Joint denoising method and system based on adaptive large neighborhood search and modal decomposition

Abstract

The invention provides a joint denoising method and system based on adaptive large neighborhood search and modal decomposition, and belongs to the technical field of signal processing and nondestructive detection.The method comprises the steps that an ultrasonic signal and a vibration signal of a detected insulator are synchronously collected and preprocessed; dynamically estimating the noise level based on the preprocessed ultrasonic signal power spectral density, and optimizing decomposition parameters by adopting an adaptive large neighborhood search algorithm; on the basis of the optimized decomposition parameters, wavelet packet decomposition and ensemble empirical mode decomposition are executed in parallel, and effective intrinsic mode function components are screened through cross-correlation verification; extracting the resonance frequency of the preprocessed vibration signal, performing target frequency band weighted enhancement on the low-frequency sub-band, and dynamically adjusting the threshold parameter of the high-frequency sub-band and the low-frequency sub-band according to the resonance frequency; and generating a preliminary de-noised signal from the fused signal, performing affine projection algorithm filtering and multi-modal cross validation, and outputting the verified ultrasonic signal as a final de-noising result.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of signal processing and non-destructive testing of electrical equipment, and in particular to a combined denoising method and system based on adaptive large neighborhood search and modal decomposition, which is especially suitable for defect detection and noise suppression of composite insulators in power systems. BACKGROUND

[0002] Composite insulators play a crucial role in power systems. Over time, insulators may be affected by factors such as contamination, dielectric breakdown, and mechanical damage, leading to performance degradation and even failure. Therefore, regular detection and monitoring of insulators is essential. The commonly used ultrasonic detection technology has the advantages of high efficiency and convenient operation, but is often disturbed by environmental noise, affecting the accuracy of the detection results.

[0003] Existing ultrasonic denoising methods include empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), complete ensemble empirical mode decomposition (CEEMDAN), hard threshold denoising, and soft threshold denoising. These methods can improve signal quality to some extent, but each has its own limitations. For example, EMD is prone to modal coupling problems, EEMD has high computational complexity, and threshold denoising methods may cause signal distortion or noise residue. Therefore, it is particularly important to develop a more efficient denoising method.

[0004] Prior art document 1 (CN110926386A) discloses a transformer winding ultrasonic three-dimensional imaging method in the field of deformation detection of transformer windings. It uses CEEMD decomposition and improved wavelet threshold joint denoising, and combines signal reconstruction to generate a three-dimensional image. However, it relies on manual experience to set the wavelet decomposition level and threshold parameters, lacks adaptive parameter optimization mechanism, and has problems such as decomposition precision being disturbed by working conditions and high-frequency effective components being easily lost. Moreover, it cannot effectively distinguish between noise and weak defect characteristics in a complex noise environment.

[0005] Prior art document 2 (CN109580787A) discloses a denoising method in the field of transformer high-voltage bushing lead ultrasonic detection technology. It uses wavelet decomposition and EEMD secondary denoising combined with correlation coefficient filtering components, and uses the "2σ rule" to filter out noise. However, it uses a fixed threshold rule to process IMF components, does not consider the dynamic change characteristics of noise levels, and has problems such as low-frequency effective signal deletion and high-frequency noise residue. SUMMARY

[0006] To solve the problems in the prior art, the present application provides a combined denoising method and system based on adaptive large neighborhood search and modal decomposition.

[0007] The present application adopts the following technical solutions.

[0008] The first aspect of the present application provides a joint denoising method based on adaptive large neighborhood search and modal decomposition, comprising the following steps:

[0009] Synchronously collect and pre-process the ultrasonic signal and the vibration signal of the insulator to be measured; dynamically estimate the noise level based on the power spectral density of the pre-processed ultrasonic signal, and optimize the decomposition parameters using an adaptive large neighborhood search algorithm; based on the optimized decomposition parameters, perform wavelet packet decomposition on the ultrasonic signal to obtain low-frequency subbands and high-frequency subbands, perform hierarchical threshold processing on the high-frequency subbands, and simultaneously perform ensemble empirical mode decomposition in parallel, and verify and select effective intrinsic mode function components through cross-correlation;

[0010] Extract the resonance frequency of the pre-processed vibration signal, weight and enhance the target frequency band of the low-frequency subband, and dynamically adjust the threshold parameters of the high-frequency and low-frequency subbands according to the resonance frequency;

[0011] Fuse the processed signals to generate a preliminary denoising signal, the processed signals including the enhanced low-frequency subband, the effective intrinsic mode function component, and the high-frequency subband processed by hierarchical threshold processing, perform affine projection algorithm filtering on the preliminary denoising signal, and perform multi-modal cross-validation, and output the validated ultrasonic signal as the final denoising result.

[0012] Optionally, the synchronous acquisition comprises:

[0013] The ultrasonic signal and the vibration signal are collected through a hardware trigger synchronization mechanism to ensure that the time alignment error is less than 1 microsecond;

[0014] Baseline drift correction is performed on the ultrasonic signal, and band-pass filtering is performed on the vibration signal to reserve the resonance frequency.

