Method and system for predicting eddy current detection data of heat transfer tube of nuclear power plant

Abstract

The invention relates to a method and a system for predicting eddy current detection data of a heat transfer tube of a nuclear power plant. The method comprises the following steps: acquiring an eddy current detection result data table; preprocessing the eddy current inspection result data table; constructing a neural network model based on data in the preprocessed eddy current examination result data table; predicting the eddy current detection data to be predicted in the eddy current detection result data table based on a neural network model to obtain a prediction result; respectively drawing an actual measurement graph and a regression curve graph based on the actual measurement value and the prediction result of the eddy current detection data to be predicted; and outputting, displaying and storing the actual measurement graph and the regression curve graph. According to the method, the non-linear relation between variables can be well processed, features can be automatically extracted from data, the method can adapt to different types of data distribution and scales, average value prediction can be carried out according to coordinate values of directions needing to be predicted by a user, and a prediction result can be rapidly given.

Description

Technical Field

[0001] The present invention relates to the technical field of eddy current inspection data processing for heat transfer tubes in nuclear power plants, and more specifically, to a prediction method and system for eddy current inspection data of heat transfer tubes in nuclear power plants. Background Art

[0002] Nuclear power plants have various types of tube heat exchangers (such as steam generators, condensers, etc.). Such tube heat exchangers usually use thin-walled tubes as heat transfer tubes, which makes the heat transfer tubes the key components of these heat exchangers. In order to confirm whether there are defects in the heat transfer tubes of these heat exchangers, the heat transfer tubes are usually inspected by eddy current regularly. Due to the huge number of heat transfer tubes, it is difficult to conduct a full inspection in actual detection, and usually a sampling method is adopted. The eddy current inspection data of the heat transfer tubes obtained are sampling sample data.

[0003] Currently, for the eddy current inspection data of heat transfer tubes obtained as sampling sample data, a regression model of the heat transfer tube distribution and the eddy current inspection results needs to be established manually. Taking the steam generator of a certain nuclear power plant as an example, its eddy current inspection results mainly count wear information, and the heat transfer tube distribution is in two dimensions: longitudinal and transverse.

[0004] Traditional methods are restricted by manual analysis and establishment of linear regression equations, and have the following deficiencies: Low efficiency: Facing large-scale complex data sets, manual operations are very time-consuming. It is very difficult to manually sort and calculate. Prone to errors: There are many manual calculation steps, and it is easy to have calculation mistakes or misinterpretation of data, affecting the accuracy of the results. Simple model: Linear regression assumes a linear relationship between variables. In fact, many relationships between the eddy current inspection results of heat transfer tubes and the distribution of heat transfer tubes are non-linear. It is very difficult to effectively capture complex relationships by manually establishing such a simple model, and prediction cannot be carried out. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a prediction method and system for eddy current inspection data of heat transfer tubes in nuclear power plants in view of the problems existing in the prior art.

[0006] The technical solution adopted by the present invention to solve its technical problems is to construct a prediction method for eddy current inspection data of heat transfer tubes in nuclear power plants, including the following steps:

[0007] Obtain the eddy current inspection result data table;

[0008] Preprocess the eddy current inspection result data table;

[0009] Construct a neural network model based on the data in the eddy current inspection result data table after the preprocessing is completed;

[0010] Predict the eddy current inspection data to be predicted in the eddy current inspection result data table based on the neural network model to obtain a prediction result;

[0011] Draw an actual measurement graph and a regression curve graph respectively based on the measured value of the eddy current detection data to be predicted and the prediction result;

[0012] Output and display the actual measurement graph and the regression curve graph and save them.

[0013] In the method for predicting eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the preprocessing of the eddy current inspection result data table includes:

[0014] Determine the specified path of the eddy current inspection result data table, and the specified path is a path string pointing to the Excel file of the eddy current inspection data table;

[0015] Split the path string, and assign the first element in the obtained list after splitting to the file name;

[0016] Read the data of the Excel file at the pointed path, and load the read data into a specified object;

[0017] Rename the column names of the specified object;

[0018] After completing the renaming, preprocess the data and establish an index for the specified object;

[0019] Group the detection results according to the column to be predicted, and calculate the average value of each group;

[0020] Extract the data in the specified object after the above processing to obtain feature data and target data.

