Self-adaptive wavelet packet fan blade acoustic emission signal noise reduction method and system

Abstract

The invention discloses a self-adaptive wavelet packet noise reduction method and system for an acoustic emission signal of a fan blade, relates to the technical field of signal processing, and aims to solve the problem of poor noise reduction effect of the traditional acoustic emission signal of the fan blade, and the method comprises the following steps: S1, decomposing the acoustic emission signal of the fan blade into a plurality of IMF components according to VMD, noise-containing IMF components and low-noise IMF components are screened through correlation coefficients and variance contribution rates; s2, carrying out adaptive redundancy processing on the noise-containing IMF component, and then carrying out wavelet packet decomposition; performing correlation coefficient analysis on the sub-components after wavelet packet decomposition, and determining whether to adopt time-space domain correlation noise reduction or threshold noise reduction according to the relationship between the correlation coefficient and the threshold; wiener filtering is carried out on the low-noise IMF component; and S3, reconstructing each IMF component, and generating a noise-reduced fan blade acoustic emission signal. The invention further comprises a corresponding system. According to the method and the system, accurate noise reduction of the acoustic emission signal of the fan blade can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method and system for reducing noise of acoustic emission signals of fan blades using adaptive wavelet packets. Background Art

[0002] The acoustic emission signal from wind turbine blades is an important diagnostic signal that can reflect the operating and health status of the blades. However, in actual use, the acoustic emission signal of wind turbine blades is often affected by multiple noise sources such as environmental noise, mechanical noise, and electrical interference, resulting in poor signal quality and difficulty in extracting effective information for fault diagnosis.

[0003] Traditional noise reduction methods such as filtering and frequency domain processing often cannot effectively distinguish between signals and noise, and may lose some useful information. For example, Chinese patent publication number CN119091904A involves a method for reducing noise in pipeline leakage acoustic emission signals. Summary of the Invention

[0004] The present invention solves the problem of poor noise reduction effect of traditional fan blade acoustic emission signals, proposes an adaptive wavelet packet fan blade acoustic emission signal noise reduction method and system, and realizes accurate noise reduction of fan blade acoustic emission signals.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: an adaptive wavelet packet fan blade acoustic emission signal noise reduction method, comprising the following steps: S1, decompose the wind turbine blade acoustic emission signal into several IMF components according to VMD, and screen the noisy IMF components and the less-noisy IMF components through correlation coefficient and variance contribution rate; S2, for the noisy IMF components, perform adaptive redundancy processing, and then perform wavelet packet decomposition; perform correlation coefficient analysis on the sub-components after wavelet packet decomposition, and determine whether to use time-space domain correlation denoising or threshold denoising based on the relationship between the correlation coefficient and the threshold; Perform Wiener filtering on the less noisy IMF components; S3, reconstructing each IMF component to generate a noise-reduced wind turbine blade acoustic emission signal.

[0006] In this technical solution, the signal is first decomposed into multiple eigencomponents through variational mode decomposition (VMD), and the noisy IMF components and the less-noisy IMF components are screened out; then, the noisy IMF components are further subjected to adaptive redundancy processing and wavelet packet decomposition, and the wavelet coefficients are classified according to the correlation coefficient, and then the time-space domain correlation denoising method or the threshold denoising method is selected. Finally, the denoised IMF components are reconstructed to generate the wind blade acoustic emission signal.

[0007] The present invention is further configured as follows: Step S1 includes: VMD optimizes the bandwidth of each modal component to converge near its center frequency. The optimization is performed using the alternating direction method. The several modal components obtained through iterative updates in the decomposition process are the eigenmode components.

[0008] In this technical solution, the goal of VMD is to minimize the sum of the bandwidths of each modal component, that is, to optimize the bandwidth of each modal component so that it converges near its center frequency.

[0009] The present invention is further configured as follows: Step S1 further includes: Combining the correlation coefficient and variance contribution rate, a comprehensive score is calculated for each IMF component, and the IMF component with a high comprehensive score is selected.

[0010] In this technical solution, a larger correlation coefficient indicates a stronger correlation between the IMF component and the noise signal, and an IMF component with a larger variance generally contains more information and energy.

