Method for extracting features of ultrasonic waves generated by typical discharge phenomenon in complex working condition environment

Abstract

The invention provides a method for extracting features of ultrasonic waves generated by a typical discharge phenomenon in a complex working condition environment, and belongs to the technical field of power equipment.The method comprises the steps that a transient feature vector is extracted by conducting synchronous compression wavelet transform on a signal component, the transient feature vector is input into a waveform propagation enhancement model, and an enhanced feature vector is output; stochastic resonance amplification is carried out on the enhanced feature vector to establish a weak feature vector, the weak feature vector and the transient feature vector are fused to establish an equivalent discharge intensity matrix, a discharge type discrimination index is calculated, the discharge type is determined according to the discrimination index, and a propagation speed parameter of a multi-path delay correction matrix is adaptively adjusted. And on the basis of the updated propagation path analysis result, a time difference of arrival matrix and a maximum likelihood estimation method are utilized to calculate the spatial position coordinates of the discharge source, and the problem of low discharge ultrasonic feature extraction accuracy caused by multi-medium propagation path interference in a complex working condition environment is solved.

Description

Technical Field

[0001] This invention belongs to the field of power equipment technology, and more specifically, relates to a method for extracting ultrasonic features generated by typical discharge phenomena under complex working conditions. Background Technology

[0002] In partial discharge detection of power equipment, ultrasonic detection methods collect the acoustic signals generated by the discharge by deploying sensor arrays, and combine time-frequency analysis and pattern recognition technologies to identify and locate the discharge type. In practical applications, power equipment typically contains multiple media such as transformer oil, insulating gases, and solid insulating materials. During propagation, ultrasonic waves undergo reflection, refraction, and mode conversion at the interfaces of different media, resulting in the sensor receiving a signal containing a superposition of direct waves and multiple reflected waves. In current power equipment discharge detection, due to the significant differences in the sound velocity of ultrasonic waves in gaseous, liquid, and solid media, and the mutual conversion between longitudinal and transverse waves at the media interfaces, traditional methods struggle to accurately separate signal components from different propagation paths. This leads to the extracted time-frequency feature parameters being severely affected by multipath interference, especially weak discharge signals which are easily submerged by multiple reflected waves. In other words, existing technologies suffer from the technical problem of low accuracy in extracting ultrasonic discharge features due to multi-media propagation path interference under complex operating conditions. Summary of the Invention

[0003] In view of this, the present invention provides a method for extracting ultrasonic features generated by typical discharge phenomena under complex working conditions, which can solve the technical problem of low accuracy of ultrasonic feature extraction due to interference from multiple media propagation paths under complex working conditions in the prior art.

[0004] This invention is implemented as follows: It provides a method for extracting ultrasonic features from typical discharge phenomena under complex operating conditions. Multiple ultrasonic sensors are deployed on the surface of the power equipment casing to collect the raw ultrasonic signals generated by the discharge, simultaneously recording the spatial coordinates of each sensor and the arrival time of the raw ultrasonic signal. The raw ultrasonic signals are input into a multi-path delay correction matrix for propagation path analysis, identifying the propagation paths of direct waves and multiple reflected waves, and separating ultrasonic signal components with different propagation modes. The ultrasonic signal components are then subjected to synchronous compressed wavelet transform processing to extract the time-frequency characteristic parameters of transient pulses and establish transient feature vectors. These transient feature vectors are input into a waveform propagation enhancement model, simulating sound waves in a gaseous medium through a waveform propagation network structure. The propagation interference process in liquid and solid media is analyzed, and the enhanced eigenvector is output. Weak signals are extracted from the enhanced eigenvector, and stochastic resonance amplification is achieved by adjusting the bistable system parameters to establish a weak eigenvector. The weak eigenvector is fused with the transient eigenvector to establish an equivalent discharge intensity matrix, and the discharge type discrimination index is calculated. The discharge type is determined based on the numerical range of the discharge type discrimination index. The propagation velocity parameters of the multipath delay correction matrix are adjusted according to the determined discharge type to update the propagation path analysis results. Based on the updated propagation path analysis results, the spatial coordinates of the discharge source are calculated using the arrival time difference matrix and the maximum likelihood estimation method, and the spatial coordinates and characteristic parameter set of the discharge source are output.

[0005] Specifically, the establishment of the multipath delay correction matrix involves establishing a three-dimensional acoustic propagation model of the power equipment, using a ray tracing algorithm to simulate the propagation path of ultrasonic waves from the discharge point to each ultrasonic sensor, calculating the medium type and propagation distance of each propagation path, calculating the theoretical propagation delay of each propagation path based on the sound velocity parameters of different medium types, and calculating the energy attenuation factor of multiple reflection paths considering the reflection coefficient and transmission coefficient of the medium interface.

[0006] The three-dimensional acoustic propagation model of the power equipment includes a gas medium region, a liquid medium region, and a solid structure region. The multipath delay correction matrix establishes a two-dimensional matrix for all propagation paths according to the ultrasonic sensor number and path sequence number. The matrix elements include three parameters: theoretical propagation delay, energy attenuation factor, and medium type weight.

