Joint denoising method and system based on multi-source noise suppression and sparse representation

Abstract

The invention provides a joint denoising method and system based on multi-source noise suppression and sparse representation, and belongs to the technical field of noise suppression in ultrasonic detection of power equipment, and the method comprises the steps: synchronously collecting an ultrasonic echo signal, a vibration signal and an electromagnetic signal of a composite insulator; decomposing the ultrasonic echo signal into a plurality of modal components and undecomposed signal margins; extracting a dominant frequency component in the vibration signal, and identifying a resonance noise feature overlapped with the frequency band of the ultrasonic signal; detecting a pulse interference event in the electromagnetic signal, marking occurrence time and a frequency range, and generating a pulse event feature; performing frequency band suppression and time domain threshold processing on the modal component according to the resonance noise characteristics and the pulse event characteristics; performing sparse representation coding on the processed modal components, executing multiple rounds of iterative screening and orthogonalization processing, and screening an atom set matched with defect echoes; and reconstructing the de-noised ultrasonic signal by using the screened atom set, and combining the de-noised ultrasonic signal with the undecomposed signal margin to output a final result.

Description

Technical Field

[0001] The present invention belongs to the technical field of noise suppression in ultrasonic detection of power equipment, and more specifically, relates to a joint denoising method and system based on multi-source noise suppression and sparse representation. Background Art

[0002] Composite insulators are the core insulating medium of high-voltage transmission lines, and their internal structural integrity is directly related to the safe operation of the power grid. Over long-term service, hidden defects such as core rod fracture and interface delamination gradually form due to the synergistic effects of mechanical stress and electrochemical corrosion, making them difficult to effectively identify using traditional visual inspection methods.

[0003] While infrared detection technology, currently widely used in the power industry, offers the advantage of non-contact detection, it lacks sensitivity for detecting internal defects and is susceptible to interference from environmental thermal disturbances. Ultrasonic detection technology, with its ability to locate deep-seated defects, is considered an ideal alternative. However, practical engineering applications face significant challenges: the complex electromagnetic environment of substations and multi-source interference from mechanical vibrations of equipment severely degrade ultrasonic echo signals, making defect feature extraction difficult.

[0004] Existing noise suppression methods are mostly based on a single signal processing dimension, making them incapable of coping with the coupled effects of multimodal interference. Traditional frequency-domain filtering can easily lose effective components when separating overlapping signals from vibration noise and defect echoes, while time-domain thresholding methods lack the adaptive processing capabilities for non-stationary electromagnetic pulse interference.

[0005] The existing technical document (CN118013191A) discloses a joint denoising method in the field of insulator anomaly detection technology, which achieves signal noise reduction through least squares filtering preprocessing, CEEMDAN modal decomposition and wavelet integrated threshold processing. However, it is only based on the frequency domain decomposition and multi-scale analysis of a single ultrasonic signal, which can easily cause the loss of effective components when suppressing mechanical resonance noise with overlapping frequency bands. In addition, the fixed-weight wavelet integrated threshold calculation method lacks dynamic adaptability to the changes in time domain noise characteristics caused by non-stationary electromagnetic interference, resulting in weak defect echoes being excessively suppressed during the reconstruction process. Summary of the Invention

[0006] In order to address the deficiencies in the prior art, the present invention provides a joint denoising method and system based on multi-source noise suppression and sparse representation.

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

[0008] A first aspect of the present invention provides a joint denoising method based on multi-source noise suppression and sparse representation, comprising the following steps:

[0009] Synchronously collect ultrasonic echo signals, vibration signals and electromagnetic signals of composite insulators;

[0010] decomposing the ultrasonic echo signal into a plurality of modal components and an undecomposed signal residual;

[0011] extracting a main frequency component in the vibration signal and identifying a resonance noise feature overlapping with a frequency band of the ultrasonic signal;

[0012] detecting a pulse interference event in the electromagnetic signal and marking a time of occurrence and a frequency range, and generating a pulse event feature;

[0013] performing frequency band suppression and time domain threshold processing on the modal components according to the resonance noise feature and the pulse event feature;

[0014] performing sparse representation coding on the processed modal components, and performing multi-round iteration screening and orthogonalization processing to screen an atomic set matching the defect echo;

[0015] reconstructing the denoised ultrasonic signal using the screened atomic set, and merging and outputting a final result with the undecomposed signal residual.

