Method for detecting internal cracks of porcelain bushing based on acoustic emission detection

Abstract

The application discloses a porcelain sleeve internal crack detection method based on acoustic emission detection and belongs to the technical field of nondestructive testing. The method comprises the following steps: acquiring initial acoustic emission signals, vibration signals, electromagnetic interference signals and spatial position parameters; generating an interference characteristic vector according to the vibration signals and the electromagnetic interference signals; performing adaptive interference cancellation processing on the initial acoustic emission signals to generate noise reduction acoustic emission signals; performing time-frequency feature extraction on the noise reduction acoustic emission signals to generate a multi-dimensional acoustic emission characteristic vector; performing dynamic characteristic analysis on the multi-dimensional acoustic emission characteristic vector to output a crack existence mark and a crack type characteristic; and performing characteristic weighted positioning calculation on the crack type characteristic and the spatial position parameters to generate crack position coordinates. The application adopts multi-source interference collaborative cancellation technology and combines dynamic characteristic analysis, so that internal cracks of a porcelain sleeve can be accurately identified, and the reliability of crack detection is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nondestructive testing, in particular to a porcelain sleeve internal crack detection method based on acoustic emission detection. BACKGROUND

[0002] As a key insulation and support component in high-voltage and ultra-high-voltage power systems, the structural integrity of the porcelain sleeve is directly related to the safe and stable operation of the entire power grid. Acoustic emission detection technology is an important nondestructive testing method, which monitors and evaluates the internal damage state of the material by passively receiving the transient elastic waves, i.e. acoustic emission signals, released when the material produces cracks or plastic deformation under stress. This technology has a broad application prospect in the health state monitoring field of power equipment such as porcelain sleeves due to its high sensitivity and ability to dynamically monitor defects.

[0003] In the prior art, acoustic emission detection of porcelain sleeves usually involves arranging acoustic emission sensors on the surface of the porcelain sleeve, and using signal amplifiers and bandpass filters to preliminarily process the collected signals to filter out noise in some frequency bands. Then, a fixed amplitude threshold is set to count or analyze the energy of signals exceeding the threshold, and when the relevant parameters accumulate to a certain extent, it is determined that there is a defect. For defect positioning, the time difference between different sensors receiving the same signal is usually measured to calculate the position.

[0004] However, the above prior art has obvious defects in actual application. The environment of the substation where the porcelain sleeve is located is usually accompanied by strong electromagnetic interference and mechanical vibration generated by equipment operation. The frequency spectrum of these interference signals often overlaps with that of the real crack acoustic emission signals, and simple bandpass filtering methods cannot effectively separate them, resulting in low signal-to-noise ratio. In addition, the judgment method relying on a fixed threshold is not sensitive to early emerging low-energy crack signals, and it is easy to misjudge some transient strong interference signals as cracks, so the reliability of the detection is not high. At the same time, the positioning method based on simple time difference is difficult to meet the requirements of accurate diagnosis in the face of the complex shape structure of the porcelain sleeve. SUMMARY

[0005] To solve the above problems, the present application provides a porcelain sleeve internal crack detection method based on acoustic emission detection, which uses multi-source interference cooperative cancellation technology and combines dynamic feature analysis to accurately identify internal cracks in the porcelain sleeve and improve the reliability of crack detection.

[0006] The above object can be achieved by the following scheme:

[0007] The method comprises the following steps: obtaining an initial acoustic emission signal, a vibration signal, an electromagnetic interference signal and a spatial position parameter; generating an interference feature vector according to the vibration signal and the electromagnetic interference signal; performing adaptive interference cancellation processing on the initial acoustic emission signal to generate a noise-reduced acoustic emission signal; performing time-frequency feature extraction on the noise-reduced acoustic emission signal to generate a multi-dimensional acoustic emission feature vector; performing dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature; and performing feature weighted positioning calculation on the crack type feature and the spatial position parameter to generate a crack position coordinate.

[0008] Optionally, the generating an interference feature vector comprises: extracting a frequency spectrum peak feature of the vibration signal and a power frequency harmonic feature of the electromagnetic interference signal; and performing feature fusion on the frequency spectrum peak feature and the power frequency harmonic feature to generate an interference feature vector.

[0009] Optionally, the adaptive interference cancellation processing comprises: generating a phase-inverted cancellation signal according to the interference feature vector; and performing real-time superposition operation on the phase-inverted cancellation signal and the initial acoustic emission signal to generate a noise-reduced acoustic emission signal.

[0010] Optionally, the dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature comprises: performing dynamic energy attenuation analysis on the multi-dimensional acoustic emission feature vector to obtain a crack existence probability value; and performing mode classification judgment on the crack existence probability value to output a crack existence flag and a crack type feature.

[0011] Optionally, the dynamic energy attenuation analysis on the multi-dimensional acoustic emission feature vector to obtain a crack existence probability value comprises: performing time-domain energy extraction on the multi-dimensional acoustic emission feature vector to obtain an energy attenuation time sequence; and performing exponential model fitting on the energy attenuation time sequence to obtain a crack existence probability value.

[0012] Optionally, the time-frequency feature extraction on the noise-reduced acoustic emission signal to generate a multi-dimensional acoustic emission feature vector comprises: extracting a time-frequency feature of the noise-reduced acoustic emission signal to obtain a complementary basis transformation parameter set; and performing feature fusion and quantization calculation on the complementary basis transformation parameter set to generate a multi-dimensional acoustic emission feature vector.

[0013] Optionally, the generating a phase-inverted cancellation signal according to the interference feature vector comprises: generating a multi-channel resonance reference waveform according to the interference feature vector; and performing nonlinear phase shift processing on the multi-channel resonance reference waveform to generate a phase-inverted cancellation signal.

