A method for detecting composite faults in wind turbine gearboxes

Abstract

The present invention discloses a method for detecting complex faults in wind turbine gearboxes. The method obtains a filtered signal from a vibration acceleration sensor above the gearbox, converts sampling from equal time to equal angle, and eliminates the periodic component of the fault signal caused by gear meshing vibration. Based on the fusion of spectral kurtosis and spectral negentropy, the narrowband components processed by maximum overlap discrete wavelet packet transform and unbiased autocorrelation are characterized and quantified to determine the appropriate demodulation frequency band. The method then extracts complex fault features and determines the detection value for complex faults in the wind turbine gearbox. The present invention has the advantage of being able to effectively distinguish different fault types.

Description

Technical Field

[0001] The present invention relates to the field of mechanical system status monitoring, and in particular to a method for detecting composite fault characteristics of a wind power gearbox. Background Art

[0002] Wind turbine gearboxes are composed of multiple gears and bearings. Bearing failures in gearboxes can easily lead to unbalanced loads on the support shaft, leading to gear failure. However, during condition monitoring and fault diagnosis, bearing fault information is easily masked by the more energetic gear mesh vibrations. This makes it difficult to extract subtle bearing fault signatures and analyze the root cause of gearbox failures.

[0003] Patent document CN102938024B discloses a performance evaluation method for a wind turbine condition monitoring system, establishes a Markov model that reflects the state of the wind turbine condition monitoring system, and provides calculation steps and corresponding mathematical formulas for the probability that the system is in various states at a certain moment. Then, a mathematical model that reflects the state of a single sensor is established, which is specifically divided into a healthy state and a failed state. A system function that reflects the state of the entire monitoring system is established, and the state of the system is divided according to the number of sensors in a healthy state and a failed state in the system, specifically into a healthy state, a risk state, and a failed state. In order to keep the system availability not less than a predetermined value, the minimum number of spare sensors and the maximum number of spare sensors allowed to be in short supply in the system are determined by calculating the steady-state value of the system availability. This helps to provide a basis for the formulation of daily maintenance plans for wind turbines, making it convenient to maintain the performance of the system to the greatest extent possible with minimal maintenance costs.

[0004] Patent document CN115577248A discloses a fault diagnosis system and method for a wind turbine generator set, which uses a convolutional neural network model based on deep learning as a feature extractor, and uses a waveform generator to generate a vibration waveform based on the implicit characteristics of the rotational speed of the shaft, that is, the self-vibration signal caused by the excitation signal generated by the rotation of the shaft, and uses the difference matrix between the generated vibration waveform and the vibration signal detected at the head end of the generator to represent the vibration characteristics at the head end of the generator after filtering out the self-vibration signal, and then characterizes the working characteristics of the wind turbine generator set based on the change characteristics between the vibration characteristics detected at the head end of the generator and the vibration characteristics of the vibration signal detected at the tail end of the generator, so as to perform fault diagnosis of the wind turbine generator set.

[0005] In field conditions, when complex faults occur, they are generally coupled together, making them more destructive than single faults. Influenced by the location of the fault excitation source and the signal transmission path, the coupled gear and bearing faults manifest themselves in vibration signals with significant non-stationarity and chaos. This method analyzes the vibration components in the signal, which is of great significance for the application of complex fault diagnosis in wind turbine gearboxes. Summary of the Invention

[0006] In response to the shortcomings and gaps in the existing technologies in the relevant fields, the present invention provides a method for extracting composite fault features of wind turbine gearboxes. The objectives of the present invention are achieved through the following technical solutions:

[0007] A method for detecting a compound fault of a wind turbine gearbox comprises:

[0008] Step 1: Receive the signal of the vibration acceleration sensor above the gearbox and perform detrending preprocessing to obtain a preprocessed signal;

[0009] Step 2: Perform cepstrum editing on the preprocessed signal, convert the edited cepstrum into a spectrum, and obtain a secondary spectrum that matches the original spectrum profile. Then, perform narrowband filtering on the signal with strict linear phase preservation based on the secondary spectrum to obtain a filtered signal.