[0016] Optionally, the decomposing the ultrasonic echo signal into a plurality of modal components and an undecomposed signal residual comprises:

[0017] adding a preset number of white noises to the original ultrasonic echo signal to generate a set of noisy signals;

[0018] performing empirical mode decomposition on each noisy signal to extract a first-order modal component;

[0019] calculating an average value of the first-order modal components of all noisy signals as a final first modal component;

[0020] removing the first modal component from the original signal to generate a first-order undecomposed signal residual;

[0021] iteratively superimposing previous modal noises in the undecomposed signal residual and repeating the decomposition process to extract subsequent modal components step by step;

[0022] stopping the decomposition when the undecomposed signal residual meets a preset termination condition, and outputting all modal components and a final undecomposed signal residual.

[0023] Optionally, the extracting a main frequency component in the vibration signal and identifying a resonance noise feature overlapping with a frequency band of the ultrasonic signal comprises:

[0024] performing spectral analysis on the vibration signal to determine its energy distribution feature, and identifying a frequency component with the maximum energy as the main frequency;

[0025] comparing the main frequency with a working frequency band range of the ultrasonic detection system;

[0026] when the main frequency is within the working frequency band range, marking it as a resonance noise source.

[0027] Optionally, detecting a pulse interference event in an electromagnetic signal and generating a pulse event feature includes:

[0028] Preprocessing the collected electromagnetic signals, including baseline correction and bandpass filtering;

[0029] Identify the time window of transient pulse signal by dynamic threshold detection method;

[0030] Extract the time domain amplitude characteristics and frequency domain main frequency bandwidth characteristics of pulse interference;

[0031] The time window is associated with the main frequency bandwidth feature and stored to generate a pulse event feature containing time-frequency features.

[0032] Optionally, the performing frequency band suppression on the modal component according to the resonance noise characteristics includes:

[0033] Screening modal components that are within a preset tolerance range with respect to the main vibration frequency;

[0034] Dynamically calculate the attenuation coefficient based on the energy ratio of the vibration signal and the modal component;

[0035] The attenuation coefficient is applied to the screened modal components to suppress resonance noise.

[0036] Optionally, performing time-domain threshold processing on the modal component according to the pulse event characteristics includes:

[0037] Identify signal segments in the high-frequency modal components that overlap with the electromagnetic pulse time window;

[0038] Calculating a noise energy statistic within the time window as a threshold reference;

[0039] The signal components with coefficients lower than a preset threshold are zeroed.

[0040] Optionally, performing sparse representation encoding on the processed modal components, performing multiple rounds of iterative screening and orthogonalization processing, and screening the set of atoms matching the defect echo includes:

[0041] Construct a time-frequency atomic preset feature matching library containing composite insulator defect echo characteristics;

[0042] Initializing the undecomposed signal residual to the processed modal component and clearing the atomic support set;

[0043] Iteratively select multiple atoms that are most correlated with the unresolved signal residue to update the support set;

[0044] The combination coefficients of atoms in the support set are optimized by the least squares method;

[0045] The undecomposed signal residue is updated to the difference between the original signal and the current reconstructed signal until the preset sparsity or residual energy threshold is met.

[0046] Optionally, the iterative selection of a plurality of atoms most correlated with the unresolved signal remainder comprises:

[0047] Calculate the inner product of the unresolved signal residual and all atoms in the preset feature matching library;

[0048] Filter the first k atoms with the highest inner product values ​​and add them to the atomic support set;

[0049] The sparse coefficients are optimized according to the atomic support set to reconstruct the signal components.

[0050] Optionally, reconstructing the denoised ultrasonic signal using the screened atomic set and combining it with the undecomposed signal residue to output a final result includes:

[0051] According to the atom set and sparse coefficients obtained by sparse representation coding, the corresponding time-frequency atoms are extracted from the preset feature matching library;

[0052] Weighted superposition of the time-frequency atoms according to the combination coefficients generates a preliminary reconstructed signal;

[0053] Performing amplitude calibration on the preliminary reconstructed signal and the decomposed undecomposed signal residue and then superimposing them;

[0054] The superimposed signal is smoothed and filtered to eliminate high-frequency oscillation artifacts and output the final denoised ultrasonic signal.