[0014] Optionally, the feature weighting positioning calculation comprises: analyzing the crack type feature to obtain a sound wave propagation attenuation factor; calculating initial time difference positioning coordinates according to the spatial position parameter; fusing the sound wave propagation attenuation factor and a signal-to-noise ratio index of the noise-reduced acoustic emission signal to generate a sensor credibility weight; and weighting and optimizing the initial time difference positioning coordinates by using the sensor credibility weight to generate crack position coordinates.

[0015] Optionally, the analyzing the crack type feature to obtain a sound wave propagation attenuation factor comprises: performing physical field mapping analysis on the crack type feature to obtain a crack equivalent geometric configuration parameter group; and obtaining the sound wave propagation attenuation factor based on the crack equivalent geometric configuration parameter group.

[0016] Based on the same inventive concept, the application also provides a porcelain bushing internal crack detection system based on acoustic emission detection, which comprises: a multi-source sensing acquisition module, used to acquire initial acoustic emission signals, vibration signals, electromagnetic interference signals and spatial position parameters; a joint interference analysis module, used to generate an interference feature vector according to the vibration signals and the electromagnetic interference signals; an adaptive cancellation execution module, used to perform adaptive interference cancellation processing on the initial acoustic emission signals to generate noise-reduced acoustic emission signals; a time-frequency feature extraction module, used to perform time-frequency feature extraction on the noise-reduced acoustic emission signals to generate a multi-dimensional acoustic emission feature vector; a crack identification module, used to perform dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature; and a weighted positioning module, used to perform feature weighting positioning calculation on the crack type feature and the spatial position parameter to generate crack position coordinates when the crack existence flag is true.

[0017] Compared with the prior art, the application has the following advantages:

[0018] The application can accurately eliminate noise from a specific source by performing feature extraction on vibration and electromagnetic interference signals in the environment and generating an interference feature vector, and then performing adaptive interference cancellation on initial acoustic emission signals, greatly improving the signal-to-noise ratio of crack signals in a complex electromagnetic and mechanical vibration environment, and enhancing the anti-interference ability of the detection method and the applicability in harsh working conditions.

[0019] The application performs dynamic feature analysis on the noise-reduced acoustic emission signals, not only judges whether a crack exists, but also outputs a crack type feature representing the properties of the crack. This deepening analysis from existence judgment to type diagnosis provides more abundant and valuable decision information for evaluating the damage degree and development trend of the porcelain bushing, and improves the intelligent level of detection.

[0020] The application innovatively combines the physical attenuation information analyzed by the crack type feature and the quality index of the signal itself when locating the crack, generates a sensor credibility weight, and optimizes the initial positioning result by weighting based on the credibility weight.

[0021] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0023] Figure 1 is a flowchart of the porcelain sleeve internal crack detection method based on acoustic emission detection of the embodiment of the present application.

[0024] Figure 2 is a dynamic energy attenuation graph of the embodiment of the present application.

[0025] Figure 3 is a crack type feature identification graph of the embodiment of the present application.

[0026] Figure 4 is a multi-dimensional acoustic emission feature vector graph of the embodiment of the present application.

[0027] Figure 5 is a crack existence probability value determination result graph of the embodiment of the present application.

[0028] Figure 6 is a structural schematic diagram of the porcelain sleeve internal crack detection system based on acoustic emission detection of the embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the objects, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0030] Referring Figure 1 One embodiment of the present application proposes a porcelain sleeve internal crack detection method based on acoustic emission detection, adopts multi-source interference cooperative cancellation technology, and combines dynamic feature analysis to accurately identify porcelain sleeve internal cracks and improve crack detection reliability.

[0031] The method of the embodiment specifically includes:

[0032] Obtain initial acoustic emission signals, vibration signals, electromagnetic interference signals, and spatial position parameters;

[0033] Generate an interference feature vector according to the vibration signals and the electromagnetic interference signals;

[0034] Perform adaptive interference cancellation processing on the initial acoustic emission signals to generate a noise-reduced acoustic emission signal;

[0035] Perform time-frequency feature extraction on the noise-reduced acoustic emission signal to generate a multi-dimensional acoustic emission feature vector;

[0036] Perform dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature;

[0037] When the crack existence flag is true, perform feature-weighted positioning calculation on the crack type feature and the spatial position parameters to generate crack position coordinates.

[0038] Specifically, first, the main environmental interference signals such as vibration and electromagnetic signals are actively collected, and their features are characterized as interference feature vectors, providing a prior model for subsequent accurate noise reduction. Then, the model is used to perform adaptive interference cancellation on the mixed initial acoustic emission signals, rather than traditional general filtering, thereby preserving the integrity of the true crack signal to the greatest extent. After obtaining the pure noise-reduced acoustic emission signal, the method extracts its multi-dimensional features through deep time-frequency analysis and further performs dynamic analysis, which not only determines the existence or nonexistence of cracks, but also preliminarily diagnoses the type of cracks. Finally, the method innovatively uses the diagnosis result, i.e., the crack type feature, as a physical constraint condition to optimize the spatial positioning process of the crack. Through feature-weighted calculation, the initial positioning result is corrected, forming a complete technical closed loop from signal purification, feature diagnosis to accurate positioning.

[0039] Optionally, the generation of the interference feature vector includes:

[0040] Extract the frequency spectrum peak feature of the vibration signal and the power frequency harmonic feature of the electromagnetic interference signal;

[0041] Feature fusion is performed on the frequency spectrum peak feature and the power frequency harmonic feature to generate an interference feature vector.