[0021] In the method for predicting eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the renaming of the column names of the specified object includes:

[0022] Name the longitudinal dimension as R;

[0023] Name the transverse dimension as C;

[0024] Name the detection result as value.

[0025] In the method for predicting eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the preprocessing of the data and the establishment of an index for the specified object include:

[0026] Remove the null values in the value column of the eddy current detection result data table;

[0027] Remove the duplicate rows in the eddy current detection result data table;

[0028] Establish an index for the specified object.

[0029] In the method for predicting eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, constructing a neural network model based on the data in the eddy current inspection result data table after preprocessing includes:

[0030] Dividing the data in the eddy current inspection result data table after preprocessing to obtain a training set and a test set;

[0031] Setting the number of training epochs of the neural network model;

[0032] Constructing a sequential model object;

[0033] Adding fully connected layers of the neural network model and determining the number of neurons and activation functions in each hidden layer;

[0034] Compiling and configuring the neural network model;

[0035] Constructing a callback object;

[0036] Training the neural network model based on the training set;

[0037] Testing the trained neural network model based on the test set;

[0038] Constructing the file name for saving the neural network model that passes the test;

[0039] Saving the neural network model that passes the test to the specified file path to complete the construction of the neural network model.

[0040] In the method for predicting eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the number of training epochs is 100; the number of neurons in each hidden layer is 512 neurons, 128 neurons, 64 neurons, and 32 neurons respectively.

[0041] In the method for predicting eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the prediction result is in the form of a two-dimensional array;

[0042] Drawing an actual measurement graph and a regression curve graph based on the measured value of the eddy current detection data to be predicted and the prediction result respectively includes:

[0043] Converting the prediction result in the form of a two-dimensional array into a one-dimensional array;

[0044] Performing a screen clearing process;

[0045] Drawing with the characteristic data of the measured value of the eddy current detection data to be predicted as the abscissa and the target data of the measured value as the ordinate to obtain the actual measurement graph;

[0046] Taking the characteristic data of the measured value of the eddy current detection data to be predicted as the abscissa and the one-dimensional array of the prediction result as the ordinate for plotting, the regression curve graph is obtained.

[0047] In the prediction method for eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the output display and saving of the measured graph and the regression curve graph include:

[0048] Construct a specified file name for saving the plotted graph;

[0049] Save the measured graph and the regression curve graph as picture files with the constructed specified file name;

[0050] Return the specified file name and the saved model file name;

[0051] Output and display the measured graph and the regression curve graph.

[0052] In the prediction method for eddy current detection data of heat transfer tubes in a nuclear power plant according to the present invention, the method further includes:

[0053] Load the neural network model saved to the specified file path;

[0054] Receive the input value to be predicted;

[0055] Convert the format of the input value;

[0056] Input the input value after format conversion into the neural network model for prediction to obtain the prediction result;

[0057] Output and display the prediction result.

[0058] The present invention also provides a prediction system for eddy current detection data of heat transfer tubes in a nuclear power plant, including:

[0059] A data table acquisition unit for acquiring an eddy current inspection result data table;

[0060] A data preprocessing unit for preprocessing the eddy current inspection result data table;

[0061] A neural network model construction unit for constructing a neural network model based on the data in the eddy current inspection result data table after preprocessing;

[0062] A data prediction unit for predicting the eddy current detection data to be predicted in the eddy current inspection result data table based on the neural network model to obtain the prediction result;

[0063] A graph plotting unit for respectively plotting a measured graph and a regression curve graph based on the measured value of the eddy current detection data to be predicted and the prediction result;

[0064] The graphic saving and displaying unit is used to output, display and save the measured graph and the regression curve graph.