[0007] Specifically, the establishment of the transient feature vector involves performing synchronous compressed wavelet transform on the ultrasonic signal component to decompose it into wavelet coefficients of multiple scales, extracting the instantaneous frequency and instantaneous amplitude corresponding to the three scales with the largest amplitude among the wavelet coefficients, calculating the pulse rise time, peak amplitude, pulse width and decay time constant of the ultrasonic signal component, and statistically analyzing the energy distribution of the ultrasonic signal component on the time-frequency plane to calculate the main frequency bandwidth and frequency centroid.

[0008] The transient feature vector is composed of seven parameters normalized: pulse rise time, peak amplitude, pulse width, decay time constant, instantaneous frequency, main frequency bandwidth, and frequency centroid.

[0009] The waveform propagation enhancement model adopts a hybrid architecture of waveform propagation network structure and multi-hop graph propagation network structure. The waveform propagation network structure contains three propagation layers, each of which simulates the propagation of sound waves in a medium. The first propagation layer simulates propagation in a gaseous medium, the second propagation layer simulates propagation in a liquid medium, and the third propagation layer simulates propagation in a solid medium.

[0010] Each propagation layer contains a longitudinal wave sub-network and a transverse wave sub-network. The longitudinal wave sub-network and the transverse wave sub-network exchange information through an interference module. The interference module uses a phase modulation mechanism to calculate the interference mode of the longitudinal wave propagation mode and the transverse wave propagation mode.

[0011] The multi-hop graph propagation network structure uses the seven parameters in the transient feature vector as graph nodes, establishes edge connections based on the physical correlation between the parameters, and propagates information in multiple hops on the graph node structure. The first hop propagation captures the correlation between time-domain parameters, the second hop propagation captures the correlation between frequency-domain parameters, and the third hop propagation captures the cross-correlation between time-domain parameters and frequency-domain parameters.

[0012] The number of propagation steps in the multi-hop graph propagation network structure is determined based on the normalized value of the peak amplitude in the transient feature vector. When the normalized value of the peak amplitude is greater than the first peak normalization threshold, two-hop propagation is used. When the normalized value of the peak amplitude is between the second peak normalization threshold and the first peak normalization threshold, three-hop propagation is used. When the normalized value of the peak amplitude is less than the second peak normalization threshold, four-hop propagation is used.

[0013] The interference intensity parameter of the waveform propagation network structure is determined based on the maximum value of the medium type weight in the multipath delay correction matrix. When the weight of the gas medium type is the largest, the interference intensity parameter is set as the first interference intensity value; when the weight of the liquid medium type is the largest, the interference intensity parameter is set as the second interference intensity value; and when the weight of the solid medium type is the largest, the interference intensity parameter is set as the third interference intensity value.

[0014] Specifically, the establishment of the weak feature vector involves performing random resonance processing on the signal components in the enhanced feature vector whose peak amplitude is less than the weak signal threshold to construct a bistable system. The signal components are input into the bistable system while the collected background noise is superimposed. The signal-to-noise ratio of the output signal of the bistable system is maximized by adjusting the potential well spacing parameter and the damping coefficient parameter of the bistable system.

[0015] Specifically, the establishment of the equivalent discharge intensity matrix involves fusing the seven parameters in the transient eigenvector and the four parameters in the weak eigenvector, performing principal component analysis on the eleven parameters after fusion to extract the first five principal components, calculating the contribution rate of each principal component as a weight coefficient, and multiplying the five principal components by their corresponding weight coefficients to form the first row of the equivalent discharge intensity matrix.

[0016] The calculation of the Euclidean distances between the enhanced eigenvector and the standard tip discharge eigenvector, the standard surface discharge eigenvector, and the standard internal discharge eigenvector, and the normalization of these three Euclidean distance values ​​constitutes the second row of the equivalent discharge intensity matrix.

[0017] The discharge type discrimination index is calculated by a discharge discrimination function. The discharge discrimination function normalizes the discharge type discrimination index by dividing the weighted sum of the five matrix elements in the first row of the equivalent discharge intensity matrix by the sum of the weighted sum and the standard discharge energy threshold, normalizes the discharge type discrimination index by dividing the minimum distance value by the sum of three Euclidean distance values, and normalizes the discharge type discrimination index by dividing the instantaneous frequency parameter by the sum of the instantaneous frequency parameter and the standard frequency threshold. The three normalized values ​​are summed according to weight coefficients of 0.5, 0.3, and 0.2.

[0018] Specifically, a discharge type discrimination index greater than the first discrimination threshold is considered a tip discharge, while a discharge type discrimination index below the first discrimination threshold is considered a tip discharge. When the discharge type discrimination index is less than the second discrimination threshold, it is determined to be surface discharge.

[0019] Specifically, for tip discharge, the gas sound velocity weight is increased to the first sound velocity weight value; for surface discharge, the solid sound velocity weight is increased to the second sound velocity weight value; and for internal discharge, the liquid sound velocity weight is increased to the third sound velocity weight value. After updating the propagation path analysis results, the spatial location coordinates of the discharge source are calculated using the arrival time difference matrix and the maximum likelihood estimation method.