[0042] Specifically, firstly, the collected vibration signal and electromagnetic interference signal are processed in parallel. For the vibration signal, the fast Fourier transform algorithm is applied to convert it from the time domain to the frequency domain to obtain its amplitude spectrum. By setting a preset energy threshold, all significant peaks exceeding the threshold are searched and identified on this amplitude spectrum. Each significant peak is represented by its corresponding frequency value and amplitude value. The set of frequencies and amplitudes of these peaks is the spectral peak feature of the vibration signal. This feature can accurately reflect the stable or transient mechanical vibration characteristics caused by environmental wind, equipment mechanical resonance, etc. For the electromagnetic interference signal, it is also subjected to fast Fourier transform processing. Considering that electromagnetic interference in power systems mainly comes from power frequency and its multiple frequency components, this step focuses on analyzing the energy distribution of the signal at the power frequency point, usually 50 Hz, and its harmonic frequency points, i.e., 100 Hz, 150 Hz, etc. The signal amplitude and phase information at these specific frequency points are extracted to form the power frequency harmonic feature of the electromagnetic interference signal. This feature accurately describes the periodic electromagnetic field interference generated by high-voltage transmission lines and surrounding electrical equipment. Finally, the spectral peak feature and the power frequency harmonic feature obtained in the previous two steps are fused. This fusion operation is usually vector splicing, i.e., integrating the feature data describing two types of interference from different physical sources into a single vector with higher dimensions and more complete information, i.e., the final generated interference feature vector. The generation of this vector can be expressed by the following concept formula:

[0043] ,

[0044] wherein, represents the final generated interference feature vector, which is a high-dimensional vector containing detailed information of the main interference sources in the environment. represents the set of spectral peak features extracted from the vibration signal. represents the set of power frequency harmonic features extracted from the electromagnetic interference signal. represents the feature fusion operation, which is usually the head-to-tail connection of the and two feature vectors to form a longer vector, thereby integrating the two into a unified interference descriptor without losing their respective information.

[0045] Optionally, the adaptive interference cancellation process includes:

[0046] generating a phase-inverted cancellation signal according to the interference feature vector;

[0047] performing real-time superposition operation on the phase-inverted cancellation signal and the initial acoustic emission signal to generate a noise-reduced acoustic emission signal.

[0048] Specifically, first, the interference feature vector is analyzed, which contains the spectral peak features extracted from the vibration signal and the power frequency harmonic features extracted from the electromagnetic interference signal. These features describe the key information of the interference, such as frequency, amplitude, and phase. Based on these feature parameters, the system generates a predicted interference waveform similar to the actual interference mixed into the initial acoustic emission signal through signal synthesis technology. For example, for each spectral peak feature, a sine wave with the corresponding frequency and amplitude is generated; for each power frequency harmonic feature, a sine wave with the corresponding frequency, amplitude, and phase is generated. Linear superposition of all these individually generated interference waveforms forms a complete predicted interference signal highly correlated with the actual environmental interference in the time domain. Next, to achieve cancellation, the process performs a phase inversion operation on the predicted interference signal. Technically, this is equivalent to multiplying the amplitude of each sample point of the predicted interference signal by -1 or shifting its overall phase by 180 degrees, thereby generating a phase-inverted cancellation signal. This cancellation signal is identical in waveform to the predicted interference signal but opposite in phase. The final step is to perform a real-time point-by-point superposition operation between the phase-inverted cancellation signal and the collected initial acoustic emission signal. Since the cancellation signal and the interference component in the initial acoustic emission signal are opposite in phase and similar in amplitude, they cancel each other out when superimposed. This process can be represented by the following formula:

[0049] ,

[0050] In this formula, represents the noise-reduced acoustic emission signal generated after the operation, which is the signal amplitude at time point . represents the amplitude of the initial acoustic emission signal obtained at the same time point . represents the amplitude of the cancellation signal at time point after phase inversion processing. Since is theoretically equal to the negative of the interference part in , this addition operation essentially subtracts the interference component from the initial signal. Through this operation, the environmental interference in the initial signal is significantly weakened, resulting in a noise-reduced acoustic emission signal with higher signal-to-noise ratio.

[0051] Optionally, the dynamic feature analysis of the multi-dimensional acoustic emission feature vector to output the crack existence flag and crack type features includes:

[0052] Performing dynamic energy decay analysis on the multi-dimensional acoustic emission feature vector to obtain a crack existence probability value;

[0053] Performing pattern classification judgment on the crack existence probability value to output the crack existence flag and crack type features.

[0054] Specifically, the process first performs dynamic energy decay analysis on the input multi-dimensional acoustic emission feature vector. The multi-dimensional acoustic emission feature vector is a structured data containing multi-dimensional information of acoustic emission signals, such as energy, frequency, etc. The purpose of dynamic energy decay analysis is to investigate the law of energy characteristics represented in the vector evolving over time. By analyzing the vector, time series data representing the distribution of signal energy in each time segment is extracted, forming an energy decay time series. Since the acoustic emission signals generated by real crack propagation follow a specific physical pattern from instantaneous burst to gradual decay, background noise or other interference does not have this regularity. Therefore, a preset mathematical model that can describe the energy decay behavior of typical crack signals is applied to the energy decay time series. By calculating the degree of agreement between the sequence and the model, a quantitative indicator, i.e., a crack existence probability value, is finally output. The higher the probability value, the more the detected signal characteristics conform to the energy release law of real cracks. Next, the process performs pattern classification judgment on the crack existence probability value obtained in the previous step to generate the final output result. This judgment process is based on a preset decision logic. For the output of the crack existence flag, the system uses a threshold comparison mechanism. This mechanism can be described by the following logical expression:

