Turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM

Abstract

The invention provides a steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM, and belongs to the technical field of thermal power plant mechanical fault diagnosis, and the method comprises the steps: obtaining an initial vibration signal of a steam turbine rotor; performing feature extraction on the initial vibration signal based on VMD kurtosis estimation and an RVM classifier, and constructing a second training set and a second test set taking a feature vector matrix of the initial vibration signal as content; taking the second training set as input data of an RVM classifier, and training the RVM classifier; optimizing and selecting kernel parameters to be optimized of the trained RVM classifier by adopting an artificial bee colony algorithm; the second test set serves as input data of the optimized RVM classifier, the fault type of the steam turbine rotor is determined according to the obtained output value of the decision function, the optimized RVM classifier is tested, and the optimized RVM classifier is used for determining the fault type of the steam turbine rotor. According to the scheme, the turbine rotor fault diagnosis accuracy and recognition efficiency are improved.

Description

Technical Field

[0001] The present application relates to the technical field of mechanical fault diagnosis in thermal power plants, and in particular to a method for diagnosing turbine rotor faults based on VMD kurtosis estimation and RVM. Background Art

[0002] Steam turbine generator sets are key production equipment in thermal power plants. They are characterized by complex structures, harsh working conditions (such as high temperature, high pressure, and high speed), and high requirements for continuous operation, and are prone to failure. The rotor system is the core component of the steam turbine generator set. The status monitoring and fault diagnosis of the rotor system ensure the safe and stable operation of the steam turbine generator set.

[0003] When analyzing the common fault characteristics of turbine rotor vibration, it is crucial to timely identify the fault type through the fault signal. In the prior art, time-frequency analysis methods are often used, such as discrete Fourier transform, short-time Fourier transform, discrete wavelet transform, S transform, empirical mode decomposition, and local mean decomposition. With the development of intelligent diagnosis technology, many intelligent algorithms have been applied to the identification and diagnosis of fault characteristics, such as artificial neural networks, support vector machines, etc. At the same time, genetic algorithms, particle swarm algorithms, and annealing algorithms are used to optimize the parameters of artificial neural networks and support vector machines.

[0004] However, the accuracy and recognition efficiency of these fault diagnosis methods are low. Therefore, how to improve the accuracy and recognition efficiency of steam turbine rotor fault diagnosis has become a problem to be solved. Summary of the invention

[0005] The purpose of this application is to provide a turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM, which can solve the problems of low accuracy and recognition efficiency of fault diagnosis methods in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for diagnosing a turbine rotor fault based on VMD kurtosis estimation and RVM, the method comprising:

[0007] Acquire an initial vibration signal of a steam turbine rotor, and divide the initial vibration signal into a first training set and a first test set;

[0008] Based on VMD kurtosis estimation and RVM classifier, feature extraction is performed on the initial vibration signal, and a second training set and a second test set are constructed with the feature vector matrix of the initial vibration signal as the content;

[0009] The feature vector in the second training set is used as the input data of the RVM classifier to train the RVM classifier;

[0010] The artificial bee colony algorithm is used to optimize the kernel parameters to be optimized of the trained RVM classifier;

[0011] The feature vector in the second test set is used as the input data of the optimized RVM classifier, the decision function of the optimized RVM classifier is obtained, and the fault type of the turbine rotor is determined according to the output value of the decision function to test the optimized RVM classifier; wherein the optimized RVM classifier is used to diagnose the fault of the turbine rotor and output the fault type of the turbine rotor.

[0012] In a possible implementation manner of the first aspect, after acquiring the initial vibration signal of the steam turbine rotor, the method further includes:

[0013] The initial vibration signal is denoised using a wavelet regional threshold denoising algorithm or a block threshold estimation algorithm.

[0014] In a possible implementation of the first aspect, a block threshold estimation algorithm is used to perform denoising processing on the initial vibration signal, including:

[0015] The initial vibration signal is processed by wavelet transform to obtain the wavelet coefficients of each scale. The expression of the wavelet coefficients is as follows:

[0016]

[0017] Among them, c j0l represents the scale factor, d jm represents the wavelet coefficient, j represents the scale, l represents the ordinal number of the scale coefficient, m represents the ordinal number of the wavelet coefficient, j0 represents the maximum scale of the scale coefficient, and J represents the maximum scale of the wavelet coefficient;

[0018] In each scale j, the wavelet coefficients are divided into several adjacent non-overlapping blocks. Block The length of is L0=floor[(logn) / 2]; wherein i represents the ordinal number of the block, floor represents the floor rounding function, and n represents the number of sampling points of the initial vibration signal;

[0019] The block is extended to a length of L1 = max(1,[1,L0 / 2]) Extending to the left and right, forming large blocks that intersect each other Big Block The length of is L=L0+2L1;

[0020] In large blocks Calculate the shrinkage factor. The shrinkage factor is expressed as follows:

[0021]

[0022] in, represents the shrinkage factor, λ represents the threshold, and λ * =4.50524, ε represents the standard deviation of the noise in the initial vibration signal, ε 2 represents the square of the standard deviation of the noise, ε>0, k represents an ordinal number;

[0023] Based on the shrinkage factor The wavelet coefficients within are thresholded, and the specific expression is as follows:

[0024]

[0025] After threshold processing, the data After performing inverse wavelet transform, the expression of the denoised vibration signal is as follows:

[0026]

[0027] in, represents the vibration signal after denoising, t represents time, φ j0k (t) represents the scaling function, ψ jk (t) represents the wavelet function.

[0028] In a possible implementation of the first aspect, based on VMD kurtosis estimation and RVM classifier, feature extraction is performed on the initial vibration signal, including:

[0029] The denoised vibration signal is decomposed into several finite bandwidth modal function components by using variational mode decomposition method. The kurtosis value of each finite bandwidth modal function component is estimated according to the rotor fault mechanism. The kurtosis value corresponding to each finite bandwidth modal function component is standardized.

[0030] In a possible implementation of the first aspect, a variational modal decomposition method is used to decompose the denoised vibration signal into a plurality of finite bandwidth modal function components, including:

[0031] Initialize each finite bandwidth modal function component, its corresponding center frequency and Lagrange operator, and set the target number of iterations. The specific expression is as follows:

[0032]

[0033] Among them, u k represents the finite bandwidth modal function component, k represents the ordinal number, ω k represents the center frequency corresponding to the bandwidth-limited modal function component, λ Lrepresents the Lagrange operator, b represents the target number of iterations, and each iteration increases by 1 based on the number of previous iterations;

[0034] By performing Hilbert transform on each finite bandwidth modal function component, the expression of the analytical signal of each finite bandwidth modal function component is obtained as follows:

[0035]

[0036] Among them, δ(t) represents the pulse function, t represents time, j represents scale, and u k (t) represents the finite bandwidth mode function, u k (t) = A k (t)cos(φ k (t)), A k (t) represents u k (t) envelope amplitude, φ k (t) represents u k The instantaneous phase of (t);

[0037] Exponential correction is performed on each center frequency to transfer the spectrum of each limited bandwidth modal function component to the corresponding baseband. The specific expression is: Calculate the square of the gradient of the analytical signal L 2 The norm is used to estimate the bandwidth of the center frequency of each finite bandwidth modal function component, and the Lagrange operator and quadratic penalty factor are introduced to transform the inequality constraint into an equality constraint. The corresponding augmented Lagrange function expression is as follows:

[0038]

[0039]

[0040] Where L represents the augmented Lagrange function, α represents the quadratic penalty factor, represents the partial derivative of the finite bandwidth modal function with respect to time t, and f(t) represents the denoised vibration signal;

[0041] The alternating direction algorithm is used to iteratively optimize the sequence, and the expressions of the finite bandwidth modal function components and center frequency are obtained as follows:

[0042]

[0043] Among them, u k ^n+1 (ω) represents the Wiener filtering of the finite bandwidth modal function component, for u k ^n+1 (ω) Perform inverse Fourier transform to get the real part u k(t),ω k n+1 (ω) represents the center frequency corresponding to the bandwidth-limited modal function component;

[0044] Repeat the steps after initialization until the termination condition is met, and finally obtain the K finite bandwidth modal function components with the smallest sum of bandwidths as u k ={u1,u2,…,u k}, and the center frequency corresponding to each finite bandwidth modal function component is ω k ={ω1,ω2,…,ω k}; The expression of the termination condition is σ represents the discrimination accuracy, and σ>0.