[0020] This invention separates direct waves and multiple reflected waves by establishing a multi-path delay correction matrix, extracts transient features by combining synchronous compressed wavelet transform, and utilizes a waveform propagation enhancement model to simulate the propagation interference process of sound waves in different media, effectively eliminating the impact of multi-path interference on feature extraction. The waveform propagation network structure employed in this invention simulates sound wave propagation in gaseous, liquid, and solid media in a layered manner, constructing longitudinal and transverse wave sub-networks within each propagation layer. An interference module models the energy conversion relationship between wave modes, accurately characterizing the physical behavior of ultrasound at multi-media interfaces. Simultaneously, the multi-hop graph propagation network structure establishes long-distance dependencies between time-domain and frequency-domain parameters, enabling weak features to be effectively amplified through the propagation process of graph nodes, significantly improving feature extraction capabilities under complex conditions. In summary, this invention solves the technical problem mentioned in the background art where multi-media propagation path interference leads to low accuracy in extracting discharge ultrasound features. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a distribution map of the normalized values ​​of the transient feature vectors in the embodiment.

[0023] Figure 3 This is a time-domain waveform diagram of the output signal of the bistable system in the embodiment.

[0024] Figure 4 This is a three-dimensional spatial coordinate diagram of the discharge source location in the embodiment. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0026] like Figure 1 The diagram shows a flowchart of a method for extracting ultrasonic features from typical discharge phenomena under complex working conditions, provided by the present invention. This method includes the following steps:

[0027] S01. Multiple ultrasonic sensors are arranged on the surface of the power equipment casing to collect the original ultrasonic signals generated by the discharge, and the spatial coordinates of each ultrasonic sensor and the arrival time of the original ultrasonic signal are recorded synchronously.

[0028] S02. Input the original ultrasound signal into the multipath delay correction matrix to perform propagation path analysis, identify the propagation paths of direct waves and multiple reflection waves, and separate the ultrasound signal components of different propagation modes.

[0029] S03. Perform synchronous compressed wavelet transform processing on the ultrasonic signal components, extract the time-frequency characteristic parameters of the transient pulse, and establish a transient feature vector;

[0030] S04. Input the transient feature vector into the waveform propagation enhancement model, simulate the propagation interference process of sound waves in gaseous, liquid and solid media through the waveform propagation network structure, and output the enhanced feature vector.

[0031] S05. Extract weak signals from the enhanced feature vector, and achieve stochastic resonance amplification by adjusting the bistable system parameters to establish a weak feature vector.

[0032] S06. The weak feature vector and the transient feature vector are fused to establish an equivalent discharge intensity matrix and the discharge type discrimination index is calculated.

[0033] S07. Determine the discharge type based on the numerical range of the discharge type discrimination index. When the discharge type discrimination index is greater than the first discrimination threshold, it is determined to be a tip discharge. When the discharge type discrimination index is between the second discrimination threshold and the first discrimination threshold, it is determined to be a surface discharge. When the discharge type discrimination index is less than the second discrimination threshold, it is determined to be an internal discharge.

[0034] S08. Adjust the propagation speed parameters of the multipath delay correction matrix according to the determined discharge type. For the tip discharge, increase the gas sound speed weight to the first sound speed weight value. For the surface discharge, increase the solid sound speed weight to the second sound speed weight value. For the internal discharge, increase the liquid sound speed weight to the third sound speed weight value. Update the propagation path analysis results.

[0035] S09. Based on the updated propagation path analysis results, calculate the spatial location coordinates of the discharge source using the arrival time difference matrix and the maximum likelihood estimation method, and output the spatial location coordinates and characteristic parameter set of the discharge source.

[0036] The process of establishing the multipath delay correction matrix includes: establishing a three-dimensional acoustic propagation model of the power equipment, which includes a gas medium region, a liquid medium region, and a solid structure region; using a ray tracing algorithm to simulate the propagation path of ultrasonic waves from the discharge point to each ultrasonic sensor, and calculating the medium type and propagation distance of each propagation path; calculating the theoretical propagation delay of each propagation path based on the sound velocity parameters of different medium types; considering the reflection coefficient and transmission coefficient of the medium interface, calculating the energy attenuation factor of multiple reflection paths; establishing a two-dimensional matrix of all propagation paths according to the ultrasonic sensor number and path sequence number, wherein the matrix elements of the two-dimensional matrix include three parameters: theoretical propagation delay, energy attenuation factor, and medium type weight; and the multipath delay correction matrix is ​​established by... Line number The matrix elements of the column represent the distance from the discharge point to the nth column. The ultrasonic sensor's first The comprehensive parameters of each propagation path.

[0037] The process of establishing the transient feature vector includes: performing synchronous compressed wavelet transform on the ultrasonic signal component to decompose the ultrasonic signal component into wavelet coefficients of multiple scales; extracting the instantaneous frequency and instantaneous amplitude corresponding to the three scales with the largest amplitude among the wavelet coefficients; calculating the pulse rise time, peak amplitude, pulse width, and decay time constant of the ultrasonic signal component; statistically analyzing the energy distribution of the ultrasonic signal component in the time-frequency plane and calculating the main frequency bandwidth and frequency centroid; and normalizing the seven parameters—pulse rise time, peak amplitude, pulse width, decay time constant, instantaneous frequency, main frequency bandwidth, and frequency centroid—to form the transient feature vector.

