Nuclear power station rotating machine fault diagnosis method based on small sample and related product

Abstract

The invention relates to the field of intelligent fault diagnosis, in particular to a nuclear power station rotating machine fault diagnosis method based on small samples and related products, and the method comprises the steps: collecting time domain data and frequency domain data of various fault types, and carrying out the range division; the method comprises the following steps: constructing a total sample set, dividing a training set and a test set, constructing a convolutional neural network, obtaining a nuclear power station rotating machinery fault diagnosis model through training, testing and optimizing iterative training through the test set until optimal convolutional neural network parameters are obtained, and constructing a final nuclear power station rotating machinery fault diagnosis model. Through the method based on small sample learning, the training data is effectively expanded, and the problem that effective diagnosis is difficult to perform under the condition that fault data samples are few in a traditional deep learning model is solved; secondly, the convolutional neural network is adopted to process fault data, so that more accurate and detailed features can be extracted, and the fault diagnosis precision is improved; and through repeated iterative training and model optimization, the stability and high efficiency of the final model are ensured.

Description

Technical Field

[0001] The present invention relates to the field of intelligent fault diagnosis, and in particular to a nuclear power plant rotating machinery fault diagnosis method based on a small sample and related products. Background Art

[0002] During the operation of a nuclear power plant, key equipment inside the reactor, such as the main pump, internal components, and electrical penetrations, needs to be continuously monitored to ensure safe, reliable, and efficient operation of the equipment. These devices are usually equipped with a condition monitoring system to monitor the operation of the equipment in real time, diagnose faults, and predict the development trend of faults, thereby providing decision support for equipment maintenance.

[0003] With the development of the Internet and computer technology, intelligent technology has gradually been applied to the monitoring and fault diagnosis of key equipment in nuclear power plants. Among them, deep learning technology has been widely used in classification, regression and prediction due to its high adaptability, self-learning and self-organizing capabilities, and has achieved remarkable results.

[0004] Taking the rotating machinery of nuclear power plants, such as the main pump, as an example, the main pump is in normal operation most of the time, and there is relatively little fault data. This leads to the problem of sample scarcity when training deep learning models, making it difficult to achieve the expected fault diagnosis effect. In the prior art, the fault diagnosis methods for rotating machinery are mostly based on big data and conventional deep learning algorithms, which fail to effectively solve the fault diagnosis problem under small sample data conditions.

[0005] Therefore, studying how to effectively expand training data under small sample conditions and improve the accuracy of fault diagnosis has become an important topic in the current field of intelligent diagnosis of nuclear power plant equipment. Summary of the invention

[0006] The technical problem to be solved by the present invention is that when there are few fault data samples of nuclear power plant rotating machinery, traditional deep learning models are difficult to effectively perform fault diagnosis. The purpose is to provide a nuclear power plant rotating machinery fault diagnosis method and related products based on small samples, which realizes the effective expansion of fault data under small sample conditions and improves the accuracy and reliability of key equipment fault diagnosis.

[0007] The present invention is achieved through the following technical solutions:

[0008] A method for diagnosing rotating machinery faults in nuclear power plants based on a small sample size, comprising:

[0009] Collect fault data of various fault types, and process the fault data to obtain time domain data and frequency domain data;

[0010] Range division is performed according to the vibration parameters corresponding to different fault types, and the divided ranges include: acceptable range, alarm range, and range requiring measures;

[0011] Construct a total sample set based on the time-domain data, frequency-domain data, and the divided ranges;

[0012] Screen the training set from the total sample set through a metric-based method; and use the remaining data as the test set;

[0013] Construct a convolutional neural network for one-dimensional fault data and train it with the training set to obtain a fault diagnosis model for nuclear power plant rotating machinery;

[0014] Test the fault diagnosis model for nuclear power plant rotating machinery with the test set and repeatedly train the convolutional neural network according to the test results;

[0015] Iteratively train until the optimal convolutional neural network parameters are obtained, and construct the final fault diagnosis model for nuclear power plant rotating machinery.

[0016] Specifically, the fault types include: rotor misalignment, rotor imbalance, component looseness, shaft friction, rotor crack, typical faults of canned motor pumps;

[0017] The time-domain data is obtained through time-domain analysis, and the time-domain analysis includes calculating the average amplitude, root mean square value, variance, and kurtosis index;

[0018] The frequency-domain data is obtained through frequency-domain analysis, and the frequency-domain analysis includes discrete Fourier transform and fast Fourier transform.

