High-efficiency nuclear main pump rotor fault diagnosis method and device

Abstract

The invention discloses a high-efficiency nuclear main pump rotor fault diagnosis method and device, and belongs to the technical field of nuclear main pump fault diagnosis. The method comprises the steps of rotor state data acquisition, data processing, fault classification and fault diagnosis result output. The data processing comprises a fixed-rotating-speed rotor data processing flow and a variable-rotating-speed rotor data processing flow which are parallel to each other; performing decomposition by adopting an improved VMD method based on a Pearson's correlation coefficient aiming at a rotor with a fixed rotating speed, and inputting a feature vector based on a singular value into a support vector machine for diagnosis classification; a three-dimensional feature matrix is obtained through short-time Fourier transform and feature level fusion for a variable-speed rotor and input into a residual neural network model for diagnosis and classification. According to the method, the fault diagnosis models of the fixed rotating speed and the variable rotating speed are cooperatively applied, so that the real-time performance of fault diagnosis can be greatly improved while high diagnosis precision is ensured, and the overall efficiency of fault diagnosis is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of nuclear main pump fault diagnosis, and particularly relates to a high-efficiency nuclear main pump rotor fault diagnosis method and device. Background Art

[0002] The nuclear reactor coolant pump (abbreviation: nuclear main pump) is the only high-speed rotating mechanical equipment in the pressurized water reactor coolant loop system, and is called the "heart" of the nuclear reactor. The rotor is one of the core components of the nuclear main pump. It rotates at a high speed in an environment of high temperature, high pressure and high radiation for a long time, and may have faults such as rotor mass imbalance, rubbing, misalignment, crack, rotor bending, etc. The occurrence of rotor faults will, in the lightest case, reduce the operation accuracy and service life of the main pump, and in the worst case, cause mechanical damage and catastrophic consequences. Therefore, carrying out effective fault diagnosis research on the rotor is of great significance for improving the operation and maintenance level of nuclear power equipment.

[0003] Existing rotor fault diagnosis systems usually only cover a single fault diagnosis model. However, a single model has certain limitations and is difficult to comprehensively and accurately identify various fault types and their complex and diverse manifestations.

[0004] In the field of rotor fault diagnosis under constant speed conditions, time-frequency analysis methods have shown good applicability. As one of the most widely used time-frequency analysis means in rotor fault diagnosis, the core of the Hilbert-Huang transform lies in empirical mode decomposition (EMD). However, when the EMD method decomposes the rotor vibration signal, it is easy to generate false components and mode mixing phenomena, which seriously restricts the performance of its play. Although improved algorithms of EMD, such as ensemble empirical mode decomposition (EEMD) and complementary ensemble empirical mode decomposition (CEEMD), have been improved to a certain extent, due to the introduction of white noise and the need for multiple iterations, they inevitably increase the signal reconstruction error and computational complexity. With the development of technology, variational mode decomposition (VMD) has gradually been widely used in the research of rotating equipment fault diagnosis. For example, Hang et al. proposed a feature extraction method based on VMD with the convergence of sample entropy as the optimization goal, and successfully extracted the fault characteristics of the micro-motor rotor imbalance; Zhu et al. applied VMD to the experimental analysis of the fault data of different rubbing degrees of the rotor, and this method can accurately obtain the frequency components of the rubbing fault signal, so as to accurately reflect the fault information; Li X et al. proposed a rolling bearing fault diagnosis method based on VMD-FRFT, and this method has strong anti-noise ability, especially suitable for low signal-to-noise ratio signals. However, the number of VMD modes needs to be set in advance and has uncertainty, and its selection plays a decisive role in the signal decomposition performance. In addition, the fault diagnosis method based on signal decomposition requires technicians to have an in-depth understanding of different fault feature manifestations, which undoubtedly increases the difficulty of diagnosis.