[0055] ,

[0056] In this expression, represents the final output crack existence flag, whose value is Boolean true or false. is the crack existence probability value calculated in the previous step, which is obtained by dynamic energy decay analysis, as shown in Figure 2 . is a pre-set probability threshold, which is determined based on a large amount of experimental data and engineering experience to distinguish high-confidence crack events from background noise. When the calculated is greater than or equal to this threshold, is set to true, indicating that the system determines that there is a crack; otherwise, it is false. For the output of crack type characteristics, the judgment process is more detailed. It not only depends on the crack existence probability value, but also uses other parameters obtained in the dynamic energy decay analysis process, such as the rate of energy decay, the size of peak energy, etc. Different crack types, such as slow-growing fatigue cracks and rapid-occurring brittle fractures, have significant differences in the shape of acoustic emission signal energy decay curves. The pattern classification judgment module has a built-in classifier that takes these energy dynamic parameters as input and maps them to a pre-defined crack type library, such as "micro-crack", "through crack", etc., to output the most likely crack type feature. Crack type feature recognition is as follows:Figure 3 As shown. The classifier construction and training process is as follows: Sample collection: Acoustic emission signal samples of different types of cracks (microcracks, through cracks, etc.) are obtained through experiments, the above input features are extracted and labeled with type labels; Training process: 70% of the samples are used as the training set, and the classifier parameters are optimized by grid search (e.g., the kernel function of SVM is RBF, and the number of trees in random forest is 100). The remaining 30% of the samples are used to verify the classification accuracy, ensuring that the accuracy is ≥90%; Mapping rules: In the predefined crack type feature library, the corresponding features of microcracks are "low energy peak value less than 50mV, slow decay rate greater than 0.8, and main frequency greater than 100kHz", and the corresponding features of through cracks are "high energy peak value greater than 200mV, fast decay rate less than 0.3, and main frequency less than 50kHz". The classifier output results are matched with the types in the library by calculating the cosine similarity of the feature vectors.

[0057] Optionally, the step of performing dynamic energy attenuation analysis on the multidimensional acoustic emission feature vector to obtain the probability value of crack existence includes:

[0058] Temporal energy extraction is performed on the multidimensional acoustic emission feature vector to obtain an energy decay time sequence;

[0059] An exponential model is used to fit the energy decay time series to obtain the probability value of crack existence.

[0060] Specifically, firstly, temporal energy extraction is performed on the input multidimensional acoustic emission feature vector, such as... Figure 4 As shown, the multidimensional acoustic emission feature vector itself already contains rich information about the signal at different time points; this step focuses on the energy dimension. By analyzing the vector in chronological order, the signal energy within a series of consecutive small time windows is calculated, thus forming a data point sequence describing the energy change over time, which is the energy decay time series sequence. This sequence intuitively depicts the energy trajectory of a potential acoustic emission event from its occurrence, reaching its peak, to its gradual dissipation. Subsequently, this method fits the obtained energy decay time series sequence to an exponential model. The exponential model is chosen because when an internal crack occurs in a material, its energy release and dissipation process physically follows a typical exponential decay law. This exponential model can be expressed as:

[0061] ,

[0062] In this formula, The model at time points The predicted theoretical energy value. Represents the initial energy amplitude of the acoustic emission event. This parameter is determined by finding the peak energy point in the energy decay time series. It is the base of the natural logarithm. This represents the energy decay constant, whose value reflects the rate of energy decay and is a key parameter that needs to be calculated through fitting. This represents the time calculated from the energy peak point. The fitting process involves adjusting the time using numerical optimization algorithms such as least squares. The value of minimizes the sum of squared errors between the theoretical energy decay curve generated by the formula and the actually observed energy decay time series. The goodness of fit directly reflects the degree of agreement between the observed signal energy change pattern and the actual crack behavior pattern. Finally, the statistical index measuring this goodness of fit, namely the coefficient of determination, is output as the crack presence probability value. The crack presence probability value determination result is as follows: Figure 5 As shown. The coefficient of determination ranges from 0 to 1. The closer its value is to 1, the higher the degree to which the model interprets the data. That is, the more the observed energy decay behavior conforms to the exponential decay law, and therefore the higher the probability of it being judged as a real crack event.

[0063] Optionally, the step of extracting time-frequency features from the noise-reduced transmitted signal to generate a multidimensional acoustic emission feature vector includes:

[0064] Extract the time-frequency features of the noise-reduced transmitted signal to obtain the complementary fundamental transform parameter set;

[0065] The complementary fundamental transformation parameter set is subjected to feature fusion and quantization calculation to generate a multidimensional acoustic emission feature vector.

[0066] Specifically, first, the noise-reduced acoustic emission signal obtained after adaptive interference cancellation processing, whose signal-to-noise ratio has been significantly improved, is executed with multi-channel parallel time-frequency feature extraction to construct a complementary basis transformation parameter set. The complementary basis transformation here refers to the simultaneous use of multiple signal processing transformation methods with different analysis advantages. Acoustic emission signals have transient and non-stationary characteristics, and a single transformation method is difficult to fully characterize all their features. Therefore, this method may use, for example, wavelet transform to capture the transient mutation characteristics and energy distribution of the signal at different frequency scales; and Hilbert-Huang transform to analyze the nonlinear and non-stationary characteristics of the signal and obtain its instantaneous frequency and instantaneous amplitude over time. At the same time, classic acoustic emission parameters such as signal peak amplitude, rise time, duration, and ring count are also directly extracted from the time domain. All these parameters obtained from different transformation domains and different analysis angles together constitute a complementary basis transformation parameter set with complementary information. Next, the complementary basis transformation parameter set containing a large number of heterogeneous parameters is executed with feature fusion and quantization calculation. Feature fusion is to integrate these parameters with different sources and different physical meanings into a unified mathematical framework. The most direct implementation is to arrange all parameters in a predetermined order and splice them into a single long vector with very high dimensions. However, the dimensions and numerical ranges of these original parameters differ greatly, and direct use will make some parameters occupy an unreasonable dominant position in subsequent analysis. Therefore, quantization calculation must be performed before or after fusion, which usually refers to the normalization or standardization of features. For example, Z-score standardization is used to convert each feature parameter to a distribution with a mean of 0 and a standard deviation of 1. This process can be expressed by the following conceptual formula:

[0067] ,

[0068] In this formula, is the final generated multi-dimensional acoustic emission feature vector. , ,..., represent each original feature parameter extracted from the complementary basis transformation parameter set.