[0010] Step 3: Estimating the instantaneous phase of the filtered signal and mapping the obtained instantaneous phase to the cumulative phase space, converting the vibration signal from equal time sampling to equal angle sampling;

[0011] Step 4. Construct an adaptive filter to eliminate the periodic components in the fault signal caused by gear meshing vibration. Step 5. Determine the appropriate demodulation frequency band and extract the composite fault features after quantizing the features of each narrowband component after maximum overlap discrete wavelet packet transform and unbiased autocorrelation processing based on the fusion spectral kurtosis and spectral negative entropy indicators. Step 6. Evaluate the composite fault situation of the wind turbine gearbox based on the composite fault features and determine the composite fault detection value of the wind turbine gearbox.

[0012] Furthermore, the cepstrum editing method in step 2 is to perform Fourier transform on the vibration acceleration signal, and obtain the cepstrum by performing inverse Fourier transform on the logarithm of its amplitude; the edited cepstrum is Fourier transformed to obtain the spectrum, and the spectrum is converted into a secondary spectrum in combination with phase information.

[0013] Furthermore, the specific steps of converting the vibration signal from equal time sampling to equal angle sampling in step 3 are as follows: after estimating the instantaneous phase of the filtered signal and mapping the obtained instantaneous phase to the cumulative phase space, a mapping relationship between time and angle is established through the instantaneous phase information corresponding to the equal time sampling point, and the vibration signal is converted from equal time sampling to equal angle sampling according to the instantaneous speed information.

[0014] Furthermore, in the process of mapping the instantaneous phase obtained in step 3 to the cumulative phase space, the cumulative phase calculation formula is:

[0015]

[0016] Among them, x k (n) is the signal after strict linear phase preservation filtering, is x k (n) is the Hilbert transform, and unwrap[] is the phase unwrapping operator.

[0017] Furthermore, the process of determining the frequency response in the process of constructing the adaptive filter in step 4 is as follows:

[0018] By applying a window function w of length L to the vibration signal x(n) L (n) Perform windowing and interception at the kth period to obtain the main vibration sequence x k (n), its expression is:

[0019] x k (n) = x(n+kT)w L (n),n=0,…,L-1

[0020] Where w L (n) is a window function of length L.

[0021] Similarly, the main vibration sequence is time-shifted by L+τ and windowed to obtain the delayed vibration sequence Right now:

[0022]

[0023] Since the time delay factor τ is strictly positive, its windowing process does not overlap, and the broadband noise components in the main vibration sequence and the delayed vibration sequence are uncorrelated;

[0024] Under this condition, the frequency response of the optimal filter that achieves the minimum mean square prediction error is:

[0025]

[0026] Where S UV is the cross power spectrum of any two signals U and V, SUU is the autopower spectrum of signal U, S VV is the autopower spectrum of signal V. p k (n) is the deterministic part of the signal, r k (n) is the non-deterministic part of the signal.

[0027] Furthermore, the steps of constructing the adaptive filter in step 4 are as follows:

[0028] x k (n) and Perform Fourier transform (the sampling point length is M, M ≥ L) to obtain X k,M (f) and The calculation formula of the frequency response of the K-th period sequence is given below:

[0029]

[0030] The above formula is obtained Performing inverse Fourier transform can obtain a noise elimination filter with a length of M.

[0031] Furthermore, the calculation formula of the fused spectrum kurtosis in step 5 is as follows:

[0032]

[0033] Where, is the unbiased autocorrelation expression of the square envelope of each signal after decomposition.

[0034]

[0035] Where X is the square envelope of the decomposed signal; τ = q / f s is the time shift parameter (q=0,1,...,N-1); f s is the sampling frequency in Hz;

[0036] The calculation formula of spectral negative entropy is as follows:

[0037]

[0038] In the formula represents the unbiased autocorrelation of the squared envelope of the signal Perform discrete Fourier transform of n points, n must be an even number, If the length is less than n, it will be padded with zeros; if it is greater than n, it will be truncated.

[0039] Furthermore, the narrowband components after maximum overlap discrete wavelet packet transform and unbiased autocorrelation processing are quantified according to the fusion of spectral kurtosis and spectral negentropy to determine the appropriate demodulation frequency band. The composite fault features are extracted by combining the square envelope spectrum. The calculation formula of the composite fault features is:

[0040]

[0041] Wherein, the weight parameter ρ represents the proportion of time domain indicators and frequency domain indicators in the overall evaluation.

[0042] Furthermore, the vibration acceleration sensor is electrically connected to a processor, and the above method is executed by the processor.

[0043] Furthermore, it also includes an input device, through which the set value is input.

[0044] The present invention can well distinguish complex fault types, thereby providing a more accurate basis for fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the calculation process of the present invention.