[0065] Implementing the prediction method and system for eddy current detection data of heat transfer tubes in nuclear power plants of the present invention has the following beneficial effects: including the following steps: obtaining a data table of eddy current inspection results; preprocessing the data table of eddy current inspection results; constructing a neural network model based on the data in the preprocessed data table of eddy current inspection results; predicting the eddy current detection data to be predicted in the data table of eddy current inspection results based on the neural network model to obtain a prediction result; respectively drawing a measured graph and a regression curve graph based on the measured value and the prediction result of the eddy current detection data to be predicted; outputting, displaying and saving the measured graph and the regression curve graph. The present invention can well handle the non-linear relationship between variables, can automatically extract features from data, can adapt to different types of data distributions and scales, and can also perform average value prediction according to the coordinate values of the prediction direction required by the user, and quickly give a prediction result. Description of the Drawings

[0066] The present invention will be further described below in conjunction with the drawings and embodiments. In the drawings:

[0067] Figure 1 is a schematic flowchart of the first embodiment of the prediction method for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention;

[0068] Figure 2 is a schematic flowchart of the second embodiment of the prediction method for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention;

[0069] Figure 3 is a logical block diagram of the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention;

[0070] Figure 4 is a schematic diagram of the user interface of the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention on the PC side;

[0071] Figure 5 is a schematic diagram of the user interface of the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention on the WEB side. Detailed Embodiments

[0072] The technical solutions 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 a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0073] In view of the problems existing in the prior art, the present invention provides a method for predicting eddy current detection data of heat transfer tubes in nuclear power plants. The method for predicting eddy current detection data of heat transfer tubes in nuclear power plants uses Python, combines pandas for data preprocessing, combines tensorflow, sklearn, and keras to build a neural network regression prediction model, and combines matplotlib to draw a regression curve for presentation to users (such as in-service inspection engineers).

[0074] Specifically, in a preferred embodiment, as Figure 1 shown, the method for predicting eddy current detection data of heat transfer tubes in nuclear power plants includes the following steps:

[0075] Step S101: Obtain the eddy current inspection result data table.

[0076] Among them, in this step, for the eddy current inspection result data table, the system can import data according to the input parameters determined by the user to directly import the eddy current inspection result data table into the system.

[0077] Step S102: Preprocess the eddy current inspection result data table.

[0078] Optionally, in the embodiment of the present invention, preprocessing the eddy current inspection result data table includes: determining the specified path (data_path) of the eddy current inspection result data table, where the specified path is a path string pointing to the Excel file of the eddy current inspection data table; splitting the path string, and assigning the first element in the obtained list after splitting to the file name (file_path_name); reading the data of the Excel file pointed to by the path and loading the read data into a specified object (i.e., a DataFrame object named df); renaming the column names of the specified object; after completing the renaming, preprocessing the data and establishing an index for the specified object; grouping the detection results according to the column to be predicted and calculating the average value of each group; extracting the data in the specified object after the above processing to obtain feature data and target data. In the embodiment of the present invention, the split method can be used for cutting the path string.

[0079] Among them, renaming the column names of the specified object includes: naming the longitudinal dimension as R; naming the transverse dimension as C; naming the detection result as value. Preprocessing the data and establishing an index for the specified object includes: removing the null values in the value column of the eddy current detection result data table; removing the duplicate rows in the eddy current detection result data table; establishing an index for the specified object.

[0080] Specifically, the preprocessing of the eddy current inspection result data table can be specifically performed according to the following detailed steps:

[0081] Step 1: For the imported eddy current inspection result data table, first define a function named mat_hg. Based on this function, determine the specified path of the eddy current inspection result data table, where the specified path can be defined as data_path. Here, data_path is a string pointing to the Excel file path of the eddy current inspection data table.

[0082] Step 2: Use the split method to split the imported file path string with '.' as the delimiter, and assign the first element in the resulting list (i.e., the file name part, excluding the extension) to file_path_name. Here, file_path_name represents the file name, which is used to save relevant files later.