[0045] In a possible implementation of the first aspect, estimating the kurtosis value of each finite bandwidth modal function component according to the rotor fault mechanism includes:

[0046] The kurtosis value is calculated according to the finite bandwidth modal function components and their average values, and the kurtosis value vector matrix T of the finite bandwidth modal function components is obtained. P}, where K1, K2, …, K P Respectively represent the kurtosis value of each finite bandwidth modal function component, P = 1, 2…, l, l represents the total number of extracted kurtosis values; the calculation formula of the kurtosis value of the finite bandwidth modal function component is as follows:

[0047]

[0048] Where K represents the kurtosis, x represents the finite bandwidth modal function component, represents the average value of the finite bandwidth modal function component, i represents the ordinal number, and n represents the number of samples;

[0049] The kurtosis values ​​corresponding to each finite bandwidth modal function component are standardized, including:

[0050] After normalizing each kurtosis value, the expression of the eigenvector matrix of the denoised vibration signal is as follows:

[0051]

[0052] Where T′ represents the eigenvector matrix of the vibration signal after denoising,

[0053] In a possible implementation of the first aspect, an artificial bee colony algorithm is used to optimize and select kernel parameters to be optimized of the trained RVM classifier, including:

[0054] Based on the first fitness value of the old food source position and the second fitness value of the new food source position of at least one scout bee, a first target food source position is determined, and the first target food source position is used as the current optimal food source position; wherein the expression of the first target food source position includes a kernel parameter to be optimized; based on the scout bee completing the search for the food source position and determining the first target food source position, the scout bee shares the information of the amount of nectar collected and the first target food source position with at least one non-scout bee in the preset search area;

[0055] The method comprises the following steps: selecting a food source location based on the fitness probability of the non-scout bees according to the amount of nectar collected by the scout bees and / or the food source location, and searching for a new food source location within a preset range of the selected food source location based on the non-scout bees; judging whether to replace the selected food source location based on the first fitness value of the selected food source location and the second fitness value of the new food source location to determine a second target food source location, and using the second target food source location as the current optimal food source location;

[0056] If a food source position in a replacement state exists, a randomly generated old food source position is used to replace the food source position in a replacement state based on scout bees; wherein the randomly generated old food source position is used to scout the food source position based on scout bees; the step of determining the first target food source position and subsequent steps are repeatedly traversed and executed until the maximum number of search cycles is reached, and the best food source position obtained when the maximum number of search cycles is reached is used as the final best food source position; wherein the expression of the final best food source position contains the optimal kernel parameter.

[0057] In a possible implementation manner of the first aspect, before determining the position of the first target food source, the method further includes:

[0058] Randomly initialize M food source locations in the preset search area, and generate the old food source location according to the following calculation formula:

[0059] x i,j =x min,j +r1×(x max,j -x min,j )

[0060] Among them, x i,j represents the location of the old food source, x max,j represents the upper bound of the jth kernel parameter to be optimized, x min,j represents the lower bound of the jth kernel parameter to be optimized, i represents the ordinal number, j = 1, 2…, D, D represents the kernel parameter to be optimized, and r1 is a random number in the range of [0, 1];

[0061] According to the objective function of the artificial bee colony algorithm, the first fitness value of each old food source position is calculated. The specific calculation formula is as follows:

[0062]

[0063] Among them, Fitness i represents the first fitness value of the i-th old food source position. The expression of the objective function of the artificial bee colony algorithm is y q represents the actual output value of the RVM classifier, represents the validation value, v represents the number of training samples in the second training set;

[0064] For the scout bees within the preset range of the current old food source location, the new food source location is generated according to the following calculation formula:

[0065]

[0066] in, represents the location of the new food source, l represents the ordinal number, l≠i, and r2 is a random number in the range of [-1, 1];

[0067] The second fitness value of each new food source location is calculated according to the objective function of the artificial bee colony algorithm. The specific calculation formula is as follows:

[0068]

[0069] in, Represents the second fitness value of the i-th new food source location.

[0070] In a possible implementation of the first aspect, determining the first target food source position according to the first fitness value of the old food source position and the second fitness value of the new food source position includes:

[0071] If the second fitness value of the new food source position is greater than the first fitness value of the old food source position, the new food source position is used to replace the old food source position, and the new food source position is used as the first target food source position; otherwise, the old food source position is retained and the old food source position is used as the first target food source position.

[0072] In a possible implementation of the first aspect, selecting the location of the food source based on the fitness probability of the non-scout bees according to the amount of nectar collected by the scout bees and / or the location of the food source includes:

[0073] Based on the non-scout bees, it is determined whether the amount of nectar collected by the scout bees is greater than the first preset threshold, and the fitness probability of the food source location is calculated according to the following calculation formula to determine whether the fitness probability is greater than the second preset threshold:

[0074]

[0075] Among them, P i Represents the fitness probability of the i-th food source location, Fitness i represents the fitness value of the i-th food source location, and M represents the number of food source locations;

[0076] If the amount of nectar collected by the scout bees at the old food source location is greater than the first preset threshold, and / or the first fitness probability of the old food source location is greater than the second preset threshold, the old food source location is selected; if the amount of nectar collected by the scout bees at the new food source location is greater than the first preset threshold, and / or the second fitness probability of the new food source location is greater than the second preset threshold, the new food source location is selected.

[0077] In a second aspect, an embodiment of the present application provides a steam turbine rotor fault diagnosis device based on VMD kurtosis estimation and RVM, the device comprising:

[0078] An acquisition unit, used for acquiring an initial vibration signal of a steam turbine rotor, and dividing the initial vibration signal into a first training set and a first test set;

[0079] A first processing unit is used for extracting features of the initial vibration signal based on VMD kurtosis estimation and RVM classifier, and constructing a second training set and a second test set with feature vector matrices of the initial vibration signal as content;

[0080] A second processing unit is used to use the feature vector in the second training set as input data of the RVM classifier to train the RVM classifier;

[0081] An optimization unit, used for optimizing and selecting kernel parameters to be optimized of the trained RVM classifier by using an artificial bee colony algorithm;

[0082] The output unit is used to obtain the decision function of the optimized RVM classifier and determine the fault type of the turbine rotor according to the output value of the decision function to test the optimized RVM classifier; wherein the optimized RVM classifier is used to diagnose the fault of the turbine rotor and output the fault type of the turbine rotor.

[0083] In a third aspect, an embodiment of the present application provides an electronic device, comprising 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 method for diagnosing a turbine rotor fault based on VMD kurtosis estimation and RVM described in any one of the first aspects is implemented.

[0084] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements any one of the turbine rotor fault diagnosis methods based on VMD kurtosis estimation and RVM in the above-mentioned first aspect.

[0085] In a fifth aspect, an embodiment of the present application provides a computer program product. When the computer program product runs on an electronic device, the electronic device executes any one of the turbine rotor fault diagnosis methods based on VMD kurtosis estimation and RVM in the above-mentioned first aspect.

