Noise processing method for transformer hot-spot temperature ultrasonic detection

Abstract

The invention relates to the technical field of transformer sound wave detection, in particular to a noise processing method for transformer hot-spot temperature ultrasonic detection, which comprises the following steps of: multiplying a pseudo-random sequence coded signal by an electric signal of a transmitting end of an ultrasonic sensor to generate a modulation signal with a pseudo-random binary coded sequence; performing time domain windowing processing on the collected ultrasonic signals to obtain a spectrogram; performing dynamic frequency domain segmentation based on the spectrogram; performing clustering analysis on the low-frequency sub-bands, distinguishing steady-state vibration noise and transient interference, and establishing a corresponding noise feature library; performing wavelet threshold denoising on each windowed signal, and retaining a high-frequency ultrasonic signal; and performing cross-correlation operation on the reconstructed and spliced complete signal and the pseudo-random sequence coded signal, and extracting a cross-correlation function peak value. By combining the pseudo-random sequence coding signal with the ultrasonic signal, effective coding of the ultrasonic signal is realized, the noise immunity is enhanced, and the noise interference problem of the transformer in the ultrasonic detection process is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer acoustic wave detection, and particularly to a noise processing method for ultrasonic detection of hot spot temperature of a transformer. Background Art

[0002] During the operation of a power transformer, hot spot temperature monitoring is crucial for equipment safety. As a method for indirectly measuring the hot spot temperature of a transformer by utilizing the functional relationship between the medium temperature and the acoustic wave propagation characteristics, the ultrasonic detection technology has the advantages of high precision, real-time continuous measurement, and convenient maintenance. As a non-invasive detection technology, it has the characteristics of being immune to electromagnetic interference and high safety, and has been gradually applied in the field of transformer internal detection.

[0003] However, during the normal operation of an oil-immersed power transformer, there are phenomena such as magnetostriction of the iron core, axial and radial vibrations of the winding, and vibrations of the cooling system. These vibrations will ultimately be transmitted to the surface of the transformer shell through the transformer oil and rigid connection parts, which will superimpose noise interference on the ultrasonic signals received by the ultrasonic sensor, resulting in a low signal-to-noise ratio of the received signal and a large calculation error of the propagation time. Existing methods mostly adopt fixed threshold filtering or single-domain processing, and it is difficult to effectively separate noise from useful signals. Summary of the Invention

[0004] To solve the above problems, the present invention provides a noise processing method for ultrasonic detection of hot spot temperature of a transformer, including: Generating a pseudo-random sequence coding signal, multiplying the pseudo-random sequence coding signal by the electrical signal at the transmitting end of the ultrasonic sensor to generate a modulation signal with a pseudo-random binary coding sequence, and loading the modulation signal to transmit the modulated ultrasonic signal; Performing time-domain windowing processing on the ultrasonic signals collected by the receiving end of the ultrasonic sensor, performing Fourier transform on the windowed signal to obtain a spectrogram; performing dynamic frequency-domain segmentation based on the spectrogram, dividing the spectrum into multiple sub-bands, and the sub-bands include high-frequency sub-bands and low-frequency sub-bands; Performing clustering analysis on the low-frequency sub-bands to distinguish steady-state vibration noise and transient interference, and establishing a corresponding noise feature library; Performing wavelet threshold denoising on each windowed signal based on the noise feature library to retain high-frequency ultrasonic signals; inverse-transforming the denoised signal into a time-domain signal, and splicing the time-domain signals after transformation of each windowed signal into a complete signal; Performing cross-correlation operation on the reconstructed and spliced complete signal and the pseudo-random sequence coding signal, extracting the peak value of the cross-correlation function; calculating the propagation duration of the ultrasonic signal according to the peak position, and inversely determining the hot spot temperature based on the sound velocity-temperature model.

[0005] The performing time-domain windowing processing on the ultrasonic signals collected by the receiving end of the ultrasonic sensor specifically includes: Estimate the time window length according to the propagation path of the ultrasonic wave. The time window is as follows: , wherein, t represents the time window; L is the propagation path length of the ultrasonic wave; v 0 represents the speed of sound; represents the offset of the speed of sound caused by temperature change; Perform windowing segmentation on the collected ultrasonic signals according to the time window length, and intercept the windowed signals through a window function.

[0006] In a specific embodiment, in order to effectively reduce spectral leakage, the window function selects the Blackman window function, which can improve the accuracy of signal processing.

[0007] In order to effectively separate the ultrasonic signal from the vibration noise, so as to adopt different processing strategies for different components, the high-frequency sub-band is the main component of the ultrasonic signal; the low-frequency sub-band is the main component of the vibration noise.