[0069] ,

[0070] is the result of the standardization of the th original feature, where and are the mean and standard deviation of the feature in a large number of background samples, which are pre-learned or online estimated. represents the fusion operation of splicing all the standardized features into a vector.

[0071] Optionally, the generating the phase-inverted cancellation signal in accordance with the interference eigenvector comprises:

[0072] generating a multi-channel resonant reference waveform in accordance with the interference eigenvector;

[0073] performing a non-linear phase shift on the multi-channel resonant reference waveform to generate the phase-inverted cancellation signal.

[0074] Specifically, first, a multi-channel resonant reference waveform is generated according to the input high-dimensional interference feature vector. The interference feature vector has detailedly described the spectral peak characteristics of the vibration signal and the power harmonic characteristics of the electromagnetic interference signal. For each of the feature components, the system will start an independent signal synthesis channel. For example, for a spectral peak feature of the vibration signal, the system will generate a sinusoidal wave corresponding to the frequency and amplitude; for a power harmonic feature of the electromagnetic interference, a sinusoidal wave corresponding to the frequency, amplitude and phase will be generated. Linear superposition of these single frequency waveforms independently generated by different feature components forms a multi-channel resonant reference waveform. This waveform is a high-precision mathematical reconstruction of the complex interference signal existing in the actual environment in the time domain. Next, the method performs nonlinear phase shift processing on the generated multi-channel resonant reference waveform to generate the final phase-inverted cancellation signal. Simple linear inversion, that is, multiplied by negative one, is the basis, but it may not be accurate enough in a complex actual environment. Because the signal may introduce nonlinear phase distortion during transmission and coupling. The nonlinear phase shift processing is just to compensate for this distortion. This processing process can be implemented through a pre-trained nonlinear mapping function or an adaptive phase-locked loop. The nonlinear mapping function training process is as follows: sample collection: in the laboratory environment, record the pure acoustic emission signal (target signal) of the porcelain sleeve crack simulation test and the common interference signal (such as pump body vibration, electromagnetic radiation) on site; data enhancement: generate mixed signal samples by randomly superimposing interference signals of different intensities (signal-to-noise ratio -10dB to 20dB) to construct a training set (70%), a validation set (20%) and a test set (10%); model training: initialize the MLP network (input layer 16 dimensions, hidden layer 32 dimensions, output layer 16 dimensions), use ReLU activation function, learning rate 0.001, training period 50 rounds, calculate the validation set error after each round, and stop training when the error does not decrease for 5 consecutive rounds; deployment verification: deploy the trained model to the processing module, test the interference cancellation effect of the test set, and require the signal-to-noise ratio of the target signal to be improved by ≥15dB. The adaptive phase-locked loop design details are as follows: the phase detector outputs an error voltage by comparing the phase difference between the input mixed signal and the VCO output signal; the PI controller converts the error voltage into a control signal, where the proportional term quickly responds to the phase deviation and the integral term eliminates the steady-state error; the VCO adjusts the output frequency according to the control signal, so that it is synchronized with the interference signal frequency, generating a reference signal with the same frequency and phase as the interference signal; interference cancellation is achieved by subtracting the reference signal from the mixed signal, and the loop dynamically adjusts the parameters to adapt to the frequency drift of the interference signal (maximum drift rate ≤5Hz / s). The input of this function or loop is the multi-channel resonant reference waveform and the more detailed phase information that may be analyzed from the interference feature vector, and the output is a waveform with precisely adjusted phase.The goal of the adjustment is to maximize the anti-correlation of the output waveform with the actually superimposed interference component in the initial acoustic emission signal. This process can be conceptually represented as:

[0075] ,

[0076] wherein, is the amplitude of the final generated phase-inverted cancellation signal at time point . is the amplitude of the multi-channel resonance reference waveform at time point , which is synthesized from the interference eigenvector. represents a nonlinear phase shift operator, which performs not only a simple phase inversion of 180 degrees, but also a dynamic, nonlinear phase fine-tuning according to the instantaneous frequency and amplitude of the reference waveform, etc. characteristics, to more accurately match and cancel the actual interference signal.

[0077] Optionally, the feature weighting positioning calculation comprises:

[0078] analyzing the crack type characteristics to obtain a sound wave propagation attenuation factor;

[0079] calculating an initial time difference positioning coordinate according to the spatial position parameter;

[0080] fusing the sound wave propagation attenuation factor and the signal-to-noise ratio index of the noise reduction acoustic emission signal to generate a sensor reliability weight;

[0081] using the sensor reliability weight to weight and optimize the initial time difference positioning coordinate to generate a crack position coordinate.

[0082] Specifically, first, the system analyzes the crack type characteristics output by the dynamic feature analysis in the previous step to obtain the acoustic wave propagation attenuation factor. Crack type characteristics, such as "micro-cracks" or "through cracks", contain information about the physical size and energy release characteristics of the crack source. The system converts this qualitative type characteristic into a quantitative physical parameter, i.e., the acoustic wave propagation attenuation factor, through a pre-set physical model library or mapping relationship. This factor describes the rate at which the energy of the acoustic emission wave generated by the specific type of crack attenuates with distance as it propagates in the porcelain sleeve material. It is a constant related to the material itself and the acoustic wave frequency. At the same time, the system uses the known spatial position parameters of each acoustic emission sensor, i.e., the precise positions of each sensor in the three-dimensional coordinate system, and the time difference between the arrival of the acoustic wave at each sensor determined from the denoised acoustic emission signal, to calculate an initial crack source location through a standard time difference positioning algorithm (TDOA). This is the initial TDOA positioning coordinate, which is a preliminary estimate based purely on geometric relationships and sound speed. The next key step is to generate sensor credibility weights. This step combines two different dimensions of information. The first is the acoustic wave propagation attenuation factor obtained from the previous step. The second is the real-time calculation of the signal-to-noise ratio (SNR) for each sensor receiving the denoised acoustic emission signal. The SNR directly reflects the clarity and quality of the signal. The system combines these two to generate the credibility weight of each sensor. A high credibility of a sensor means that the signal it receives is clear (high SNR) and conforms to the physical attenuation law predicted by the acoustic wave propagation attenuation factor from the preliminary positioning point to the sensor location. Conversely, if a sensor's signal-to-noise ratio is low or its signal strength deviates significantly from the predicted value of the physical attenuation model, its credibility weight will be reduced. Finally, the initial TDOA positioning coordinates are optimized using a set of sensor credibility weights tailored for each sensor. This optimization process is an iterative calculation that re-solves the positioning equation set, but in the solving process, the contribution of each sensor's measurement data (i.e., the time difference) to the final result is adjusted by its corresponding sensor credibility weight. This optimization process can be conceptually represented as:

[0083] ,

[0084] In this formula, is the final optimized crack location coordinate. is the crack location coordinate variable to be solved. is the credibility weight of the th sensor, which is calculated by combining the signal-to-noise ratio and the acoustic wave propagation attenuation factor. is the actual measured time difference between the signal arriving at the th sensor and the reference sensor. is the theoretically calculated signal arrival time difference. The goal of this formula is to find a location such that the weighted sum of squared errors between all sensor measurements and the theoretically calculated data is minimized.

[0085] Optionally, the analyzing the crack type feature to obtain the sound wave propagation attenuation factor comprises:

[0086] physically field-mapping analyzing the crack type feature to obtain a crack equivalent geometric configuration parameter set;

[0087] obtaining the sound wave propagation attenuation factor based on the crack equivalent geometric configuration parameter set.

[0088] ​Specifically, first, the input crack type feature is analyzed by physical field mapping. The crack type feature, such as "micro crack", "fatigue crack" or "brittle fracture", is a qualitative or semi-quantitative classification label output by the previous dynamic feature analysis module. The physical field mapping analysis is to construct an expert knowledge base or a physical simulation database through a pre-constructed expert knowledge base or a physical simulation database. The construction method of the expert knowledge base or the physical simulation database is as follows: Database core content: store the mapping relationship between different crack type features (such as micro crack, through crack) and the corresponding equivalent geometric configuration parameter group (including crack length, depth, orientation angle, cross section shape coefficient), and the correction coefficient under different porcelain bushing materials (such as alumina ceramic, zirconia ceramic) and size specifications; Data source and construction steps: expert knowledge base: through the collection of 500+ groups of measured data of known cracks (including laboratory artificial sample preparation and field failure cases), the crack type and geometric parameters are labeled by more than 3 experts in the field, and after consistency verification (Kappa coefficient ≥ 0.85), they are stored in the database; Physical simulation database: a three-dimensional model of the porcelain bushing is established by using finite element software (such as ANSYS), 1000+ kinds of cracks with preset geometric parameters (length 0.1-5mm, depth 0.05-2mm, orientation angle 0°-90°) are simulated, the acoustic emission signal propagation characteristics are calculated, the characteristic parameters corresponding to the crack type are extracted, and the simulation data group is formed. Database updating mechanism: every 100 new measured data is accumulated, the mapping relationship is updated through incremental learning to ensure coverage of new crack morphology. This classification label is "translated" into a group of parameters that can describe the physical morphology of the crack source, i.e. the crack equivalent geometric configuration parameter group. This parameter group may include the equivalent length, width, orientation angle of the crack, and the coefficient representing the surface roughness, etc. For example, "micro crack" may be mapped to a small defect with a size in microns and an ellipsoidal shape; while "through crack" may be mapped to an elongated rectangular plane across the wall thickness of the porcelain bushing. This mapping relationship is pre-established through a large number of finite element simulation analysis, or combined with metallographic experimental data. Next, based on the crack equivalent geometric configuration parameter group obtained in the previous step, the system calculates the final acoustic wave propagation attenuation factor based on acoustic theory. The attenuation of acoustic wave propagation in medium is mainly composed of two parts, namely the absorption attenuation of material itself and the geometric attenuation caused by scattering. The material absorption attenuation is related to the medium properties and the frequency of the acoustic wave, and for a specific porcelain bushing, it can be regarded as a background constant. The geometric attenuation is closely related to the characteristics of the sound source, i.e. the geometric configuration of the crack, and the frequency of the acoustic wave. The system uses acoustic scattering theory models, such as Kirchhoff approximation or Born approximation, to take the crack equivalent geometric configuration parameter group as input. These models can calculate the attenuation law of the sound field radiated by a sound source with a specific geometric shape during propagation due to the diffusion to the surrounding. The calculation process finally outputs one or a group of frequency-dependent acoustic wave propagation attenuation factors. This factor can be represented as:

[0089] ,

[0090] In the formula, represents the final obtained sound wave propagation attenuation factor. represents a crack equivalent geometry parameter group containing crack equivalent length, width, orientation and other information, which is obtained by crack type feature mapping analysis. represents the main frequency or characteristic frequency of the acoustic emission signal, which can be obtained from the spectrum analysis of the noise-reduced acoustic emission signal. represents a calculation function that maps crack geometry and sound wave frequency to attenuation factor based on acoustic scattering theory.

[0091] Based on the same inventive concept, as shown in Figure 6 The application also provides a porcelain sleeve internal crack detection system based on acoustic emission detection, which comprises:

[0092] A multi-source sensing acquisition module is configured to acquire initial acoustic emission signals, vibration signals, electromagnetic interference signals and spatial position parameters.

[0093] A joint interference analysis module is configured to generate an interference feature vector according to the vibration signals and the electromagnetic interference signals.

[0094] An adaptive cancellation execution module is configured to perform adaptive interference cancellation processing on the initial acoustic emission signals to generate noise-reduced acoustic emission signals.

[0095] A time-frequency feature extraction module is configured to perform time-frequency feature extraction on the noise-reduced acoustic emission signals to generate a multi-dimensional acoustic emission feature vector.