[0019] Optionally, two vibration limits T1 and T2 are set in sequence according to the vibration parameters of the rotating machinery, and T1 < T2; if the vibration parameter T < T1, it is divided into the acceptable range; if T1 ≤ T < T2, it is divided into the alarm range; if T ≥ T2, it is divided into the range requiring measures.

[0020] Optionally, three warning values t1, t2, and t3 are also set in sequence according to the vibration parameters of the rotating machinery, and t1 < T1 < t2 < T2 < t3; if t1 < T < T1, it is determined as a warning for the conversion from the acceptable range to the alarm range; if t2 < T < T2, it is determined as a warning for the conversion from the alarm range to the range requiring measures; if T > t2, it is determined as a warning for urgent handling.

[0021] Optionally, the vibration parameters include vibration intensity, peak vibration displacement, amplitude at the fundamental frequency in the vibration spectrum, and amplitude at the second harmonic of the fundamental frequency in the vibration spectrum; if any vibration parameter reaches the vibration limit, it is divided into the corresponding range.

[0022] Specifically, the method of training the convolutional neural network with the training set includes:

[0023] Initialize the convolutional neural network parameters, set the initial weights and initial biases of the convolutional neural network;

[0024] Convert the data in the training set into a one-dimensional vector; then convert the one-dimensional vector into a matrix form;

[0025] Perform the first convolution operation on the matrix to obtain the convolution feature map;

[0026] Perform maximum pooling on the convolution feature map to obtain the pooled feature map;

[0027] Compress the pooled feature map into a one-dimensional vector and use it as the input of the fully connected layer;

[0028] The input vector is processed through the fully connected layer to obtain the output result;

[0029] Calculate the error between the output and the expected output;

[0030] Based on the error, the weights and biases of the convolutional neural network are adjusted through the back-propagation algorithm to minimize the error;

[0031] Repeat the above steps and iteratively train the convolutional neural network until the error converges to the preset range;

[0032] Obtain a nuclear power plant rotating machinery fault diagnosis model.

[0033] Specifically, the method for testing the nuclear power plant rotating machinery fault diagnosis model includes:

[0034] Input the fault data in the test set into the initially trained nuclear power plant rotating machinery fault diagnosis model and calculate the model's prediction results;

[0035] Compare the model's predictions with the actual fault labels in the test set and calculate the test error;

[0036] Analyze test errors to determine model performance metrics;

[0037] If the performance index does not meet the preset requirements, the fault data in the test set is added to the training set to expand the training set;

[0038] Retrain the convolutional neural network.

[0039] A nuclear power plant rotating machinery fault diagnosis terminal based on a small sample includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the nuclear power plant rotating machinery fault diagnosis method based on a small sample as described above is implemented.

[0040] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample as described above is implemented.

[0041] A computer program product includes a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the nuclear power plant rotating machinery fault diagnosis method based on a small sample as described above.

[0042] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0043] The present invention collects initial fault data of various fault types and converts them into one-dimensional vectors, and then converts the one-dimensional vectors into matrix form; performs convolution operation on the matrix to obtain a convolution feature map, and performs maximum pooling processing on the convolution feature map to obtain a pooled feature map; compresses the pooled feature map into a one-dimensional vector, and uses it as the input of the fully connected layer, and processes the input vector through the fully connected layer to obtain an output result; then, calculates the error between the output result and the expected output, and adjusts the weight and bias of the convolution neural network through the back propagation algorithm according to the error to minimize the error; this process is repeated, and iterative training is performed until the optimal convolution neural network parameters are obtained, and finally a nuclear power plant rotating machinery fault diagnosis model is constructed. In addition, the model is tested through a test set, and the model is retrained and optimized according to the test results to ensure the high accuracy and reliability of the model.