[0008] The noise feature library includes a steady-state noise feature library and a transient noise feature library, which can comprehensively cover different types of noise features, making noise recognition and classification more accurate.

[0009] Among them, the steady-state noise feature library is the eigenvector corresponding to the steady-state cluster center, including energy distribution and spectral shape. By analyzing the energy distribution and spectral shape, the steady-state vibration noise can be more accurately separated from the signal.

[0010] The transient noise feature library is the eigenvector corresponding to the transient cluster center, including time-domain features, which are helpful for capturing the suddenness and irregularity of transient noise.

[0011] Perform wavelet threshold denoising on each windowed signal based on the noise feature library. The specific operation is as follows: Select a wavelet basis that matches the characteristics of the ultrasonic signal, and determine the decomposition level according to the signal sampling frequency and the ultrasonic main frequency range; Set the adaptive threshold based on the noise feature library. For the detail coefficients of each layer of the wavelet decomposition, judge its dominant component according to the noise feature library, match the steady-state noise feature library for the low-frequency layer, and match the transient noise feature library for the high-frequency layer; Perform discrete wavelet transform on each windowed signal to obtain the detail coefficients of each layer; perform denoising according to the detail thresholds of each layer of the set wavelet decomposition layer.

[0012] The symbol width of the pseudo-random sequence coding signal is determined by the ultrasonic frequency. The formula for the symbol width is: , In the formula, Tc represents the symbol width; f 0 is the ultrasonic frequency; k is a positive integer value.

[0013] Beneficial effects: The present invention is a noise processing method for ultrasonic detection of the hot spot temperature of a transformer. By combining a pseudo-random sequence coding signal with an ultrasonic signal, effective coding of the ultrasonic signal is achieved, enhancing the anti-noise performance; sub-bands are dynamically divided according to the spectral characteristics of the vibration noise, and non-stationary noise is suppressed by combining time-domain windowing. A vibration noise feature library is established through clustering analysis to achieve dynamic matching of noise components, providing a basis for subsequent denoising; by performing wavelet threshold denoising on each windowed signal, high-frequency ultrasonic signals are effectively retained and noise interference is removed; finally, by performing cross-correlation operation on the received and processed signal and the pseudo-random sequence coding signal, the propagation duration of the acoustic wave signal can be accurately calculated, significantly improving the signal-to-noise ratio, thus effectively solving the problem of noise interference existing in the ultrasonic detection process of oil-immersed power transformers and improving the accuracy of hot spot temperature detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] By reading the detailed description of the preferred embodiments below, the solutions and advantages of the present application will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0015] Figure 1 is a flowchart of the noise processing method for ultrasonic detection of the hot spot temperature of a transformer. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings.

[0017] This embodiment provides a noise processing method for ultrasonic detection of the hot spot temperature of a transformer, which is applied to a transformer ultrasonic detection system. The transformer ultrasonic detection system includes a signal generation end, an ultrasonic signal transmission end, an ultrasonic signal reception end, and a processing terminal. By arranging the transformer ultrasonic detection system outside the power transformer box body, the acoustic wave signal propagation in the box body is collected through the ultrasonic signal transmission end and the ultrasonic signal reception end, and the hot spot temperature of the transformer is detected according to the changes in the propagation speed, propagation path, and propagation time of the acoustic wave signal.

[0018] In order to process the influence of vibration noise superposition when the ultrasonic signal propagates in the power transformer box body, refer to Figure 1 , and the specific implementation steps of the method are as follows: Step 1: Generate a pseudo-random sequence coding signal, multiply the pseudo-random sequence coding signal by the electrical signal at the transmitting end of the ultrasonic sensor to generate a modulation signal with a pseudo-random binary coding sequence, and load the modulation signal to transmit the modulated ultrasonic signal; The signal generating end generates a periodic pseudo-random sequence coding signal, and multiplies the pseudo-random sequence coding signal by the electrical signal at the transmitting end of the ultrasonic sensor to generate a modulation signal with a pseudo-random binary coding.

[0019] Among them, the symbol width of the pseudo-random sequence coding signal is determined according to the ultrasonic frequency at the transmitting end of the ultrasonic sensor, and the formula for the symbol width is: , In the formula, Tc represents the symbol width; f 0 is the ultrasonic frequency; k is a positive integer value.

[0020] Transmit the generated modulation signal to the transmitting end of the ultrasonic sensor, and drive the ultrasonic sensor at the transmitting end to transmit ultrasonic signals into the power transformer box.