[0015] Optionally, the dynamic estimation of the noise level and the optimization of the decomposition parameters comprise:

[0016] Power spectral density analysis is performed on the pre-processed ultrasonic signal, the average value of the power spectral density is calculated in a pre-set noise frequency band, and the square root of the average value is taken as the real-time noise level estimation value;

[0017] The decomposition parameters are dynamically optimized using an adaptive large neighborhood search algorithm, the objective function is to maximize the signal-to-noise ratio, the decomposition parameters include the wavelet packet decomposition layer number, the low-frequency subband threshold coefficient, and the ensemble number and white noise amplitude of EEMD, and the hierarchical threshold parameters include the low-frequency subband threshold coefficient and the high-frequency subband hierarchical threshold;

[0018] Based on the optimal low-frequency subband threshold coefficient, the high-frequency subband hierarchical threshold parameters are generated according to a pre-set proportion rule to obtain the optimal decomposition parameter combination.

[0019] Optionally, the optimization of the decomposition parameters by the adaptive large neighborhood search algorithm comprises:

[0020] Randomly increasing and decreasing the number of decomposition layers, and randomly perturbing and adjusting the low-frequency sub-band threshold coefficient and the white noise amplitude;

[0021] Applying a value range constraint to the parameter combination, including limiting the maximum value of the number of decomposition layers, the upper and lower limits of the low-frequency sub-band threshold coefficient, and the reasonable interval of the white noise amplitude;

[0022] Dynamically adjusting the weight distribution of the destruction operator and the repair operator according to the historical optimization performance;

[0023] Generating 10 sets of candidate parameter combinations in each iteration, and selecting the optimal solution by calculating the signal-to-noise ratio and mean square error of each combination;

[0024] If there is no signal-to-noise ratio improvement for 5 consecutive iterations or the preset number of iterations is reached, the optimization is terminated and the optimal wavelet packet decomposition layer number, low-frequency sub-band threshold coefficient, and EEMD set number and white noise amplitude combination are output.

[0025] Optionally, the screening of effective intrinsic mode function components comprises:

[0026] Calculating the normalized cross-correlation coefficient of the high-frequency sub-band signal obtained by wavelet packet decomposition and all intrinsic mode function components generated by ensemble empirical mode decomposition;

[0027] Dynamically setting a first screening threshold and a second screening threshold according to the real-time noise level estimate, wherein the first threshold is used to retain high-correlation components, and the second threshold is used to eliminate low-correlation components;

[0028] Retaining the intrinsic mode function components whose normalized cross-correlation coefficients are greater than the first screening threshold, and eliminating the components whose normalized cross-correlation coefficients are less than the second screening threshold;

[0029] Performing wavelet threshold reprocessing on the components whose normalized cross-correlation coefficients are between the first threshold and the second threshold, and the threshold strength is negatively correlated with the current noise level.

[0030] Optionally, the target frequency band enhancement and threshold adjustment comprises:

[0031] Extracting the resonance main frequency of the vibration signal, and determining the target enhancement frequency band of the ultrasonic signal based on the resonance main frequency and the insulator sheath thickness, wherein the width of the target frequency band is inversely proportional to the sheath thickness;

[0032] Selectively weighting and amplifying the low-frequency sub-band of the ultrasonic signal in the target enhancement frequency band, and the weighting coefficient is negatively correlated with the number of effective intrinsic mode function components selected.

[0033] Optionally, the filtering according to the preliminary de-noised signal comprises:

[0034] initializing filter order, projection order, step parameter and regularization factor, the filter order being dynamically adjusted according to the sampling rate of the preliminary de-noised signal;

[0035] constructing an input signal matrix and a projection matrix, and updating filter weights through adaptive iteration;

[0036] dynamically adjusting the step parameter according to the local gradient of the preliminary de-noised signal;

[0037] outputting the filtered de-noised signal after a preset number of iterations.

[0038] Optionally, the multi-modal cross-validation on the filtered de-noised signal comprises:

[0039] time domain verification, calculating the time domain dynamic time warping distance between the filtered de-noised signal and a standard discharge pulse template, and verifying whether the dynamic time warping distance is less than a preset time domain tolerance value;

[0040] frequency domain verification, calculating the energy proportion of the filtered de-noised signal in a target enhanced frequency band, and verifying whether the energy proportion is higher than a preset frequency domain threshold value;

[0041] energy verification, calculating the peak-to-peak value retention rate of the filtered de-noised signal and the original ultrasonic signal, and verifying whether the peak-to-peak value retention rate is higher than a preset energy threshold value.

[0042] Optionally, the processing on the multi-modal cross-validation result comprises:

[0043] if all the three verifications pass, marking as a high-confidence signal and outputting the final de-noised result;

[0044] if there is at least one verification that does not pass and the deviation of the non-passing item is not more than 20%, triggering adaptive adjustment of decomposition parameters and re-executing the de-noising and verification process, the parameter adjustment comprising reducing the hierarchical threshold intensity coefficient, relaxing the eigenmode function screening threshold or updating the standard discharge pulse template;

[0045] if there is at least one verification that does not pass and the deviation of the non-passing item is more than 20%, discarding the current filtered de-noised signal and triggering device self-checking.