[0038] The waveform propagation enhancement model is structured as follows: It employs a hybrid architecture of waveform propagation network and multi-hop graph propagation network. The waveform propagation network contains three propagation layers, each simulating sound wave propagation in a different medium: the first layer simulates propagation in a gaseous medium, the second in a liquid medium, and the third in a solid medium. Each layer contains longitudinal wave sub-networks and transverse wave sub-networks, which interact via an interference module. This interference module uses a phase modulation mechanism to calculate the interference patterns of the longitudinal and transverse wave propagation modes. The multi-hop graph propagation network uses the seven parameters in the transient feature vector as graph nodes, establishing edge connections based on the physical relationships between the parameters. Information propagates in multiple hops across the graph node structure: the first hop captures the relationships between time-domain parameters, the second hop captures the relationships between frequency-domain parameters, and the third hop captures the cross-relationships between time-domain and frequency-domain parameters. The multi-hop graph... The propagation steps of the propagation network structure are determined based on the normalized value of the peak amplitude in the transient feature vector. Two-hop propagation is used when the normalized value of the peak amplitude is greater than the first peak normalization threshold; three-hop propagation is used when the normalized value of the peak amplitude is between the second peak normalization threshold and the first peak normalization threshold; and four-hop propagation is used when the normalized value of the peak amplitude is less than the second peak normalization threshold. The interference intensity parameter of the waveform propagation network structure is determined based on the maximum value of the medium type weight in the multipath delay correction matrix. When the gas medium type weight is the largest, the interference intensity parameter is set to the first interference intensity value; when the liquid medium type weight is the largest, the interference intensity parameter is set to the second interference intensity value; and when the solid medium type weight is the largest, the interference intensity parameter is set to the third interference intensity value. The output layer of the waveform propagation enhancement model concatenates and fuses the feature vectors of the waveform propagation network structure and the multi-hop graph propagation network structure to output the enhanced feature vector.

[0039] The steps for establishing the waveform propagation enhancement model training dataset include: building a simulated discharge platform in a laboratory environment containing three media: gaseous, liquid, and solid; fabricating three typical discharge types: tip discharge, surface discharge, and internal discharge, with different discharge voltages and gaps set for each type; using multiple ultrasonic sensors to collect ultrasonic signals generated during the discharge process, recording the ultrasonic signal waveforms and ultrasonic sensor position information; labeling the collected ultrasonic signals, including the typical discharge type, discharge location coordinates, discharge intensity level, and medium propagation path; performing data enhancement processing on the labeled ultrasonic signals by adding Gaussian noise of different intensities, changing the sampling rate, and adjusting the signal amplitude to simulate different working conditions; and dividing the processed ultrasonic signals into a training set, a validation set, and a test set in an 8:1:1 ratio.

[0040] The training steps of the waveform propagation enhancement model include: training the waveform propagation enhancement model using supervised learning, with the transient feature vector as input and the enhanced feature vector and typical discharge type labels as output; setting the loss function as a weighted sum of feature reconstruction loss and classification loss, where the feature reconstruction loss is calculated using mean squared error and the classification loss is calculated using cross-entropy, with a weight ratio of 0.6:0.4; using an adaptive moment estimation optimizer to update parameters, with the initial learning rate set to the initial learning rate value, and the learning rate decaying to 0.5 times the original value every 20 training epochs; setting the batch size to 32 and the total number of training epochs to 150; monitoring the classification accuracy on the validation set during training, and stopping training early when the classification accuracy no longer improves for 10 consecutive epochs; and saving the model parameters with the highest classification accuracy as the final model parameters.

[0041] The waveform propagation network structure, by simulating the physical propagation process of ultrasound in different media, can accurately characterize the reflection and refraction behavior of sound waves at the interfaces of gas and solid media, and liquid and solid media. The interference modules of the longitudinal wave sub-network and the transverse wave sub-network realize the energy conversion relationship between wave modes, enabling the waveform propagation network structure to learn the wave interference laws in real physical scenarios. The multi-hop graph propagation network structure, by constructing the graph node structure relationship between feature parameters, realizes long-distance dependency capture between different physical quantities. The first hop propagation, the second hop propagation, and the third hop propagation establish internal correlations of time-domain parameters, internal correlations of frequency-domain parameters, and cross-correlations between time-domain parameters and frequency-domain parameters, respectively, so that weak discharge characteristics can be captured through graph nodes. The propagation process is amplified and identified; the medium-layer design of the waveform propagation network structure and the multi-hop propagation mechanism of the multi-hop graph propagation network structure work together. The waveform propagation network structure enhances the spatial dimension information of the enhanced feature vector from the perspective of physical propagation, while the multi-hop graph propagation network structure enhances the semantic dimension information of the enhanced feature vector from the perspective of parameter correlation. The fusion of the two significantly improves the detection capability of weak discharge signals; the adaptive adjustment mechanism of the interference intensity parameter and the number of propagation steps enables the waveform propagation enhancement model to dynamically optimize the feature extraction process according to the intensity of the input signal and the characteristics of the propagation path, and automatically select the optimal feature enhancement strategy for different typical discharge types and medium environments, avoiding the problem of insufficient adaptability of fixed parameter models when facing diverse working conditions.

[0042] The process of establishing the weak feature vector includes: performing random resonance processing on the signal components in the enhanced feature vector whose peak amplitude is less than the weak signal threshold; constructing a bistable system, wherein the potential function of the bistable system is a double-well structure, the two wells have equal depths and the barrier height is adaptively adjusted according to the noise intensity; inputting the signal components into the bistable system, while simultaneously superimposing the acquired background noise; adjusting the well spacing parameter and damping coefficient parameter of the bistable system to maximize the signal-to-noise ratio of the output signal of the bistable system; calculating the peak amplitude, pulse width, and spectral energy of the output signal of the adjusted bistable system; performing fourth-order cumulant calculation on the output signal of the bistable system to suppress Gaussian noise components; extracting the peak amplitude, pulse width, spectral energy, and fourth-order cumulant amplitude of the noise-suppressed bistable system output signal, normalizing them, and then forming the weak feature vector.

