, characterized in that: Get the dataset, including: Collect SEM images of fuels with different chemical compositions and preparation process conditions; Based on the SEM images of the fuel, a SEM photo dataset corresponding to different chemical compositions, sintering temperatures and holding times was established.

3. The fuel microstructure prediction method based on a generative deep learning model according to claim 1, characterized in that: The data set is preprocessed, including: Preprocessing the images and experimental data in the data set respectively; Preprocessing the image includes: measuring the length of the ruler on the image, calculating the scale of the image, scaling the image based on the scale, converting it into a grayscale image, and performing histogram equalization on the image to enhance the image contrast and image visual effect; then converting the image into an RGB image, and finally completing the image preprocessing by random cropping and horizontal and vertical flipping operations; The experimental data were preprocessed, including: using the mean-variance normalization method to process the sintering temperature, Ti content and holding time variables, calculating the mean and standard deviation of each column and performing data conversion.

4. The fuel microstructure prediction method based on a generative deep learning model according to claim 1, characterized in that: Build a generative deep learning model based on the conditional variational autoencoder model; The data of the training set is input into the generative deep learning model to perform model training to obtain a trained generative deep learning model, including: Based on the conditional variational autoencoder model, increase the depth of the encoder and decoder, optimize the conditional variational autoencoder model, and build a generative deep learning model; The training parameters are set, and the generative deep learning model is trained and optimized based on the training set to obtain a trained generative deep learning model.

5. The fuel microstructure prediction method based on a generative deep learning model according to claim 4, characterized in that: Increased encoder and decoder depth, including: The attention mechanism is used to embed different process conditions and increase the depth of the encoder and decoder. The encoder and decoder are expanded to 6 units, each of which includes a convolutional layer, a pooling layer, and a normalization layer. The last two units are connected with a residual structure.

6. The fuel microstructure prediction method based on a generative deep learning model according to claim 5, characterized in that: When training the generative deep learning model, an encoder is used for each process condition image separately, and a decoder is shared by all images; the input process conditions are combined into an embedding vector with a length of 10 through a linear layer, and the embedding vector and the latent vector of the encoder are concatenated and then input into the decoder; The loss function during training consists of reconstruction loss and KL divergence loss.

7. The fuel microstructure prediction method based on a generative deep learning model according to claim 6, characterized in that: The reconstruction loss is expressed using cross entropy loss, and the corresponding procedure is: F.BCE_loss(image, generated_image) The corresponding procedure for KL divergence loss is: kl_loss = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp()) Among them, mu is the mean of the latent variable and logvar is the standard deviation of the latent variable.

8. The fuel microstructure prediction method based on a generative deep learning model according to claim 1, characterized in that: Process conditions are fed into a trained generative deep learning model to generate fuel-specific microstructures, including: The input component process conditions are converted into input embedding vectors through a linear layer, and the embedding vectors and latent variables generated by training are combined to generate new latent variables using the attention mechanism; The new latent variable is input into the trained decoder to generate the corresponding image.

9. The fuel microstructure prediction method based on a generative deep learning model according to claim 8, characterized in that: The input component process conditions are converted into input embedding vectors through a linear layer. The embedding vectors and latent variables generated by training are combined to generate new latent variables using the attention mechanism, including: Let the input variable be a i , the latent variable is recorded as z i , where i is the sequence number, z i The latent variable generated for training consists of 64 numbers; the calculation formula of the latent variable is: Among them, z4 is the latent variable of the test set data with sequence number 4, and a4 is the input variable.

10. The fuel microstructure prediction method based on a generative deep learning model according to claim 8, characterized in that: The new latent variables are input into the trained decoder to generate corresponding images, which are used to show the grain shape and size of the microstructure.

Images