[0086] This application first extracts features of the initial vibration signal through VMD kurtosis estimation and RVM classifier, and constructs the feature vector matrix of the initial vibration signal as the second training set and the second test set of the RVM classifier. Secondly, the RVM classifier is trained based on the second training set, and then the artificial bee colony algorithm is used to optimize the kernel parameters to be optimized of the trained RVM classifier. Finally, based on the second test set, the fault type of the turbine rotor is determined according to the output value of the decision function to test the optimized RVM classifier. The optimized RVM classifier is used to determine the fault type of the turbine rotor.

[0087] The solution of the present application adopts an artificial bee colony algorithm to optimize and select the kernel parameters of the RVM classifier, which greatly improves the classification accuracy of the RVM classifier and improves the generalization ability of the RVM classifier. At the same time, the model structure is simple, the search ability is strong, and the convergence speed is fast, thereby improving the accuracy and recognition efficiency of turbine rotor fault diagnosis, and has strong ease of use and practicality.

[0088] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0090] Figure 1 It is a schematic diagram of an application scenario of a steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in an embodiment of the present application;

[0091] Figure 2 It is a schematic diagram of the steps of a steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in an embodiment of the present application;

[0092] Figure 3is a schematic diagram of a vibration waveform of a fault state signal provided in an embodiment of the present application;

[0093] Figure 4 It is a schematic diagram of the overall process of a steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in an embodiment of the present application;

[0094] Figure 5 It is a structural schematic diagram of a steam turbine rotor fault diagnosis device based on VMD kurtosis estimation and RVM provided in an embodiment of the present application;

[0095] Figure 6 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0096] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0097] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0098] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0099] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0100] As used in this specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0101] In addition, in the description of the present application, the terms "first", "second", "third", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0102] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0103] Steam turbine generator sets are key production equipment in thermal power plants. They are characterized by complex structures, harsh working conditions (such as high temperature, high pressure, and high speed), and high requirements for continuous operation, and are prone to failure. The rotor system is the core component of the steam turbine generator set. The status monitoring and fault diagnosis of the rotor system ensure the safe and stable operation of the steam turbine generator set. There have been many major accidents caused by steam turbine rotor failures at home and abroad, causing significant economic losses to the society and economy. Therefore, accurate and rapid fault diagnosis research on steam turbine rotors is of great practical significance.

[0104] Steam turbine rotor faults mainly include: rotor imbalance, rotor misalignment, rotor cracks, bearing looseness, bearing wear, oil film oscillation, oil film vortex, and compound faults caused by the above faults. Fault diagnosis is generally divided into two steps. The first step is fault feature extraction, and the second step is to identify various types of faults. When analyzing the common fault features of steam turbine rotor vibration, it is very important to timely identify the fault type through the fault signal.

[0105] In the prior art, time-frequency analysis methods are often used, such as Discrete Fourier Transform (DFT), Short Time Fourier Transform (STFT), Discrete Wavelet Transform (DWT), S-transform (ST), Empirical Mode Decomposition (EMD), Local Mean Decomposition (LMD) and other methods.

[0106] At present, there are few studies and patents on turbine rotor fault diagnosis methods. In recent years, the rapidly developed intelligent algorithms and optimization algorithms have been widely and successfully applied in this regard. They are fast, robust, and highly portable, and are suitable for industrial fault diagnosis.

[0107] With the development of intelligent diagnosis technology, many intelligent algorithms have been applied to the recognition and diagnosis of fault features, such as artificial neural network (ANN), support vector machine (SVM), and genetic algorithm (GA), particle swarm optimization (PSO) and annealing algorithm (SA) to optimize the parameters of ANN and SVM, but these algorithms have their own shortcomings. In recent years, the research on convolutional neural network (CNN) has become a new trend in the field of artificial intelligence, but the turbine rotor fault data set is a small sample set, and the use of convolutional neural network is limited.

[0108] However, the accuracy and recognition efficiency of these fault diagnosis methods are low. Therefore, how to improve the accuracy and recognition efficiency of steam turbine rotor fault diagnosis has become a problem to be solved.

[0109] In response to the above-mentioned defects, an embodiment of the present application provides a turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM. First, the initial vibration signal is feature extracted through VMD kurtosis estimation and RVM classifier, and the feature vector matrix of the initial vibration signal is constructed as the second training set and the second test set of the RVM classifier. Secondly, the RVM classifier is trained based on the second training set, and then the artificial bee colony algorithm is used to optimize the kernel parameters to be optimized of the trained RVM classifier. Finally, based on the second test set, the fault type of the turbine rotor is determined according to the output value of the decision function to test the optimized RVM classifier. The optimized RVM classifier is used to determine the fault type of the turbine rotor.

[0110] The scheme of the present application uses an artificial bee colony algorithm to optimize the kernel parameters of the RVM classifier, which greatly improves the classification accuracy of the RVM classifier and improves the generalization ability of the RVM classifier. At the same time, the model structure is simple, the search ability is strong, and the convergence speed is fast, thereby improving the accuracy and recognition efficiency of the turbine rotor fault diagnosis. After determining the fault type of the turbine rotor, the maintenance personnel can perform corresponding maintenance on the turbine rotor according to the specific fault type, so that the turbine generator set can operate safely and stably, and it has strong ease of use and practicality.

[0111] See also Figure 1 , Figure 1 It is a schematic diagram of the application scenario of the steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in an embodiment of the present application, that is, a schematic diagram of the steam turbine rotor structure.

[0112] In some embodiments, Figure 1 As shown, the probe of the fault diagnosis sensor should be installed at the horizontal and vertical positions of the bearing bracket of the turbine rotor as much as possible to facilitate the measurement of signals in the horizontal and vertical directions. The initial vibration signal is continuously collected through the data acquisition software in the sensor, and the initial vibration signal is saved in the form of a text file for subsequent processing of the initial vibration signal.

[0113] Based on the above application scenario, the embodiment of the present application provides a method for diagnosing turbine rotor faults based on VMD kurtosis estimation and RVM. The specific process of implementing the method is introduced below through a specific embodiment.

[0114] See also Figure 2 , Figure 2 Schematic diagram of the steps of the steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in the embodiment of the present application. Figure 2 As shown, the method may include the following steps:

[0115] S201, acquiring an initial vibration signal of a steam turbine rotor, and dividing the initial vibration signal into a first training set and a first test set.

[0116] In some embodiments, the initial vibration signal of the steam turbine rotor can be obtained from the distributed control system (DCS) of the thermal power plant, or can be obtained through data acquisition software in a laboratory simulation of a steam turbine rotor fault experiment. The initial vibration signal includes: a fault state signal and a normal state signal. The fault state signal is the vibration signal of the steam turbine rotor in a fault state, mainly including: rotor imbalance, rotor misalignment, rotor cracks, bearing looseness, bearing wear, oil film oscillation, and oil film vortex. The vibration waveform diagram of the fault state signal is shown in FIG. Figure 3As shown in the figure, the vibration waveforms of these fault state signals are not stable and fluctuate greatly. The normal state signal is the vibration signal of the turbine rotor in the normal state, and the vibration waveform is relatively stable.

[0117] In one embodiment, the fault state signal and the normal state signal can be combined by unified normalization processing, and randomly divided into a training set and a test set according to a preset ratio. The preset ratio can be determined according to the specific situation in the actual application scenario, and is not specifically limited here.

[0118] According to an embodiment of the present application, after obtaining the initial vibration signal of the steam turbine rotor, the method may further include the following steps:

[0119] The initial vibration signal is denoised using a wavelet block-thresholding (WBT) algorithm or a block threshold estimation algorithm.

[0120] In some embodiments, the discrete noise model of the initial vibration signal is: x(i) = f(i) + g(i). Where x(i) represents the initial vibration signal, f(i) represents the useful signal, g(i) represents the noise signal, i is a natural number, i = 0, 1, 2, ..., n-1, and n represents the number of sampling points. In order to simplify the wavelet block threshold denoising, the number of sampling points is selected as n = 2 J , J∈Z, Z is a positive integer.