[0011] The present invention is further configured as follows: Step S2 includes: Calculate the linear correlation coefficient ρ between the reference signal and the decomposed signal, and set the threshold H. If ρ < H, then use the spatiotemporal domain correlation denoising method; if ρ ≥ H, then use the threshold denoising method.

[0012] In this technical solution, when ρ<H, it indicates that the decomposed signal contains fewer time domain features of the signal, and time-space domain correlation noise reduction can be safely performed at this time; when ρ≥H, it indicates that the decomposed signal contains more signal time domain features, and only threshold noise reduction processing can be performed.

[0013] The present invention is further configured such that: the spatiotemporal domain correlation noise reduction includes: A time-space domain matrix is constructed to convert the original signal into a T×N matrix, where T is the number of time points and N is the number of channels. The correlation matrix is calculated and its eigenvalue decomposition is performed. The first K eigencomponents are selected according to the size of the eigenvalues, and the signal is reconstructed based on the selected main eigencomponents.

[0014] In this technical solution, the above correlation matrix is subjected to eigenvalue decomposition to extract the main features.

[0015] The present invention is further configured as follows: the threshold function of the threshold denoising includes a hard threshold function and a soft threshold function, the hard threshold function specifically: retaining coefficients greater than the threshold and setting coefficients less than the threshold to zero; the soft threshold function specifically: shrinking coefficients greater than the threshold based on the hard threshold.

[0016] In this technical solution, threshold processing is performed on high-frequency detail coefficients to remove noise.

[0017] The present invention is further configured as follows: the Wiener filtering process in step S2 includes: Get the noisy signal X and noise statistics Perform Fourier transform on the signal to obtain the frequency domain signal X(f); Calculate the transfer function H(f) of the Wiener filter based on the statistical characteristics of the signal and noise; Apply H(f) to the frequency domain signal The filtered frequency domain signal Perform inverse Fourier transform to obtain the denoised signal.

[0018] In this technical solution, Wiener filtering aims to estimate the original signal S(f) by minimizing the mean square error (MSE).

[0019] The present invention is further configured such that: the wind turbine blade acoustic emission signal reconstructed and generated in step S3 is the cumulative sum of multiple IMF components.

[0020] The present invention is further configured such that: the correlation matrix includes a time correlation matrix and a space correlation matrix.

[0021] An adaptive wavelet packet fan blade acoustic emission signal noise reduction system is applicable to the above-mentioned adaptive wavelet packet fan blade acoustic emission signal noise reduction method, comprising: The first module decomposes the wind turbine blade acoustic emission signal into several IMF components, and filters out the noisy IMF components and the less-noisy IMF components; The second module performs adaptive redundancy processing on the noisy IMF components and then performs wavelet packet decomposition. The subcomponents after wavelet packet decomposition are subjected to correlation coefficient analysis and the noise reduction method is determined based on the relationship between the correlation coefficient and the threshold. The Wiener filter is performed on the less noisy IMF components. The third module combines the IMF components to reconstruct the original signal.

[0022] An adaptive wavelet packet fan blade acoustic emission signal denoising system of the present technical solution includes a first module, a second module and a third module. The above three modules are connected in sequence. The first module decomposes the signal into multiple intrinsic mode components (IMFs) through VMD, and screens the noisy IMF components and the less-noisy IMF components through correlation coefficient and variance contribution rate; the second module performs corresponding denoising processing on the noisy IMF components and the less-noisy IMF components respectively, and finally reconstructs the signal in the third module.

[0023] The present invention can bring the following beneficial effects: The present invention discloses an adaptive wavelet packet noise reduction method for fan blade acoustic emission signals. By combining VMD, adaptive redundancy, wavelet packet decomposition, and threshold noise reduction technology, the method achieves efficient noise reduction of fan blade acoustic emission signals in complex noise environments. The method can not only effectively remove noise, but also maximize the preservation of signal characteristics and improve signal quality. The method of the present invention can flexibly adapt to different working conditions of fan blades and automatically adjust the noise reduction strategy to cope with changes in the noise environment, thereby improving the intelligence level and adaptability of the noise reduction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of an adaptive wavelet packet noise reduction method for fan blade acoustic emission signals in the present application.