[0005] In the field of rotor fault diagnosis under variable speed conditions, the amplitude and characteristic frequency of rotor vibration signals are constantly changing, which undoubtedly poses greater challenges to fault diagnosis work. Relevant research shows that convolutional neural networks (CNNs) have powerful learning abilities, flexible model structures, and good transferability, and have achieved remarkable results in the fault diagnosis of variable speed rotating equipment. For example, Wei Z et al. proposed a wide convolutional kernel deep convolutional neural network model and achieved the fault diagnosis of rolling bearings under variable conditions with the help of original vibration signals; Qiu G Q et al. constructed a bearing variable speed condition fault diagnosis model based on WSO-VMDA and ResNet-SWIN and verified its effectiveness through experiments; Wang F et al. constructed a motor fault diagnosis model under variable speed conditions using a cascaded CNN with a stepwise optimization function; Zhao X et al. combined a new intra-class and inter-class constraint (IIC) with an integrated adaptive activation function with CNN, enabling it to be applied to the fault diagnosis of gearboxes under variable speed conditions. However, most of the existing variable speed rotating equipment fault diagnosis models focus on the bearing field, and typical CNNs may experience a decline in performance as the number of network layers increases.

[0006] The research results show that the constant speed diagnosis model features rapid diagnosis speed, high accuracy, and a small demand for samples; while the variable speed diagnosis model presents the advantages of high accuracy and excellent generalization ability. Each of these two models has its own unique strengths and significant characteristics. Summary of the Invention

[0007] The present invention aims to solve at least one of the technical problems in the above related technologies to some extent.

[0008] To this end, the object of the present invention is to provide a high-efficiency nuclear main pump rotor fault diagnosis method and device, which can greatly improve the real-time performance of fault diagnosis while ensuring high diagnostic accuracy, and improve the overall efficiency of fault diagnosis.

[0009] To solve the above technical problems, the present invention is implemented as follows:

[0010] The embodiment of the present invention provides a high-efficiency nuclear main pump rotor fault diagnosis method, and the method includes rotor state data acquisition, data processing and fault classification, and outputting fault diagnosis results;

[0011] The data processing includes a parallel constant speed rotor data processing process and a variable speed rotor data processing process;

[0012] The constant speed rotor data processing process includes:

[0013] S11. Based on the data category, perform data-level fusion on each data;

[0014] S12. Perform normalization processing on the fused data;

[0015] S13. Use the improved VMD method based on the Pearson correlation coefficient to decompose the normalized data, obtaining a series of modal components;

[0016] S14. Compose an initial feature matrix with the obtained modal components, and perform singular value decomposition on the initial feature matrix to obtain a series of singular values;

[0017] S15. Compose a feature vector with the obtained singular values, input it into the constant-speed fault classification model for classification processing, and obtain the fault diagnosis type;

[0018] The variable-speed rotor data processing flow includes:

[0019] S21. Perform normalization processing on the state data of the collected variable-speed rotor respectively;

[0020] S22. Perform short-time Fourier transform on each state data respectively to obtain several two-dimensional time-frequency diagrams;

[0021] S23. Perform feature-level fusion on the two-dimensional time-frequency diagrams to obtain a three-dimensional feature matrix;

[0022] S24. Input the three-dimensional feature matrix into the variable-speed fault classification model for classification processing, and obtain the fault diagnosis type.

[0023] In addition, according to the high-efficiency nuclear main pump rotor fault diagnosis method of the present invention, the following additional technical features may also be included:

[0024] In some of the embodiments, the rotor state data acquisition is specifically: using a plurality of sensors to collect the vibration data of each position of the nuclear main pump rotor respectively.

[0025] In some of the embodiments, the specific steps of step S13 include:

[0026] S131. Set the initial value of the number of modes K, and perform VMD decomposition on the vibration signal of the rotor to obtain K modal components with center frequencies;

[0027] S132. Let i = 1 and i ≤ K, and calculate the Pearson correlation coefficient between the i-th modal component and the original signal;

[0028] S133. Judge whether the Pearson correlation coefficient of the current modal component is less than the threshold. If so, stop the iteration and enter S135; otherwise, let i = i + 1 and return to S132;

[0029] S134. When all K modal components are not less than the threshold, let K = K + 1, and return to S131;

[0030] S135. Eliminate the modal components with Pearson correlation coefficients less than the threshold, and retain the existing modal components.

[0031] In some embodiments, the constant-speed fault classification model is a trained support vector machine.

[0032] In some embodiments, the initial value of K is 2.

[0033] In some embodiments, the calculation method of the Pearson correlation coefficient is as follows:

[0034]

[0035] where p is the Pearson correlation coefficient, and are the means of x and y respectively, x and y represent two variables respectively, and n represents the dimension of these two variables.