[0046] The second aspect of the present application provides a joint de-noising system based on adaptive large neighborhood search and modal decomposition, based on the joint de-noising method based on adaptive large neighborhood search and modal decomposition according to the first aspect of the present application, the system comprising:

[0047] A signal acquisition and preprocessing module is configured to synchronously acquire ultrasonic signals and vibration signals, and perform baseline correction and band-pass filtering on the signals;

[0048] A parameter optimization module is configured to dynamically optimize wavelet packet decomposition layers, threshold parameters and noise amplitude of ensemble empirical mode decomposition based on an adaptive large neighborhood search algorithm;

[0049] A signal decomposition module is configured to perform wavelet packet decomposition and ensemble empirical mode decomposition in parallel, and perform hierarchical threshold processing on high-frequency subbands;

[0050] A component screening module is configured to screen effective intrinsic mode function components through cross-correlation verification, and eliminate noise interference components;

[0051] A frequency band enhancement module is configured to extract vibration signal resonance frequencies and guide weighted amplification of target frequency bands of ultrasonic signals;

[0052] A post-processing filtering module is configured to perform affine projection algorithm filtering on reconstructed signals;

[0053] A multi-modal verification module is configured to perform time-domain, frequency-domain and energy multi-dimensional cross-verification on denoised signals.

[0054] Compared with the prior art, the present application has at least the following beneficial effects:

[0055] 1. The present application dynamically optimizes signal decomposition parameters through an adaptive large neighborhood search algorithm, solves the problem of strong dependence on artificial experience and insufficient decomposition precision in the prior art, and significantly improves the separation efficiency of noise and effective components in ultrasonic signals.

[0056] 2. The present application iteratively filters reconstructed signals through an affine projection algorithm, solves the defect of residual noise interference in traditional threshold denoising methods, and realizes the synchronous improvement of signal smoothness and fidelity.

[0057] 3. The present application solves the problems of modal aliasing and invalid component interference through parallel cross-correlation verification of wavelet packet decomposition and ensemble empirical mode decomposition, and improves the integrity and reliability of signal reconstruction.

[0058] 4. The present application solves the problem of difficult defect feature extraction in complex noise environments through vibration signal resonance frequency guided ultrasonic signal frequency band enhancement, and significantly improves the detection precision of internal defects of composite insulators. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are one embodiment of the present application, and other drawings can also be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0060] Figure 1 A flow chart of the method in an embodiment of the present application;

[0061] Figure 2 An ultrasonic signal to be denoised in an embodiment of the present application;

[0062] Figure 3 A signal after denoising processing in an embodiment of the present application. DETAILED DESCRIPTION

[0063] The present application will be further described in detail below in combination with the drawings and specific embodiments. The advantages and features of the present application will be more apparent according to the following description. It should be noted that the drawings are greatly simplified and all use non-precise proportions, only to facilitate, clear auxiliary purpose of explaining the embodiments of the present application. In order to make the purpose, features and advantages of the present application more apparent and easy to understand, please refer to the drawings. It should be noted that the structure, proportion, size and the like shown in the drawings attached to the present specification are only used to cooperate with the content disclosed in the specification, so as to be understood and read by those skilled in the art, and not to limit the defined conditions for implementing the present application, so it does not have the technical essence, any modification of structure, change of proportion relationship or adjustment of size, without affecting the effect and purpose that can be produced by the present application, should still fall within the scope of the technical content disclosed by the present application.

[0064] The present application provides a joint denoising method based on adaptive large neighborhood search and modal decomposition in embodiment 1, as shown in formula (1), including the following steps: Figure 1

[0065] Step 1, synchronously collecting ultrasonic signals and vibration signals of the measured insulator and pre-processing.

[0066] Preferably, in step 1, synchronously collecting ultrasonic signals and vibration signals of the measured insulator includes:

[0067] Deploying ultrasonic sensors and vibration sensors to synchronously collect ultrasonic signals S u (t) and vibration signals S v (t), wherein the vibration signal S v (t) contains the resonance frequency and amplitude.

[0068] Further preferably, collecting ultrasonic signals of the measured insulator further includes: ​

[0069] According to the actual appearance data and internal structure of the composite insulator, an ultrasonic transducer suitable for detecting the composite insulator is selected;

[0070] The transmitter transmits an ultrasonic signal, and the receiver receives the ultrasonic signal after propagating through the interior of the composite insulator.

[0071] Further preferably, collecting the vibration signal of the measured insulator further comprises:

[0072] A piezoelectric acceleration sensor is deployed to collect the mechanical vibration signal of the equipment, with a frequency response range of 0.1-10 kHz and a sensitivity of 50 mV / g.

[0073] Further preferably, synchronously collecting the ultrasonic signal and the vibration signal of the measured insulator further comprises using a hardware-triggered synchronization time synchronization mechanism to ensure that the time alignment error of the two signals is <1 μs.