[0025] Figure 2 This is a schematic diagram of an adaptive wavelet packet fan blade acoustic emission signal noise reduction system of the present application. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0027] In actual use, the acoustic emission signals of wind turbine blades are often affected by multiple noise sources such as environmental noise, mechanical noise, and electrical interference, resulting in poor signal quality and difficulty in extracting effective information for fault diagnosis. Traditional noise reduction methods such as filtering and frequency domain processing are often unable to effectively distinguish between signals and noise, and may even lose some useful information.

[0028] To improve signal clarity and accuracy, an effective noise reduction method is urgently needed. Variational mode decomposition (VMD), a new adaptive signal decomposition method, can decompose complex non-stationary signals into several intrinsic mode functions (IMFs), thereby separating the noise from the useful components. However, some of the IMF components obtained by VMD decomposition may still contain noise.

[0029] Wavelet packet transform is a multi-scale time-frequency analysis tool that can perform local analysis on signals at different scales and remove noise through adaptive threshold processing.

[0030] Combining the advantages of VMD and wavelet packets, the noise components can be effectively removed without losing useful signals, thereby improving the quality of the signal.

[0031] Since the acoustic emission signal is a non-stationary time-varying signal, the denoising method is required to have a high local analysis capability to preserve its characteristics as much as possible during the signal extraction process.

[0032] Commonly used denoising methods include wavelet transform, empirical mode decomposition, and independent component analysis. Wavelet transform has become a mainstream denoising method due to its mature theory and fast computational speed. Wavelet-based denoising methods primarily include wavelet modulus maximum denoising, wavelet spatial correlation denoising, and wavelet threshold denoising. Wavelet threshold denoising, in particular, has been widely used in research and engineering applications due to its computational simplicity and excellent results. However, in the wavelet threshold denoising process, the selection of the wavelet basis, the number of wavelet decomposition levels, the threshold criterion, and the threshold function can significantly affect the denoising effect. To address these issues, several improvements have been proposed, such as using wavelet packets instead of wavelets to improve the adaptability of mid- and high-frequency signals, determining the optimal number of decomposition levels through singular spectrum analysis to avoid signal distortion, improving soft and hard threshold functions, and using scale-based threshold selection methods to avoid over-suppression.

[0033] However, most of these methods only process individual wavelet coefficients and fail to fully consider the correlation between them. To improve denoising accuracy, a block thresholding method has been proposed. By considering the correlation between adjacent wavelet coefficients in the time domain, it reduces misjudgments and improves denoising accuracy. Furthermore, a spatial neighborhood joint denoising method based on dual-tree complex wavelet packets exploits the correlation between wavelet coefficients in the time and frequency domains, achieving excellent denoising results.

[0034] Example 1 In view of the problems existing in the prior art, this embodiment proposes an adaptive wavelet packet fan blade acoustic emission signal noise reduction method, referring to Figure 1 , which mainly includes the following steps.

[0035] In step S1, the wind turbine blade sound generation signal is decomposed by VMD to generate several IMF components, and the noisy IMF components and the low-noise IMF components are screened according to the correlation coefficient and the variance contribution rate.

[0036] For step S1 above, the goal of VMD is to minimize the sum of the bandwidths of each modal component, that is, to optimize the bandwidth of each modal component so that it converges around its center frequency. The specific optimization objective formula is: Among them, u k (t) represents the kth modal component; w k is u k The center frequency of (t); is the time derivative; δ(t) is the unit impulse function; j is the imaginary unit; * is the convolution calculation.

[0037] By optimizing the alternating direction method (ADMM), the decomposition process can be iteratively updated in the following two steps: Update the modal component u k (t): Update center frequency w k : Among them, U k (w) is u k Fourier transform of (t) Update the Lagrange multiplier λ(t): Where τ is the step size parameter After iterative update, K modal components u are obtained k (t) is the final intrinsic mode component (IMFs).

[0038] Next, the IMF components with high noise content are screened through correlation coefficient and variance contribution rate.

