A wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD

Abstract

The invention discloses a composite fault diagnosis method for a wind turbine gearbox based on AVMD and improved RSSD, and relates to the technical field of fault diagnosis of large-scale rotating mechanical equipment. The method comprises the following steps: obtaining a vibration acceleration signal of a wind turbine gearbox, and determining VMD parameters by using a RWSSA optimization algorithm; performing VMD decomposition on an original signal, selecting an optimal component according to a minimum spread entropy-frequency domain mutual correlation coefficient criterion, and reconstructing the optimal component to obtain a reconstructed signal; performing RSSD parameter optimization on the reconstructed signal by using the RWSSA algorithm, and performing RSSD decomposition on the reconstructed signal based on the RSSD parameters to obtain high and low resonance components containing fault components; performing 1.5-dimensional spectrum analysis on the high and low resonance components, and comparing them with theoretical fault characteristic frequencies to determine the composite fault type; the RWSSA algorithm is used to overcome the problem that the algorithm is prone to fall into a local minimum in parameter optimization, VMD is used to achieve signal denoising, and RSSD is used to achieve decoupling and separation of composite faults, thereby achieving good diagnostic effect.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis of large-scale rotating mechanical equipment, and more specifically to a composite fault diagnosis method for a wind turbine gearbox based on AVMD and improved RSSD. Background Art

[0002] As a renewable and clean energy, wind energy is being developed and utilized by more and more countries around the world. Among them, the gearbox, as the core component of the wind turbine transmission system, works for a long time in a complex and harsh natural environment, and it is easy to have local failures and cause serious accidents. Therefore, accurate fault diagnosis and effective status monitoring of wind turbine gearboxes are of great significance to avoid major safety accidents and reduce operating costs. In actual engineering applications, the failure of wind turbine gearboxes usually manifests as a composite fault composed of gear and bearing failures. The collected original vibration signal of the gearbox is generally a non-stationary and nonlinear signal, and the multiple characteristic frequencies in the signal are coupled and aliased with each other and mixed with strong environmental noise, which brings great difficulties to the extraction, separation and identification of composite fault feature information in early gearbox damage.

[0003] In order to effectively detect composite fault signals, many studies based on vibration signal processing methods have been carried out at home and abroad. For example, wavelet packet decomposition, empirical mode decomposition (EMD), variational mode decomposition (VMD), local mean decomposition (LMD), maximum correlation kurtosis deconvolution (MCKD), etc. The above methods have achieved good results in the field of fault diagnosis, but there are still certain limitations: the basis function in wavelet packet decomposition is difficult to determine, which affects the accurate extraction of composite fault features; EMD and LMD methods are prone to mode aliasing and boundary effects; VMD method overcomes the mode aliasing problem well, but a single VMD method is difficult to obtain good diagnostic results in processing complex multi-component non-stationary signals; MCKD focuses on the selection of single fault resonance bands, which is easily affected by noise interference and missed diagnosis. These limitations restrict the application of signal processing methods in rotating machinery fault diagnosis. Resonance sparse decomposition (RSSD) is a nonlinear signal analysis method based on the resonance properties of vibration signals proposed in 2012. Compared with other signal processing methods, the RSSD method does not use frequency band division to decompose the signal, but considers the frequency and bandwidth of the signal and decomposes the signal sparsely based on the difference in quality factors of different components in the signal. It has strong adaptability and accuracy. In recent years, the RSSD method has been gradually applied to the fault diagnosis of wind turbine gearboxes.

[0004] In existing research, the RSSD method mainly determines the optimal quality factor based on manual selection or the use of traditional intelligent optimization algorithms. The process of obtaining the quality factor takes a long time to calculate and is inefficient. It is also difficult to effectively separate the gear fault signal and the bearing fault signal in the composite signal in the RSSD decomposition, highlighting the fault characteristics. At the same time, strong background noise and interference signals seriously affect the decomposition effect of RSSD. Summary of the invention

[0005] The purpose of the present invention is to solve the above technical problems and provide a wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD.