[0074] Preferably, in step 1, the pre-processing of the collected ultrasonic signal and vibration signal comprises:

[0075] Baseline drift correction is performed on the ultrasonic signal S u (t);

[0076] Band-pass filtering is performed on the vibration signal S v (t) to retain the frequency band of the resonance frequency ±5 kHz.

[0077] It is worth noting that, in order to solve the problems of poor synchronization of multi-modal data and serious noise interference in the traditional signal collection process, the present application realizes high-precision alignment and preliminary noise reduction of ultrasonic and vibration signals through a hardware-triggered synchronization mechanism and baseline drift correction technology, providing high-quality input signals for subsequent processing.

[0078] Step 2: Dynamically estimate the noise level based on the power spectral density of the pre-processed ultrasonic signal, and dynamically optimize the decomposition parameters using an adaptive large neighborhood search (ALNS) algorithm.

[0079] Preferably, step 2 comprises:

[0080] Step 2.1: Perform power spectral density analysis on the pre-processed ultrasonic signal, calculate the average power spectral density in the pre-set noise frequency band, and take the square root of the average value as the real-time noise level estimate;

[0081] Further preferably, step 2.1 comprises:

[0082] Perform power spectral density analysis on the pre-processed ultrasonic signal S u (t) in the pre-set noise dominant frequency band [fl ,f h ] is used to calculate the noise standard deviation σ as the real-time noise level estimation value, and the formula is:

[0083]

[0084] Wherein, PSD(k) is the power spectrum density value of the signal at the kth frequency point; is the frequency band indicator function. When the frequency belongs to [f l ,f h ] is 1, otherwise it is 0; M is the number of frequency domain sampling points, that is, the number of frequency points for power spectrum density (PSD) analysis.

[0085] Further preferably, exemplarily, the noise dominant frequency band [f l ,f h ] is calibrated in the following way:

[0086] A background noise signal is collected during an initial period without ultrasonic excitation;

[0087] Calculate its power spectrum density peak interval and set f l =0.1Mhz, f h =0.3Mhz.

[0088] In step 2.2, the adaptive large neighborhood search (ALNS) algorithm is used to dynamically optimize the decomposition parameters. The objective function is to maximize the signal-to-noise ratio. The decomposition parameters include the number of wavelet packet decomposition layers, the low-frequency subband threshold coefficient, the number of EEMD sets, and the white noise amplitude. The layered threshold parameters include the low-frequency subband threshold coefficient and the high-frequency subband layered threshold.

[0089] Further preferably, the step 2.2 includes:

[0090] The destruction operator randomly increases or decreases the number of decomposition layers and randomly perturbs the low-frequency subband threshold coefficient and white noise amplitude;

[0091] The repair operator imposes value range constraints on parameter combinations;

[0092] Dynamically adjust the operator selection probability based on historical optimization performance.

[0093] Specifically, the objective function of ALNS is:

[0094]

[0095] Specifically, the ALNS optimization process includes:

[0096] (1) Destruction operator design

[0097] Randomly adjust the number of decomposition layers J±1 (J≥3);

[0098] Low-frequency sub-band threshold coefficient λ low and white noise amplitude A noise Gaussian perturbation:

[0099] λ' low = λ low + N(0, 0.1σ)

[0100] A' noise = A noise + N(0, 0.05σ)

[0101] Set the number of times K dynamically according to the noise level:

[0102]

[0103] (2) Repair operator design

[0104] Forced constraints:

[0105] λ low ∈ [0.4σ, 1.0σ];

[0106] A noise ∈ [0.1σ, 0.5σ];

[0107] K ∈ [10, 100].

[0108] (3) Weight adaptive mechanism

[0109] The initial weights include, the destruction operator weight w d = 1, the repair operator w r = 1;

[0110] After each iteration, the weight is updated as:

[0111] w d = w d + a·ΔSNR

[0112] w r = w r + b·(1-MSE)

[0113] where a and b are learning rate coefficients, calibrated based on experiments;

[0114] (4) Iterative optimization

[0115] Generate 10 groups of candidate parameter combinations each round, calculate the SNR and MSE of each combination, and keep the best three solutions into the next round;

[0116] The termination condition is no SNR improvement for 5 consecutive rounds or reaching 50 iterations.

[0117] It should be noted that, in order to solve the problem of insufficient decomposition accuracy caused by the dependence of traditional parameter adjustment on manual experience, the adaptive large neighborhood search algorithm is used for global optimization of the decomposition layer number, the threshold coefficient and the noise amplitude, so that the parameter adaptability and the signal decomposition accuracy are improved, and the defects of mode aliasing and noise residue are overcome.

[0118] Step 2.3, based on the optimal low-frequency sub-band threshold coefficient output in step 2.2, generating each high-frequency sub-band hierarchical threshold parameter according to a preset proportion rule, and obtaining an optimal decomposition parameter combination.

[0119] Exemplarily, in step 2.3, generating each high-frequency sub-band hierarchical threshold parameter according to a preset proportion rule includes:

[0120] When the optimal decomposition layer number is 3,

[0121] The D2 sub-band threshold is: λ D,J-1 = 1.8·λ low .

[0122] The D3 sub-band threshold is: λ D,J = 2.2·λ low .