[0039] Correlation coefficient ρ imf，noise It reflects the linear relationship between the IMF component and the noise signal. The larger the correlation coefficient, the stronger the correlation between the IMF component and the noise signal.

[0040] The formula of the correlation coefficient is specifically expressed as: in: cov(imf, noise) is the covariance of the IMF component and the noise signal.

[0041] var(imf) is the variance of the IMF component var(noise) is the variance of the noise signal The IMF component with a larger correlation coefficient is selected as the noise component and filtered by threshold setting.

[0042] Variance Contribution Rate VC imf It reflects the proportion of the variance of each IMF component in the total variance. IMF components with larger variance usually contain more information and energy.

[0043] The formula for variance contribution rate is specifically expressed as: in: var(imf) is the variance of the IMF component It is the sum of the variances of all IMF components.

[0044] Combining the correlation coefficient and the variance contribution rate, a comprehensive score is calculated for each IMF component, and the IMF components with high comprehensive scores are selected. These components make great contributions both in the noise signal and the energy distribution.

[0045] For step S2, its main process is as follows: for the noisy IMF component, perform adaptive redundancy processing on it, and then perform wavelet packet decomposition; perform correlation coefficient analysis on the sub-components after wavelet packet decomposition, and determine whether to use space-time domain correlation noise reduction or threshold noise reduction according to the relationship between the correlation coefficient and the threshold.

[0046] The linear correlation coefficient ρ between the reference signal and the decomposed signal is calculated. At the same time, a threshold H is preset in advance. The specific noise reduction method is determined by comparing the relationship between the linear correlation coefficient ρ and the threshold H. If ρ < H, then space-time domain correlation noise reduction is used; if ρ ≥ H, then the threshold noise reduction method is used.

[0047] More specifically, the processing process of the noisy IMF component is described below.

[0048] First, through adaptive redundant lifting wavelet packet noise reduction, interpolation and zero-padding are performed on the initial prediction and its update to keep the number of sample points of the signal unchanged during the decomposition process, thereby avoiding information loss and frequency aliasing caused by downsampling; the specific formula is as follows: where, c l is the approximation signal of the l-th layer, d l is the approximation signal of the l-th layer, P is the redundant predictor, and U is the redundant updater. The traditional wavelet transform only further decomposes the low-frequency part during the decomposition process, and does not decompose the high-frequency part anymore, which results in relatively low frequency resolution of the high-frequency signal. The wavelet packet transform improves the frequency resolution of the signal in the high-frequency part by decomposing the high-frequency signal. Combining the wavelet packet transform of the adaptive lifting wavelet can maximize the retention of the space-time characteristics of the signal and enhance the detailed analysis of the signal.

[0049] From the definition of wavelet space domain correlation, the correlation coefficient can be known: or(n) = W(j, n) * W(k, n) To make the correlation coefficient and the wavelet coefficient comparable, it is transformed into a dimensionless form: When |N j (n)| > W(j, n), the coefficient is the signal wavelet coefficient at this time; when |Nj When (n)|≤W(j,n), the coefficient is the noise wavelet coefficient.

[0050] The local gradient of the signal is used as the decision map to replace the original wavelet coefficients, and then the correlation processing is performed. The local gradient has more time domain features than the single wavelet coefficients. For the original lifted wavelet coefficients w(j, k) of the jth layer, there is a gradient decision map D(j, k) with: D(j,k)=(|W(j,k)-W(j,k-1)|+|W(j,k+1)-W(j,k)|) / 2 In the subsequent correlation processing, gradient maps D(j,n) and D(k,n) are used instead of wavelet coefficients W(j,n) and W(k,n) for calculation.

[0051] Secondly, the optimal number of decomposition layers is determined. In this embodiment, the noisy signal is decomposed layer by layer through discrete wavelet transform, and the correlation between each layer of decomposition signal and the original signal is used to determine the optimal number of decomposition layers to achieve effective separation of signal and noise.