[0006] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical scheme: a wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD, comprising the following steps:

[0007] Step 1, collecting vibration signals generated by a gear-bearing composite fault in a wind turbine gearbox;

[0008] Step 2, set the preset parameters of the random walk sparrow algorithm, take the minimum scatter entropy value of each IMF component after VMD decomposition of the gearbox vibration signal as the objective function, use the random walk sparrow algorithm to optimize the VMD core parameter penalty factor α and the number of modal components k, and obtain the optimal penalty factor and modal component number combination [α0, k0];

[0009] Step 3: Using the optimal penalty factor α0 and the modal component number k0 as preset parameters, VMD decomposition is performed on the wind turbine gearbox vibration signal to obtain a series of IMF components with different center frequencies;

[0010] Step 4: Combine the spread entropy-frequency domain cross-correlation coefficient criterion to remove the decomposed high-frequency noise and false interference components, and select the remaining IMF components for reconstruction to obtain the reconstructed signal;

[0011] Step 5: Set the preset parameters of the random walk sparrow algorithm, take the minimum scatter entropy-frequency domain mutual correlation coefficient criterion of the high-quality component after the reconstructed signal resonance sparse decomposition as the objective function, use the random walk sparrow algorithm to optimize the high and low quality factors, and obtain the optimal high quality factor and low quality factor combination [Q H , Q L ];

[0012] Step 6: Optimize the quality factor Q H and low quality factor Q L As preset parameters, resonant sparse decomposition is performed on the reconstructed signal to obtain high and low resonance components containing fault components;

[0013] Step 7: Perform 1.5-dimensional spectrum analysis on the high and low resonance components of the wind turbine gearbox fault signal to extract the fault characteristic frequency and complete the decoupling and effective identification of the composite fault characteristics.

[0014] Specifically, the random walk sparrow algorithm in step 2 is an intelligent algorithm that introduces a random walk strategy on the basis of the sparrow search algorithm to perturb and update the initial optimal solution, and obtains an accurate optimal solution through multiple iterations.

[0015] Specifically, the preset parameters of the random walk sparrow algorithm in step 2 include: population size, maximum number of iterations, variable interval, random walk coefficient, VMD core parameter penalty factor α and modal component number k preset variable interval α∈(100,5000), k∈(3,10), step size is 1, population size Pop is set to 20, maximum number of iterations M is set to 30, and random walk coefficient I is set to 1.

[0016] Specifically, in step 4, the spread entropy is combined with the frequency domain mutual correlation coefficient to propose a spread entropy-frequency domain mutual correlation coefficient criterion, and the calculation formula is as follows:

[0017]

[0018] Where: Z represents the criterion function of the spread entropy-frequency domain cross-correlation coefficient, DE(x,m,c,d) represents the spread entropy value in the signal x, m, d and c are the embedding dimension, time delay and number of categories in the spread entropy, respectively, and c is m represents the number of dispersive entropy modes, Represents each scatter mode The probability of C f Represents the signal x in the frequency domain i and i The mutual correlation coefficient, G x and G y are the power spectra of signal x and signal y respectively, f a To analyze the frequency.

[0019] Specifically, in step 5, the minimum spread entropy-frequency domain correlation coefficient criterion of high-quality components is used as the objective function, and the random walk sparrow algorithm is used to optimize the high and low quality factors in the resonance sparse decomposition. The objective function calculation formula is as follows:

[0020]

[0021] Where: fitness represents fitness, Z High-IMF The dispersion entropy and frequency domain correlation coefficient index of the high-quality resonance component are expressed. According to previous experience, the high-quality factor Q H The optimization range is set to [5,50], with a low quality factor QL The optimization range is set to [1,5].

[0022] Specifically, in step 6, the high and low quality factors Q are obtained by using step 5. H , Q L Substitute it into the resonance sparse decomposition to obtain high and low resonance components. The resonance sparse decomposition process is as follows:

[0023] (1) The RSSD method uses morphological component analysis (MCA) to decompose the signal. The signal is decomposed into signals x1 and x2 by using the different resonance properties between faults in the reconstructed signal. Assuming that signals x1 and x2 can be sparsely represented by basis function libraries S1 and S2, respectively, the calculation formula of the sparse decomposition objective function constructed by MCA is as follows:

[0024]

[0025] Where: λ1 and λ2 are weight coefficients, S1 and S2 are the wavelet basis function libraries with adjustable quality factors of high and low resonance components, respectively, W1 and W2 are the transformation coefficients of sub-signals x0 and x1 in the framework S1 and S2, respectively;

[0026] (2) The split augmented Lagrangian search algorithm (SALSA) is used for the formula in (1) to iteratively update the transformation coefficient W1 * , Obtain the minimum objective function J and calculate the high resonance component of the input signal and low resonance components The calculation formula is as follows:

[0027]

[0028] The beneficial effects of the present invention are as follows:

[0029] 1. The present invention adopts random walk sparrow algorithm to optimize VMD and RSSD parameters, which has the advantages of few control variables, fast convergence speed and high global optimization solution accuracy;

[0030] 2. The present invention adopts the spread entropy-frequency domain mutual correlation coefficient criterion, which can effectively identify the modal components with rich fault feature information, further reduce noise interference, and preserve fault feature information;

[0031] 3. The present invention adopts the RWSSA-VMD method to perform noise reduction preprocessing on the original signal containing the composite fault of gear and bearing, and then combines the RWSSA-RSSD method to perform composite fault decoupling and separation on the noise-reduced signal, and finally uses the 1.5-dimensional spectrum to extract the fault characteristic frequency and identify the fault characteristics. The method is an effective method for diagnosing composite faults of wind turbine gearboxes. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of the diagnostic method of the present invention;

[0033] Figure 2 Flowchart for optimizing VMD and RSSD parameters for RWSSA algorithm;

[0034] Figure 3 It is a time domain diagram and envelope spectrum diagram of the simulation signal of the embodiment;

[0035] Figure 4 It is the Z value index diagram of each I MF component after VMD decomposition;

[0036] Figure 5 Iterative change curve of RSSD parameters optimized for RWSSA algorithm;

[0037] Figure 6 This is the high resonance component diagram obtained by RSSD decomposition in this embodiment;

[0038] Figure 7 This is the low resonance component diagram obtained by RSSD decomposition in this embodiment;

[0039] Figure 8 The high resonance component graph obtained by optimizing RSSD decomposition with traditional genetic algorithm;

[0040] Fig. 9 This is the low resonance component graph obtained by optimizing RSSD decomposition using the traditional genetic algorithm. DETAILED DESCRIPTION

[0041] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention.

[0042] When a local fault occurs in the gear of a wind turbine gearbox, the rotational frequency and its frequency multiples will modulate the meshing frequency and its harmonics of the gear, and the vibration signal will show an amplitude modulation-modulation characteristic. When a local fault occurs in the rolling bearing of a wind turbine gearbox, the natural frequency of the faulty element will be modulated by the fault frequency, and the vibration signal will show a periodic impact characteristic. The amplitude modulation-modulation component is a narrowband signal with high frequency aggregation, and the periodic impact component is a broadband signal with high time aggregation. The RSSD method uses the quality factor Q (defined as the ratio of the center frequency and the frequency bandwidth) to evaluate the resonance properties of the signal, and realizes the effective decoupling and separation of the transient non-oscillating component and the periodic pulse component of the vibration signal based on the resonance properties of the signal. At the same time, the vibration signal of the wind turbine gearbox collected has strong background noise and interference, which seriously affects the decomposition effect of RSSD.

[0043] Example 1

[0044] Based on the above reasons, this embodiment provides a wind turbine gearbox composite fault diagnosis method, which includes four modules: vibration signal acquisition, noise reduction, composite fault decoupling, and fault feature extraction. Figure 1 As shown, the specific implementation method includes the following steps:

[0045] A. Vibration signal acquisition module

[0046] Step 1: Collect the vibration signal generated by the gear-bearing composite fault of the wind turbine gearbox through the acceleration sensor.

[0047] B. Vibration signal noise reduction module

[0048] Step 2: Set the preset parameters of the random walk sparrow algorithm, take the minimum scattered entropy value of each IMF component after VMD decomposition of the gearbox vibration signal as the objective function, use the random walk sparrow algorithm to optimize the VMD core parameter penalty factor α and the number of modal components k, and obtain the optimal penalty factor and modal component number combination [α0, k0].

[0049] Step 3: Using the optimal penalty factor α0 and the modal component number k0 as preset parameters, VMD decomposition is performed on the wind turbine gearbox vibration signal to obtain a series of IMF components with different center frequencies.