[0046] Figure 2 This is the data envelope demodulation result after resampling the time domain vibration data and the angle domain in the example of the present invention.

[0047] Figure 3 The enhanced square envelope order spectrum display result is drawn by using the proposed DRS-improved Autogram method on the actual wind turbine gearbox vibration data in the present invention.

[0048] Figure 4 The DRS-Autogram method is used to compare the actual wind turbine gearbox compound fault operating state in the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clear, the following examples are described in detail with reference to the accompanying drawings.

[0050] Reference Figure 1 , Figure 1 The calculation process of the present invention is shown in FIG. The specific implementation process of the present invention is as follows:

[0051] The vibration acceleration of the gearbox structure is detected by a vibration accelerometer installed above the gearbox, and the vibration signal is preprocessed by detrending terms.

[0052] The signal is cepstrum edited by performing a Fourier transform on the vibration acceleration signal, then performing an inverse Fourier transform on the logarithm of its amplitude to obtain the real cepstrum. The real cepstrum lines are edited by setting a threshold, and the edited real cepstrum is Fourier transformed, combining the phase information to convert it into a secondary spectrum to extract the spectrum profile. Based on the secondary spectrum, the signal is then narrowband filtered with a strictly linear phase. The upper and lower cutoff frequencies of the narrowband filter are determined by searching for the minimum points at both ends near the maximum value within the secondary spectrum. Theoretically, an ideal filter cannot be implemented in physical hardware, but based on the FFT / IFFT transform pair, an "ideal" FIR digital filter can be implemented from a software perspective, ensuring a "0" transition band while maintaining the advantages of the FIR digital filter's strictly linear phase.

[0053] Assuming the acceleration signal is x(t), the calculation formula is as follows:

[0054] C(τ)=F -1 [lg(|Y(f)|]

[0055] Y(f)=F[x(t)]=A(f)exp(jφ(f))

[0056] Where F[·] and F -1 [·] denotes the Fourier transform and inverse Fourier transform, respectively. C(τ) is the inverse spectrum of the vibration signal x(t). A(f) and φ(f) denote the amplitude and phase, respectively.

[0057] The frequency response function expression of an ideal bandpass filter is:

[0058]

[0059] By calculating the FFT of x(n), the FFT components of the passband [f1, f2] and its corresponding negative frequencies [-f1, -f2] are retained, the FFT components of the remaining frequencies are set to zero, and an inverse Fourier transform is performed on them.

[0060] The instantaneous phase of the filtered signal is estimated and mapped to the cumulative phase space. The instantaneous phase information corresponding to the equal time sampling points is used to establish a mapping relationship between time and angle, and the angle domain is resampled. The instantaneous phase calculation formula is:

[0061] Filtered time domain waveform x k After (n) is extracted from the original signal, its Hilbert transform is as follows:

[0062]

[0063] Its instantaneous phase It can be calculated by the following formula

[0064]

[0065] Where, is x k (n) Hilbert transform, unwrap[] is the phase unwrapping operator, which can be x k The instantaneous phase of (n) is mapped to the cumulative phase space.

[0066] Figure 2 This is the data envelope demodulation result after resampling the time domain vibration data and the angle domain in the example of the present invention.

[0067] The angle domain signal is discretely and randomly separated. By eliminating the periodic components generated by gear meshing vibration and highlighting the random components, the purpose of enhancing the characteristics of the composite fault is achieved. By recovering the fault source data, the relatively clean signal components are extracted and their signal characteristics are obtained. The core is to obtain the frequency response function between the main vibration sequence and the delayed vibration sequence, and then obtain the inverse Fourier transform to obtain the separation filter. The implementation process of the filter is briefly described as follows:

[0068] By applying a window function w of length L to the vibration signal x(n) L (n) Perform windowing and interception at the kth period to obtain the main vibration sequence x k (n), its expression is:

[0069] x k (n) = x(n+kT)w L (n),n=0,…,L-1

[0070] Where w L (n) is a window function of length L.