[0121] In some embodiments, the steps of wavelet block threshold denoising are: (1) obtaining wavelet coefficients w by performing discrete wavelet transform (DWT) on the observed signal x; (2) processing the wavelet coefficients w using a threshold function to obtain new wavelet coefficients: Among them, η λ (·) represents the useful signal, λ represents the threshold; (3) Perform inverse wavelet transform on the wavelet coefficients to obtain the estimation function of the real signal f

[0122] Among them, commonly used threshold functions include hard threshold function and soft threshold function, but these two threshold functions cannot find the optimal denoising threshold. Therefore, the embodiment of the present application adopts the block threshold estimation method to denoise the initial vibration signal, and its denoising goal is to calculate the estimated value of the useful signal f Make the mean square error as small as possible. The expression of mean square error is:

[0123] In this embodiment, a block threshold estimation algorithm is used to pre-process the initial vibration signal, thereby reducing external interference such as noise, thereby improving the authenticity of feature extraction.

[0124] According to an implementation of the present application, the denoising process of the initial vibration signal using the block threshold estimation algorithm may include the following steps:

[0125] The initial vibration signal is processed by wavelet transform to obtain the wavelet coefficients of each scale. The expression of the wavelet coefficients is as follows:

[0126]

[0127] Among them, c j0l represents the scale factor, d jm represents the wavelet coefficient, j represents the scale (i.e., the wavelet decomposition layer), l represents the ordinal number of the scale coefficient, m represents the ordinal number of the wavelet coefficient, j0 represents the maximum scale of the scale coefficient, and J represents the maximum scale of the wavelet coefficient. j0 and J may be equal or unequal.

[0128] In each scale j, the wavelet coefficients are divided into several adjacent non-overlapping blocks. (a data block consisting of wavelet coefficients), block The length of is L0=floor[(logn) / 2]; where i represents the ordinal number of the block, floor represents the floor rounding function, taking the maximum integer not greater than the input value, and n represents the number of sampling points of the initial vibration signal;

[0129] The block is extended to a length of L1 = max(1,[1,L0 / 2]) Extending to the left and right, forming large blocks that intersect each other Big Block The length of is L=L0+2L1;

[0130] In large blocks Calculate the shrinkage factor. The shrinkage factor is expressed as follows:

[0131]

[0132] in, represents the shrinkage factor, λ represents the threshold, and λ * =4.50524, ε represents the standard deviation of the noise in the initial vibration signal, ε 2 represents the square of the standard deviation of the noise, ε>0, k represents an ordinal number;

[0133] Based on the shrinkage factor The wavelet coefficients within are thresholded, and the specific expression is as follows:

[0134]

[0135] The threshold processing is as follows: if the current noise value is greater than the threshold, the noise is deleted, that is, the amplitude of the noise is set to 0; if the current noise value is less than or equal to the threshold, the noise is retained.

[0136] After threshold processing, the data After performing inverse wavelet transform, the expression of the denoised vibration signal is as follows:

[0137]

[0138] in, represents the vibration signal after denoising, t represents time, φ j0k (t) represents the scaling function, ψ jk (t) represents the wavelet function.

[0139] S202, based on VMD kurtosis estimation and RVM classifier, feature extraction is performed on the initial vibration signal, and a second training set and a second test set are constructed with feature vector matrices of the initial vibration signal as content.

[0140] In some embodiments, feature extraction can be performed on the initial vibration signal in the first training set based on VMD (Variational Mode Decomposition) kurtosis estimation and relevance vector machine (RVM) classifier, and then a second training set with the eigenvector matrix of the initial vibration signal as the content is constructed as input data for the RVM classifier; feature extraction can also be performed on the initial vibration signal in the first test set based on VMD kurtosis estimation and RVM classifier, and then a second test set with the eigenvector matrix of the initial vibration signal as the content is constructed as input data for the RVM classifier.

[0141] According to an embodiment of the present application, based on VMD kurtosis estimation and RVM classifier, feature extraction of the initial vibration signal may include the following steps:

[0142] The denoised vibration signal is decomposed into several finite bandwidth modal function components (i.e., modal components or VMD components) by using the variational modal decomposition method. The kurtosis value of each finite bandwidth modal function component is estimated according to the rotor fault mechanism. The kurtosis value corresponding to each finite bandwidth modal function component is standardized.

[0143] It should be noted that there are two constraints on variational mode decomposition: (1) the sum of the bandwidths of the center frequencies of each modal component must be minimized, and (2) the sum of all modal components must be equal to the denoised vibration signal. The specific calculation formula is as follows:

[0144]

[0145] Among them, u k represents the finite bandwidth modal function component, k represents the ordinal number, ω k represents the center frequency corresponding to the finite bandwidth modal function component, δ(t) represents the pulse function, t represents time, j represents the scale, and u k (t) represents the finite bandwidth mode function, u k (t) = A k (t)cos(φ k (t)), A k (t) represents u k (t) envelope amplitude, φ k (t) represents u k (t), st represents the constraint condition, and f(t) represents the vibration signal after denoising.

[0146] According to an embodiment of the present application, the denoised vibration signal is decomposed into a plurality of finite bandwidth modal function components using a variational modal decomposition method, which may include the following steps:

[0147] Initialize each finite bandwidth modal function component, its corresponding center frequency, and Lagrange operator (i.e., Lagrange multiplier, Lagrange multiplier), and set the target number of iterations. The specific expression is as follows:

[0148]

[0149] Among them, u k represents the finite bandwidth modal function component, k represents the ordinal number, ω k represents the center frequency corresponding to the bandwidth-limited modal function component, λ L represents the Lagrange operator, b represents the target number of iterations, and each iteration increases by 1 based on the number of previous iterations.

[0150] By performing Hilbert transform on each finite bandwidth modal function component, the expression of the analytical signal of each finite bandwidth modal function component is obtained as follows:

[0151]

[0152] Among them, δ(t) represents the pulse function, t represents time, j represents scale, and u k (t) represents the finite bandwidth mode function, u k (t) = A k (t)cos(φ k (t)), A k (t) represents u k (t) envelope amplitude, φ k(t) represents u k The instantaneous phase of (t).

[0153] Exponential correction is performed on each center frequency to transfer the spectrum of each limited bandwidth modal function component to the corresponding baseband. The specific expression is:

[0154] Calculate the square of the gradient of the analytical signal L 2 The norm is used to estimate the bandwidth of the center frequency of each finite bandwidth modal function component, and the Lagrange operator and quadratic penalty factor are introduced to transform the inequality constraint into an equality constraint. The corresponding augmented Lagrange function expression is as follows:

[0155]

[0156]

[0157] Where L represents the augmented Lagrange function, α represents the quadratic penalty factor, represents the partial derivative of the finite bandwidth modal function with respect to time t, and f(t) represents the denoised vibration signal.

[0158] The alternating direction method of multipliers (ADMM) is used to iteratively optimize the sequence, and the expressions of the finite bandwidth modal function components and center frequency are obtained as follows:

[0159]

[0160] Among them, u k ^n+1 (ω) represents the Wiener filtering of the finite bandwidth modal function component, for u k ^n+1 (ω) Perform inverse Fourier transform to get the real part u k (t),ω k n+1 (ω) represents the center frequency corresponding to the bandwidth-limited modal function component.

[0161] Repeat the steps after initialization until the termination condition is met, and finally obtain the K finite bandwidth modal function components with the smallest sum of bandwidths as u k ={u1,u2,…,u k}, and the center frequency corresponding to each finite bandwidth modal function component is ω k ={ω1,ω2,…,ω k}. The expression of the termination condition is σ represents the discrimination accuracy, and σ>0.