[0052] The specific steps are: at each decomposition layer, calculate the correlation coefficients of all decomposed signals and the noisy signal, and use the decomposition signal with the maximum correlation coefficient as the reference signal to maximize the time domain characteristics of the signal. When the number of decomposition layers increases, the signal information and the noise information are gradually separated, and the difference between the maximum correlation coefficient and the minimum correlation coefficient Δr j will gradually increase, indicating that the separation effect of signal and noise is enhanced. When the optimal decomposition layer is reached, Δr j When the signal starts to decrease, it means that the signal characteristics begin to be lost, and then it is no longer possible to select an effective reference signal. At this time, decomposition should be stopped. By dynamically adjusting the number of decomposition layers, the present invention can automatically select the optimal decomposition level according to the signal characteristics, thereby improving the noise reduction effect.

[0053] If you want to perform time-space domain noise reduction on a signal, you must first find the best reference signal. For wavelet packet decomposition, a reference signal search method based on correlation coefficient is proposed to adaptively find the most suitable reference signal at each scale as the decomposition signal. Assume there is an additive noise interference signal X: X i =A i +B i Among them, A i is the signal, B i is Gaussian white noise, and X is transformed by discrete wavelet to obtain the decomposition signal Y: Y i =a i +b i Among them, a iis the wavelet coefficient of the changed signal, b i is the wavelet coefficient of the changed noise, and the linear correlation coefficient is: Perform dimensionless processing on the above denominator, decompose the numerator, and then obtain the following formula: Cov(X, Y) = Cov(A, a) + Cov(A, b) + Cov(a, B) + Cov(B, b) The signal A and the noise B are not correlated, so the signal A and the noise wavelet transform b, and the noise B and the signal wavelet transform a are also not correlated. Then there will be Cov(A, b) ≈ 0 and Cov(a, B) ≈ 0. Determine whether to perform correlation noise reduction by calculating the correlation coefficient.

[0054] In the actual application process, when the decomposition scale is large, there may be a situation where a large number of time-domain features of the signal are included in multiple decomposed signals. For these signals, directly using correlation noise reduction is likely to cause misfiltering of signal features. Therefore, it is necessary to perform threshold noise reduction on them for preliminary processing. To avoid misjudgment, it is possible to judge whether it is suitable to use the correlation noise reduction method by calculating the linear correlation coefficient ρ between the reference signal and the decomposed signal.

[0055] In this embodiment, there is a threshold H. When ρ < H, it means that the time-domain features of the signal in the decomposed signal are less, and at this time, spatio-temporal domain correlation noise reduction can be safely performed; when ρ ≥ H, it indicates that the decomposed signal contains more signal time-domain features, and only threshold noise reduction processing can be performed.

[0056] Experimental verification shows that when the reference signal and the decomposed signal are weakly correlated or uncorrelated, setting the threshold H = 0.4 is reasonable. In this case, the time-domain features of the signal contained in the decomposed signal can be ignored, and performing correlation noise reduction will not have a significant impact on the final result.

[0057] For spatio-temporal domain correlation noise reduction, specifically: first, construct the corresponding spatio-temporal domain matrix, convert the original signal into a matrix form of T×N, where the above T is the number of time points and N is the number of channels; when the signal is single-channel, expand the signal into a multi-line matrix by frame division or windowing.

[0058] Secondly, calculate the correlation matrix, perform eigenvalue decomposition on it, and select the first K eigencomponents according to the size of the eigenvalues; finally, reconstruct the signal according to the selected main eigencomponents.

[0059] Specifically, the correlation matrix includes a time correlation matrix and a space correlation matrix.

[0060] Among them, the time correlation matrix can be expressed as: Among them, R t It is a T×T symmetric matrix, representing the temporal correlation.

[0061] The spatial correlation matrix can be expressed as: where R s It is an N×N symmetric matrix, representing the spatial correlation.

[0062] Perform eigenvalue decomposition on the above correlation matrix to extract the main features.

[0063] Specifically, set R as the correlation matrix (time and / or space domain) and perform eigenvalue decomposition, which is specifically expressed as follows: R=UΛU T Where U is the eigenvector matrix and Λ is the diagonal eigenvalue matrix. i It reflects the energy of the information contained in the corresponding eigenvector. The signal component corresponds to a larger eigenvalue, and the noise component corresponds to a smaller eigenvalue.