[0036] In some embodiments, the threshold in step S133 is 0.2.

[0037] In some embodiments, the normalization method in step S21 is as follows:

[0038]

[0039] where x max and x min represent the maximum and minimum values of the sample x respectively.

[0040] In some embodiments, the variable-speed fault classification model is a trained residual neural network model.

[0041] The embodiment of the present invention also provides a high-efficiency nuclear main pump rotor fault diagnosis device, including a processor and a memory. A software program is stored on the memory, and when the processor runs the software program, it can implement the steps of the high-efficiency nuclear main pump rotor fault diagnosis method described in any one of the above.

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

[0043] In the embodiments of the present invention, the provided high-efficiency fault diagnosis method for the rotor of a nuclear main pump improves the overall efficiency of fault diagnosis in terms of the collaborative application of dual models. The constant-speed diagnosis model features rapid diagnosis speed, high accuracy, and a relatively small demand for samples. The variable-speed diagnosis model, on the other hand, exhibits high accuracy and excellent generalization ability. Each of these two models has its own unique strengths and remarkable characteristics. By synergistically applying the constant-speed and variable-speed fault diagnosis models, the advantages of different models can be integrated to complement each other's deficiencies, thereby effectively improving the accuracy and reliability of fault diagnosis work and accelerating the diagnosis speed.

[0044] In the prior art, fault diagnosis of the rotor of a nuclear main pump can be achieved through conventional signal processing and machine learning techniques. However, due to the limitations of a single fault diagnosis model, it is difficult to comprehensively, accurately, and quickly identify various fault types and their complex manifestations. The method proposed in the present invention covers technologies related to multi-sensor information fusion, time-frequency analysis, machine learning, deep residual network, and multi-model collaborative application, enabling high-precision and high-efficiency fault diagnosis of the main pump rotor under all operating conditions.

[0045] The high-efficiency fault diagnosis device for the rotor of a nuclear main pump of the present invention can implement the steps of the high-efficiency fault diagnosis method for the rotor of a nuclear main pump, and thus has at least all the features and advantages of the high-efficiency fault diagnosis method for the rotor of a nuclear main pump, which will not be elaborated here. The additional aspects and advantages of the present invention will be partially given in the following description, partially become apparent from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0046] Figure 1 It is a flowchart of the high-efficiency fault diagnosis method for the rotor of a nuclear main pump disclosed in an embodiment of the present invention.

[0047] Figure 2 It is a flowchart of the improved VMD method based on the Pearson correlation coefficient disclosed in an embodiment of the present invention. Detailed Embodiments

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. 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.

[0049] Next, the embodiments of the present invention will be described in detail in conjunction with the drawings, through specific embodiments and their application scenarios.

[0050] Both the constant-speed diagnostic model and the variable-speed diagnostic model have their own unique advantages and remarkable characteristics. If the constant-speed and variable-speed fault diagnostic models are used in cooperation, the advantages of different models can be integrated to make up for each other's deficiencies, thereby effectively improving the accuracy and reliability of fault diagnosis work and accelerating the diagnostic speed. In this way, the progress and innovation of the nuclear main pump fault diagnosis technology can be vigorously promoted, laying a solid foundation for the continuous, stable and safe operation of the nuclear main pump rotor.

[0051] Please refer to Figure 1 As shown, in some embodiments of the present invention, a high-efficiency nuclear main pump rotor fault diagnosis method is provided, including two parallel rotor fault diagnostic models. The specific implementation steps are as follows:

[0052] Under the constant-speed working condition:

[0053] Step 1: Under the constant-speed working condition, use multiple sensors to jointly observe the operating state of the main pump rotor, and multiple sensors simultaneously collect the rotor vibration signals. Many pieces of information about the rotor operating state are included in the time-domain and frequency-domain information of the signals.

[0054] Step 2: Perform data-level information fusion on the data obtained by different sensors in Step 1, and then perform standardization processing on the fused data to facilitate subsequent fault classification. Multi-sensor information fusion can enhance the integrity of information and thus improve the accuracy of information. Standardization processing can improve the comparability of data and thus enhance the accuracy of data analysis.