[0071] Similarly, the main vibration sequence is time-shifted by L+τ and windowed to obtain the delayed vibration sequence Right now:

[0072]

[0073] DRS is a process of constructing an adaptive filter by The predictable part of the optimal prediction of the main vibration sequence x k The deterministic component in (n). Figure 3 The enhanced square envelope order spectrum display result is drawn by using the proposed DRS-Autogram method on the actual wind turbine gearbox vibration data in the present invention. Figure 4 The present invention uses the DRS-Autogram method to identify the actual wind turbine gearbox composite fault operating state. k(n) and During the windowing process, the deterministic components in the original signal remain unchanged. Since the time delay factor τ is strictly positive, the windowing process does not overlap, and the broadband noise components in the main vibration sequence and the delayed vibration sequence are uncorrelated.

[0074] Under these conditions, the frequency response of the optimal filter that achieves the minimum mean squared prediction error is given by:

[0075]

[0076] Where S UV is the cross power spectrum of any two signals U and V, S UU is the autopower spectrum of signal U, S VV is the autopower spectrum of signal V. p k (n) is the deterministic part of the signal, r k (n) is the non-deterministic part of the signal.

[0077] x k (n) and Perform Fourier transform (the sampling point length is M, M ≥ L) to obtain X k,M (f) and The calculation formula of the frequency response of the K-th period sequence is given below:

[0078]

[0079] The above formula is obtained Performing an inverse Fourier transform yields a noise cancellation filter of length M. Its phase is corrected by subtracting the time delay factor τ, which can then be applied directly to the original vibration signal. In practical applications, the number of points in the vibration sequence computed by the Fourier transform is L, which also determines the effective length of the filter. When the filter length M is greater than the vibration sequence length L, it is padded with zeros. The length L of the short-term vibration sequence determines the amount of information in the filter and its frequency resolution.

[0080] The signal is decomposed in frequency by using the maximum overlap discrete wavelet packet transform, and the unbiased autocorrelation analysis is used to pre-process the narrowband signals after the maximum overlap discrete wavelet packet transform, which can eliminate the noise interference and make the output signal purer. A new index that integrates spectral kurtosis and spectral negentropy is designed to quantify the characteristics of each narrowband component after the maximum overlap discrete wavelet packet transform and unbiased autocorrelation processing to select the optimal filtering band. The prominent harmonic components in the square envelope spectrum are evaluated based on spectral kurtosis to detect the repetitive transient components, i.e., the pulse components, in the time domain signal; the spectral negentropy is used to evaluate the prominent harmonic components in the square envelope spectrum to detect the repetitive transient components, i.e., the pulse components ... repetitive harmonic components in the time domain signal. To characterize the energy distribution of repetitive transients in the frequency domain:

[0081]

[0082] In the formula represents the unbiased autocorrelation of the squared envelope of the signal Perform discrete Fourier transform of n points, n must be an even number, When the length is less than n, it will be padded with zeros, and when it is greater than n, it will be truncated. And because the spectral kurtosis is actually the sum of the harmonics in the square envelope spectrum normalized by the mean square energy, its evaluation value is returned as a measure of impulse in the time domain. Therefore, the spectral kurtosis in the time domain and the spectral negentropy in the spectrum share the same unit and can be weighted. The final fusion index SP is as follows:

[0083]

[0084] Here, the weight parameter ρ represents the contribution of time-domain and frequency-domain indicators to the overall evaluation. The weighted fusion index is analyzed by calculating the unbiased autocorrelation function of each node signal. Narrowband signals corresponding to the maximum value or the index value within a specific range are selected as the data source for fault feature extraction. Fault diagnosis is then performed by combining the square envelope spectra.

[0085] Finally, it should be noted that the above is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting compound faults in a wind turbine gearbox, characterized in that: include: Step 1: Receive the signal of the vibration acceleration sensor above the gearbox and perform detrending preprocessing to obtain a preprocessed signal; Step 2: Perform cepstrum editing on the preprocessed signal, convert the edited cepstrum into a spectrum, and obtain a secondary spectrum that matches the original spectrum profile. Then, perform narrowband filtering on the signal with strict linear phase preservation based on the secondary spectrum to obtain a filtered signal. Step 3: Estimating the instantaneous phase of the filtered signal and mapping the obtained instantaneous phase to the cumulative phase space, converting the vibration signal from equal time sampling to equal angle sampling; Step 4: construct an adaptive filter to eliminate the periodic components in the fault signal caused by gear meshing vibration; Step 5: Based on the fusion of spectral kurtosis and spectral negentropy, the characteristics of each narrowband component after maximum overlap discrete wavelet packet transform and unbiased autocorrelation processing are quantified to determine the appropriate demodulation frequency band and extract the composite fault characteristics; Step 6: Evaluate the wind turbine gearbox composite fault condition based on the composite fault characteristics and determine the wind turbine gearbox composite fault detection value.