[0064] Filter the first K eigenvalues (larger eigenvalues) according to the size of the eigenvalues: Where K is the number of signal components retained, usually set according to the cumulative energy ratio: Cumulative energy ratio: Reconstruct the signal based on the filtered main characteristic components: Among them, U signal is the eigenvector matrix corresponding to the first K eigencomponents, is the signal matrix after noise reduction.

[0065] For threshold denoising: the detail coefficient d for high frequencies j，k Perform threshold processing to remove noise.

[0066] Threshold functions include hard threshold functions and soft threshold functions.

[0067] Among them, the hard threshold function: retain the corresponding coefficients greater than the threshold, and set the corresponding coefficients less than the threshold to zero, which is specifically expressed as: Soft threshold function: Based on the hard threshold, the coefficients greater than the threshold are further shrunk. Specifically expressed as: where sgn(·) is the sign function.

[0068] For threshold selection: The fixed threshold is: Where, σ: noise standard deviation; N: signal length Adaptive threshold: dynamically adjusted according to the statistical characteristics of each layer's detail coefficients.

[0069] The coefficients after processing and Perform wavelet transform (IDWT) to reconstruct the denoised signal, which is specifically expressed as:

[0070] For the less noisy IMF component, Wiener filtering is performed. Wiener filtering aims to estimate the original signal S(f) by minimizing the mean square error (MSE).

[0071] Assume that the observed signal X(f) is the superposition of the signal S(f) and the noise N(f): X(f)=S(f)+N(f) The transfer function of the Wiener filter is: Among them, P S (f) is the signal power spectrum density, P N (f) is the noise power spectral density; The output signal after filtering is:

[0072] The specific steps are: 1. Obtain the noisy signal X and noise statistics through experiments or estimation methods; 2. Then perform Fourier transform on the signal to obtain the frequency domain signal X(f); 3. Calculate the transfer function of the Wiener filter based on the statistical characteristics of the signal and noise; 4. Apply the transfer function H(f) to the frequency domain signal: 5. For the filtered frequency domain signal Perform inverse Fourier transform and return to the time domain to obtain the denoised signal.

[0073] Step S3: reconstructing each IMF component to generate a noise-reduced wind turbine blade acoustic emission signal.

[0074] Through the above steps S1 to S2, a number of processed IMF components are obtained, and these IMF components are combined to reconstruct the original signal.

[0075] Assume that K IMF components are obtained by VMD (IMF1, IMF2, ...IMF after processing) K )The reconstruction format of the original signal x(t) is:

[0076] Example 2 Based on Example 1, this example also proposes an adaptive wavelet packet fan blade acoustic emission signal noise reduction system, referring to Figure 2 , which mainly includes a first module, a second module and a third module, wherein the first module, the second module and the third module are connected in sequence.

[0077] The first module can decompose the acoustic emission signal of the wind turbine blade and filter out the noisy IMF component and the less-noisy IMF component.

[0078] The first module specifically includes a decomposition unit and a screening unit connected to the decomposition unit, which respectively perform the above-mentioned signal decomposition and IMF component screening functions.

[0079] For the second module, the functions it can achieve are: for the noisy IMF components, adaptive redundancy processing is performed, and then wavelet packet decomposition is performed; correlation coefficient analysis is performed on the sub-components after wavelet packet decomposition, and the noise reduction method is determined according to the relationship between the correlation coefficient and the threshold; for the less noisy IMF components, Wiener filtering is performed.

[0080] The second module includes a first processing unit and a second processing unit. The first processing unit includes an adaptive redundancy block, a wavelet packet decomposition block, a spatiotemporal correlation noise reduction block, a threshold noise reduction block, and a signal reconstruction block. These blocks can perform the noise reduction process on the noisy IMF components. The second processing unit is equipped with a Wiener filter block to filter the less noisy IMF components.

[0081] The third module can realize the function of combining the IMF components to reconstruct the original signal.

[0082] The third module comprises a joint reconstruction unit in which signal reconstruction can be achieved.