[0055] A second aspect of the present invention provides a joint denoising system based on multi-source noise suppression and sparse representation, based on the joint denoising method based on multi-source noise suppression and sparse representation described in the first aspect of the present invention, the system comprising:

[0056] Multi-source signal acquisition module, used to synchronously acquire ultrasonic echo signals, vibration signals and electromagnetic signals;

[0057] A signal decomposition module, used for decomposing ultrasonic signals into modal components;

[0058] Vibration analysis module, used to extract the main vibration frequency and generate resonance noise characteristic parameters;

[0059] Electromagnetic interference analysis module, used to detect and mark the time-frequency characteristics of electromagnetic pulse events;

[0060] Modal fusion denoising module, used to perform frequency band suppression and time domain threshold processing on modal components;

[0061] Sparse coding module, used to perform multiple rounds of iterative screening and orthogonalization processing to screen defect echo atoms;

[0062] The signal reconstruction module is used to generate a denoised ultrasonic signal.

[0063] Compared with the prior art, the beneficial effects of the present invention include at least:

[0064] 1. The present invention solves the problem of blind spots in noise suppression of single sensor data in complex substation environments through a multi-source signal synchronous acquisition mechanism, realizes the spatiotemporal alignment and collaborative analysis of ultrasonic, vibration and electromagnetic signals, and provides a data basis for multimodal interference separation.

[0065] 2. The adaptive decomposition method provided by the present invention overcomes the modal aliasing defect caused by the fixed basis function of traditional wavelet decomposition. It reduces the modal aliasing index by more than 60% through the noise injection mechanism, effectively improving the separation accuracy of high-frequency noise and defect echoes.

[0066] 3. The sparse representation coding system constructed by the present invention solves the problem that the traditional threshold method does not adequately retain the features of non-stationary signals. Through data-driven optimization of the preset feature matching library, it improves the noise suppression capability while enhancing the generalization of detection of new composite insulator materials.

[0067] 4. The vibration-electromagnetic dual-dimensional noise suppression strategy proposed in the present invention breaks through the technical limitations of single frequency domain or time domain denoising. Through the synergistic mechanism of vibration main frequency suppression and electromagnetic pulse time domain threshold processing, it realizes the joint attenuation of multi-source interference while fully retaining the time delay and amplitude characteristics of the defect echo.

[0068] 5. The hardware synchronization acquisition module designed in this invention solves the problem of feature misalignment caused by multi-sensor signal delay deviation. Through GPS clock synchronization and three-dimensional spatial calibration technology, it ensures the precise temporal and spatial correlation of ultrasonic detection data with vibration and electromagnetic interference events. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 is a flow chart of a method provided according to an embodiment of the present invention;

[0070] Figure 2 3 is a schematic diagram of denoising comparison provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0071] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0072] In embodiment 1, the present invention provides a joint denoising method based on multi-source noise suppression and sparse representation, such as Figure 1 As shown, the following steps are included:

[0073] Step 1: synchronously collect ultrasonic echo signals, vibration signals and electromagnetic signals of the composite insulator.

[0074] Preferably, in step 1, collecting the ultrasonic echo signal of the composite insulator includes:

[0075] Step 1.1, select an ultrasonic transducer suitable for actual testing based on the actual appearance data and internal structure of the composite insulator to be tested;

[0076] Exemplarily, the step 1.1 includes:

[0077] Select an ultrasonic transducer suitable for insulator inspection. The frequency selection needs to balance resolution and penetration depth. For composite insulators, the sound velocity is about 1000-1500m / s, and 2MHz can be used with a detection depth of >50mm, or 5MHz for thin-walled insulators with a resolution of 0.3mm.

[0078] Secondly, the probe type should preferably be a contact angle probe with a wear-resistant protective layer and a refraction angle of 45°-60° to detect inclined cracks, or an immersion probe with a focal length of 50-100mm to avoid surface wear;

[0079] Finally, the interface must match the ultrasonic detector's excitation voltage of 100-400V and a sampling rate ≥5 times the center frequency, and use silicone grease coupling agent to optimize acoustic energy transmission;

[0080] In addition, for UHV insulators, phased array probes with 16-32 array elements can be used to achieve electronic scanning and improve detection efficiency.