[0043] The process of establishing the equivalent discharge intensity matrix includes: fusing the seven parameters in the transient feature vector and the four parameters in the weak feature vector; performing principal component analysis on the eleven parameters after fusion to extract the first five principal components; calculating the contribution rate corresponding to each principal component and using the contribution rate as a weight coefficient; multiplying the five principal components by their corresponding weight coefficients to form the first row of the equivalent discharge intensity matrix; calculating the Euclidean distance between the enhanced feature vector and the standard tip discharge feature vector, the standard surface discharge feature vector, and the standard internal discharge feature vector, and normalizing the three Euclidean distance values ​​to form the second row of the equivalent discharge intensity matrix; the equivalent discharge intensity matrix is ​​a two-row, five-column matrix, the first row of the equivalent discharge intensity matrix reflects the energy distribution characteristics of the discharge signal, and the second row of the equivalent discharge intensity matrix reflects the type similarity characteristics of the discharge signal.

[0044] The discharge type discrimination index is calculated using a discharge discrimination function. This function comprehensively evaluates the energy distribution characteristics and type similarity characteristics in the equivalent discharge intensity matrix. Inputs include the weighted sum of the five matrix elements in the first row of the equivalent discharge intensity matrix, the minimum distance value among the three Euclidean distance values ​​in the second row of the equivalent discharge intensity matrix, and the instantaneous frequency parameter in the transient feature vector. The output is the discharge type discrimination index. The calculation process involves normalizing the weighted sum of the five matrix elements in the first row by dividing the weighted sum by the sum of the weighted sum and the standard discharge energy threshold; normalizing the minimum distance value by dividing the minimum distance value by the sum of the three Euclidean distance values; normalizing the instantaneous frequency parameter by dividing the instantaneous frequency parameter by the sum of the instantaneous frequency parameter and the standard frequency threshold; and summing the three normalized values ​​using weighting coefficients of 0.5, 0.3, and 0.2 to obtain the discharge type discrimination index.

[0045] The process of establishing the arrival time difference matrix includes: selecting the ultrasonic sensor that receives the original ultrasonic signal earliest as the reference sensor; calculating the time difference between the time when each other ultrasonic sensor receives the original ultrasonic signal and the time when the reference sensor receives the original ultrasonic signal; performing propagation path correction on each time difference according to the updated multipath delay correction matrix, deducting the additional delay introduced by multiple reflection paths; arranging the corrected time differences in the order of ultrasonic sensor numbers to form an arrival time difference vector; establishing an objective function using the maximum likelihood estimation method, the objective function including the spatial location coordinates of the discharge source, the location coordinates of the ultrasonic sensors, the arrival time difference vector, and the sound velocity parameter; solving for the spatial location coordinates of the discharge source by optimizing the objective function, the optimization process using the gradient descent algorithm for iterative calculation; and combining the arrival time difference vector and the solved spatial location coordinates of the discharge source to form the arrival time difference matrix.

[0046] The method for obtaining the first discrimination threshold is as follows: 100 sets of tip discharge samples and 100 sets of surface discharge samples are created in a laboratory environment, and ultrasonic signals of the tip discharge samples and the surface discharge samples are collected; the discharge type discrimination index is calculated for the ultrasonic signals according to the methods of steps S01 to S06; the minimum value of the discharge type discrimination index of the tip discharge samples and the maximum value of the discharge type discrimination index of the surface discharge samples are counted; the arithmetic mean of the minimum value of the discharge type discrimination index of the tip discharge samples and the maximum value of the discharge type discrimination index of the surface discharge samples is taken as the first discrimination threshold.

[0047] The method for obtaining the second discrimination threshold is as follows: 100 sets of surface discharge samples and 100 sets of internal discharge samples are created in a laboratory environment, and ultrasonic signals of the surface discharge samples and the internal discharge samples are collected; the discharge type discrimination index is calculated for the ultrasonic signals according to the methods of steps S01 to S06; the minimum value of the discharge type discrimination index of the surface discharge samples and the maximum value of the discharge type discrimination index of the internal discharge samples are calculated; the arithmetic mean of the minimum value of the discharge type discrimination index of the surface discharge samples and the maximum value of the discharge type discrimination index of the internal discharge samples is taken as the second discrimination threshold.

[0048] The method for obtaining the first, second, and third sound speed weight values ​​is as follows: A three-dimensional acoustic propagation mathematical model of the power equipment is established, which includes three sound speed parameters: gas medium sound speed, liquid medium sound speed, and solid medium sound speed. For the physical process of ultrasonic waves generated by the tip discharge in the gas medium, the proportion of the gas medium propagation path to the total propagation path is calculated using the three-dimensional acoustic propagation mathematical model of the power equipment, and this proportion is used as the first sound speed weight value. For the physical process of ultrasonic waves generated by surface discharge on the surface of the solid medium, the proportion of the solid medium propagation path to the total propagation path is calculated using the three-dimensional acoustic propagation mathematical model of the power equipment, and this proportion is used as the second sound speed weight value. For the physical process of ultrasonic waves generated by internal discharge in the liquid medium, the proportion of the liquid medium propagation path to the total propagation path is calculated using the three-dimensional acoustic propagation mathematical model of the power equipment, and this proportion is used as the third sound speed weight value.