[0083] Step 3: Use the read_excel function of the pandas library to read the Excel file at the specified path (i.e., data_path), and load the data into a DataFrame object named df. Here, df is mainly used to facilitate subsequent data processing and analysis.

[0084] Step 4: Rename the column names of df to ['R', 'C', 'value']. That is, rename the column names in the vertical and horizontal directions. The column name in the vertical dimension is R, the column name in the horizontal dimension is C, and the column name of the inspection result (wear value) is value.

[0085] Step 5: Use the dropna method, and specify to only focus on the 'value' column through the subset parameter, and remove the rows containing null values (such as NaN) in this column to ensure data integrity and the effectiveness of subsequent processing. That is, first, use the pandas library to remove the null values in the value column of the original data table. Then, use the drop_duplicates method to remove duplicate rows in the dataset according to the combination of the ['R', 'C'] columns to ensure data uniqueness. Finally, call the reset_index method to re-index the DataFrame.

[0086] Step 6: Group the 'value' column in df according to the column specified by the global variable (AI_axis) (i.e., the column to be predicted), and calculate the average value of each group. By grouping and calculating the average value of each group, df can finally become a Series object with the grouping basis column (the column specified by AI_axis) as the index and the average value of 'value' for each group.

[0087] Step 7: Extract the feature data (independent variable) and target data (dependent variable) from the df object processed in Step 6. Specifically: df.index.values is used to obtain the values of the index as the feature X. Here, the index is actually the value of the grouping basis column (according to the previous groupby operation). df.values is used to obtain the data of the corresponding average value of each group (i.e., the target variable) and assign it to y.

[0088] Step S103: Build a neural network model based on the data in the eddy current inspection result data table after preprocessing is completed.

[0089] Optionally, in the embodiment of the present invention, building a neural network model based on the data in the eddy current inspection result data table after preprocessing is completed includes: dividing the data in the eddy current inspection result data table after preprocessing is completed to obtain a training set and a test set; setting the number of training epochs of the neural network model; building a Sequential model object; adding fully connected layers of the neural network model and determining the number of neurons and activation functions of each hidden layer; performing compilation configuration on the neural network model; building a callback object; training the neural network model based on the training set; testing the trained neural network model based on the test set; building the file name for saving the neural network model that passes the test; saving the neural network model that passes the test to the specified file path to complete the construction of the neural network model. Among them, the number of training epochs is 100; the number of neurons in each hidden layer is 512 neurons, 128 neurons, 64 neurons, and 32 neurons respectively.

[0090] Specifically, the construction of the neural network model can be executed through the following detailed steps:

[0091] Step 1: Use the train_test_split function (from the sklearn package) to divide the dataset into a training set and a test set in a ratio of 8:2. X and y are the features and target variable data extracted previously. test_size = 0.2 indicates that the test set accounts for 20% of the total dataset, and random_state = 42 is used to specify the random number seed to ensure the repeatability of each division result.

[0092] Step 2: Set the number of training epochs epochs to 100. That is, the training dataset will be traversed completely 100 times.

[0093] Step 3: Create a Sequential model object model. Among them, this Sequential model object is a way in keras (a library commonly used to build deep learning models) to build a simple linearly stacked neural network model.

[0094] Step 4: Then, use the add method to sequentially add fully connected layers (Dense layers) to the model. The number of neurons in each layer is specified as 512, 128, 64, and 32 neurons respectively. Through verification, the number of neurons in each layer determined by the present invention is more suitable for the regression prediction of the eddy current inspection results of heat transfer tubes and the distribution of heat transfer tubes. At the same time, the activation function uses the tanh hyperbolic tangent function, and the input dimension of the input data is specified as 1 by input_dim = 1 for the first layer (because the previously extracted feature X is one-dimensional data), and the output dimension of the last layer is 1, indicating that a predicted value is output (for regression tasks).