[0096] A crack identification module is configured to perform dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature.

[0097] A weighted positioning module is configured to perform feature weighted positioning calculation on the crack type feature and the spatial position parameters to generate crack position coordinates when the crack existence flag is true.

[0098] To verify the feasibility of the application in practice, the application is applied to the health state online monitoring of a high-voltage side porcelain bushing of a main transformer in a certain 500 kV substation. The substation environment is complex, with strong mechanical vibration caused by the transformer body and cooling fans, and strong electromagnetic field interference generated by the high-voltage bus, which poses a severe challenge to the detection of weak acoustic emission signals inside the porcelain bushing. In this embodiment, 4 acoustic emission sensors, 1 three-axis vibration sensor and 1 electromagnetic interference probe are arranged at key positions on the surface of the porcelain bushing, and the three-dimensional spatial position parameters of each acoustic emission sensor are accurately calibrated. The system continuously monitors from August 1 to September 30, 2023 to verify the effectiveness of the application.

[0099] During the monitoring period, the system collects the initial acoustic emission signal, vibration signal and electromagnetic interference signal in real time. For the vibration signal, the system extracts its frequency spectrum peak value feature through fast Fourier transform, and identifies the significant peaks of 100 Hz and 150 Hz caused by device resonance. For the electromagnetic interference signal, the system extracts the energy and phase of 50 Hz power frequency and its 3rd harmonic (150 Hz) and 5th harmonic (250 Hz) as power frequency harmonic features. Subsequently, the system fuses these two groups of features to generate an interference feature vector that accurately describes the current environment. Based on this vector, the system generates a phase-inverted cancellation signal and superimposes it with the initial acoustic emission signal in real time to obtain a denoised acoustic emission signal with significantly improved signal-to-noise ratio.

[0100] On September 5, 2023, at 11:22:15, the system captured a group of high-energy transient signals. After the above-mentioned adaptive interference cancellation processing, the signal-to-noise ratio of the signal was improved from the original 6.2 dB to 23.5 dB. Then, the system extracts the time-frequency features of the denoised acoustic emission signal through complementary methods such as wavelet transform and Hilbert-Huang transform, and combines parameters such as peak amplitude and ring count to generate a multi-dimensional acoustic emission feature vector containing 64 dimensions.

[0101] The system then performs dynamic feature analysis on the multi-dimensional acoustic emission feature vector. First, the time-domain energy is extracted to obtain the energy decay time sequence of the event, and an exponential model fitting is performed on it. The determination coefficient R 2 of the fitting result (i.e. the crack existence probability value) is 0.97, which is much higher than the preset judgment threshold of 0.90. Therefore, the system determines that the crack existence flag is true, and according to the energy decay rate, peak energy and other parameters, the crack type feature is identified as "internal micro-crack".

[0102] Since the crack existence flag is true, the system immediately starts the feature-weighted positioning calculation. First, according to the type feature of "internal micro-crack", the corresponding sound wave propagation attenuation factor is obtained from the built-in physical model library. At the same time, according to the time difference of signals received by each sensor, the initial time difference positioning coordinates are calculated as (125.3, 88.1, 450.6) mm. Subsequently, the system fuses the sound wave propagation attenuation factor and the signal-to-noise ratio index of signals received by each sensor to generate the sensor credibility weight. For example, the signal-to-noise ratio of the No. 2 sensor closest to the crack source is the highest and the attenuation conforms to the physical model, so its credibility weight is given as 0.95; while the signal quality of the No. 4 sensor far away and affected by the edge effect is poor, so its weight is only 0.32. Finally, the initial coordinates are weighted and optimized using this set of weights to generate the final crack position coordinates as (121.7, 85.9, 452.1) mm.

[0103] To verify the accuracy of the detection result, the substation arranged for maintenance in October, and the porcelain sleeve was disassembled and checked. The inspection result found that there was indeed an internal micro-crack with a length of about 3 mm at the coordinates (121.1, 86.4, 453.5) mm, and the error of the positioning result of the method was less than 5 mm, which verified the high accuracy of the method.

[0104] Table 1: Data table of porcelain sleeve interference signal features and noise reduction effects

[0105]

[0106] Table 2: Data table of porcelain sleeve crack dynamic feature analysis and type judgment

[0107]

[0108] Table 3: Data table of porcelain sleeve crack positioning calculation process and precision comparison

[0109]

[0110] From the above table data, it can be seen that the application shows excellent performance in a complex industrial field environment. Table 1 data shows that the adaptive interference cancellation method of the application can specifically eliminate the main mechanical vibration and electromagnetic interference in the field, and the signal signal-to-noise ratio is improved by more than 17dB on average, providing a high-quality signal basis for subsequent accurate analysis. Table 2 data shows that through dynamic energy attenuation analysis, the application can effectively distinguish between real crack signals and incidental noise. In the event of September 5, the fitting degree as high as 0.97 accurately triggered the crack alarm, while other noise events with higher energy were correctly excluded because their energy attenuation patterns did not conform to the physical model, proving the high reliability and low false alarm rate of the application in crack identification. Table 3 data clearly shows the advantages of feature weighting positioning calculation. By introducing the physical attenuation model corresponding to the crack type and signal quality evaluation, the system intelligently gives different weights to different sensors, effectively correcting the initial positioning result based purely on geometry, and the final positioning error is only 4.14mm. If only the initial time difference positioning is used, the error will exceed 10mm, which shows that the positioning accuracy of the application has been significantly improved compared with the traditional method.

[0111] It should be noted that the electrical connection between the above-mentioned units does not necessarily represent the direct connection of the line, and the indirect connection mode can be applied to the embodiments of the application as long as the purpose of the application is achieved. The above-described is only an exemplary embodiment of the application, and cannot limit the scope of the application.