[0049] The method for obtaining the first peak normalization threshold is as follows: statistically analyze the distribution of the normalized values ​​of the peak amplitude in the training set, and calculate the 80th percentile of the normalized values ​​of the peak amplitude as the first peak normalization threshold.

[0050] The method for obtaining the second peak normalization threshold is as follows: statistically analyze the distribution of the normalized values ​​of the peak amplitude in the training set, and calculate the 40th percentile of the normalized values ​​of the peak amplitude as the second peak normalization threshold.

[0051] The method for obtaining the first interference intensity value, the second interference intensity value, and the third interference intensity value is as follows: in a laboratory environment, the energy conversion ratio of the longitudinal wave propagation mode to the transverse wave propagation mode of the ultrasonic wave is measured in a gaseous medium, a liquid medium, and a solid medium, respectively; the energy conversion ratio in the gaseous medium is taken as the first interference intensity value; the energy conversion ratio in the liquid medium is taken as the second interference intensity value; and the energy conversion ratio in the solid medium is taken as the third interference intensity value.

[0052] The initial learning rate value is obtained by using a learning rate range test method, setting the learning rate on the training set from a range of... Gradually increase to Perform test training and record the rate of loss function descent for different learning rates; select the learning rate with the fastest loss function descent and the most stable training process as the initial learning rate value.

[0053] The method for obtaining the weak signal threshold is as follows: statistically analyze the peak amplitude distribution of normal operation noise in the training set, calculate the 90th percentile of the peak amplitude of the normal operation noise; and take 1.5 times the 90th percentile of the peak amplitude of the normal operation noise as the weak signal threshold.

[0054] The method for obtaining the standard discharge energy threshold is as follows: a typical discharge type sample is created in a laboratory environment, and the ultrasonic signal of the typical discharge type sample is collected; the energy of the ultrasonic signal is calculated, and the energy distribution of the typical discharge type sample is statistically analyzed; the median of the energy distribution is calculated as the standard discharge energy threshold.

[0055] The method for obtaining the standard frequency threshold is as follows: a typical discharge type sample is created in a laboratory environment, and the ultrasonic signal of the typical discharge type sample is collected; the ultrasonic signal is subjected to spectrum analysis, and the main frequency distribution of the typical discharge type sample is statistically analyzed; the median of the main frequency distribution is calculated as the standard frequency threshold.

[0056] The method for obtaining the standard tip discharge feature vector, the standard surface discharge feature vector, and the standard internal discharge feature vector is as follows: 50 sets of tip discharge samples, 50 sets of surface discharge samples, and 50 sets of internal discharge samples are fabricated in a laboratory environment; the enhanced feature vectors are extracted from the tip discharge samples, the surface discharge samples, and the internal discharge samples according to steps S01 to S04; the average value of the enhanced feature vectors of the tip discharge samples is calculated as the standard tip discharge feature vector; the average value of the enhanced feature vectors of the surface discharge samples is calculated as the standard surface discharge feature vector; and the average value of the enhanced feature vectors of the internal discharge samples is calculated as the standard internal discharge feature vector.

[0057] The specific implementation methods of the above steps are described in detail below.

[0058] The specific implementation of step S01 involves selecting several placement points on the outer surface of the power equipment to install ultrasonic sensors. The selected sensors should meet the requirement of a wide frequency response range covering 20kHz to 500kHz, with a sensitivity of not less than -60dB. During placement, a three-dimensional coordinate measurement system is used to accurately record the spatial coordinates of each sensor, with the coordinate measurement accuracy reaching the millimeter level. The acquisition system uses a multi-channel synchronous acquisition card to digitize the analog signals output by each sensor, and the sampling frequency is set to above 2MHz to ensure accurate capture of high-frequency ultrasonic signals. The synchronous triggering mechanism achieves time alignment of signal acquisition from each channel by setting a threshold trigger or external triggering method, with the time synchronization error controlled within 1μs. The recorded raw ultrasonic signal should contain complete pulse waveform data and corresponding timestamp information.

[0059] The specific implementation of step S02 involves inputting the acquired raw ultrasonic signal into a pre-established multipath delay correction matrix for propagation path analysis and identification. This matrix is ​​constructed based on a three-dimensional acoustic propagation model of power equipment. The model includes gaseous medium regions such as air gaps, liquid medium regions such as insulating oil, and solid structure regions such as metal shells and insulating materials. A ray tracing algorithm is used to simulate all possible propagation paths of ultrasonic waves from the assumed discharge point to each sensor. The algorithm calculates the sequence of medium types traversed by each path and the propagation distance in each medium. Based on the parameters of approximately 340 m / s for gas sound, approximately 1400 m / s for liquid sound, and approximately 5000 m / s for solid sound, the theoretical propagation delay of each path is calculated. At the same time, the influence of the reflection coefficient and transmission coefficient of the medium interface on energy is considered, and the energy attenuation factor of multiple reflection paths is calculated. By comparing the arrival time of the measured signal with the theoretical delay of each path, the direct wave path and multiple reflection wave path are identified, and the ultrasonic signal components of different propagation modes are separated. The separation process is achieved using a time-delay-based signal decomposition technique.