[0095] Step 5: Use the compile method to configure the compilation of the model, specify the optimizer as 'adam' (a commonly used optimization algorithm with adaptive learning rate), and the loss function as'mse' (mean squared error, commonly used in regression problems to measure the error between the predicted value and the true value).

[0096] Step 6: Create an EarlyStopping callback object, which is used to stop the training in advance according to specific conditions during the training process to prevent overfitting. Here, the monitoring metric is set to 'val_loss' (the loss value on the validation set), and patience = 5 means that when the validation set loss has not improved for 5 consecutive times, the training will stop.

[0097] Step 7: Use the fit method to train the model, input the features X_train and the target y_train of the training set, specify the number of training epochs, and put the previously defined EarlyStopping callback object early_stop into the callbacks list and pass it in, so that the early stopping mechanism can be implemented during the training process to prevent overfitting.

[0098] Step 8: Use the evaluate method to evaluate the trained model on the test set (X_test and y_test), calculate the loss value (the specific loss calculation here is based on the'mse' mean squared error loss function specified when compiling the model before), and assign the loss value to the loss variable.

[0099] Step 9: Construct the file name for saving the model, and obtain the complete model file name by adding the.keras extension to the previously extracted file name part (file_path_name).

[0100] Step 10: Then call the save method of the model to save the trained model to the specified file path (the position corresponding to the file name), which is convenient for subsequent reuse of the model.

[0101] Step S104: Predict the eddy current detection data to be predicted in the eddy current inspection result data table based on the neural network model to obtain a prediction result.

[0102] Optionally, in the embodiment of the present invention, the prediction result is in the form of a two-dimensional array.

[0103] Specifically, use the trained neural network model to predict all the feature data X (including the feature parts corresponding to the training set and the test set) to obtain the prediction result y_pred.

[0104] Step S105: Draw a measured value graph and a regression curve graph respectively based on the measured value and the prediction result of the eddy current detection data to be predicted.

[0105] Optionally, in the embodiment of the present invention, drawing a measured value graph and a regression curve graph respectively based on the measured value and the prediction result of the eddy current detection data to be predicted includes: converting the prediction result in the form of a two-dimensional array into a one-dimensional array; performing a screen clearing process; using the feature data of the measured value of the eddy current detection data to be predicted as the abscissa and the target data of the measured value as the ordinate for drawing to obtain the measured value graph; using the feature data of the measured value of the eddy current detection data to be predicted as the abscissa and the one-dimensional array of the prediction result as the ordinate for drawing to obtain the regression curve graph.

[0106] Specifically, the drawing of the measured value graph and the regression curve graph can be performed through the following detailed steps:

[0107] Step 1: Call the ravel method to flatten the two-dimensional prediction result array into a one-dimensional array for convenient subsequent visualization and other operations.

[0108] Step 2: Call the clf method of the matplotlib library (a commonly used library for data visualization) to clear the current graph (clear the screen) (if there were other graphs drawn before) and prepare to draw a new graph.

[0109] Step 3: Use the scatter method to draw a scatter plot (i.e., the measured value graph), with X as the abscissa and y (the true value) as the ordinate, and add a label to indicate that it is the true sampling value.

[0110] Step 4: Then use the plot method to draw a curve, with X as the abscissa and y_pred (the predicted value) as the ordinate, set the color to red, and add a label to indicate that it is the predicted value of the artificial intelligence regression curve.

[0111] Step 5: Finally, call the legend method to add a legend, and let the legend automatically select a suitable position to display through loc='best' to distinguish the graphical elements corresponding to the true value and the predicted value.

[0112] Step S106: Output and display the measured graph and the regression curve graph and save them.

[0113] Optionally, in the embodiment of the present invention, outputting, displaying, and saving the measured graph and the regression curve graph includes: constructing a specified file name for saving the drawn graph; saving the measured graph and the regression curve graph as picture files with the constructed specified file name; returning the specified file name and the saved model file name; outputting and displaying the measured graph and the regression curve graph.