[0162] According to an implementation of the present application, estimating the kurtosis value of each finite bandwidth modal function component according to the rotor fault mechanism may include the following steps:

[0163] The kurtosis value is calculated according to the finite bandwidth modal function components and their average values, and the kurtosis value vector matrix T of the finite bandwidth modal function components is obtained. P}. Among them, K1, K2, …, K P Respectively represent the kurtosis value of each finite bandwidth modal function component, P = 1, 2…, l, l represents the total number of extracted kurtosis values. Kurtosis can measure the flatness of data distribution, that is, the statistic of the steepness of the data value distribution. The calculation formula of the kurtosis value of the finite bandwidth modal function component is as follows:

[0164]

[0165] Where K is the kurtosis (dimensionless), x is the finite bandwidth modal function component, represents the average value of the finite bandwidth modal function component, i represents the ordinal number, and n represents the number of samples.

[0166] In one embodiment, the standardization process may be a normalization process. The kurtosis values ​​corresponding to each finite bandwidth modal function component are standardized, including: normalizing each kurtosis value to obtain the expression of the eigenvector matrix of the denoised vibration signal as follows:

[0167]

[0168] Wherein, T′ represents the eigenvector matrix of the denoised vibration signal (i.e., the kurtosis value vector matrix of each VMD component of the denoised vibration signal),

[0169] S203, using the feature vector in the second training set as input data of the RVM classifier to train the RVM classifier.

[0170] In one embodiment, the feature vector of the initial vibration signal in the second training set can be used as input data of the RVM classifier, and the RVM classifier outputs the fault type of the turbine rotor. If the fault type output by the RVM classifier is accurate, it means that the accuracy of the RVM classifier is high; if the fault type output by the RVM classifier is inaccurate, it means that the accuracy of the RVM classifier is low and further training is required to improve the accuracy.

[0171] S204, using an artificial bee colony algorithm to optimize and select kernel parameters to be optimized of the trained RVM classifier.

[0172] It should be noted that RVM is an intelligent learning technology based on a sparse Bayesian framework, which is more convenient than SVM because there is no need to set penalty parameters in RVM. In addition, the number of support vectors in RVM is much smaller than that in SVM, making RVM have a sparser representation than SVM. Therefore, RVM has better generalization ability than SVM.

[0173] The regression function of RVM consists of a linear combination of weighted kernel functions. The specific expression is as follows:

[0174]

[0175] Where x=[x1,x2,…,x N ] is the feature set input vector (i.e., the feature vector of the initial vibration signal in the second training set), N represents the number of samples, w=[w1,w2,…,w N ] is the weight vector of the RVM classifier, w0 represents the bias, K(x,x i ) represents the kernel function, and y represents the recognition result output by the RVM classifier (in the embodiment of the present application, it is the fault type of the turbine rotor).

[0176] The Gaussian RBF kernel (Radial Basis Function, RBF kernel) has been widely used in RVM classifiers and can be expressed as follows:

[0177]

[0178] Among them, K RBF (x i ,x j ) represents the Gaussian RBF kernel function, x i and x j Represents the feature set input vector, exp represents the natural exponential function, and γ represents the kernel parameter of the Gaussian RBF kernel function.

[0179] Since the selection of kernel parameters has a certain influence on the prediction results of the RVM classifier, an artificial bee colony algorithm (ABC) or other algorithms can be used to select appropriate kernel parameters. The kernel parameters are optimized by updating the location of the food source for the bees, and the optimal kernel parameters can be determined by finding the best food source location. According to one embodiment of the present application, the kernel parameters to be optimized of the trained RVM classifier are optimized using an artificial bee colony algorithm, which can include the following steps:

[0180] Based on the first fitness value of the old food source position and the second fitness value of the new food source position by at least one scout bee, the first target food source position is determined, and the first target food source position is used as the current optimal food source position. The expression of the first target food source position includes the kernel parameter to be optimized. When the scout bee completes the search for the food source position and determines the first target food source position, the scout bee shares the information of the amount of nectar collected and the first target food source position with at least one non-scout bee (other bees) in the preset search area. There can be one or more scout bees and non-scout bees, and the information of the first target food source position can be the coordinates of the first target food source position.

[0181] The method selects a food source location based on the fitness probability of the non-scout bees according to the amount of nectar collected by the scout bees and / or the food source location, and searches for a new food source location within a preset range of the selected food source location based on the non-scout bees; determines whether to replace the selected food source location based on the first fitness value of the selected food source location and the second fitness value of the new food source location, so as to determine a second target food source location, and uses the second target food source location as the current optimal food source location.

[0182] Among them, the preset range can be determined according to the specific circumstances in the actual application scenario and is not specifically limited here.

[0183] If a food source position in a replacement state (i.e., an abandoned old food source position) exists, the randomly generated old food source position is used to replace the food source position in a replacement state based on the scout bees; wherein, the randomly generated old food source position is used to scout the food source position based on the scout bees. It should be noted that this step is to prevent the occurrence of a local optimal situation. New food source positions cannot be selected all the time, and some old food source positions should also be appropriately selected for continued scout.

[0184] Repeat the steps of determining the first target food source position and its subsequent steps until the maximum number of search cycles is reached, and take the best food source position obtained when the maximum number of search cycles is reached as the final best food source position, and the expression of the final best food source position contains the optimal kernel parameters.

[0185] Among them, the maximum number of cycles can be determined according to the specific circumstances in the actual application scenario and is not specifically limited here.

[0186] In this embodiment, an artificial bee colony algorithm is used to optimize and select the kernel parameters of the RVM classifier, which greatly improves the classification accuracy of the RVM classifier and improves the generalization ability of the RVM classifier.

[0187] According to an embodiment of the present application, before determining the location of the first target food source, the method may further include the following steps:

[0188] Randomly initialize M food source locations in the preset search area, and generate the old food source location according to the following calculation formula:

[0189] x i,j =x min,j +r1×(x max,j -x min,j )

[0190] Among them, x i,j represents the location of the old food source, x max,j represents the upper bound of the jth kernel parameter to be optimized, x min,j represents the lower bound of the jth kernel parameter to be optimized, i represents the ordinal number; j = 1, 2…, D, D represents the kernel parameter to be optimized, in one embodiment, D = 1, there is only one parameter to be optimized, and accordingly, there is only one food source position to be updated; r1 is a random number in the range of [0, 1]. The range of the preset search area can be determined according to the specific situation in the actual application scenario, and is not specifically limited here.

[0191] According to the objective function of the artificial bee colony algorithm, the first fitness value of each old food source position is calculated. The specific calculation formula is as follows:

[0192]

[0193] Among them, Fitness i represents the first fitness value of the i-th old food source position. The expression of the objective function of the artificial bee colony algorithm is y q represents the actual output value of the RVM classifier, represents the validation value, and v represents the number of training samples in the second training set.

[0194] For the scout bees within the preset range of the current old food source location, the new food source location is generated according to the following calculation formula:

[0195]

[0196] in, represents the new food source location, l represents the ordinal number, l≠i, and r2 is a random number in the range of [-1, 1]. The preset range of the current old food source location can be determined according to the specific situation in the actual application scenario, and is not specifically limited here.

[0197] The second fitness value of each new food source location is calculated according to the objective function of the artificial bee colony algorithm. The specific calculation formula is as follows:

[0198]

[0199] in, Represents the second fitness value of the i-th new food source location.

[0200] According to an embodiment of the present application, determining the first target food source position according to the first fitness value of the old food source position and the second fitness value of the new food source position may include the following steps:

[0201] If the second fitness value of the new food source position is greater than the first fitness value of the old food source position, the new food source position is used to replace the old food source position, and the new food source position is used as the first target food source position; otherwise, the old food source position is retained and the old food source position is used as the first target food source position.