[0021] Step 2: Perform time-domain windowing processing on the ultrasonic signals collected by the receiving end of the ultrasonic sensor, perform Fourier transform on the windowed signal to obtain a spectrogram; perform dynamic frequency-domain segmentation based on the spectrogram, and divide the spectrum into multiple sub-bands, where the sub-bands include high-frequency sub-bands and low-frequency sub-bands; When the ultrasonic signals are collected by the receiving end of the ultrasonic sensor, the receiving end of the ultrasonic sensor will transmit the ultrasonic signals to the processing terminal, and the processing terminal performs time-domain windowing processing on the ultrasonic signals collected by the receiving end of the ultrasonic sensor. The specific operation is as follows: Estimate the time window length according to the propagation path of the ultrasonic wave. The time window is: , Among them, t represents the time window; L is the ultrasonic propagation path length; v 0 represents the sound speed; represents the sound speed offset caused by temperature change; Perform windowing segmentation on the collected ultrasonic signals according to the time window length; select the Blackman window function as the window function to intercept the windowed signal.

[0022] Perform fast Fourier transform (FFT) on the windowed signal, take the amplitude energy of the transform coefficients, transform the frequency to the Mel scale, take the logarithm, and then obtain the spectrogram of the ultrasonic signal through discrete cosine transform.

[0023] Perform dynamic frequency-domain segmentation on the spectrogram according to the vibration and noise energy distribution, and divide the spectrum into multiple sub-bands. Among them, the high-frequency sub-band is the main component of the ultrasonic signal; the low-frequency sub-band is the main component of the vibration and noise.

[0024] Step 3: Perform clustering analysis on the low-frequency sub-band to distinguish steady-state vibration noise from transient interference, and establish a corresponding noise feature library; Take the low-frequency sub-band as the object of noise analysis, and extract features for each frequency point in the low-frequency sub-band, including energy value, time-domain statistic, and frequency-domain statistic; the energy value is calculated as the square of the amplitude of each frequency point, representing the energy distribution; the time-domain statistic is the mean, variance, and peak factor of the signal within the window; the frequency-domain statistic is the bandwidth and flatness of the sub-band spectrum.

[0025] Use the K-means clustering algorithm to perform clustering analysis on the low-frequency sub-band, and divide the low-frequency sub-band data into two categories, including steady-state vibration noise and transient interference. In this embodiment, randomly select 2 initial cluster centers (corresponding to the two types of noise); calculate the Euclidean distance from each data point in the low-frequency sub-band to the cluster center, and assign it to the nearest cluster; recalculate the cluster center; recalculate the Euclidean distance from each data point to the cluster center for reallocation until the change in the cluster center is less than the threshold or the maximum number of iterations is reached; based on the cluster, mark the two types of cluster labels, and mark each data point as belonging to steady-state vibration noise or transient interference noise. It should be noted that the K-means clustering algorithm belongs to the prior art, and the specific steps will not be elaborated.

[0026] Based on the distinguished steady-state vibration noise and transient interference, establish a corresponding noise feature library, including a steady-state noise feature library and a transient noise feature library; establish a steady-state noise feature library by extracting the feature vectors corresponding to the steady-state cluster center, including energy distribution and spectral shape; construct a transient noise feature library by extracting the feature vectors corresponding to the transient cluster center, including time-domain features.

[0027] Step 4: Perform wavelet threshold denoising on each windowed signal based on the noise feature library, and retain the high-frequency ultrasonic signal; inverse-transform the denoised signal into a time-domain signal, and splice the time-domain signals after transformation of each windowed signal into a complete signal; The specific operation of performing wavelet threshold denoising on each windowed signal based on the noise feature library is as follows: Select a wavelet basis that matches the characteristics of the ultrasonic signal, and determine the decomposition level according to the signal sampling frequency and the ultrasonic main frequency range; set the wavelet decomposition level to 5 layers, and ensure that the high-frequency ultrasonic component is located in the lower decomposition layer; Set the adaptive threshold based on the noise feature library. For the detail coefficients of each layer of the wavelet decomposition, judge its dominant component according to the noise feature library, match the steady-state noise feature library for the low-frequency layer, and match the transient noise feature library for the high-frequency layer; Perform discrete wavelet transform on each windowed signal to obtain the detail coefficients of each layer; denoise according to the detail thresholds of each layer of the preset wavelet decomposition layer.

[0028] Step 5: Perform cross-correlation operation on the reconstructed and spliced complete signal and the pseudo-random sequence coded signal, and extract the peak value of the cross-correlation function; calculate the propagation duration of the ultrasonic signal according to the peak position, and inversely determine the hot spot temperature based on the sound speed-temperature model; After the ultrasonic sensor transmitter emits an ultrasonic signal, it will transmit the pseudo-random sequence coded signal to the processing terminal for storage; when the processing terminal obtains the reconstructed and spliced complete signal through Step 2 - Step 4, it calls the pseudo-random sequence coded signal to perform cross-correlation operation.