[0044] The present invention effectively expands the training data through a method based on small sample learning, and solves the problem that traditional deep learning models are difficult to effectively diagnose when there are few fault data samples; secondly, a convolutional neural network is used to process the fault data, which can extract more accurate and detailed features and improve the accuracy of fault diagnosis; through repeated iterative training and model optimization, the stability and efficiency of the final model are ensured; finally, the model is verified and adjusted through a test set, which further improves the reliability of the model in practical applications, ensures the safe and reliable operation of rotating machinery equipment in nuclear power plants, thereby reducing the risks and maintenance costs caused by equipment failures and improving the overall operation efficiency of nuclear power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings illustrate exemplary embodiments of the present invention and, together with the description thereof, are used to explain the principles of the present invention. These drawings are included to provide a further understanding of the present invention, and the accompanying drawings are included in and constitute a part of this specification and do not constitute a limitation of the embodiments of the present invention.

[0046] Figure 1It is a flow chart of the method for diagnosing nuclear power plant rotating machinery faults based on small samples according to the present invention.

[0047] Figure 2 It is the convolutional neural network model according to the present invention. DETAILED DESCRIPTION

[0048] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and implementation methods. It is understood that the specific implementation methods described herein are only used to explain the relevant content, rather than to limit the present invention.

[0049] It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the drawings.

[0050] In the absence of conflict, the embodiments and features of the embodiments of the present invention may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0051] Embodiment 1

[0052] like Figure 1 As shown, a method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample is provided, comprising:

[0053] Collect fault data of various fault types, and process the fault data to obtain time domain data and frequency domain data; Collecting fault data of various fault types refers to obtaining operating data under different fault conditions from rotating mechanical equipment in nuclear power plants (such as main pumps, in-core components, etc.). These data include but are not limited to vibration signals, temperature changes, noise levels, etc. of the equipment. Time domain data refers to signal data directly recorded in a time series, such as changes in vibration amplitude over time. Frequency domain data is converted from time domain data to frequency domain through methods such as Fourier transform, thereby analyzing the characteristics of the signal at different frequencies, such as the spectrum characteristics of vibration signals. These two data processing methods can fully reflect the operating status and fault characteristics of the equipment.

[0054] According to the vibration parameters corresponding to different fault types, the ranges are divided into: acceptable range, alarm range and range of measures to be taken; the range of vibration parameters corresponding to different fault types is to classify the collected vibration data. Specifically, vibration parameters such as vibration intensity (effective value of vibration velocity), vibration displacement peak-to-peak value, vibration 1× amplitude and 2× amplitude, etc., according to the values ​​of these parameters, the fault data are divided into three categories: the acceptable range represents the vibration level of the equipment under normal operating conditions; the alarm range indicates that there may be potential problems with the equipment, which need to be monitored but no immediate measures need to be taken; the range of measures to be taken indicates that the equipment failure is serious and immediate measures such as repair or shutdown must be taken.

[0055] The total sample set is constructed based on the time domain data, frequency domain data and the divided range; constructing the total sample set means integrating the processed time domain data, frequency domain data and their corresponding range classification results together to form a complete data set. This total sample set contains the data features and classification information of the equipment under various operating conditions, and is the basis for model training and testing.

[0056] The training set is selected from the total sample set through a metric-based method, and the remaining data is used as a test set. The metric-based method refers to the use of a mathematical metric (such as Euclidean distance, cosine similarity, etc.) to select the training set data. The purpose of selecting the training set is to select the most representative data to improve the training effect of the model. The remaining data is used as a test set to evaluate the performance of the model.

[0057] Convolutional neural network for one-dimensional fault data is constructed and trained through training sets to obtain a nuclear power plant rotating machinery fault diagnosis model; Convolutional neural network (CNN) is a deep learning model that is good at processing images and time series data. Convolutional neural network for one-dimensional fault data refers specifically to a CNN model that processes one-dimensional time series signals (such as vibration signals). CNN is trained through training sets, that is, the training data is input for forward propagation, the error between the output result and the expected output is calculated, and the weights and biases of the network are adjusted through the back-propagation algorithm to gradually reduce the error, and finally a model that can accurately diagnose faults is obtained.

[0058] The nuclear power plant rotating machinery fault diagnosis model is tested through the test set, and the convolutional neural network is repeatedly trained based on the test results; the test model uses the test set data to verify the accuracy and generalization ability of the model. The test set is input into the CNN model, the model's prediction results are calculated, and compared with the actual results to evaluate the error and performance. According to the test results, the model is repeatedly trained, that is, the network parameters are further adjusted and optimized to improve the model performance.