2. The method for detecting compound faults of wind turbine gearboxes according to claim 1, characterized in that: The cepstrum editing method in step 2 is to perform Fourier transform on the vibration acceleration signal, and obtain the cepstrum by performing inverse Fourier transform on the logarithm of its amplitude; perform Fourier transform on the edited cepstrum to obtain the spectrum, and convert the spectrum into a secondary spectrum in combination with phase information.

3. The method for detecting compound faults of wind turbine gearboxes according to claim 1, characterized in that: The specific steps of converting the vibration signal from equal time sampling to equal angle sampling in step 3 are as follows: after estimating the instantaneous phase of the filtered signal and mapping the obtained instantaneous phase to the cumulative phase space, a mapping relationship between time and angle is established through the instantaneous phase information corresponding to the equal time sampling points, and the vibration signal is converted from equal time sampling to equal angle sampling based on the instantaneous speed information.

4. The method for detecting compound faults of wind turbine gearboxes according to claim 1, characterized in that: In the process of mapping the instantaneous phase obtained in step 3 to the cumulative phase space, the cumulative phase calculation formula is: Among them, x k (n) is the signal after strict linear phase preservation filtering, is x k (n) is the Hilbert transform, and unwrap[] is the phase unwrapping operator.

5. The method for detecting compound faults of wind turbine gearboxes according to claim 1, characterized in that: The process of determining the frequency response in the process of constructing the adaptive filter in step 4 is as follows: By applying a window function w of length L to the vibration signal x(n) L (n) Perform windowing and interception at the kth period to obtain the main vibration sequence x k (n), its expression is: x k (n) = x(n+kT)w L (n),n=0,…,L-1;where w L (n) is a window function of length L; The above main vibration sequence is time-shifted by L+τ and windowed to obtain the delayed vibration sequence Right now: Since the time delay factor τ is strictly positive, its windowing process does not overlap, and the broadband noise components in the main vibration sequence and the delayed vibration sequence are uncorrelated; Under this condition, the frequency response of the optimal filter that achieves the minimum mean square prediction error is: Where S UV is the cross power spectrum of any two signals U and V, S UU is the autopower spectrum of signal U, S VV is the autopower spectrum of signal V, p k (n) is the deterministic part of the signal, r k (n) is the non-deterministic part of the signal.

6. The method for detecting compound faults of a wind turbine gearbox according to claim 5, characterized in that: The steps of constructing the adaptive filter in step 4 are: x k (n) and Perform Fourier transform, the sampling point length is M, M ≥ L, and obtain X k,M (f) and The calculation formula of the frequency response of the K-th period sequence is given below: The above formula is obtained Performing inverse Fourier transform can obtain a noise elimination filter with a length of M.

7. The method for detecting compound faults of a wind turbine gearbox according to claim 1, characterized in that: The calculation formula of the fused spectrum kurtosis in step 5 is as follows: Where, is the unbiased autocorrelation expression of the square envelope of each signal after decomposition; Where X is the square envelope of the decomposed signal; is the time shift parameter, q=0,1,…,N-1; f s is the sampling frequency in Hz; The calculation formula of spectral negative entropy is as follows: ； In the formula represents the unbiased autocorrelation of the squared envelope of the signal Perform discrete Fourier transform of n points, n must be an even number, If the length is less than n, it will be padded with zeros; if it is greater than n, it will be truncated.

8. The method for detecting compound faults of a wind turbine gearbox according to claim 7, characterized in that: According to the fusion of spectral kurtosis and spectral negentropy, the narrowband components after maximum overlap discrete wavelet packet transform and unbiased autocorrelation processing are quantified to determine the appropriate demodulation frequency band. Then, the composite fault feature is extracted by combining the square envelope spectrum. The calculation formula of the composite fault feature is: ； Wherein, the weight parameter ρ represents the proportion of time domain indicators and frequency domain indicators in the overall evaluation.

9. The method for detecting compound faults of a wind turbine gearbox according to any one of claims 1 to 8, characterized in that: The vibration acceleration sensor is electrically connected to a processor, and the method according to any one of claims 1 to 8 is executed by the processor.

10. The method for detecting compound faults of a wind turbine gearbox according to claim 9, characterized in that: It also includes an input device, through which the set value is input.

Images