[0050] Step 4: Combine the spread entropy-frequency domain cross-correlation coefficient criterion to eliminate the decomposed high-frequency noise and false interference components, and select the remaining IMF components for reconstruction to obtain the reconstructed signal. The spread entropy-frequency domain cross-correlation coefficient criterion calculation formula is as follows:

[0051]

[0052] Where: Z represents the criterion function of the spread entropy-frequency domain cross-correlation coefficient, DE(x,m,c,d) represents the spread entropy value in the signal x, m, d and c are the embedding dimension, time delay and number of categories in the spread entropy, respectively, and c is m represents the number of dispersive entropy modes, Represents each scatter mode The probability of C f Represents the signal x in the frequency domain i and i The mutual correlation coefficient, G x and G y are the power spectra of signal x and signal y respectively, f a To analyze the frequency.

[0053] B. Vibration signal composite fault decoupling and separation module

[0054] Step 5: Set the preset parameters of the random walk sparrow algorithm, take the minimum scatter entropy-frequency domain mutual correlation coefficient criterion of the high-quality component after the reconstructed signal resonance sparse decomposition as the objective function, use the random walk sparrow algorithm to optimize the low-quality factor, and obtain the optimal high-quality factor and low-quality factor combination [Q H , Q L ]. The objective function calculation formula is as follows:

[0055]

[0056] Where: fitness represents fitness / objective function, Z High-IMF The dispersion entropy and frequency domain correlation coefficient index of the high-quality resonance component are expressed. According to previous experience, the high-quality factor Q H The optimization range is set to [5,50], with a low quality factor Q L The optimization range is set to [1,5].

[0057] Step 6: Optimize the quality factor Q H and low quality factor Q L As the preset parameters, the reconstructed signal is subjected to resonance sparse decomposition to obtain the high and low resonance components containing the fault components. The resonance sparse decomposition process is as follows:

[0058]

[0059] Where: λ1 and λ2 are weight coefficients, S1 and S2 are the wavelet basis function libraries with adjustable quality factors for high and low resonance components, respectively, and W1 and W2 are the transformation coefficients of sub-signals x0 and x1 under the frameworks S1 and S2, respectively. * and W2 * are iteratively updating the transformation coefficients, and They are the high resonance component and the low resonance component of the input signal respectively.

[0060] C. Vibration signal fault identification and extraction module

[0061] Step 7: Perform 1.5-dimensional spectrum analysis on the high and low resonance components of the wind turbine gearbox fault signal, extract the fault characteristic frequency, and complete the decoupling and effective identification of the composite fault characteristics.

[0062] Embodiment 2 (Simulation Example)

[0063] In this embodiment, combined with the wind turbine gearbox gear-bearing fault vibration model, a simulation signal containing the fault gear periodic pulse signal, the rolling bearing outer ring fault impact signal, the gear meshing component and the shaft component and other narrow-band interference signals and Gaussian white noise is constructed. The key parameter values ​​are as follows: the sampling frequency is 5120Hz, and the number of sampling points is 8192. The shaft rotation frequency f r is 7.5Hz, and the characteristic frequency of the bearing outer ring fault is f n is 40Hz, the gear fault characteristic frequency f s is 75Hz, gear meshing frequency f m is 500Hz, the outer ring fault resonance frequency f n The frequency is 1000Hz, the damping index is 400, and the Gaussian white noise signal-to-noise ratio is -8dB.

[0064] The first step is to obtain the composite fault signal of the gearbox gear-bearing of the wind turbine through the acceleration sensor. The present invention constructs a simulation signal including the periodic pulse signal of the faulty gear, the outer ring fault impact signal of the rolling bearing and Gaussian white noise. The time domain diagram and envelope spectrum of the original signal are shown in the figure. Figure 3 As shown in the figure, the time domain waveform of the vibration signal is affected by noise, and the periodic impact signal is almost completely covered. The envelope spectrum can find the fault characteristic frequencies and their multiples of the bearings and gears, but there are many prominent interference components, and the fault characteristics cannot be effectively identified and extracted.

[0065] Step 2: Specify the optimal penalty factor α and the range of modal component number k in VMD. In this case, α∈(100,5000), k∈(3,10), set the preset parameters of the RWSSA algorithm, take the minimum scatter entropy as the objective function, and use the RWSSA algorithm to adaptively obtain the optimal parameter combination of VMD. In this example, the optimal parameter combination [6,1658] is obtained, and the minimum scatter entropy is 3.32.