[0081] Step 1.2: The transmitter transmits an ultrasonic signal, and the receiver receives and filters the ultrasonic echo signal after propagating through the composite insulator.

[0082] Exemplarily, the step 1.2 includes:

[0083] First, select an ultrasonic probe with a center frequency that matches the insulator material characteristics, usually 0.5-10MHz, and fit it tightly to the insulator surface using a coupling agent such as silicone grease;

[0084] Secondly, configure the pulse transmitter / receiver, set the pulse width to 1-3 cycles, the voltage to 100-400V, and the trigger sampling rate to more than 5 times the probe center frequency;

[0085] Then, the probe is moved along a grid path on the surface of the insulator, each measurement point emits an ultrasonic wave and receives a return signal, and the position coordinates are recorded synchronously.

[0086] Then, a band-pass filter is used to suppress environmental noise, with a passband range of 0.8-1.2 times the center frequency of the probe, and the signal-to-noise ratio is improved by time-domain averaging, typically 16-64 times of superposition.

[0087] Step 1.3, standardizing the filtered ultrasonic echo signal;

[0088] Further preferably, the step 1.3 comprises:

[0089] The filtered signal is normalized to eliminate amplitude drift and baseline offset.

[0090] The key feature points of the signal are labeled, such as the time delay and amplitude of the defect echo, and stored in a standardized format such as MAT or TDMS.

[0091] Exemplarily, the step 1.3 comprises:

[0092] The original signal, position information and acquisition parameters such as gain and sampling rate are stored synchronously in a standardized format such as MAT or TDMS, and the time domain feature echo delay time, amplitude decay and other information of possible defect areas such as cracks and pores are labeled; the whole process needs to be carried out in a shielding environment to avoid electromagnetic interference, and the probe sensitivity needs to be calibrated regularly to ensure data consistency.

[0093] It should be noted that the standard preprocessing of the ultrasonic echo signal before denoising can effectively improve the quality of the original signal, by suppressing random noise, eliminating hardware interference and correcting signal drift, to provide clean input for subsequent processing;

[0094] The adaptive noise injection mechanism overcomes the mode aliasing problem of traditional EMD, and obtains purer modal components by integrating the decomposition results under multiple noise disturbances, realizing the natural separation of signal and noise in time-frequency domain;

[0095] This combination not only preserves the non-stationary characteristics of the signal (such as sudden defect echoes), but also significantly improves the accuracy of subsequent denoising methods such as sparse representation through adaptive frequency band decomposition, especially for weak ultrasonic echoes with a signal-to-noise ratio below 10dB, which can achieve an SNR improvement of more than 15dB while perfectly preserving the time delay and amplitude information of defect features, providing a reliable analysis foundation for industrial non-destructive testing.

[0096] Step 2, decomposing the standardized ultrasonic echo signal into multiple modal components and undecomposed signal residuals.

[0097] Step 2.1: Add white noise to the original ultrasonic signal x(t). In order to construct a stable mode, it is necessary to add noise to the signal multiple times and average it to generate a noisy signal.

[0098]

[0099] Among them, e (k) (t) is the k-th added white noise sequence, ε is the amplitude of the noise, k = 1, 2, ..., K is the number of noise additions;

[0100] Step 2.2, first mode extraction, use standard EMD for each noisy signal Decompose and extract the first modal component IMF1 (k) (t), after multiple iterations, calculate the average first mode:

[0101]

[0102] Step 2.3, residual calculation, remove the first modal component from the original ultrasonic signal and calculate the residual r1′(t) of the remaining signal:

[0103] r1′(t)=x(t)-IMF1(t)

[0104] The residual is the signal after subtracting the first mode from the original signal;

[0105] Step 2.4, adaptive noise introduction and next mode extraction. For subsequent modes, new modes are extracted by introducing adaptive noise on the residual.

[0106] In each step, the preceding modal noise is superimposed on the undecomposed signal residual r1′(t) to generate a new undecomposed signal residual:

[0107]

[0108] The new unresolved signal residue Perform EMD decomposition to extract the second modal component And calculate its mean mode:

[0109]

[0110] Step 2.5, repeat step 2.4 and iteratively calculate the n-th order residual:

[0111] r n (t) = r n-1 (t)-IMF n (t)

[0112] Until the unresolved signal residue r n(t) is a monotonic function or its energy is lower than the preset threshold, and the denoised signal is output.