[0055] Step 3: Use the improved VMD method based on the Pearson correlation coefficient to decompose the data processed in Step 2 to obtain a series of modal components (IMFs). The process of decomposing using the improved VMD method is as Figure 2 shown. The specific steps are as follows:

[0056] 1) Set the initial value of the number of modes K to 2, perform VMD decomposition on the rotor vibration signal, and obtain K modal components μ i (i = 1, 2,...K).

[0057] 2) Calculate the Pearson correlation coefficient p(μ i between each decomposed modal component μ i and the original signal f). The calculation process of the Pearson correlation coefficient is expressed as:

[0058]

[0059] In the formula, p is the Pearson correlation coefficient, x and y are x = [x1, x2,..., x n , y = [y1, y2,..., yn The mean value of ], a positive correlation coefficient indicates a positive correlation between x and y, and a negative correlation coefficient indicates a negative correlation between x and y. x and y represent two variables respectively, and n represents the dimension of these two variables. The larger |p| is, the stronger the correlation between x and y is, and vice versa, the weaker the correlation is.

[0060] 3) If all the correlation coefficients p(μ i , f), (i = 1, 2,... K) calculated in step 2) are greater than 0.2, this means that the signal is under-decomposed. Update the value of K to make K = K + 1, and repeat the previous steps 1) and 2).

[0061] 4) When the correlation coefficient p(μ i , f) is less than 0.2 for the first time, stop the iterative update process, and eliminate the modal components with correlation coefficients less than 0.2. The modal components with correlation coefficients less than 0.2 are spurious components, so as to obtain K - 1 effective modal components p(μ i , f), (i = 1, 2,... K - 1).

[0062] Step 4: Combine the effective modal components obtained in step 3) to form an initial feature matrix, and perform singular value decomposition on the matrix to obtain a series of singular values.

[0063] Singular Value Decomposition (SVD for short) is a commonly used matrix decomposition algorithm, which decomposes a matrix into the product of three matrices. In the present invention, the SVD of matrix A is defined as:

[0064] A = UXV T (2)

[0065] In the formula, U represents a two-dimensional matrix of m×m, X represents a two-dimensional matrix of m×m, the elements outside the main diagonal are all 0, and the elements on the main diagonal are the singular values of the SVD decomposition. V represents an n×n matrix, and U and V satisfy U T U = I, V T V = I. For the square matrix A T Perform eigenvalue decomposition on A to obtain formula (3):

[0066] (A T A)v i = λ i v i (3)

[0067] From this, n eigenvalues of A T A and the corresponding n eigenvectors can be obtained. Combine the n eigenvectors of A T A to form an n×n matrix V, which is the V matrix in formula (2). For the square matrix AAT Perform eigenvalue decomposition and obtain formula (4):

[0068] (AA T ) i =λ i u i (4)

[0069] AA T The m eigenvectors of form an m×m matrix U, which is the U matrix in formula (2). From this, we can get formula (5):

[0070]

[0071] From this we can find each singular value σ i , and then find the singular value matrix X.

[0072] Step 5: Input the singular value feature vector obtained in step 4 into the support vector machine for fault classification to determine the operating status of the nuclear main pump rotor, and feed the results back to the human-computer interaction interface so that the staff can make management decisions.

[0073] The above-mentioned support vector machine is a pre-trained support vector machine model. The original training data is the historical operating status of the main pump rotor collected by the sensor. It has undergone the fault diagnosis processing under the above-mentioned constant speed condition to obtain a feature vector composed of singular values ​​based on the historical data. Different labels are set for the feature vectors of different fault states. The feature vectors and labels are used as training data sets to train the support vector machine to obtain a trained support vector machine.

[0074] The reason why the constant speed rotor fault diagnosis in the present invention is of high accuracy and high speed lies in the overall implementation of steps 2, 3, and 5. Step 2 performs data-level information fusion on the data obtained by different sensors, which can obtain more accurate and complete information than a single sensor, and reduce the probability of misjudgment of faults caused by missing information from a single sensor. Step 3 uses an improved VMD method based on the Pearson correlation coefficient to decompose the signal and eliminate invalid components, thereby retaining effective fault feature information to the greatest extent and improving the accuracy of the diagnosis model. Step 5 uses a support vector machine to classify faults. The support vector machine is suitable for small sample fault classification, and its computational complexity is low, so that the diagnosis speed can be improved while ensuring the fault classification effect.