[0202] In one embodiment, if the second fitness value of the new food source position is less than or equal to the first fitness value of the old food source position, the old food source position may be retained and used as the first target food source position.

[0203] According to one embodiment of the present application, selecting the location of a food source based on the fitness probability of non-scout bees according to the amount of nectar collected by scout bees and / or the location of the food source may include the following steps:

[0204] Based on the non-scout bees, it is determined whether the amount of nectar collected by the scout bees is greater than the first preset threshold, and the fitness probability of the food source location is calculated according to the following calculation formula to determine whether the fitness probability is greater than the second preset threshold:

[0205]

[0206] Among them, P i Represents the fitness probability of the i-th food source location, Fitness i represents the fitness value of the i-th food source location, and M represents the number of food source locations.

[0207] If the amount of nectar collected by the scout bees at the old food source location is greater than the first preset threshold, and / or the first fitness probability of the old food source location is greater than the second preset threshold, the old food source location is selected; if the amount of nectar collected by the scout bees at the new food source location is greater than the first preset threshold, and / or the second fitness probability of the new food source location is greater than the second preset threshold, the new food source location is selected.

[0208] In one embodiment, if the amount of nectar collected by the scout bees at the old food source location is greater than a first preset threshold, then the non-scout bees can select the old food source location. In another embodiment, if the first fitness probability of the old food source location is greater than a second preset threshold, then the non-scout bees can select the old food source location. In another embodiment, if the amount of nectar collected by the scout bees at the old food source location is greater than the first preset threshold, and the first fitness probability of the old food source location is greater than the second preset threshold, then the non-scout bees can select the old food source location.

[0209] In one embodiment, if the amount of nectar collected by the scout bees at the new food source location is greater than a first preset threshold, then the non-scout bees can select the new food source location. In another embodiment, if the second fitness probability of the new food source location is greater than a second preset threshold, then the non-scout bees can select the new food source location. In another embodiment, if the amount of nectar collected by the scout bees at the new food source location is greater than the first preset threshold, and the second fitness probability of the new food source location is greater than the second preset threshold, then the non-scout bees can select the new food source location.

[0210] Among them, the first preset threshold and the second preset threshold can be determined according to the specific situation in the actual application scenario, and are not specifically limited here.

[0211] It should be noted that in order to gradually iterate and find the optimal food source location, non-scout bees must search for a new food source location regardless of whether they choose the new food source location or the old food source location determined by the scout bees, and then determine the current optimal food source location based on the first fitness value of the selected food source location and the second fitness value of the new food source location.

[0212] According to an embodiment of the present application, judging whether to replace the selected food source position according to the first fitness value of the selected food source position and the second fitness value of the new food source position to determine the second target food source position may include the following steps:

[0213] If the second fitness value of the new food source position is greater than the first fitness value of the selected food source position, the new food source position is used to replace the selected food source position, and the new food source position is used as the first target food source position; if the second fitness value of the new food source position is less than or equal to the first fitness value of the selected food source position, the selected food source position is retained and the selected food source position is used as the second target food source position.

[0214] In one embodiment, when non-scout bees select a new food source location, if the second fitness value of the new food source location found is greater than the first fitness value of the selected new food source location, then the new food source location found is used to replace the selected new food source location, and the new food source location found is used as the second target food source location; if the second fitness value of the new food source location found is less than or equal to the first fitness value of the selected new food source location, then the selected new food source location is retained and the selected new food source location is used as the second target food source location.

[0215] In another embodiment, when non-scout bees select the old food source location, if the second fitness value of the new food source location found is greater than the first fitness value of the selected old food source location, then the new food source location found is used to replace the selected old food source location, and the new food source location found is used as the second target food source location; if the second fitness value of the new food source location found is less than or equal to the first fitness value of the selected old food source location, then the selected old food source location is retained and the selected old food source location is used as the second target food source location.

[0216] S205, using the feature vector in the second test set as the input data of the optimized RVM classifier, obtaining the decision function of the optimized RVM classifier, and determining the fault type of the turbine rotor according to the output value of the decision function to test the optimized RVM classifier; wherein the optimized RVM classifier is used to diagnose faults of the turbine rotor and output the fault type of the turbine rotor.

[0217] In one embodiment, the feature vector in the second test set is input into the optimized RVM classifier, and the optimized RVM classifier can be tested to verify the accuracy of the optimized RVM classifier. If the fault type output by the optimized RVM classifier is consistent with the actual fault type, it proves that the accuracy of the optimized RVM classifier is high; if the fault type output by the optimized RVM classifier is inconsistent with the actual fault type, it proves that the accuracy of the optimized RVM classifier is low and further training and optimization are required.

[0218] In one embodiment, after testing the optimized RVM classifier, the optimized RVM classifier can be used to diagnose the fault of the steam turbine rotor, and then output the fault type of the steam turbine rotor. The fault type of the steam turbine rotor output by the RVM classifier with higher accuracy is more accurate, and the fault type of the steam turbine rotor output by the RVM classifier with lower accuracy is inaccurate.

[0219] In one embodiment, the fault types of the steam turbine rotor include: rotor imbalance, rotor misalignment, rotor cracks, bearing looseness, bearing wear, oil film oscillation, oil film vortex, etc. Different output values ​​of the decision function correspond to different fault types, and the fault type or fault types of the steam turbine rotor can be determined according to the specific output value of the decision function. After determining the fault type of the steam turbine rotor, the maintenance personnel can perform corresponding maintenance on the steam turbine rotor according to the specific fault type, so that the steam turbine generator set can operate safely and stably.

[0220] In one embodiment, the output format of the decision function is abcdefgh, where a, b, c, d, e, f, g, and h represent the eight states of normal, rotor imbalance, rotor misalignment, rotor crack, bearing looseness, bearing wear, oil film oscillation, and oil film vortex, respectively, and the values ​​are 0 or 1 (0 means non-existence, 1 means existence). When the output value of the decision function is 01000000, it can be determined that the turbine rotor has a rotor imbalance fault; when the output value of the decision function is 00100000, it can be determined that the turbine rotor has a rotor misalignment fault. Other situations are similar and will not be repeated here.

[0221] See also Figure 4 , Figure 4 Schematic diagram of the overall process of the steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in the embodiment of the present application. Figure 4 As shown, the method may include the following steps:

[0222] S401, data set acquisition.

[0223] In some embodiments, a data acquisition system in the fault diagnosis sensor acquires an initial vibration signal of the steam turbine rotor, and then divides the initial vibration signal into a first training set and a first test set.

[0224] S402, feature extraction.

[0225] In some embodiments, the fault diagnosis sensor extracts features from the initial vibration signal. The fault diagnosis sensor first performs WBT denoising on the initial vibration signals in the first training set and the first test set, and then performs VMD decomposition on the denoised vibration signals, decomposing the denoised vibration signals into a number of finite bandwidth modal function components (i.e., VMD components), and then performs kurtosis estimation on each VMD component, and standardizes the kurtosis values ​​corresponding to each VMD component, and finally constructs a second training set and a second test set with the eigenvector matrix of the denoised vibration signal as the content, and uses the eigenvectors in the second training set as input data of the RVM classifier to train the RVM classifier.

[0226] S403, fault identification.

[0227] In some embodiments, the fault diagnosis sensor performs fault identification on the turbine rotor based on the association vector machine. The artificial bee colony algorithm (ABC) is used to optimize and select the kernel parameters to be optimized of the trained association vector machine, and then the feature vectors in the second test set are used as input data of the optimized association vector machine, and the fault diagnosis result, i.e., the fault type of the turbine rotor, is output based on the optimized association vector machine to test the optimized association vector machine, and the optimized association vector machine is used to perform fault diagnosis on the turbine rotor and output the fault type of the turbine rotor.

[0228] It should be noted that the specific implementation principles of steps S401 to S403 have been described in the above embodiments and will not be repeated here.