[0066] Step 3: Using the optimal parameter combination as the preset parameters, perform VMD decomposition on the signal to obtain 6 IMF components. Calculate the Z value of the criterion index of the spread entropy-frequency domain mutual correlation coefficient of each modal component, and remove the modal components greater than the average Z value, such as Figure 4 As shown, IMF1, IMF4, IMF5, and IMF6 are removed, and IMF2 and IMF3 modal components are selected for reconstruction to obtain the reconstructed signal.

[0067] Step 4: Specify the high quality factor Q in RSSD H and low quality factor Q L The value range of Q H ∈(5,100), Q L∈(1,5), redundancy is set to r1 = r2 = 3.5, set the preset parameters of the RWSSA algorithm, and use the minimum spread entropy-frequency domain mutual correlation coefficient criterion in the low resonance component as the objective function to obtain the optimal high quality factor and low quality factor combination. In this example, the optimal high and low quality factor combination [13.65, 3.78] is obtained, and the convergence curve is shown in the figure Figure 5 As shown in the figure, it shows that the RWSSA algorithm converges in the second generation and obtains the minimum objective function value of 4.57.

[0068] Step 5: Optimize the quality factor Q H and low quality factor Q L As the preset parameters, the reconstructed signal is subjected to resonance sparse decomposition to obtain the high and low resonance components containing the fault components, and a 1.5-dimensional spectrum analysis is performed. The results are shown in Figure 6 ,like Figure 7 shown.

[0069] Depend on Figure 6 It can be seen that the gear local fault frequency f can be clearly observed in the 1.5-dimensional spectrum of the high resonance component s and its frequency doubled components, given by Figure 7 It can be seen that the bearing local fault frequency f can be clearly observed in the 1.5-dimensional spectrum of the low resonance component o Therefore, it can be determined that there is a gear-bearing composite fault in the planetary gearbox of the fixed wind turbine. The diagnosis results are consistent with the experimental results, which proves the effectiveness of the proposed method.

[0070] In order to verify the noise reduction effect of the spread entropy-frequency domain cross-correlation coefficient criterion index Z in the reconstructed signal, the envelope spectrum peak factor E is used. c , correlation coefficient r, and scatter entropy DE single index selected VMD modal component reconstruction for analysis and comparison. At the same time, the root mean square error (RMSE) and signal-to-noise ratio (SNR) were used as denoising effect evaluation indicators. If the RMSE is smaller and the SNR is larger, the denoising effect is better. The results are shown in Table 1:

[0071] Table 1 (mean square error and signal-to-noise ratio values ​​under each indicator)

[0072] Indicator Type RMSE SNR Initial signal 0 -8 Z 0.238 -2.36 <![CDATA[E c ]]> 0.481 -3.18 DE 0.548 -3.37 r 0.913 -6.74

[0073] From the comparison of various indicators in Table 1, the root mean square error obtained by using the spread entropy-frequency domain cross-correlation coefficient criterion indicator Z is the smallest and the signal-to-noise ratio is the largest, which proves the effectiveness of the method proposed in this paper in denoising effect and performance indicators.

[0074] In order to verify the superiority of the method of the present invention, without VMD denoising, the minimum scatter entropy-frequency domain mutual correlation coefficient criterion value is used as the target value function, and the traditional genetic algorithm is used to optimize the RSSD parameter combination [Q H ,QL ], the high and low resonance components of the obtained resonance sparse decomposition are analyzed by 1.5-dimensional spectrum, and the results are as follows Figure 8 ,like Fig. 9 shown.

[0075] Depend on Figure 8 It can be seen that the 1.5-dimensional spectrum of the high resonance component cannot find the local fault frequency f of the gear s and its multiples, given by Fig. 9 It can be seen that although the 1.5-dimensional spectrum of the low resonance component can find the local fault frequency f of the bearing o and its multiples, but the interference component amplitude is too large to complete the effective decoupling and separation of the planetary gearbox compound fault. Therefore, this embodiment is more effective in diagnosing the compound fault of the wind turbine gearbox.