[0075] Under variable speed conditions:

[0076] Step 1: Use multiple sensors to jointly observe the operating status of the main pump rotor under variable speed conditions. Multiple sensors simultaneously collect rotor vibration signals, such as radial displacement signals of rotor vibration. Much of the information about the rotor's operating status is contained in the signal's time domain and frequency domain information.

[0077] Step 2: Normalize the rotor vibration data collected by different sensors to eliminate the influence of data dimension differences on model training. The data normalization process can be expressed as:

[0078]

[0079] where x max and x min represent the maximum and minimum values of sample x, respectively.

[0080] Step 3: Perform short-time Fourier transform on the data from n different sensors under the same fault mode respectively, and n two-dimensional time-frequency diagrams can be obtained. Feature-level fusion of the n two-dimensional time-frequency diagrams can obtain a three-dimensional feature matrix. Then, the above processing is performed on the data under different fault modes respectively, and several three-dimensional feature matrices can be obtained. Feature-level fusion is to perform information fusion during the feature extraction process. Here, the n two-dimensional time-frequency diagrams are the extracted information, and merging the n two-dimensional time-frequency diagrams represents feature-level information fusion.

[0081] The short-time Fourier transform is a commonly used time-frequency analysis algorithm. It can divide the signal into multiple short-time windows, perform Fourier transform on each window, and obtain the spectral information of the window. By sliding the window, the spectral information of the entire signal at different time periods can be obtained

[69] . The definition formula of the short-time Fourier transform is shown in Equation (4-1):

[0082]

[0083] where S(ω,τ) is the amplitude at frequency ω at time τ; g(t - τ) represents the window function with the central time at τ; f(t) represents the input signal; * represents the complex conjugate symbol; j represents the complex number symbol. Since the nuclear main pump rotor data is a discrete time series, the discrete form of STFT is adopted in this chapter, and its definition formula is shown in Equation (8):

[0084]

[0085] where f(k) represents the input signal; g(kT - mT) represents the window function; m, n, and k are all integers; T represents the sampling period of time; F represents the sampling period of frequency.

[0086] Step 4: Normalize the three-dimensional feature matrix obtained in Step 3, and then input it into a pre-designed residual neural network for fault classification to determine the operating state of the main pump rotor, and feedback the diagnosis result to the human-computer interaction interface for staff to make management decisions.

[0087] In one embodiment of the present invention, the formula for data normalization is as follows:

[0088]

[0089] where x is the original data, x min and x max are the minimum and maximum values of the data respectively, and is the normalized data. Usually, the data is normalized to the interval [0, 1].

[0090] The above residual neural network is a pre-trained model. The original data for training is the historical operating state of the main pump rotor collected by sensors, and after being processed by the above variable-speed rotor process, a three-dimensional feature matrix is obtained. Classification labels are respectively set for the feature matrices corresponding to different fault modes. The three-dimensional feature matrix and the corresponding classification labels are used as the training data set to train the residual neural network, and a trained residual neural network is obtained.

[0091] The variable-speed rotor fault diagnosis model in the present invention has high accuracy and strong generalization ability, which is due to the overall implementation of Step 3 and Step 4. In Step 3, the short-time Fourier transform is used to analyze the rotor vibration data. The short-time Fourier transform can simultaneously display the information of the signal in two dimensions of the time domain and the frequency domain, and it can also accurately track the dynamic change process of the signal frequency; by performing feature-level data fusion on the rich information extracted from different sensor data through the short-time Fourier transform, the accuracy, integrity and reliability of the information can be improved. Thus, higher accuracy can be achieved when performing fault diagnosis. In Step 4, the residual neural network is used to extract features. The residual neural network can effectively solve the problem of gradient disappearance, and can construct a very deep network to extract complex features; it can reuse features, improve the learning ability, and make the network converge faster; it also has strong generalization ability, can extract high-level information, and combat overfitting.

[0092] Aiming at the problem that the existing rotor fault diagnosis system usually only covers a single fault diagnosis model. However, a single model has certain limitations and it is difficult to comprehensively and accurately distinguish various fault types and their complex and diverse manifestations. The present invention conducts research on improving the accuracy of the nuclear main pump rotor fault diagnosis while enhancing the overall efficiency of the diagnosis. For this purpose, fault diagnosis models adapted to the fixed-speed and variable-speed conditions of the rotor are respectively designed. By means of the coordinated operation of the two models, their respective advantages are integrated and their mutual defects are made up, thereby improving the speed, accuracy, reliability and generalization ability of the fault diagnosis.