[0060] The specific implementation of step S03 involves performing time-frequency domain joint analysis on the separated ultrasonic signal components using synchronous compressed wavelet transform. This transform method overcomes the limitation of time-frequency resolution by the uncertainty principle in traditional wavelet transform. By redistributing the energy distribution on the time-frequency plane, a clearer time-frequency representation is obtained. In the transform process, Morlet wavelet is selected as the mother wavelet function, and the signal is decomposed into wavelet coefficients of multiple scales. The three scales with the largest amplitude are identified from the wavelet coefficient matrix, and the instantaneous frequency and instantaneous amplitude parameters of the corresponding scales are extracted. At the same time, the pulse rise time (the duration of the rising edge from the baseline to the peak), peak amplitude (the maximum absolute value of the waveform), pulse width (full width at half maximum), and decay time constant (the time required for the amplitude to decay to 36.8% of the initial value) are measured on the time-domain waveform. The distribution characteristics of signal energy are statistically analyzed on the time-frequency plane, and the main frequency bandwidth and frequency centroid position are calculated. The seven parameters extracted—pulse rise time, peak amplitude, pulse width, decay time constant, instantaneous frequency, main frequency bandwidth, and frequency centroid—are normalized and then used to form a transient feature vector.

[0061] The specific implementation of step S04 involves inputting the transient feature vector into the waveform propagation enhancement model for feature enhancement processing. This model employs a hybrid architecture design combining a waveform propagation network structure and a multi-hop graph propagation network structure. The waveform propagation network structure contains three parallel propagation layers that respectively simulate the physical processes of sound wave propagation in gaseous, liquid, and solid media. Each propagation layer internally constructs a longitudinal wave sub-network and a transverse wave sub-network. Information interaction between the two wave modes is achieved through an interference module. The interference module uses a phase modulation mechanism to calculate the interference effect between the longitudinal and transverse waves. The interference intensity parameter is based on the maximum weight of the medium type in the multipath delay correction matrix. The value is adaptively determined: 0.3 for gaseous media, 0.5 for liquid media, and 0.7 for solid media. The multi-hop graph propagation network structure uses the seven parameters of the transient feature vector as graph nodes and establishes edge connections based on the physical correlation between parameters. Information is propagated in the graph in multiple hops to capture long-distance dependencies. The number of propagation steps is adaptively selected based on the peak amplitude normalization value. Two-hop propagation is used when the normalization value is greater than 0.7, three-hop propagation is used when it is between 0.4 and 0.7, and four-hop propagation is used when it is less than 0.4. The output layer concatenates and fuses the feature vectors of the two network structures to obtain the enhanced feature vector.

[0062] The specific implementation of step S05 involves performing random resonance processing on signal components in the enhanced feature vector whose peak amplitude is less than the weak signal threshold to achieve weak signal detection and amplification. The weak signal threshold is determined by multiplying the 90th percentile of the peak amplitude of normal operating noise in the statistical training set by 1.5. When constructing the bistable system, a dual-potential-well potential function structure is adopted, with the two potential wells having equal depths and the barrier height adaptively adjusted according to the intensity of the collected background noise. The weak signal component is superimposed with the background noise and input into the bistable system. The signal-to-noise ratio of the system output signal is maximized by adjusting the potential well spacing parameter and the damping coefficient parameter. The adjustment process uses an traversal search method to find the optimal parameter combination within the preset parameter range. The output peak amplitude, output pulse width, and output spectral energy of the optimized system output signal are calculated. The fourth-order cumulant is calculated on the output signal to suppress Gaussian noise components. The fourth-order cumulant has zero-value characteristics for Gaussian processes but remains sensitive to non-Gaussian signals. The four parameters of the noise-suppressed signal—output peak amplitude, output pulse width, output spectral energy, and fourth-order cumulant amplitude—are extracted, normalized, and then used to form the weak feature vector.

[0063] The specific implementation of step S06 involves fusing the seven parameters of the transient eigenvector with the four parameters of the weak eigenvector to form an eleven-dimensional feature space. Principal component analysis is then performed on the fused parameters to reduce the feature dimension and extract key information. The eigenvalues ​​and eigenvectors of the covariance matrix are calculated, and the first five principal components and their corresponding contribution rates are extracted. These contribution rates are used as weighting coefficients and multiplied by each principal component to form the first row of the equivalent discharge intensity matrix. This row reflects the energy distribution characteristics of the discharge signal. The Euclidean distance between the enhanced eigenvector and the standard tip discharge eigenvector, standard surface discharge eigenvector, and standard internal discharge eigenvector is calculated. The Euclidean distance is calculated using the distance between two points in the feature space. The straight-line distances are normalized to form the second row of the equivalent discharge intensity matrix. This row reflects the similarity between the discharge signal and typical discharge types. The discharge type discrimination index is calculated using a discharge discrimination function. The function input includes the weighted sum of the five elements in the first row of the equivalent discharge intensity matrix, the minimum value among the three distance values ​​in the second row, and the instantaneous frequency parameter in the transient feature vector. The weighted sum is normalized by dividing it by the sum of the standard discharge energy threshold, the minimum distance value is normalized by dividing it by the sum of the three distance values, and the instantaneous frequency parameter is normalized by dividing it by the sum of the standard frequency threshold. The three normalized values ​​are summed with weights of 0.5, 0.3, and 0.2 to obtain the discharge type discrimination index.