[0113]

[0114] Step 3: Extract the main frequency component in the vibration signal and identify the resonance noise characteristics that overlap with the ultrasonic signal frequency band.

[0115] Preferably, the step 3 includes:

[0116] Step 3.1, perform Fourier transform on the vibration signal and calculate its spectral energy distribution;

[0117] Specifically, the vibration signal v(t) is subjected to Fourier transform to calculate the spectral energy:

[0118]

[0119] Step 3.2, extract the frequency component with the largest energy in the spectrum as the main frequency;

[0120] Specifically, determining the main frequency component includes:

[0121]

[0122] Among them, f s is the vibration signal sampling frequency.

[0123] It is worth emphasizing that, in order to address the interference residue caused by the overlap of mechanical vibration noise and defect echo frequency bands, the present invention realizes the accurate identification of the resonance noise source through main frequency feature extraction and frequency band correlation analysis.

[0124] Step 3.3: When the main frequency is within the operating frequency band of the ultrasonic detection system, it is determined to be a resonance noise source and marked; the lower limit and upper limit of the ultrasonic detection frequency band are determined according to the ultrasonic transducer calibration parameters.

[0125] Specifically, the conditions for determining the source of resonance noise are:

[0126] f min ≤f V ≤f max

[0127] where f min and f max are the minimum operating frequency and maximum operating frequency of the ultrasonic detection system respectively.

[0128] Step 4: Detect pulse interference events in electromagnetic signals, mark the occurrence time and frequency range, and generate pulse event characteristics.

[0129] Preferably, step 4 includes:

[0130] Step 4.1, preprocessing the collected electromagnetic signals;

[0131] Specifically, the electromagnetic signal is subjected to m (t) Baseline correction and bandpass filtering to filter out low-frequency environmental interference and high-frequency thermal noise, and obtain the pre-processed electromagnetic signal e' m (t).

[0132] Step 4.2, identifying electromagnetic pulse events through sliding window energy detection method;

[0133] Specifically, the preprocessed electromagnetic signal is divided into segments of window duration, and the window energy is calculated:

[0134]

[0135] Among them, T ω is the window duration;

[0136] To set dynamic thresholds:

[0137] Threshold(k)=μ E +3σ E

[0138] Among them, μ E is the mean energy of the first N windows, σ E is the standard deviation;

[0139] If the window energy exceeds the dynamic threshold, it is determined to be a pulse interference event, and its start and end times are recorded as the time window:

[0140] T pulse =[kT ω ,(k+1)T ω ]

[0141] It should be noted that, in order to address the problem of missed detection of transient pulse interference by the traditional threshold method, the present invention significantly improves the capture sensitivity of pulse events in strong electromagnetic environments through dynamic threshold adjustment and time-frequency joint analysis.

[0142] Step 4.3, extracting the time domain features of the pulse interference event;

[0143] Specifically, calculate the peak amplitude:

[0144]

[0145] Calculate the pulse duration:

[0146] T dur =t end -t start

[0147] Among them, t start and t end The pulse amplitude exceeds and falls below 0.5A for the first time peak time point.

[0148] Step 4.4, perform spectrum analysis on the pulse interference event;

[0149] Specifically, the pulse segment is subjected to a short-time Fourier transform (STFT):

[0150]

[0151] Among them, the pulse segment is e' m (t)| t∈Tpulse , w(t) is the Hanning window function;

[0152] Extract the main frequency:

[0153]

[0154] Calculate the main frequency bandwidth B main It is the frequency range when the spectrum energy drops by 3dB.

[0155] Step 4.5: associate the time window of the pulse interference event with the main frequency range and store it to generate the pulse event feature for subsequent denoising processing;

[0156] Specifically, the storage time window T pulse , main frequency f main and bandwidth B main .

[0157] Step 5: Perform modal fusion denoising on the modal components of the ultrasonic signal according to the resonance noise characteristics and the pulse event characteristics.