[0076] The technical solutions provided by the embodiments of the present invention are introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the embodiments of the present invention. The description of the above embodiments is only applicable to help understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, according to the embodiments of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD, characterized in that: The following steps are involved: Step 1, collecting vibration signals generated by a gear-bearing composite fault in a wind turbine gearbox; Step 2, setting the preset parameters of the random walk sparrow algorithm, taking the minimum scattered entropy value of each IMF component after VMD decomposition of the gearbox vibration signal as the objective function, using the random walk sparrow algorithm to optimize the VMD core parameter penalty factor α and the number of modal components k, and obtaining the optimal penalty factor and modal component number combination [α0, k0]; the random walk sparrow algorithm is an intelligent algorithm that introduces a random walk strategy on the basis of the sparrow search algorithm to perturb and update the initial optimal solution, and obtains the precise optimal solution through multiple iterations; Step 3: Using the optimal penalty factor α0 and the modal component number k0 as preset parameters, VMD decomposition is performed on the wind turbine gearbox vibration signal to obtain a series of IMF components with different center frequencies; Step 4: Combine the spread entropy-frequency domain mutual correlation coefficient criterion to eliminate the decomposed high-frequency noise and false interference components, select the remaining IMF components for reconstruction to obtain the reconstructed signal; combine the spread entropy with the frequency domain mutual correlation coefficient, and propose the spread entropy-frequency domain mutual correlation coefficient criterion. The calculation formula is as follows: Where: Z represents the criterion function of the spread entropy-frequency domain cross-correlation coefficient, DE(x,m,c,d) represents the spread entropy value in the signal x, m, d and c are the embedding dimension, time delay and number of categories in the spread entropy, respectively, and c is m represents the number of dispersive entropy modes, Represents each scatter mode The probability of C f Represents the signal x in the frequency domain i and i The mutual correlation coefficient, G x and G y are the power spectra of signal x and signal y respectively, f a To analyze the frequency; Step 5: Set the preset parameters of the random walk sparrow algorithm, take the minimum scatter entropy-frequency domain mutual correlation coefficient criterion of the high-quality component after the reconstructed signal resonance sparse decomposition as the objective function, use the random walk sparrow algorithm to optimize the high and low quality factors, and obtain the optimal high quality factor and low quality factor combination [Q H , Q L ]; Step 6: Optimize the quality factor Q H and low quality factor Q L As preset parameters, resonant sparse decomposition is performed on the reconstructed signal to obtain high and low resonance components containing fault components; Step 7: Perform 1.5-dimensional spectrum analysis on the high and low resonance components of the wind turbine gearbox fault signal to extract the fault characteristic frequency and complete the decoupling and effective identification of the composite fault characteristics.

2. According to claim 1, a wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD is characterized in that: The preset parameters of the random walk sparrow algorithm in step 2 include: population size, maximum number of iterations, variable interval, random walk coefficient, VMD core parameter penalty factor α and modal component number k preset variable interval α∈(100,5000), k∈(3,10), step size is 1, population size Pop is set to 20, maximum number of iterations M is set to 30, and random walk coefficient I is set to 1.

3. The wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD according to claim 1 is characterized in that: In step 5, the minimum scatter entropy-frequency domain correlation coefficient criterion of high-quality components is used as the objective function, and the random walk sparrow algorithm is used to optimize the high and low quality factors in the resonance sparse decomposition. The objective function calculation formula is as follows: Where: fitness represents fitness, Z High-IMF The dispersion entropy and frequency domain correlation coefficient index of the high-quality resonance component are expressed. According to previous experience, the high-quality factor Q H The optimization range is set to [5,50], with a low quality factor Q L The optimization range is set to [1,5].

4. The wind turbine gearbox composite fault diagnosis method based on AVMD and improved RSSD according to claim 1 is characterized in that: In step 6, high and low quality factors Q are obtained by using step 5. H , Q L Substitute it into the resonance sparse decomposition to obtain high and low resonance components. The resonance sparse decomposition process is as follows: (1) The RSSD method uses morphological component analysis (MCA) to decompose the signal. The signal is decomposed into signals x1 and x2 by using the different resonance properties between faults in the reconstructed signal. Assuming that signals x1 and x2 can be sparsely represented by basis function libraries S1 and S2, respectively, the calculation formula of the sparse decomposition objective function constructed by MCA is as follows: Where: λ1 and λ2 are weight coefficients, S1 and S2 are the wavelet basis function libraries with adjustable quality factors of high and low resonance components, respectively, W1 and W2 are the transformation coefficients of sub-signals x0 and x1 in the framework S1 and S2, respectively; (2) The split augmented Lagrangian search algorithm (SALSA) is used for the formula in (1) to iteratively update the transformation coefficient W1 * , Obtain the minimum objective function J and calculate the high resonance component of the input signal and low resonance components The calculation formula is as follows:

Images