[0059] Iterative training is performed until the optimal convolutional neural network parameters are obtained, and the final nuclear power plant rotating machinery fault diagnosis model is constructed. Iterative training is a continuous optimization process. Through multiple cycles of training and testing, the model parameters are adjusted until the model error converges and reaches the preset performance standard, and the optimal network parameters are obtained. This process ensures that the final nuclear power plant rotating machinery fault diagnosis model has high accuracy and reliability, and can effectively identify and diagnose equipment faults in practical applications.

[0060] The above steps can complete the training of the model, and when the model is trained, it can predict the external signals collected in real time.

[0061] After the model is trained, it can diagnose the externally collected rotating machinery signals. After the external collected signals enter the server, the diagnostic model judges the data. When the data is normal, it is directly sent to the database for storage; when the data is abnormal, the fault type is first determined, and then the signal is transmitted to the early warning unit for alarm.

[0062] The server receives various signal data collected from rotating mechanical equipment in real time. These data are usually collected by sensors and transmitted to the server for processing.

[0063] The received signal data is input into a pre-trained convolutional neural network diagnostic model. The model will make judgments based on the input data and identify the current operating status of the device.

[0064] If the diagnostic model determines that the data is within the acceptable range, it indicates that the equipment is operating normally. At this time, the data will be directly sent to the database for storage. The stored normal data can be used for subsequent model optimization and equipment operation history analysis. If the diagnostic model determines that the data is beyond the acceptable range, it indicates that the equipment may have a fault or abnormal situation. At this time, the model will further determine the type of fault.

[0065] The diagnostic model analyzes the abnormal data and determines the specific fault type, such as rotor misalignment, rotor imbalance, loose components, etc. After determining the fault type, the abnormal signal will be transmitted to the early warning unit. After receiving the signal, the early warning unit will issue an alarm to remind the operator to check the equipment in time to prevent the fault from further deteriorating.

[0066] Through the above process, the trained convolutional neural network diagnostic model can monitor the operating status of rotating machinery in real time and issue timely warnings for abnormal situations, thereby improving the operating safety and reliability of the equipment. The storage of normal data also provides data support for subsequent analysis and model optimization.

[0067] Embodiment 2

[0068] Fault types include: rotor misalignment, rotor imbalance, loose components, shaft friction, rotor cracks, and typical canned motor pump faults.

[0069] Rotor misalignment: refers to the fact that the axis of the rotor is not in the same straight line as the axis of the bearing or coupling, causing vibration and imbalance. Common symptoms include large vibration amplitude of the equipment at low frequency.

[0070] Rotor imbalance: The uneven distribution of rotor mass causes centrifugal force during rotation, which causes vibration. It is usually characterized by a significant vibration peak at a frequency (1X) corresponding to the rotational speed.

[0071] Component Looseness: Some components inside the equipment (such as bolts, nuts, etc.) become loose, resulting in abnormal noises and unstable vibrations during equipment operation. The loose components may cause vibration signals with multiple frequency components.

[0072] Shaft Friction: Friction occurs between the shaft and other components (such as bearings, seals, etc.), leading to wear and overheating. This type of fault usually shows higher harmonic components in the vibration signal.

[0073] Rotor Crack: Cracks appear inside or on the surface of the rotor, resulting in a decrease in structural strength and irregular vibrations and noises during operation. The characteristic signals of crack faults may be manifested as new frequency components or harmonics appearing in the spectrum.

[0074] Typical Faults of Canned Motor Pumps: Canned motor pumps are commonly used rotating mechanical equipment in nuclear power plants, and their typical faults include motor overheating, bearing damage, seal leakage, etc. These faults will show obvious abnormalities in vibration signals and other monitoring parameters.

[0075] Time-domain data is obtained through time-domain analysis, and the time-domain analysis includes calculating the average amplitude, root mean square value, variance, and kurtosis index; time-domain data is the signal data recorded from the time series. Average Amplitude: The average value of the signal, reflecting the overall offset degree of the signal. Root Mean Square Value (RMS): The effective value of the signal, indicating the energy level of the signal. Variance: The degree of dispersion of the signal data, reflecting the size of the signal fluctuation. Kurtosis Index: A measure of the sharpness of the signal, reflecting the degree of sudden changes in the signal.