[0158] Preferably, the step 5 includes:

[0159] Step 5.1, perform frequency band suppression on the modal component corresponding to the main frequency of vibration;

[0160] Specifically, filter out modal components with frequencies close to the main vibration frequency and set the frequency tolerance range:

[0161] IMF i ∈S vib If and only if

[0162] Where Δf = 0.1f v is the frequency tolerance;

[0163] The attenuation coefficient is dynamically calculated based on the ratio of the vibration signal energy to the modal component energy:

[0164]

[0165] Among them, β∈[0.1,0.5] is the adjustable attenuation factor, E v is the energy of the vibration signal in the resonance frequency band, is the energy of the modal component in the same frequency band, specifically:

[0166]

[0167] Apply damping to the modal components:

[0168] IMF' i (t) = λ i IMF i (t)

[0169] Step 5.2, performing time domain threshold enhancement on the high-frequency modal components within the electromagnetic pulse interference time window;

[0170] Specifically, filter out high-frequency modal components:

[0171] IMF j ∈S em If and only if

[0172] Wherein, for example, the preset threshold value f th =0.8f center , f center is the center frequency of the ultrasound probe;

[0173] Calculate the standard deviation of the noise within the EMP window:

[0174]

[0175] Time domain thresholding:

[0176]

[0177] Wherein, χ=3 is the threshold coefficient.

[0178] Step 5.3: record the processed modal components for subsequent sparse coding and reconstruction;

[0179] Specifically, the processed IMF' i (t) and the IMF” j (t) Merge into a new modal set {IMF proc (t)}.

[0180] It is worth noting that in order to address the feature distortion caused by a single denoising dimension, the present invention uses a synergistic mechanism of vibration frequency band suppression and electromagnetic time domain threshold processing to fully preserve the time delay characteristics of the defect echo while suppressing noise.

[0181] Step 6: Perform sparse representation encoding on the denoised modal components, perform multiple rounds of iterative screening and orthogonalization processing, and screen the set of atoms that match the defect echo.

[0182] Preferably, step 6 includes:

[0183] Step 6.1, constructing a preset feature matching library, which contains time-frequency atoms that match the echo characteristics of composite insulator defects;

[0184] Specifically, construct the Gabor preset feature matching library D = {g γ (t)} γ∈Γ , where the atomic parameters γ = (s,u,f) represent scale, time shift, and frequency respectively.

[0185] Step 6.2, initialize the undecomposed signal residual to the processed modal component and clear the atomic support set;

[0186] Specifically, initialize the residual r0 = IMF proc (t), support set Iteration number k = 1;

[0187] Step 6.3, iteratively select multiple atoms that are most correlated with the unresolved signal residue;

[0188] Further preferably, the step 6.3 includes:

[0189] Calculate the inner product of the residual and all atoms in the dictionary, and select the top k atoms with the highest correlation;

[0190] Add the selected atoms to the support set and record their index;

[0191] Specifically, take the first two maximum values, calculate the inner product and filter the first k = 2 atoms:

[0192]

[0193] Update support set:

[0194]

[0195] Step 6.4, optimize the combination coefficients of atoms in the current support set by the least squares method;

[0196] Specifically, optimize the combination coefficient:

[0197]

[0198] Among them, the IMF proc is the modal component after processing in step 5, The index in the dictionary belongs to the support set Sk submatrix of .

[0199] Step 6.5, update the undecomposed signal residual to the difference between the original signal and the current reconstructed signal;

[0200] Specifically, update the residuals:

[0201]

[0202] Step 6.6: Repeat steps 6.3 to 6.5 until the preset sparsity or residual energy threshold is met;

[0203] Specifically, the termination condition is:

[0204] If k ≥ L0 (L0 is the preset sparsity) or ‖r k ||2≤τ||IMF proc || 2 (residual energy ratio threshold), stop the iteration, where τ is the residual convergence threshold.

[0205] Step 6.7: Output the sparse coefficient vector and the corresponding atom set for signal reconstruction.

[0206] Specifically, the output sparse coefficient α = ∪α k And the atomic index set S = S k .

[0207] It should be pointed out that in order to address the problem of mismatch between the atomic library and defect characteristics in traditional sparse coding, the present invention significantly improves the matching probability of defect echo atoms by embedding prior knowledge of the composite insulator acoustic propagation model.