[0093] Aiming at the deficiencies of the existing VMD in the background technology, when constructing the fixed-speed rotor fault model of the present invention, an improved VMD algorithm based on the Pearson coefficient is combined with the support vector machine (SVM). In this way, not only can the performance of the VMD method be improved, but also the diagnostic results can be directly output, effectively reducing the difficulty for technicians to classify faults.

[0094] Aiming at the problem of variable-speed rotor fault diagnosis, the present invention innovatively proposes a fault diagnosis model that combines the STFT transform and ResNet. This model can fully exploit multi-sensor information. After taking the rotor vibration signal as a sample input, it can automatically complete a series of processes such as feature extraction, model training, and diagnostic result output, providing an efficient and reliable solution for variable-speed rotor fault diagnosis.

[0095] For the parts not described in detail in the present invention, reference can be made to the existing technologies in the field or the well-known technologies to those skilled in the art, and the present invention will not be elaborated too much.

[0096] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the purpose of the present invention and the scope protected by the claims, and all of them belong to the protection scope of the present invention.

Claims

1. A high-efficiency fault diagnosis method for the rotor of a nuclear main pump, characterized in that, The method includes rotor state data collection, data processing and fault classification, and output of fault diagnosis results; The data processing includes a parallel fixed speed rotor data processing flow and a variable speed rotor data processing flow; The constant speed rotor data processing process includes: S11. Based on data categories, data-level fusion is performed on each data; S12, standardizing the fused data; S13, using the improved VMD method based on Pearson correlation coefficient to decompose the standardized data and obtain a series of modal components; S14, using the obtained modal components to form an initial characteristic matrix, and performing singular value decomposition on the initial characteristic matrix to obtain a series of singular values; S15, using the obtained singular values ​​to form a feature vector, inputting it into a constant speed fault classification model for classification processing, and obtaining a fault diagnosis type; The variable speed rotor data processing process includes: S21, normalizing the collected state data of the variable speed rotor respectively; S22, performing short-time Fourier transform on each state data to obtain a number of two-dimensional time-frequency graphs; S23, performing feature-level fusion on the two-dimensional time-frequency graph to obtain a three-dimensional feature matrix; S24, input the three-dimensional feature matrix into the variable speed fault classification model for classification processing to obtain the fault diagnosis type.

2. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 1, characterized in that The rotor status data collection specifically includes: using a plurality of sensors to respectively collect vibration data of each position of the nuclear main pump rotor.

3. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 1, wherein, The specific steps of step S13 include: S131, setting an initial value of the modal number K, and performing VMD decomposition on the vibration signal of the rotor to obtain K modal components with center frequencies; S132, let i=1 and i≤K, calculate the Pearson correlation coefficient between the i-th modal component and the original signal; S133, determine whether the Pearson correlation coefficient of the current modal component is less than the threshold value, if so, stop the iteration and enter S135; otherwise, set i=i+1 and return to S132; S134. When all K modal components are not less than the threshold, set K=K+1 and return to S131; S135. Eliminate the modal components whose Pearson correlation coefficient is less than the threshold value, and retain the existing modal components.

4. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 1, wherein, The constant speed fault classification model is a trained support vector machine.

5. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 3, characterized in that The initial value of K is 2.

6. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 3, characterized in that The Pearson correlation coefficient is calculated as: where p is the Pearson correlation coefficient, and are the means of x and y respectively, where x and y represent two variables, and n represents the dimension of these two variables.

7. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 3, characterized in that The threshold in step S133 is 0.

2.

8. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 1, characterized in that The normalization process in step S21 is as follows: where x max and x min represent the maximum and minimum values of the sample x, respectively.

9. The high-efficiency nuclear main pump rotor fault diagnosis method according to claim 1, wherein The variable speed fault classification model is a trained residual neural network model.

10. A high-efficiency nuclear main pump rotor fault diagnosis device, comprising a processor and a memory, wherein a software program is stored on the memory, and is characterized in that, When the processor runs the software program, it can implement the steps of the high-efficiency nuclear main pump rotor fault diagnosis method as described in any one of claims 1 to 9.

Images