[0076] Frequency-domain data is obtained through frequency-domain analysis, and the frequency-domain analysis includes discrete Fourier transform and fast Fourier transform.

[0077] Two vibration limits T1 and T2 are set in sequence according to the vibration parameters of the rotating machinery, and T1 < T2; if the vibration parameter T < T1, it is classified into the acceptable range, indicating that the equipment is operating normally and no measures need to be taken; if T1 ≤ T < T2, it is classified into the alarm range, indicating that the equipment may have potential problems and needs to be closely monitored, but no specific measures need to be taken temporarily; if T ≥ T2, it is classified into the range where measures need to be taken, indicating that the equipment failure is relatively serious and measures such as repair or shutdown must be taken immediately to avoid further damage.

[0078] Three warning values t1, t2, and t3 are also set in sequence according to the vibration parameters of the rotating machinery, and t1 < T1 < t2 < T2 < t3; the warning values are used to give early warnings before the vibration parameters approach or exceed the limits, so as to take preventive measures in a timely manner.

[0079] If t1 < T < T1, it is determined as a warning for the conversion from the acceptable range to the alarm range, indicating that the vibration level of the equipment is approaching the alarm range and an early warning is needed to prompt the operator to pay attention and check.

[0080] If \(t_2 < T < T_2\), it is determined as a pre-warning for the conversion from the alarm range to the range where measures need to be taken, indicating that the vibration level of the equipment is approaching the range where measures need to be taken. Further vigilance is required and corresponding measures should be prepared.

[0081] If \(T > t_2\), it is determined as a pre-warning that requires urgent handling, indicating that the vibration level of the equipment has exceeded the safe range and immediate handling is needed to prevent serious failures or damages.

[0082] By setting vibration limit values and pre-warning values, the operating state of rotating machinery can be effectively monitored, and pre-warnings can be given in advance before problems occur, thus ensuring the stable operation of the equipment, reducing downtime and maintenance costs caused by failures, and improving the overall operating efficiency of the nuclear power plant.

[0083] Vibration parameters include vibration intensity, peak vibration displacement, amplitude at the fundamental frequency in the vibration spectrum, and amplitude at the second harmonic of the fundamental frequency in the vibration spectrum; if any vibration parameter reaches the vibration limit value, it is classified into the corresponding range.

[0084] If the parameter value of the vibration intensity reaches the pre-warning range, regardless of whether the other three parameters reach the pre-warning range, it is determined that the pre-warning has been reached. If the peak vibration displacement reaches the range where measures need to be taken, regardless of the ranges of the other three parameters, it is determined that the range where measures need to be taken has been reached.

[0085] Embodiment III

[0086] As Figure 2 shown, a method for training a convolutional neural network through a training set is provided, including:

[0087] Initialize the parameters of the convolutional neural network, setting the initial weights and initial biases of the convolutional neural network; before the start of training, the weights and biases of the convolutional neural network need to be initialized. Commonly used methods include random initialization, small-value initialization, or using the parameters of a pre-trained model.

[0088] Convert the data in the training set into one-dimensional vectors; then convert the one-dimensional vectors into matrix form; this is data preprocessing. The original fault data in the training set is usually time series signals. These time series signals need to be converted into one-dimensional vectors and then rearranged into a matrix form suitable for convolutional operations to ensure that the data format meets the input requirements of the convolutional neural network.

[0089] Perform the first convolutional operation on the matrix to obtain a convolutional feature map; the convolutional layer extracts local features in the data by applying multiple convolutional kernels (filters) to the input matrix.

[0090] Perform maximum pooling on the convolution feature map to obtain the pooled feature map;

[0091] The pooled feature map is compressed into a one-dimensional vector and used as the input of the fully connected layer; the pooled feature map is a two-dimensional matrix, and in order to be input into the fully connected layer, it needs to be flattened into a one-dimensional vector.

[0092] The input vector is processed through the fully connected layer to obtain the output result; the fully connected layer takes the compressed one-dimensional vector as input, performs linear transformation and nonlinear activation operations, and generates the final output result.

[0093] Calculate the error between the output and the expected output; compare the output of the convolutional neural network with the expected output (label) in the training set and calculate the error.