[0029] The cross-correlation operation of the reconstructed and spliced complete signal and the pseudo-random sequence coded signal has the following formula: , In the formula, r(t) represents the spliced complete signal; s(t) represents the pseudo-random sequence coded signal; t represents the time window; R rs (τ) is the cross-correlation function value, indicating the similarity degree of the two signals at different time offsets; Determine the peak position of the cross-correlation function according to the cross-correlation function value τ peak , and determine the propagation duration of the ultrasonic signal according to the peak position t d = τ peak + t .

[0030] Determine the propagated sound wave according to the propagation duration and propagation path, and inversely derive the hot spot temperature based on the sound speed-temperature model.

Claims

1. A noise processing method for ultrasonic detection of hot spot temperature of transformer, characterized in that: include: Generate a pseudo-random sequence coding signal, multiply the pseudo-random sequence coding signal with the electrical signal of the ultrasonic sensor transmitting end to generate a modulation signal with a pseudo-random binary coding sequence, load the modulation signal to transmit the modulated ultrasonic signal; Performing time domain windowing processing on the ultrasonic signal collected by the receiving end of the ultrasonic sensor, performing Fourier transform on the windowed signal, and obtaining a spectrum diagram; performing dynamic frequency domain segmentation based on the spectrum diagram, and dividing the spectrum into multiple sub-bands, wherein the sub-bands include high-frequency sub-bands and low-frequency sub-bands; Perform cluster analysis on low-frequency sub-bands to distinguish steady-state vibration noise from transient interference, and establish a corresponding noise feature library; Based on the noise feature library, wavelet threshold denoising is performed on each windowed signal to retain the high-frequency ultrasonic signal; the denoised signal is inversely transformed into a time domain signal, and the transformed time domain signals of each windowed signal are spliced ​​into a complete signal; The reconstructed and spliced ​​complete signal is cross-correlated with the pseudo-random sequence coded signal to extract the peak of the cross-correlation function; the propagation time of the ultrasonic signal is calculated according to the peak position, and the hot spot temperature is determined based on the inversion of the sound speed-temperature model.

2. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 1 is characterized in that: The time domain windowing process of the ultrasonic signal collected by the ultrasonic sensor receiving end is specifically as follows: The time window length is estimated based on the propagation path of the ultrasonic wave. The time window is: ， in, t represents a time window; L is the ultrasonic wave propagation path length; v 0 represents the speed of sound; Indicates the sound velocity deviation caused by temperature change; The collected ultrasonic signal is windowed and segmented according to the length of the time window, and the windowed signal is intercepted by the window function.

3. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 2 is characterized in that: The window function is selected as the Blackman window function.

4. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 1 is characterized in that: The high-frequency sub-band is the main component of the ultrasonic signal; the low-frequency sub-band is the main component of the vibration noise.

5. The noise processing method for transformer hot spot temperature ultrasonic detection according to claim 1 is characterized in that: The noise feature library includes a steady-state noise feature library and a transient noise feature library.

6. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 5 is characterized in that: The steady-state noise feature library is a feature vector corresponding to the steady-state cluster center, including energy distribution and spectrum shape.

7. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 5 is characterized in that: The transient noise feature library is a feature vector corresponding to the center of the transient cluster, including time domain features.

8. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 5 is characterized in that: The wavelet threshold denoising is performed on each windowed signal based on the noise feature library, and the specific operation is as follows: Select a wavelet basis that matches the characteristics of the ultrasonic signal, and determine the number of decomposition layers based on the signal sampling frequency and the ultrasonic main frequency range; Based on the noise feature library, adaptive threshold is set, and the dominant component of each detail coefficient of the wavelet decomposition layer is determined according to the noise feature library. The steady-state noise feature library is matched to the low-frequency layer, and the transient noise feature library is matched to the high-frequency layer. Discrete wavelet transform is performed on each windowed signal to obtain detail coefficients of each layer; denoising is performed according to the detail thresholds of each layer of the set wavelet decomposition layer.

9. The noise processing method for ultrasonic detection of hot spot temperature of transformer according to claim 1 is characterized in that: The symbol width of the pseudo-random sequence encoding signal is determined by the ultrasonic frequency, and the formula for the symbol width is: ， In the formula, Tc Indicates the width of the code element; f 0 is the ultrasonic frequency; k A positive integer value.

Images