[0114] Specifically, the output display and saving of the measured graph and the regression curve graph can be executed through the following steps:

[0115] Step 1: Construct the file name for saving the drawn graph. The complete graph file name hg_curve_name is obtained by concatenating the file name part (file_path_name), AI_axis, and the fixed suffix _AI_hg.jpg.

[0116] Step 2: Call the savefig method to save the drawn graph as a picture file with the specified file name, and set the resolution to 600 dpi (dots per inch, used to control quality parameters such as picture clarity).

[0117] Step 3: Finally, the function returns the saved graph file name hg_curve_name and the saved model file name model_name so that the relevant saved file information can be obtained at the place where the function is externally called.

[0118] Further, as Figure 2 shown, in another preferred embodiment, after step S106, the following steps are further included:

[0119] Step S107: Load the neural network model saved to the specified file path.

[0120] Step S108: Receive the input value to be predicted.

[0121] Step S109: Convert the format of the input value.

[0122] Step S110: Pass the input value after format conversion into the neural network model for prediction to obtain the prediction result.

[0123] Step S111: Output and display the prediction result.

[0124] Specifically, first, load the neural network model in the specified file path. Then, obtain an integer value from the text box (entry_AI_axis_num) input by the user as the row number (longitudinal R) or column number (lateral C) to be predicted, that is, the input value to be predicted. Next, convert the input value into the numpy array format and then pass it into the neural network model for prediction to obtain the prediction result. After retaining two decimal places for the prediction result, insert it into another text box (through the insert_AI_text function: receive the text content to be displayed, first clear the existing content in the specified text box (res_AI_text), and then insert the passed-in text at the end of the text box and start a new line) for display.

[0125] Reference Figure 3 , Figure 3 is the logic block diagram of the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention.

[0126] As Figure 3 shown, the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants includes:

[0127] A data table acquisition unit 301, which is used to acquire the eddy current inspection result data table.

[0128] A data preprocessing unit 302, which is used to preprocess the eddy current inspection result data table.

[0129] A neural network model construction unit 303, which is used to construct a neural network model based on the data in the eddy current inspection result data table after the preprocessing is completed.

[0130] A data prediction unit 304, which is used to predict the eddy current detection data to be predicted in the eddy current inspection result data table based on the neural network model to obtain the prediction result.

[0131] A graph drawing unit 305, which is used to draw an actual measurement graph and a regression curve graph respectively based on the measured value and the prediction result of the eddy current detection data to be predicted.

[0132] A graph saving and display unit 306, which is used to output and display and save the actual measurement graph and the regression curve graph.

[0133] Specifically, for the specific cooperation operation process among the units in the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants here, reference can be made to the above-mentioned prediction method for eddy current detection data of heat transfer tubes in nuclear power plants, which will not be elaborated here.

[0134] Reference Figure 4 and Figure 5 , Figure 4 is the schematic diagram of the usage interface of the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention on the PC side; Figure 5It is a schematic diagram of the usage interface of the prediction system for eddy current detection data of heat transfer tubes in nuclear power plants provided by the present invention on the WEB side. Among them, the PC side and the WED side only have different user interaction forms but the same method, which can meet the different needs of users in different demand scenarios.

[0135] Taking the PC side as an example below, the neural network regression prediction of the eddy current inspection results of the heat transfer tubes of a certain heat exchanger in a certain nuclear power plant will be described.

[0136] As Figure 4 shown, first open the PC side version, enter the sub-module of "ISI-Toolkit in-service inspection toolbox", select R or C, and determine the eigenvalue for the neural network regression model (that is, determine the input value to be predicted). If the column number C is selected as the eigenvalue, as Figure 4 shown as "C".

[0137] Next, the user clicks the "AI regression curve" button, and a file selection dialog box will pop up. The user can upload the eddy current inspection result table of the heat transfer tubes to be predicted as needed. At this time, the system can automatically read the eddy current inspection result table of the heat transfer tubes.