[0123] Exemplarily, the preset proportion rule can also be adjusted based on the real-time noise level. When the real-time noise level is lower than the preset low-noise threshold, the D3 and D4 sub-band threshold proportions are adjusted to 1.2 and 1.6, respectively.

[0124] Step 3, based on the decomposition parameters optimized in step 2, performing wavelet packet decomposition and EEMD decomposition in parallel, and screening effective IMF components through cross-correlation verification.

[0125] Preferably, in step 3, performing wavelet packet decomposition and EEMD decomposition in parallel includes:

[0126] (1) Wavelet packet decomposition

[0127] Performing wavelet packet decomposition on the ultrasonic signal preprocessed in step 1, and the decomposition layer number is dynamically determined according to step 2, wherein the high-frequency sub-band is processed using the hierarchical threshold parameters set in step 2;

[0128] Specifically, according to the decomposition layer number J determined in step 2 (exemplarily, when J = 3), using db4 wavelet basis to perform 3-layer wavelet packet decomposition on the preprocessed ultrasonic signal, and obtaining a low-frequency approximation sub-band and high-frequency detail sub-bands:

[0129] {A3, D1, D2, D3} = WPD(S' u (t), db4, J = 3)

[0130] Performing wavelet packet decomposition on the high-frequency sub-bands D J-1 and DJ Apply the high-frequency sub-band hierarchical threshold parameter λ D,J-1 and λ D,J set in step 2 respectively

[0131] (2) EEMD decomposition

[0132] Synchronously perform ensemble empirical mode decomposition, wherein the amplitude of the added white noise is proportional to the noise level estimation value in step 2, and the ensemble average number is dynamically adjusted based on the noise level;

[0133] Specifically, add white noise with amplitude A noise , perform K times of ensemble average to obtain m intrinsic mode function components IMF1, IMF m , …, IMF

[0134] Preferably, in step 3, the effective IMF component is screened by cross-correlation verification, which includes:

[0135] Calculate the time delay cross-correlation between the high-frequency sub-band of wavelet packet decomposition and each intrinsic mode function component obtained by ensemble empirical mode decomposition;

[0136] Retain the components with cross-correlation coefficients greater than a first threshold value, and eliminate the components with cross-correlation coefficients less than a second threshold value;

[0137] The first threshold value and the second threshold value are dynamically adjusted according to the noise level estimation value in step 2, and the higher the noise level, the lower the first threshold value.

[0138] Specifically, calculate the normalized cross-correlation coefficient of each intrinsic mode function component with the high-frequency sub-band D hi = D J-1 + D J :

[0139]

[0140] The rule for screening effective IMF components is to retain IMF m with ρ m > 0.6, eliminate IMF m with ρ m < 0.3, and perform wavelet thresholding on the remaining components, with the threshold strength being negatively correlated with ρ m :

[0141] Further, the first threshold value and the second threshold value of the normalized cross-correlation coefficient are dynamically fine-tuned according to the noise level estimation value σ in step 2, and when σ > 0.5σ max , the first threshold value is reduced to 0.55, and the second threshold value is increased to 0.35.

[0142] In view of the signal distortion problem caused by inaccurate modal component screening in the prior art, the present application effectively eliminates non-correlated noise components through parallel cross-correlation verification of wavelet packet decomposition and ensemble empirical mode decomposition, and ensures the integrity and fidelity of signal reconstruction.

[0143] Step 4: Extract the resonance frequency of the vibration signal to guide the frequency band enhancement of the ultrasonic signal.

[0144] Preferably, the step 4 comprises:

[0145] (1) Resonance frequency band calibration of the vibration signal

[0146] Based on the frequency distribution characteristics of the effective intrinsic mode function components screened out in step 3, the characteristic frequency band related to the resonance of the composite insulator structure is extracted from the vibration signal, and the range of the characteristic frequency band is dynamically determined according to the insulator sheath thickness and the material damping coefficient;

[0147] Verify the stability of the resonance frequency, and determine it as effective when the frequency fluctuation in a plurality of consecutive time windows is less than a preset tolerance;

[0148] Specifically, the vibration signal S v (t) is subjected to FFT to extract the main resonance frequency f v of the composite insulator, and the dynamic range thereof is determined by the following formula:

[0149]

[0150] Wherein, f v is in kHz, d is the sheath thickness in mm, and R is the resonance frequency characteristic value of the insulator sheath per unit thickness in kHz·mm;

[0151] The verification condition is that if L≥5 consecutive frames satisfy , it is determined to be effective.

[0152] (2) Cross-modal frequency band enhancement

[0153] Determine the target enhancement frequency band of the ultrasonic signal according to the resonance frequency band, and the width of the target frequency band is inversely proportional to the sheath thickness;

[0154] The low-frequency sub-band of the ultrasonic signal decomposed in step 3 is selectively weighted and amplified, and the weighting coefficient is negatively correlated with the number of effective intrinsic mode function components in step 3;

[0155] Specifically, the target enhancement frequency band (unit: kHz) is defined, wherein κ is the damping coefficient of the insulator sheath material (dimensionless), and the molecular constant 200 is a dimensionless composite constant with a unit of kHz·mm -1 ; the low-frequency sub-band A J(f) performing weighting:

[0156]

[0157] wherein the weighting coefficient a = 1.8 + 0.2E -1 , E is the number of effective IMF components reserved in step 3.