[0229] The turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM provided in the embodiment of the present application first performs feature extraction on the initial vibration signal through VMD kurtosis estimation and RVM classifier, and constructs the feature vector matrix of the initial vibration signal as the second training set and the second test set of the RVM classifier, and then trains the RVM classifier based on the second training set, and then uses the artificial bee colony algorithm to optimize the kernel parameters to be optimized of the trained RVM classifier, and finally determines the fault type of the turbine rotor according to the output value of the decision function based on the second test set to test the optimized RVM classifier, and the optimized RVM classifier is used to determine the fault type of the turbine rotor.

[0230] The scheme of the present application adopts an artificial bee colony algorithm to optimize and select the kernel parameters of the RVM classifier, which greatly improves the classification accuracy of the RVM classifier and improves the generalization ability of the RVM classifier. At the same time, the model structure is simple, the search ability is strong, and the convergence speed is fast, thereby improving the accuracy and recognition efficiency of the turbine rotor fault diagnosis. After determining the fault type of the turbine rotor, the maintenance personnel can perform corresponding maintenance on the turbine rotor according to the specific fault type, so that the steam turbine generator set can operate safely and stably. In addition, the block threshold estimation algorithm is used to pre-process the initial vibration signal, which reduces external interference such as noise, thereby improving the authenticity of feature extraction, and has strong ease of use and practicality.

[0231] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0232] Corresponding to the method of the above embodiment, Figure 5It is a structural schematic diagram of a steam turbine rotor fault diagnosis device based on VMD kurtosis estimation and RVM provided in an embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0233] Reference Figure 5 , the device comprises:

[0234] An acquisition unit 501 is used to acquire an initial vibration signal of a steam turbine rotor and divide the initial vibration signal into a first training set and a first test set;

[0235] The first processing unit 502 is used to extract features from the initial vibration signal based on VMD kurtosis estimation and RVM classifier, and construct a second training set and a second test set with feature vector matrices of the initial vibration signal as content;

[0236] The second processing unit 503 is used to use the feature vector in the second training set as input data of the RVM classifier to train the RVM classifier;

[0237] An optimization unit 504 is used to optimize the kernel parameters to be optimized of the trained RVM classifier by using an artificial bee colony algorithm;

[0238] The output unit 505 is used to obtain the decision function of the optimized RVM classifier and determine the fault type of the turbine rotor according to the output value of the decision function to test the optimized RVM classifier; wherein the optimized RVM classifier is used to perform fault diagnosis on the turbine rotor and output the fault type of the turbine rotor.

[0239] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0240] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0241] Figure 6 This is a schematic diagram of the structure of the electronic device 6 provided in the embodiment of the present application. Figure 6 As shown, the electronic device 6 of this embodiment includes: at least one processor 601 ( Figure 6 Only one is shown in the figure), a memory 603, and a computer program 602 stored in the memory 603 and executable on at least one processor 601, wherein the processor 601 implements the steps in the above method embodiment when executing the computer program 602.

[0242] The electronic device 6 may be a sensor such as a mechanical fault diagnosis sensor of a thermal power plant. The electronic device 6 may include, but is not limited to, a processor 601 and a memory 603. Those skilled in the art will appreciate that Figure 6 It is only an example of the electronic device 6 and does not constitute a limitation on the electronic device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0243] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0244] In some embodiments, the memory 603 may be an internal storage unit of the electronic device 6, such as a hard disk or memory of the electronic device 6. In other embodiments, the memory 603 may also be an external storage device of the electronic device 6, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital card (SecureDigital, SD), a flash card (Flash Card), etc. equipped on the electronic device 6. Further, the memory 603 may also include both an internal storage unit of the electronic device 6 and an external storage device. The memory 603 is used to store an operating system, an application program, a boot loader (Boot Loader), data, and other programs, such as program codes of a computer program. The memory 603 may also be used to temporarily store data that has been output or is to be output.

[0245] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, when the present application implements all or part of the processes in the above-mentioned embodiment method, it can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps applied to the method embodiment can be implemented. Among them, the computer program includes a computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may at least include: any entity or device that can carry the computer program code to the computing device / electronic device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electric carrier signal, a telecommunication signal, and a software distribution medium, such as a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, a computer-readable storage medium cannot be an electric carrier signal and a telecommunication signal.

[0246] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0247] An embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device executes the steps in the above-mentioned method embodiments.

[0248] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0249] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0250] In the embodiments provided in the present application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. The device / electronic device embodiments described above are merely schematic, and the division of the above modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, and some features can be ignored and not executed. Another point is that the indirect coupling, direct coupling or communication connection between each other shown or discussed can be an indirect coupling, direct coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0251] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0252] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM, characterized in that: The method comprises: Acquire an initial vibration signal of a steam turbine rotor, and divide the initial vibration signal into a first training set and a first test set; Based on VMD kurtosis estimation and RVM classifier, feature extraction is performed on the initial vibration signal, and a second training set and a second test set are constructed with feature vector matrices of the initial vibration signal as content; Using the feature vector in the second training set as input data of the RVM classifier to train the RVM classifier; The artificial bee colony algorithm is used to optimize the kernel parameters to be optimized of the trained RVM classifier; The feature vector in the second test set is used as the input data of the optimized RVM classifier, the decision function of the optimized RVM classifier is obtained, and the fault type of the turbine rotor is determined according to the output value of the decision function to test the optimized RVM classifier; wherein the optimized RVM classifier is used to diagnose faults of the turbine rotor and output the fault type of the turbine rotor.

2. The steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to claim 1 is characterized in that: After obtaining the initial vibration signal of the steam turbine rotor, the method further includes: The initial vibration signal is denoised using a wavelet region threshold denoising algorithm or a block threshold estimation algorithm.

3. The steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to claim 2 is characterized in that: The initial vibration signal is subjected to denoising processing using a block threshold estimation algorithm, including: The initial vibration signal is processed by wavelet transform to obtain wavelet coefficients of various scales. The expression of wavelet coefficients is as follows: Among them, c j0l represents the scale factor, d jm represents the wavelet coefficient, j represents the scale, l represents the ordinal number of the scale coefficient, m represents the ordinal number of the wavelet coefficient, j0 represents the maximum scale of the scale coefficient, and J represents the maximum scale of the wavelet coefficient; In each scale j, the wavelet coefficients are divided into several adjacent non-overlapping blocks. Block The length of is L0=floor[(logn) / 2]; wherein i represents the ordinal number of the block, floor represents the floor rounding function, and n represents the number of sampling points of the initial vibration signal; The block is extended to a length of L1 = max(1,[1,L0 / 2]) Extending to the left and right, forming large blocks that intersect each other Big Block The length of is L=L0+2L1; In large blocks Calculate the shrinkage factor. The shrinkage factor is expressed as follows: in, represents the shrinkage factor, λ represents the threshold, and λ * =4.50524, ε represents the standard deviation of the noise in the initial vibration signal, ε 2 represents the square of the standard deviation of the noise, ε>0, k represents an ordinal number; Based on the shrinkage factor The wavelet coefficients within are thresholded, and the specific expression is as follows: After threshold processing, the data After performing inverse wavelet transform, the expression of the denoised vibration signal is as follows: in, represents the vibration signal after denoising, t represents time, φ j0k (t) represents the scaling function, ψ jk (t) represents the wavelet function.

4. The steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to claim 3 is characterized in that: Based on VMD kurtosis estimation and RVM classifier, feature extraction is performed on the initial vibration signal, including: Decomposing the denoised vibration signal into a plurality of finite bandwidth modal function components by using a variational modal decomposition method; The kurtosis value of each finite bandwidth modal function component is estimated according to the rotor fault mechanism; The kurtosis values ​​corresponding to each finite bandwidth modal function component are normalized.