[0208] Sparse representation uses a preset feature matching library to perform atomic-level decomposition of each modal component, which can accurately match the transient features in the signal (such as the time-frequency structure of ultrasonic defect echoes), thereby effectively distinguishing between real signal components and noise; feature extraction based on sparse coefficients can quantify the physical meaning of each modal component, that is, the large-value sparse coefficients in the high-frequency mode correspond to the defect echo, and the low-frequency modal coefficients reflect the background characteristics of the material, providing interpretable quantitative indicators for defect identification; in addition, sparse representation can achieve noise filtering and avoid signal distortion caused by traditional threshold methods by retaining significant coefficients and suppressing small coefficients, which is particularly suitable for retaining the time delay and amplitude characteristics of weak defects; finally, the feature vector composed of sparse coefficients has low dimension and clear physical meaning, and can be directly used as the input of machine learning classifiers, significantly improving the accuracy and efficiency of defect detection.

[0209] Step 7: Use the screened atomic set to reconstruct the denoised ultrasonic signal and combine it with the undecomposed signal residue to output the final result.

[0210] Preferably, the step 7 includes:

[0211] Step 7.1, extracting corresponding atoms from a preset feature matching library based on the sparse coefficient vector and the atom index set;

[0212] Step 7.2, linearly combine the sparse coefficients with the atoms to reconstruct the denoised modal components;

[0213] Specifically, reconstruct the denoised modal components:

[0214]

[0215] Among them, S is the atomic index set output in step 6, α i The sparse coefficients optimized in step 6, is the index γ in the dictionary D i The corresponding atoms.

[0216] Step 7.3, add the reconstructed modal components to the undecomposed signal residue retained in step 2 to generate a preliminary denoised signal;

[0217] Step 7.4, post-processing the preliminary denoised signal, including amplitude calibration and smoothing filtering;

[0218] Step 7.5, output the final denoised ultrasonic signal to complete the noise suppression of composite insulator defect detection. For example, Figure 2 As shown, the signal denoising comparison is performed using the method of the present invention.

[0219] In embodiment 2, the present invention provides a joint denoising system based on multi-source noise suppression and sparse representation. Based on the joint denoising method based on multi-source noise suppression and sparse representation described in embodiment 1, the system includes:

[0220] Multi-source signal acquisition module, used to synchronously collect ultrasonic echo signals, vibration signals and electromagnetic signals;

[0221] A signal decomposition module, for decomposing the ultrasonic signal into modal components;

[0222] Vibration analysis module, used to extract the main vibration frequency and identify the resonant noise frequency band;

[0223] Electromagnetic interference analysis module, used to detect electromagnetic pulse interference events and mark time-frequency characteristics;

[0224] Modal fusion denoising module, used to perform frequency band suppression and time domain threshold processing on modal components based on the main vibration frequency and electromagnetic pulse characteristics;

[0225] Sparse coding module, used to perform multiple rounds of iterative screening and orthogonalization processing, build a preset feature matching library and screen the atom set;

[0226] The signal reconstruction module is used to reconstruct the denoised ultrasonic signal.

[0227] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0228] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A joint denoising method based on multi-source noise suppression and sparse representation, characterized in that: The steps include: Synchronously collect ultrasonic echo signals, vibration signals and electromagnetic signals of composite insulators; Decomposing the ultrasonic echo signal into multiple modal components and undecomposed signal residue; Extract the main frequency component in the vibration signal and identify the resonance noise characteristics that overlap with the ultrasonic signal frequency band; Detect pulse interference events in electromagnetic signals and mark the occurrence time and frequency range to generate pulse event characteristics; Perform frequency band suppression and time domain threshold processing on the modal components according to the characteristics of resonance noise and pulse events; Perform sparse representation encoding on the processed modal components, perform multiple rounds of iterative screening and orthogonalization processing, and select the atomic set that matches the defect echo; The denoised ultrasonic signal is reconstructed using the screened atomic set and merged with the undecomposed signal residue to output the final result.

2. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 1, characterized in that: Decomposing the ultrasonic echo signal into multiple modal components and an undecomposed signal residue includes: Adding a preset number of white noises to the original ultrasonic echo signal to generate a noisy signal set; Perform empirical mode decomposition on each noisy signal to extract the first-order modal component; Calculate the average value of the first-order modal components of all noisy signals as the final first modal component; removing the first modal component from the original signal to generate a first-order unresolved signal residual; Iteratively superimposing preceding modal noise on the undecomposed signal residue and repeating the decomposition process to extract subsequent modal components step by step; When the undecomposed signal residue meets the preset termination condition, the decomposition is stopped and all modal components and the final undecomposed signal residue are output.

3. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 1, characterized in that: The extracting of the main frequency component in the vibration signal and identifying the resonance noise feature overlapping with the ultrasonic signal frequency band includes: Perform spectrum analysis on the vibration signal to determine its energy distribution characteristics and identify the frequency component with the largest energy as the main frequency; Comparing the main frequency with the operating frequency range of the ultrasonic detection system; When the main frequency is within the operating frequency range, it is marked as a resonance noise source.

4. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 1, characterized in that: Detecting a pulse interference event in an electromagnetic signal and generating a pulse event feature includes: Preprocessing the collected electromagnetic signals, including baseline correction and bandpass filtering; Identify the time window of transient pulse signal by dynamic threshold detection method; Extract the time domain amplitude characteristics and frequency domain main frequency bandwidth characteristics of pulse interference; The time window is associated with the main frequency bandwidth feature and stored to generate a pulse event feature containing time-frequency features.

5. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 3, characterized in that: The frequency band suppression of the modal component according to the resonance noise characteristics includes: Screening modal components that are within a preset tolerance range with respect to the main vibration frequency; Dynamically calculate the attenuation coefficient based on the energy ratio of the vibration signal and the modal component; The attenuation coefficient is applied to the screened modal components to suppress resonance noise.

6. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 4, characterized in that: The performing time domain threshold processing on the modal component according to the pulse event characteristics includes: Identify signal segments in the high-frequency modal components that overlap with the electromagnetic pulse time window; Calculating a noise energy statistic within the time window as a threshold reference; The signal components with coefficients lower than a preset threshold are zeroed.

7. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 1, characterized in that: The sparse representation encoding is performed on the processed modal components, and multiple rounds of iterative screening and orthogonalization are performed to screen the atomic set that matches the defect echo, including: Construct a time-frequency atomic preset feature matching library containing composite insulator defect echo characteristics; Initializing the undecomposed signal residual to the processed modal component and clearing the atomic support set; Iteratively select multiple atoms that are most correlated with the unresolved signal residue to update the support set; The combination coefficients of atoms in the support set are optimized by the least squares method; The undecomposed signal residue is updated to the difference between the original signal and the current reconstructed signal until the preset sparsity or residual energy threshold is met.

8. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 7, characterized in that: The iterative selection of the plurality of atoms most correlated with the unresolved signal remainder comprises: Calculate the inner product of the unresolved signal residual and all atoms in the preset feature matching library; Filter the first k atoms with the highest inner product values ​​and add them to the atomic support set; The sparse coefficients are optimized according to the atomic support set to reconstruct the signal components.

9. The joint denoising method based on multi-source noise suppression and sparse representation according to claim 8, characterized in that: The method of reconstructing the denoised ultrasonic signal using the screened atomic set and combining it with the undecomposed signal residue to output the final result includes: According to the atom set and sparse coefficients obtained by sparse representation coding, the corresponding time-frequency atoms are extracted from the preset feature matching library; Weighted superposition of the time-frequency atoms according to the combination coefficients generates a preliminary reconstructed signal; Performing amplitude calibration on the preliminary reconstructed signal and the decomposed undecomposed signal residue and then superimposing them; The superimposed signal is smoothed and filtered to eliminate high-frequency oscillation artifacts and output the final denoised ultrasonic signal.

10. A joint denoising system based on multi-source noise suppression and sparse representation, based on the joint denoising method based on multi-source noise suppression and sparse representation according to any one of claims 1 to 9, characterized in that: The system includes: Multi-source signal acquisition module, used to synchronously acquire ultrasonic echo signals, vibration signals and electromagnetic signals; A signal decomposition module, used for decomposing ultrasonic signals into modal components; Vibration analysis module, used to extract the main vibration frequency and generate resonance noise characteristic parameters; Electromagnetic interference analysis module, used to detect and mark the time-frequency characteristics of electromagnetic pulse events; Modal fusion denoising module, used to perform frequency band suppression and time domain threshold processing on modal components; Sparse coding module, used to perform multiple rounds of iterative screening and orthogonalization processing to screen defect echo atoms; The signal reconstruction module is used to generate a denoised ultrasonic signal.

Images