[0158] Further, the dynamic adjustment rule of the weighting coefficient is:

[0159]

[0160] The width of the target enhanced frequency band satisfies the constraint:

[0161]

[0162] For the problem of blind signal frequency band enhancement in a complex noise environment, the present application realizes the cooperative optimization of accurate enhancement of the target frequency band and noise suppression through vibration signal resonance frequency extraction and cross-modal frequency band linkage weighting, and significantly improves the extraction ability of defect features.

[0163] Step 5, the multi-modal signal after fusion processing generates a preliminary denoising signal, and after performing APA filtering post-processing and multi-modal cross-validation on it, the denoising ultrasonic signal that passes the verification is output as the denoising result.

[0164] Preferably, the step 5 comprises:

[0165] The enhanced low-frequency sub-band, the reserved effective intrinsic modal function component and the threshold-processed high-frequency sub-band are multi-scale fused and reconstructed to generate a denoised ultrasonic signal.

[0166] Further preferably, the multi-scale fusion reconstruction comprises wavelet packet inverse transform and weighted superposition of empirical mode components, and the weight is positively correlated with the signal-to-noise ratio of each component.

[0167] Specifically, the multi-scale fusion reconstruction comprises:

[0168] The enhanced low-frequency sub-band, the reserved effective intrinsic modal function component and the threshold-processed high-frequency sub-band are fused by wavelet packet inverse transform (IWPT) to generate a preliminary denoising signal:

[0169]

[0170] wherein the weight coefficient γ e = p m / ∑p m , p m is the cross-correlation coefficient calculated in step 3.

[0171] Preferably, the step 3, the APA filter post-processing of the preliminary denoising signal comprises:

[0172] The reconstructed preliminary denoising signal is input into an affine projection algorithm (APA) filter, the filter weight is updated by adaptive iteration, and a high-fidelity signal is output.

[0173] Specifically, it comprises:

[0174] (1) APA filter initialization

[0175] Initialize the filter order L, the projection order M, the step size parameter μ and the regularization factor δ; wherein the filter order is dynamically adjusted according to the signal sampling rate.

[0176] (2) Input signal matrix construction

[0177] Define the input signal vector:

[0178] S(n)=[S sd (n),S sd (n-1),K S sd (n-L+1)] T

[0179] Construct the projection matrix:

[0180] U(n)=[X(n),X(n-1),K,X(n-L+1)]

[0181] (3) Error calculation and weight update

[0182] The desired signal d(n)=S sd (n), and the filter output is:

[0183] y(n)=w T (n)S(n)

[0184] Error vector:

[0185] e(n)=d(n)-U T (n)w(n)

[0186] Weight update formula:

[0187] w(n+1)=w(n)+μ·U(n)[U T (n)U(n)+δI] -1 e(n)

[0188] (4) Dynamic step size adjustment

[0189] According to the local gradient of the signal Adaptive adjustment of step size:

[0190]

[0191] wherein, is preset as 50% of the maximum amplitude of the signal, and

[0192] (5) Output the filtered signal

[0193] Output the de-noised signal after filtering through N iterations, N is the length of the signal:

[0194]

[0195] Preferably, in step 3, performing multi-modal cross-validation on the filtered de-noised signal comprises:

[0196] (1) Time domain verification

[0197] Calculate the dynamic time warping distance between the de-noised filtered ultrasound signal and the standard discharge pulse template:

[0198] D DTW = min π ∑ (i,j)∈π |S APA (t i )-S ref (t j )|

[0199] The verification condition is D DTW ≤0.3·max(S ref ), max(S ref ) is the maximum amplitude of the signal in the standard discharge pulse template.

[0200] (2) Frequency domain verification

[0201] Calculate the energy proportion of the de-noised filtered ultrasound signal in the target enhanced frequency band [f v -Δf, f v +Δf]:

[0202]

[0203] The verification condition is R freq ≥60%.

[0204] (3) Energy verification

[0205] Calculate the peak-to-peak value retention rate of the de-noised and filtered ultrasound signal:

[0206]

[0207] The verification condition is 0.7≤R eg ≤1.2.

[0208] Preferably, the following processing is performed on the verification result:

[0209] If all three verifications pass, it is marked as a high-confidence signal and output;

[0210] If both verifications pass and the deviation of the failed items does not exceed the threshold of 20%, the parameter adaptive adjustment and re-verification are triggered;

[0211] If one or zero items pass the verification, or the deviation of the failed items exceeds 20%, the current signal is discarded and the device self-test is triggered.

[0212] Exemplarily, the parameter adaptive adjustment includes:

[0213] When only the time domain verification fails, update the standard discharge pulse template library;

[0214] When frequency domain or energy verification fails, the following parameter adjustments are triggered:

[0215] Reduce the layer threshold intensity coefficient and adjust the threshold parameter of the low-frequency subband from 0.6 times the noise level estimate to 0.4 times;

[0216] Reduce the segmentation threshold ratio in the segmentation threshold function and adjust 1.8σ to 1.2σ;

[0217] Relaxing the IMF screening threshold will retain ρ m IMF >0.6 m Instead, retain ρ m IMF >0.5 m .