5. The steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to claim 4 is characterized in that: The denoised vibration signal is decomposed into a number of finite bandwidth modal function components using a variational modal decomposition method, including: Initialize each finite bandwidth modal function component, its corresponding center frequency and Lagrange operator, and set the target number of iterations. The specific expression is as follows: Among them, u k represents the finite bandwidth modal function component, k represents the ordinal number, ω k represents the center frequency corresponding to the bandwidth-limited modal function component, λ L represents the Lagrange operator, b represents the target number of iterations, and each iteration increases by 1 based on the number of previous iterations; By performing Hilbert transform on each finite bandwidth modal function component, the expression of the analytical signal of each finite bandwidth modal function component is obtained as follows: Among them, δ(t) represents the pulse function, t represents time, j represents scale, and u k (t) represents the finite bandwidth mode function, u k (t) = A k (t)cos(φ k (t)), A k (t) represents u k (t) envelope amplitude, φ k (t) represents u k The instantaneous phase of (t); Exponential correction is performed on each center frequency to transfer the spectrum of each limited bandwidth modal function component to the corresponding baseband. The specific expression is: The square of the gradient of the analytical signal is calculated as 2 The norm is used to estimate the bandwidth of the center frequency of each finite bandwidth modal function component, and the Lagrange operator and quadratic penalty factor are introduced to transform the inequality constraint into an equality constraint. The corresponding augmented Lagrange function expression is as follows: Where L represents the augmented Lagrange function, α represents the quadratic penalty factor, represents the partial derivative of the finite bandwidth modal function with respect to time t, and f(t) represents the denoised vibration signal; The alternating direction algorithm is used to iteratively optimize the sequence, and the expressions of the finite bandwidth modal function components and center frequency are obtained as follows: Among them, u k ^n+1 (ω) represents the Wiener filtering of the finite bandwidth modal function component, for u k ^n+1 (ω) Perform inverse Fourier transform to get the real part u k (t),ω k n+1 (ω) represents the center frequency corresponding to the bandwidth-limited modal function component; Repeat the steps after initialization until the termination condition is met, and finally obtain the K finite bandwidth modal function components with the smallest sum of bandwidths as u k ={u1,u2,…,u k }, and the center frequency corresponding to each finite bandwidth modal function component is ω k ={ω1,ω2,…,ω k }; The expression of the termination condition is: σ represents the discrimination accuracy, and σ>

0.

6. The method for diagnosing steam turbine rotor faults based on VMD kurtosis estimation and RVM according to claim 4, characterized in that: The kurtosis value of each finite bandwidth modal function component is estimated according to the rotor fault mechanism, including: The kurtosis value is calculated according to the finite bandwidth modal function component and its average value, and the kurtosis value vector matrix T of the finite bandwidth modal function component is obtained. P }, where K1, K2, …, K P Respectively represent the kurtosis value of each finite bandwidth modal function component, P = 1, 2…, l, l represents the total number of extracted kurtosis values; the calculation formula of the kurtosis value of the finite bandwidth modal function component is as follows: Where K represents the kurtosis, x represents the finite bandwidth modal function component, represents the average value of the finite bandwidth modal function component, i represents the ordinal number, and n represents the number of samples; The kurtosis values ​​corresponding to each finite bandwidth modal function component are standardized, including: Each kurtosis value is normalized to obtain the expression of the eigenvector matrix of the denoised vibration signal as follows: Where T′ represents the eigenvector matrix of the vibration signal after denoising, 7. The steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to claim 1, characterized in that: The artificial bee colony algorithm is used to optimize and select the kernel parameters to be optimized of the trained RVM classifier, including: Determine a first target food source position based on a first fitness value of the old food source position and a second fitness value of the new food source position by at least one scout bee, and use the first target food source position as the current optimal food source position; wherein the expression of the first target food source position includes the kernel parameter to be optimized; Based on the scout bees completing the search for the food source location and determining the first target food source location, sharing the information of the amount of nectar collected and the first target food source location with at least one non-scout bee in a preset search area; selecting a food source location based on the non-scout bees’ fitness probability based on the amount of nectar collected by the scout bees and / or the location of the food source, and searching for a new food source location based on the non-scout bees’ finding a new food source location within a preset range of the selected food source location; Determine whether to replace the selected food source position according to the first fitness value of the selected food source position and the second fitness value of the new food source position to determine the second target food source position, and use the second target food source position as the current optimal food source position; If the food source position in the replacement state exists, the randomly generated old food source position is used to replace the food source position in the replacement state based on the scout bees; wherein the randomly generated old food source position is used to scout the food source position based on the scout bees; Repeat the step of determining the position of the first target food source and subsequent steps until the maximum number of search cycles is reached, and use the best food source position obtained when the maximum number of search cycles is reached as the final best food source position; Among them, the expression of the final optimal food source position contains the optimal kernel parameters.

8. The method for diagnosing steam turbine rotor faults based on VMD kurtosis estimation and RVM according to claim 7, characterized in that: Before determining the location of the first target food source, the method further includes: Randomly initialize M food source locations in the preset search area, and generate the old food source location according to the following calculation formula: x i,j =x min,j +r1×(x max,j -x min,j ) Among them, x i,j represents the location of the old food source, x max,j represents the upper bound of the jth kernel parameter to be optimized, x min,j represents the lower bound of the jth kernel parameter to be optimized, i represents the ordinal number, j = 1, 2…, D, D represents the kernel parameter to be optimized, and r1 is a random number in the range of [0, 1]; According to the objective function of the artificial bee colony algorithm, the first fitness value of each old food source position is calculated. The specific calculation formula is as follows: Among them, Fitness i represents the first fitness value of the i-th old food source position. The expression of the objective function of the artificial bee colony algorithm is y q represents the actual output value of the RVM classifier, represents the validation value, v represents the number of training samples in the second training set; For the scout bees within the preset range of the current old food source location, the new food source location is generated according to the following calculation formula: in, represents the location of the new food source, l represents the ordinal number, l≠i, and r2 is a random number in the range of [-1, 1]; The second fitness value of each new food source location is calculated according to the objective function of the artificial bee colony algorithm. The specific calculation formula is as follows: in, Represents the second fitness value of the i-th new food source location.

9. The method for diagnosing steam turbine rotor faults based on VMD kurtosis estimation and RVM according to claim 7 or 8, characterized in that: Determining a first target food source position according to a first fitness value of an old food source position and a second fitness value of a new food source position includes: If the second fitness value of the new food source position is greater than the first fitness value of the old food source position, the new food source position is used to replace the old food source position, and the new food source position is used as the first target food source position; otherwise, the old food source position is retained and the old food source position is used as the first target food source position.

10. The steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to claim 8, characterized in that: The location of the food source is selected based on the fitness probability of the non-scout bees according to the amount of nectar collected by the scout bees and / or the location of the food source, including: Based on the non-scout bees, it is determined whether the amount of nectar collected by the scout bees is greater than the first preset threshold, and the fitness probability of the food source location is calculated according to the following calculation formula to determine whether the fitness probability is greater than the second preset threshold: Among them, P i Represents the fitness probability of the i-th food source location, Fitness i represents the fitness value of the i-th food source location, and M represents the number of food source locations; If the amount of nectar collected by the scout bees at the old food source location is greater than a first preset threshold, and / or the first fitness probability of the old food source location is greater than a second preset threshold, then the old food source location is selected; If the amount of nectar collected by the scout bees at the new food source location is greater than a first preset threshold, and / or the second fitness probability of the new food source location is greater than a second preset threshold, the new food source location is selected.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM described in any one of claims 1-10 is implemented.

12. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steam turbine rotor fault diagnosis method based on VMD kurtosis estimation and RVM according to any one of claims 1 to 10 is implemented.

Images