[0094] According to the error, the weights and biases of the convolutional neural network are adjusted through the back propagation algorithm to minimize the error; the back propagation algorithm adjusts the weights and biases of the convolutional neural network layer by layer by calculating the gradient of the error. The goal is to minimize the error and make the model's prediction results closer to the true value. Commonly used optimization algorithms include stochastic gradient descent (SGD), Adam, etc.

[0095] Repeat the above steps and iteratively train the convolutional neural network until the error converges to the preset range; the training process is an iterative optimization process, repeating forward propagation, error calculation and back propagation, and gradually adjusting the model parameters until the error converges to the preset range. Error convergence means that the performance of the model has reached an ideal state, and the prediction results are stable and accurate.

[0096] After multiple iterations of training, a nuclear power plant rotating machinery fault diagnosis model was obtained.

[0097] In addition, after building the intelligent diagnosis model, fault data that has not been included in the model training is selected from the data model library as test data, and the small sample convolutional neural network diagnosis model is functionally verified. On this basis, multiple batches of repeated training and verification are carried out. At the same time, the diagnosis model parameters are adjusted by increasing or decreasing training samples, adjusting convolutional neural network parameters, etc. The training time, classification accuracy and other indicators of the diagnosis model are comprehensively statistically analyzed to determine the optimal small sample convolutional neural network diagnosis model parameters.

[0098] The specific method for testing the nuclear power plant rotating machinery fault diagnosis model includes:

[0099] The fault data in the test set is input into the preliminarily trained nuclear power plant rotating machinery fault diagnosis model to calculate the model's prediction results; the fault data in the test set that has not participated in the model training is input into the trained nuclear power plant rotating machinery fault diagnosis model. After the model receives the test data, it generates a prediction result through forward propagation calculation. The prediction result is usually a probability distribution of the fault type or a specific fault classification label.

[0100] Compare the model's prediction results with the actual fault labels in the test set and calculate the test error. The accuracy of the model's prediction is measured by calculating the error. Common error calculation methods include mean square error (MSE) and cross-entropy loss.

[0101] Analyze the test error to determine the model performance indicators; Analyzing the test error can derive a series of performance indicators, such as accuracy, precision, recall, F1 score, etc. These indicators can be used to comprehensively evaluate the performance of the model and determine the effectiveness of the model in fault detection and classification tasks.

[0102] If the performance index does not meet the preset requirements, the fault data in the test set is added to the training set to expand the training set; if the performance index of the model does not meet the preset requirements, it means that the model may still be insufficient. At this time, the fault data used for evaluation in the test set can be added back to the training set to increase the diversity and quantity of training data. By expanding the training set, the model can be exposed to more fault samples, thereby improving its generalization ability and prediction accuracy.

[0103] Retrain the convolutional neural network. Input the expanded training set into the convolutional neural network again and retrain it. This process includes steps such as forward propagation, error calculation, and back propagation.

[0104] Embodiment 4

[0105] A nuclear power plant rotating machinery fault diagnosis terminal based on a small sample includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the nuclear power plant rotating machinery fault diagnosis method based on a small sample as described above is implemented.

[0106] The memory can be used to store software programs and modules. The processor executes various functional applications and data processing of the terminal by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an execution program required for at least one function, etc.

[0107] The data storage area can store data created according to the use of the terminal, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0108] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample as described above is implemented.

[0109] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer readable instruction data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storage technology, CD-ROM, DVD or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory and mass storage devices can be collectively referred to as memory.

[0110] A computer program product includes a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the nuclear power plant rotating machinery fault diagnosis method based on a small sample as described above.

[0111] A computer program product includes a computer program or set of instructions for performing specific tasks or implementing specific functions. These programs or instructions are designed to be executed by a processor to implement a series of predefined steps or operations. The program product may be stored in various forms of computer storage media, such as memory, hard disk, solid-state drive, optical disk or other forms of digital storage devices. It may exist in the form of compiled binary code or in the form of scripts or bytecodes that can be executed by an interpreter. The program product uses carefully designed algorithms and logical instructions to enable the processor to process data in a specific order and manner to complete various functions such as data analysis, user interaction, device control, etc.

[0112] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments / methods or examples described in this specification and the features of the different embodiments / methods or examples, unless they are contradictory.