[0218] It is worth noting that in order to address the problems of residual noise interference and verification simplicity in traditional denoising methods, the present invention achieves a dual improvement in signal smoothness and reliability through iterative filtering using an affine projection algorithm combined with multi-dimensional cross-validation in the time domain, frequency domain, and energy, providing a high-confidence basis for defect identification.

[0219] For example, in one embodiment of the present invention, the denoising method proposed by the present invention is applied, and the signals before and after denoising are as follows: Figure 2 、 3 As shown in Table 1, the comparison of denoising effects using different methods is shown in Table 1:

[0220] Table 1 Evaluation indicators

[0221]

[0222] In embodiment 2, the present invention provides a joint denoising system based on adaptive large neighborhood search and modal decomposition. Based on the joint denoising method based on adaptive large neighborhood search and modal decomposition described in embodiment 1, the system includes:

[0223] a signal acquisition and preprocessing module, configured to synchronously acquire ultrasonic signals and vibration signals, and perform baseline correction and band-pass filtering on the signals;

[0224] a parameter optimization module, configured to dynamically optimize a wavelet packet decomposition layer number, a threshold parameter, and a noise amplitude of ensemble empirical mode decomposition based on an adaptive large neighborhood search algorithm;

[0225] a signal decomposition module, configured to perform wavelet packet decomposition and ensemble empirical mode decomposition in parallel, and perform hierarchical threshold processing on high-frequency subbands;

[0226] a component screening module, configured to screen valid intrinsic mode function components by cross-correlation verification, and eliminate noise interference components;

[0227] a frequency band enhancement module, configured to extract a resonance frequency of the vibration signal and guide weighted amplification of a target frequency band of the ultrasonic signal;

[0228] a post-processing filtering module, configured to perform affine projection algorithm filtering on a reconstructed signal, eliminate residual noise, and improve signal smoothness;

[0229] a multi-modal verification module, configured to perform time-domain, frequency-domain, and energy multi-dimensional cross-verification on a denoising signal, and ensure reliability of an output signal.

[0230] The present disclosure can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.

[0231] It should be noted that the above examples are only used to illustrate the technical solutions of the present application, and not to limit it. Although the present application has been described in detail with reference to the above examples, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application. Any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.

Claims

1. A joint denoising method based on adaptive large neighborhood search and modal decomposition, characterized in that: The steps include: The ultrasonic and vibration signals of the insulator under test are synchronously collected and preprocessed. The noise level is dynamically estimated based on the power spectral density of the preprocessed ultrasonic signal, and the decomposition parameters are optimized using an adaptive large neighborhood search algorithm. Based on the optimized decomposition parameters, the ultrasonic signal is subjected to wavelet packet decomposition to obtain low-frequency and high-frequency sub-bands. The high-frequency sub-band is subjected to hierarchical threshold processing, and ensemble empirical mode decomposition is performed in parallel. Effective intrinsic mode function components are screened through cross-correlation verification. Extract the resonant frequency of the preprocessed vibration signal, perform weighted enhancement on the low-frequency sub-band, and dynamically adjust the high and low frequency sub-band threshold parameters according to the resonant frequency. The fused signal generates a preliminary denoised signal, which includes an enhanced low-frequency subband, an effective intrinsic mode function component and a high-frequency subband after hierarchical threshold processing. The preliminary denoised signal is filtered using an affine projection algorithm and subjected to multimodal cross-validation. The verified ultrasound signal is output as the final denoising result.

2. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 1, characterized in that: The synchronous acquisition includes: Ultrasonic and vibration signals are collected through a hardware-triggered synchronization mechanism, ensuring that the time alignment error is less than 1 microsecond. The ultrasonic signal is corrected for baseline drift, and the vibration signal is subjected to band-pass filtering with resonance frequency preservation.

3. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 1, characterized in that: The dynamic estimation of noise level and optimization of decomposition parameters include: Perform power spectral density analysis on the preprocessed ultrasonic signal, calculate the average power spectral density within the preset noise frequency band, and use the square root of the average value as the real-time noise level estimation value; An adaptive large neighborhood search algorithm is used to dynamically optimize the decomposition parameters. The objective function is to maximize the signal-to-noise ratio. The decomposition parameters include the number of wavelet packet decomposition layers, the low-frequency subband threshold coefficient, the number of EEMD sets, and the white noise amplitude. The layered threshold parameters include the low-frequency subband threshold coefficient and the high-frequency subband layered threshold. Based on the optimal low-frequency sub-band threshold coefficient, the high-frequency sub-band layered threshold parameters are generated according to the preset proportion rule to obtain the optimal decomposition parameter combination.

4. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 1, characterized in that: The optimization of decomposition parameters by using the adaptive large neighborhood search algorithm includes: Randomly increase or decrease the number of decomposition layers, and randomly perturb and adjust the low-frequency subband threshold coefficient and white noise amplitude; Imposing value range constraints on parameter combinations, including limiting the maximum number of decomposition layers, the upper and lower limits of the low-frequency sub-band threshold coefficients, and the reasonable range of white noise amplitude; Dynamically adjust the weight distribution of destruction operators and repair operators based on historical optimization performance; Each round of iteration generates 10 sets of candidate parameter combinations, and the optimal solution is selected by calculating the signal-to-noise ratio and mean square error of each combination; If there is no improvement in the signal-to-noise ratio after five consecutive iterations or the preset number of iterations is reached, the optimization is terminated and the optimal wavelet packet decomposition layer number, low-frequency subband threshold coefficient, EEMD set number and white noise amplitude combination are output.

5. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 1, characterized in that: The screening of effective intrinsic mode function components includes: Calculate the normalized cross-correlation coefficient between the high-frequency subband signal obtained by wavelet packet decomposition and all the intrinsic mode function components generated by the ensemble empirical mode decomposition; Dynamically setting a first screening threshold and a second screening threshold according to a real-time noise level estimation value, wherein the first threshold is used to retain high-correlation components and the second threshold is used to eliminate low-correlation components; retaining the intrinsic mode function components whose normalized cross-correlation coefficients are greater than a first screening threshold, and removing the components whose normalized cross-correlation coefficients are less than a second screening threshold; The components whose normalized cross-correlation coefficients are between the first threshold and the second threshold are subjected to wavelet threshold reprocessing, wherein the threshold intensity is negatively correlated with the current noise level.

6. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 1, characterized in that: The target frequency band enhancement and threshold adjustment include: Extracting the main resonant frequency of the vibration signal, and determining a target enhanced frequency band of the ultrasonic signal based on the main resonant frequency and the thickness of the insulator sheath, wherein the width of the target frequency band is inversely proportional to the sheath thickness; The low-frequency sub-band of the ultrasonic signal within the target enhanced frequency band is selectively weighted amplified, and the weighting coefficient is negatively correlated with the number of the screened effective intrinsic mode function components.

7. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 1, characterized in that: The performing of affine projection algorithm filtering according to the preliminary denoising signal comprises: Initializing a filter order, a projection order, a step size parameter, and a regularization factor, wherein the filter order is dynamically adjusted according to the sampling rate of the preliminary denoised signal; Construct the input signal matrix and projection matrix, and update the filter weights through adaptive iteration; Dynamically adjusting a step size parameter according to a local gradient of the preliminary denoised signal; After a preset number of iterations, the filtered denoised signal is output.

8. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 6, characterized in that: The performing multimodal cross validation on the filtered denoised signal comprises: Time domain verification, calculating the time domain dynamic time warping distance between the filtered denoised signal and the standard discharge pulse template, and verifying whether the dynamic time warping distance is less than a preset time domain tolerance value; Frequency domain verification, calculating the energy proportion of the filtered denoised signal in the target enhancement frequency band, and verifying whether the energy proportion is higher than a preset frequency domain threshold; Energy verification, calculating the peak-to-peak value retention rate of the filtered denoised signal and the original ultrasonic signal, and verifying whether the peak-to-peak value retention rate is higher than a preset energy threshold.

9. The joint denoising method based on adaptive large neighborhood search and modal decomposition according to claim 8, characterized in that: The processing of multimodal cross-validation results includes: If all three verifications are passed, it is marked as a high-confidence signal and the final denoising result is output; If at least one verification item fails and the deviation of the failed items does not exceed 20%, the decomposition parameter adaptive adjustment is triggered and the denoising and verification process is re-executed. The parameter adjustment includes lowering the layer threshold intensity coefficient, relaxing the intrinsic mode function screening threshold, or updating the standard discharge pulse template; If at least one verification item fails and the deviation of the failed items exceeds 20%, the current filtered denoised signal is discarded and the device self-test is triggered.

10. A joint denoising system based on adaptive large neighborhood search and modal decomposition, based on the joint denoising method based on adaptive large neighborhood search and modal decomposition according to any one of claims 1 to 9, characterized in that: The system includes: Signal acquisition and preprocessing module, used to synchronously acquire ultrasonic signals and vibration signals, and perform baseline correction and bandpass filtering on the signals; Parameter optimization module, which is used to dynamically optimize the number of wavelet packet decomposition layers, threshold parameters, and noise amplitude of ensemble empirical mode decomposition based on an adaptive large neighborhood search algorithm; Signal decomposition module, which is used to perform wavelet packet decomposition and ensemble empirical mode decomposition in parallel, and perform hierarchical threshold processing on high-frequency subbands; Component screening module, used to screen effective intrinsic mode function components through cross-correlation verification and eliminate noise interference components; Frequency band enhancement module, used to extract the resonance frequency of the vibration signal and guide the weighted amplification of the target frequency band of the ultrasonic signal; Post-processing filtering module, used to perform affine projection algorithm filtering on the reconstructed signal; The multimodal verification module is used to perform multi-dimensional cross-validation of the denoised signal in the time domain, frequency domain, and energy domain.

Images