[0112] That is, any equivalent changes and modifications made according to the teachings of the application are still within the scope of the application. Other embodiments of the application will be readily apparent to those skilled in the art upon considering the specification and practice of the true principles disclosed. The present application is intended to cover any variations, uses, or adaptive changes of the application following the general principles of the application and including common knowledge or conventional technical means in the art not disclosed by the application.

Claims

1. A method for detecting internal cracks in porcelain bushings based on acoustic emission detection, characterized in that, The method comprises: acquiring an initial acoustic emission signal, a vibration signal, an electromagnetic interference signal, and a spatial position parameter; generating an interference feature vector according to the vibration signal and the electromagnetic interference signal; performing adaptive interference cancellation processing on the initial acoustic emission signal to generate a noise-reduced acoustic emission signal; performing time-frequency feature extraction on the noise-reduced acoustic emission signal to generate a multi-dimensional acoustic emission feature vector; performing dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature, wherein the dynamic feature analysis comprises: performing dynamic energy attenuation analysis on the multi-dimensional acoustic emission feature vector to obtain a crack existence probability value; and performing pattern classification judgment on the crack existence probability value to output the crack existence flag and the crack type feature; when the crack existence flag is true, performing feature weighted positioning calculation on the crack type feature and the spatial position parameter to generate a crack position coordinate, wherein the feature weighted positioning calculation comprises: analyzing the crack type feature to obtain an acoustic wave propagation attenuation factor; calculating an initial time difference positioning coordinate according to the spatial position parameter; fusing the acoustic wave propagation attenuation factor and a signal-to-noise ratio index of the noise-reduced acoustic emission signal to generate a sensor credibility weight; and performing weighted optimization on the initial time difference positioning coordinate by using the sensor credibility weight to generate the crack position coordinate.

2. The method for detecting internal crack of porcelain bushing based on acoustic emission detection according to claim 1, characterized in that, The generation of the interference feature vector comprises: extracting a frequency spectrum peak value feature of the vibration signal and a power frequency harmonic feature of the electromagnetic interference signal; performing feature fusion on the frequency spectrum peak value feature and the power frequency harmonic feature to generate the interference feature vector.

3. The method for detecting internal crack of porcelain bushing based on acoustic emission detection according to claim 1, characterized in that, The adaptive interference cancellation processing comprises: generating a phase-inverted cancellation signal according to the interference feature vector; performing real-time superposition operation on the phase-inverted cancellation signal and the initial acoustic emission signal to generate the noise-reduced acoustic emission signal.

4. The method for detecting internal crack of porcelain bushing based on acoustic emission detection according to claim 1, characterized in that, The dynamic energy attenuation analysis on the multi-dimensional acoustic emission feature vector to obtain the crack existence probability value comprises: performing time-domain energy extraction on the multi-dimensional acoustic emission feature vector to obtain an energy attenuation time sequence; performing exponential model fitting on the energy attenuation time sequence to obtain the crack existence probability value.

5. The method for detecting internal crack of porcelain bushing based on acoustic emission detection according to claim 1, characterized in that, The time-frequency feature extraction on the noise-reduced acoustic emission signal to generate the multi-dimensional acoustic emission feature vector comprises: extracting a time-frequency feature of the noise-reduced acoustic emission signal to obtain a complementary basis transformation parameter set; performing feature fusion and quantization calculation on the complementary basis transformation parameter set to generate the multi-dimensional acoustic emission feature vector.

6. The method for detecting internal crack of porcelain bushing based on acoustic emission detection according to claim 3, characterized in that, The generation of the phase-inverted cancellation signal according to the interference feature vector comprises: generating a multi-channel resonance reference waveform according to the interference feature vector; performing nonlinear phase shift processing on the multi-channel resonance reference waveform to generate the phase-inverted cancellation signal.

7. The method for detecting internal crack of porcelain bushing based on acoustic emission detection according to claim 1, characterized in that, The analysis of the crack type feature to obtain the acoustic wave propagation attenuation factor comprises: performing physical field mapping analysis on the crack type feature to obtain a crack equivalent geometric configuration parameter group; obtaining the acoustic wave propagation attenuation factor based on the crack equivalent geometric configuration parameter group.

8. A porcelain sleeve internal crack detection system based on acoustic emission detection, applied to the porcelain sleeve internal crack detection method based on acoustic emission detection according to any one of claims 1 to 7, characterized in that, The system comprises: a multi-source sensing acquisition module configured to acquire an initial acoustic emission signal, a vibration signal, an electromagnetic interference signal, and a spatial position parameter; The joint interference analysis module is configured to generate an interference feature vector according to the vibration signal and the electromagnetic interference signal; The adaptive cancellation execution module is configured to perform adaptive interference cancellation processing on the initial acoustic emission signal to generate a noise-reduced acoustic emission signal; The time-frequency feature extraction module is configured to perform time-frequency feature extraction on the noise-reduced acoustic emission signal to generate a multi-dimensional acoustic emission feature vector; The crack identification module is configured to perform dynamic feature analysis on the multi-dimensional acoustic emission feature vector to output a crack existence flag and a crack type feature, including: performing dynamic energy attenuation analysis on the multi-dimensional acoustic emission feature vector to obtain a crack existence probability value; performing pattern classification judgment on the crack existence probability value to output the crack existence flag and the crack type feature; The weighted positioning module is configured to perform feature weighted positioning calculation on the crack type feature and the spatial position parameter to generate a crack position coordinate when the crack existence flag is true, including: analyzing the crack type feature to obtain an acoustic wave propagation attenuation factor; calculating an initial time difference positioning coordinate according to the spatial position parameter; fusing the acoustic wave propagation attenuation factor and a signal-to-noise ratio index of the noise-reduced acoustic emission signal to generate a sensor credibility weight; and performing weighted optimization on the initial time difference positioning coordinate by using the sensor credibility weight to generate the crack position coordinate.

Images