[0113] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0114] It should be understood by those skilled in the art that the above embodiments are only for the purpose of clearly illustrating the present invention, and are not intended to limit the scope of the present invention. For those skilled in the art, other changes or modifications may be made based on the above invention, and these changes or modifications are still within the scope of the present invention.

Claims

1. A method for diagnosing rotating machinery faults in nuclear power plants based on small samples, characterized in that: include: Collect fault data of various fault types, and process the fault data to obtain time domain data and frequency domain data; According to the vibration parameters corresponding to different fault types, the ranges are divided into: acceptable range, alarm range and range requiring measures; Construct a total sample set based on time domain data, frequency domain data and divided ranges; The training set is selected from the total sample set by a metric-based method; and the remaining data is used as the test set; Construct a convolutional neural network for one-dimensional fault data and train it with a training set to obtain a nuclear power plant rotating machinery fault diagnosis model; The nuclear power plant rotating machinery fault diagnosis model is tested using the test set, and the convolutional neural network is repeatedly trained based on the test results; Iterative training is performed until the optimal convolutional neural network parameters are obtained, and the final nuclear power plant rotating machinery fault diagnosis model is constructed.

2. A method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample according to claim 1, characterized in that: Fault types include: rotor misalignment, rotor imbalance, loose components, shaft friction, rotor cracks, and typical canned motor pump faults; The time domain data is obtained by time domain analysis, which includes calculating the average amplitude, root mean square value, variance and kurtosis index; The frequency domain data is obtained through frequency domain analysis, which includes discrete Fourier transform and fast Fourier transform.

3. A method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample according to claim 1, characterized in that: Two vibration limit values ​​T1 and T2 are set in sequence according to the vibration parameters of the rotating machinery, and T1<T2; if the vibration parameter T<T1, it is classified as an acceptable range; if T1≤T<T2, it is classified as an alarm range; T≥T2, it is classified as a range where measures need to be taken.

4. A method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample according to claim 3, characterized in that: According to the vibration parameters of the rotating machinery, three warning values ​​t1, t2 and t3 are set in sequence, and t1<T1<t2<T2<t3; if t1<T<T1, it is determined as a warning for the conversion from the acceptable range to the alarm range; if t2<T<T2, it is determined as a warning for the conversion from the alarm range to the range where measures need to be taken; if T>t2, it is determined as a warning that needs to be dealt with urgently.

5. A method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample according to claim 3 or 4, characterized in that: Vibration parameters include vibration severity, vibration displacement peak, amplitude at the fundamental frequency in the vibration spectrum, and amplitude at the second harmonic of the fundamental frequency in the vibration spectrum; if any vibration parameter reaches the vibration limit, it is divided into the corresponding range.

6. A method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample according to claim 1, characterized in that: Methods for training convolutional neural networks using training sets include: Initialize the convolutional neural network parameters, set the initial weights and initial biases of the convolutional neural network; Convert the data in the training set into a one-dimensional vector; then convert the one-dimensional vector into a matrix form; Perform the first convolution operation on the matrix to obtain the convolution feature map; Perform maximum pooling on the convolution feature map to obtain the pooled feature map; Compress the pooled feature map into a one-dimensional vector and use it as the input of the fully connected layer; The input vector is processed through the fully connected layer to obtain the output result; Calculate the error between the output and the expected output; Based on the error, the weights and biases of the convolutional neural network are adjusted through the back-propagation algorithm to minimize the error; Repeat the above steps and iteratively train the convolutional neural network until the error converges to the preset range; Obtain a nuclear power plant rotating machinery fault diagnosis model.

7. A method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample according to claim 1, characterized in that: The method for testing the fault diagnosis model of nuclear power plant rotating machinery includes: The fault data in the test set is input into the initially trained nuclear power plant rotating machinery fault diagnosis model, and the prediction results of the model are calculated; Compare the model's predictions with the actual fault labels in the test set and calculate the test error; Analyze test errors to determine model performance metrics; If the performance index does not meet the preset requirements, the fault data in the test set is added to the training set to expand the training set; Retrain the convolutional neural network.

8. A nuclear power plant rotating machinery fault diagnosis terminal based on a small sample, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample as described in any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a small sample-based nuclear power plant rotating machinery fault diagnosis method as described in any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the method for diagnosing rotating machinery faults in a nuclear power plant based on a small sample as described in any one of claims 1 to 7 is implemented.

Images