[0083] An adaptive wavelet packet fan blade acoustic emission signal denoising system according to the present technical solution includes a first module, a second module and a third module. The first module decomposes the signal into multiple intrinsic mode components (IMFs) through VMD, and screens the noisy IMF components and the less-noisy IMF components through correlation coefficient and variance contribution rate; the second module performs corresponding denoising processing on the noisy IMF components and the less-noisy IMF components respectively, and finally reconstructs the signal in the third module.

[0084] Finally, it should be noted that the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention and are intended to be encompassed by the claims of the present invention.

Claims

1. An adaptive wavelet packet noise reduction method for fan blade acoustic emission signals, characterized in that: The following steps are involved: S1, decompose the wind turbine blade acoustic emission signal into several IMF components according to VMD, and screen the noisy IMF components and the less-noisy IMF components through correlation coefficient and variance contribution rate; S2, for the noisy IMF components, perform adaptive redundancy processing, and then perform wavelet packet decomposition; perform correlation coefficient analysis on the sub-components after wavelet packet decomposition, and determine whether to use time-space domain correlation denoising or threshold denoising based on the relationship between the correlation coefficient and the threshold; Perform Wiener filtering on the less noisy IMF components; S3, reconstructing each IMF component to generate a noise-reduced wind turbine blade acoustic emission signal.

2. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 1 is characterized in that: The step S1 comprises: VMD optimizes the bandwidth of each modal component to converge near its center frequency. The optimization is performed using the alternating direction method. The several modal components obtained through iterative updates in the decomposition process are the eigenmode components.

3. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 1 or 2, characterized in that: The step S1 further includes: Combining the correlation coefficient and variance contribution rate, a comprehensive score is calculated for each IMF component, and the IMF component with a high comprehensive score is selected.

4. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 3 is characterized in that: The step S2 comprises: Calculate the linear correlation coefficient ρ between the reference signal and the decomposed signal, and set the threshold H. If ρ < H, then use the spatiotemporal domain correlation denoising method; if ρ ≥ H, then use the threshold denoising method.

5. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 1 or 4, characterized in that: The time-space domain correlation noise reduction includes: A time-space domain matrix is constructed to convert the original signal into a T×N matrix, where T is the number of time points and N is the number of channels. The correlation matrix is calculated and its eigenvalue decomposition is performed. The first K eigencomponents are selected according to the size of the eigenvalues, and the signal is reconstructed based on the selected main eigencomponents.

6. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 1, characterized in that: The threshold function of the threshold denoising includes a hard threshold function and a soft threshold function. The hard threshold function specifically retains coefficients greater than the threshold and sets coefficients less than the threshold to zero; the soft threshold function specifically shrinks coefficients greater than the threshold based on the hard threshold.

7. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 1, 4 or 6, characterized in that: The Wiener filtering process of step S2 includes: Get the noisy signal X and noise statistics Perform Fourier transform on the signal to obtain the frequency domain signal X(f); Calculate the transfer function H(f) of the Wiener filter based on the statistical characteristics of the signal and noise; Apply H(f) to the frequency domain signal The filtered frequency domain signal Perform inverse Fourier transform to obtain the denoised signal.

8. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 1, characterized in that: The wind turbine blade acoustic emission signal reconstructed and generated in step S3 is the cumulative sum of multiple IMF components.

9. The adaptive wavelet packet noise reduction method for fan blade acoustic emission signals according to claim 5, characterized in that: The correlation matrix includes a time correlation matrix and a space correlation matrix.

10. An adaptive wavelet packet fan blade acoustic emission signal denoising system, applicable to the adaptive wavelet packet fan blade acoustic emission signal denoising method according to any one of claims 1 to 7, characterized in that: include: The first module decomposes the wind turbine blade acoustic emission signal into several IMF components, and filters out the noisy IMF components and the less-noisy IMF components; The second module performs adaptive redundancy processing on the noisy IMF components and then performs wavelet packet decomposition. The subcomponents after wavelet packet decomposition are subjected to correlation coefficient analysis and the noise reduction method is determined based on the relationship between the correlation coefficient and the threshold. The Wiener filter is performed on the less noisy IMF components. The third module combines the IMF components to reconstruct the original signal.

Images