[0138] Then, the loaded neural network model is used to train the read data. After the training is completed, the measured graph (that is, the scatter plot of the original heat transfer tube roaming inspection results and the eigenvalue) and the predicted regression curve graph (that is, the regression curve graph of the predicted value) are automatically drawn and displayed in the PC side interface, as Figure 4 shown, and saved as a picture file at the same time.

[0139] Finally, perform the mean prediction. Specifically: the user enters the eigenvalue data of the feature dimension to be predicted in the input box (such as predicting the mean value of the eddy current inspection results of the heat transfer tubes with column number C being 10, the user only needs to enter 10, as Figure 4 shown), and then clicks the "AI predict mean" button to obtain the predicted result mean, as Figure 4 shown as 5.15.

[0140] Based on the actual engineering, there is a correlation between the eddy current inspection results of the heat transfer tubes and the distribution of the heat transfer tubes, and the degree of correlation in the horizontal and vertical directions is different. The present invention classifies the eigenvalue of the eddy current inspection results of the heat transfer tubes into column number C and row number R, which can more effectively improve the accuracy and training effect for the subsequent construction and training of the neural network model. At the same time, important parameters of the neural network model are set, and the number of training epochs is 100. The number of neurons in each hidden layer: there are 512, 128, 64, and 32 neurons respectively, which are more suitable for the regression prediction of the eddy current inspection results of the heat transfer tubes and the distribution of the heat transfer tubes); activation function: use the tanh hyperbolic tangent function, which is more suitable for non-linear regression.

[0141] Through the present invention, the deficiencies of the prior art can be avoided, and at the same time, the following advantages are also possessed: It can handle complex relationships: It can handle the non-linear relationships between variables well. The neural network artificial intelligence model can perform more accurate modeling. It can achieve automatic feature learning: The neural network can automatically extract features from the data. It has high adaptability: It can adapt to different types of data distributions and scales. Whether it is a small data set or a large data set, whether it is a condenser with a huge number of heat transfer tubes or a small heat exchanger with only hundreds of heat transfer tubes, it can be highly adaptable. The interface integration is friendly: This device adopts a friendly human-computer interaction interface. The in-service inspection engineer only needs to upload the eddy current detection result data table, and this device will automatically perform data analysis and establish a neural network regression model, and perform average value prediction according to the coordinate value in a certain direction that the user needs to predict, and quickly give the prediction result.

[0142] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description in the method part for the relevant parts.

[0143] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0144] The steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0145] The above embodiments are only used to illustrate the technical concept and characteristics of the present invention, and their purpose is to enable those who are familiar with this technology to understand the content of the present invention and implement it accordingly, and cannot limit the protection scope of the present invention. All equivalent changes and modifications made to the scope of the claims of the present invention shall fall within the scope covered by the claims of the present invention.

Claims

1. A method for predicting eddy current detection data of heat transfer tubes in nuclear power plants, characterized in that: The following steps are involved: Obtain eddy current inspection result data sheet; Preprocessing the eddy current inspection result data table; Building a neural network model based on the data in the eddy current inspection result data table after preprocessing; Predicting the eddy current detection data to be predicted in the eddy current inspection result data table based on the neural network model to obtain a prediction result; Based on the measured value of the eddy current detection data to be predicted and the predicted result, a measured graph and a regression curve graph are respectively drawn; The measured graph and the regression curve graph are output, displayed and saved.

2. The method for predicting eddy current detection data of heat transfer tubes in nuclear power plants according to claim 1, characterized in that: The preprocessing of the eddy current inspection result data table comprises: Determine a designated path of the eddy current inspection result data table, wherein the designated path is a path string pointing to the Excel file of the eddy current inspection data table; The path string is segmented, and the first element in the list obtained after segmentation is assigned to the file name; Read the data of the Excel file pointed to by the path, and load the read data into the specified object; Rename the column name of the specified object; After the renaming is completed, preprocessing the data and establishing an index of the specified object; Group the test results according to the column to be predicted and calculate the average value of each group; The data in the designated object that has completed the above processing is extracted to obtain feature data and target data.

3. The method for predicting eddy current testing data of heat transfer tubes in nuclear power plants according to claim 2, characterized in that: The renaming of the column name of the specified object includes: Name the vertical dimension R; Name the horizontal dimension C; Name the test result value.

4. The method for predicting eddy current detection data of heat transfer tubes in nuclear power plants according to claim 3, characterized in that: The preprocessing of data and establishing the index of the specified object comprises: Remove the empty values ​​in the value column of the eddy current test result data table; Removing duplicate rows in the eddy current test result data table; Creates an index for the specified object.

5. The method for predicting eddy current detection data of heat transfer tubes in nuclear power plants according to claim 1, characterized in that: The method of constructing a neural network model based on the data in the eddy current inspection result data table after preprocessing includes: Divide the data in the eddy current inspection result data table after preprocessing to obtain a training set and a test set; Set the number of training rounds for the neural network model; Construct a sequential model object; Adding a fully connected layer of the neural network model and determining the number of neurons and activation functions of each hidden layer; Compiling and configuring the neural network model; Construct callback object; Training the neural network model based on the training set; Testing the trained neural network model based on the test set; Build the file name of the neural network model that passes the test; The neural network model that passes the test is saved to a specified file path to complete the construction of the neural network model.

6. The method for predicting eddy current detection data of heat transfer tubes in nuclear power plants according to claim 5, characterized in that: The number of training rounds is 100; the number of neurons in each hidden layer is 512 neurons, 128 neurons, 64 neurons and 32 neurons respectively.

7. The method for predicting eddy current testing data of heat transfer tubes in nuclear power plants according to claim 1, characterized in that: The prediction result is in the form of a two-dimensional array; The step of drawing a measured graph and a regression curve graph based on the measured value of the eddy current detection data to be predicted and the predicted result comprises: Convert the prediction result in the two-dimensional array into a one-dimensional array; Execute screen clearing process; The measured graph is obtained by plotting the characteristic data of the measured values ​​of the eddy current detection data to be predicted as the abscissa and the target data of the measured values ​​as the ordinate; The regression curve graph is obtained by plotting the characteristic data of the measured values ​​of the eddy current detection data to be predicted as the horizontal coordinate and the one-dimensional array of the prediction results as the vertical coordinate.

8. The method for predicting eddy current testing data of heat transfer tubes in nuclear power plants according to claim 1, characterized in that: The outputting, displaying and saving the measured graph and the regression curve graph comprises: Construct the specified file name to save the drawn graphics; The measured graph and the regression curve graph are saved as a constructed picture file with a specified file name; Returns the specified file name and the saved model file name; The measured graph and the regression curve graph are output and displayed.

9. The method for predicting eddy current testing data of heat transfer tubes in nuclear power plants according to any one of claims 1 to 8, characterized in that: The method further comprises: Load the neural network model saved to the specified file path; Receive the input value to be predicted; Convert the input value into a new format; The input value after format conversion is passed into the neural network model for prediction to obtain a prediction result; The prediction results are output and displayed.

10. A prediction system for eddy current testing data of heat transfer tubes in nuclear power plants, characterized in that: include: A data table acquisition unit, used for acquiring an eddy current inspection result data table; A data preprocessing unit, used for preprocessing the eddy current inspection result data table; A neural network model building unit, used to build a neural network model based on the data in the eddy current inspection result data table after preprocessing; A data prediction unit, used for predicting the eddy current detection data to be predicted in the eddy current inspection result data table based on the neural network model to obtain a prediction result; A graph drawing unit, used for drawing a measured graph and a regression curve graph respectively based on the measured value of the eddy current detection data to be predicted and the predicted result; The graphics storage and display unit is used to output, display and store the measured graph and the regression curve graph.

Images