[0064] The specific implementation of step S07 is to classify and determine the discharge type based on the calculated discharge type discrimination index value. The first discrimination threshold is determined by calculating the discrimination index of 100 sets of tip discharge samples and 100 sets of surface discharge samples, and taking the arithmetic mean of the minimum value of the tip discharge discrimination index and the maximum value of the surface discharge discrimination index, with a reference value of approximately 0.75. The second discrimination threshold is determined by calculating the discrimination index of 100 sets of surface discharge samples and 100 sets of internal discharge samples, and taking the arithmetic mean of the minimum value of the surface discharge discrimination index and the maximum value of the internal discharge discrimination index, with a reference value of approximately 0.45. The discrimination logic is as follows: when the discrimination index is greater than the first discrimination threshold, it is determined to be tip discharge; when the discrimination index is between the second discrimination threshold and the first discrimination threshold, it is determined to be surface discharge; and when the discrimination index is less than the second discrimination threshold, it is determined to be internal discharge.

[0065] The specific implementation of step S08 is to adjust the propagation speed parameter in the multipath delay correction matrix according to the determined discharge type to improve positioning accuracy. For tip discharge, the gas sound speed weight is increased to the first sound speed weight value, which is determined by solving the proportion of the gas medium propagation path to the total path through the three-dimensional acoustic propagation mathematical model, with a reference value of about 0.8. For surface discharge, the solid sound speed weight is increased to the second sound speed weight value, which is the proportion of the solid medium propagation path, with a reference value of about 0.7. For internal discharge, the liquid sound speed weight is increased to the third sound speed weight value, which is the proportion of the liquid medium propagation path, with a reference value of about 0.75. After the weight adjustment, the theoretical time delay of each propagation path is recalculated, and the propagation path analysis results are updated to reflect the differences in sound wave propagation characteristics of different discharge types.

[0066] The specific implementation of step S09 is to calculate the spatial coordinates of the discharge source based on the updated propagation path analysis results, select the sensor that receives the earliest signal as the reference sensor, calculate the arrival time difference between other sensors and the reference sensor, perform path correction on the time difference according to the updated multipath delay correction matrix to deduct the additional time delay introduced by multiple reflections, arrange the corrected time differences according to the sensor number to form the arrival time difference vector, and establish the objective function using the maximum likelihood estimation method. The function includes the discharge source position coordinates to be solved, the known sensor position coordinates, the arrival time difference vector, and the sound speed parameter. The objective function is iteratively optimized using the gradient descent algorithm to obtain the spatial coordinates of the discharge source. During the iteration process, the coordinate estimate is continuously updated until the objective function converges, and the final spatial coordinates of the discharge source and a complete set of feature parameters including discharge type, discharge intensity, and feature parameters are output.

[0067] It should be noted that the key technical ideas of this invention include a multi-path delay correction mechanism based on a physical propagation model, a hybrid architecture design of a waveform propagation enhancement model, and a weak signal extraction method based on adaptive random resonance. These three technical ideas form a complete feature extraction and recognition system. The multi-path delay correction mechanism establishes an acoustic propagation model encompassing gas, liquid, and solid media, and uses a ray tracing algorithm to accurately calculate the propagation paths of direct waves and multiple reflected waves. This solves the positioning error problem caused by neglecting multipath effects in traditional methods. Compared to traditional methods that rely on a single sound velocity assumption, this mechanism can accurately characterize the actual propagation process of sound waves under complex operating conditions, significantly improving the accuracy of discharge source positioning. The waveform propagation enhancement model adopts a hybrid architecture of a waveform propagation network structure and a multi-hop graph propagation network structure. The former simulates the longitudinal and transverse wave interference behavior of sound waves in different media from a physical propagation perspective, while the latter captures the long-distance dependencies between feature parameters from a parameter correlation perspective. Compared to traditional single neural network structures, this hybrid architecture simultaneously enhances the physical and semantic dimensions of the features, making the model both conform to physical laws and possess powerful feature extraction capabilities. The adaptive stochastic resonance method constructs a bistable system and adaptively adjusts system parameters based on background noise intensity. It utilizes noise energy to nonlinearly amplify weak signals and combines this with fourth-order cumulants to suppress Gaussian noise. Compared to traditional linear filtering methods, this approach can effectively detect weak discharge signals in strong noise environments. The synergistic effect of these three techniques is reflected in the multi-path delay correction mechanism providing accurate signal components for subsequent feature extraction, the waveform propagation enhancement model achieving in-depth feature mining under physical constraints, and the adaptive stochastic resonance method further enhancing the detectability of weak features. These three elements form a complete chain from signal preprocessing to feature enhancement and weak signal extraction. Compared to the independent processing of each stage in traditional methods, this synergistic mechanism significantly improves the recognition accuracy and positioning precision of discharge phenomena under complex conditions through step-by-step information enhancement and feedback optimization.

[0068] It should be noted that this invention also solves the following technical problem: in complex working conditions, weak discharge signals are easily submerged by background noise and multiple reflected wave interference, making it difficult to detect early discharge faults in a timely manner. This invention constructs a bistable system by performing random resonance processing on signal components with peak amplitudes less than the weak signal threshold in the enhanced feature vector. It also maximizes the signal-to-noise ratio of the system output signal by adaptively adjusting the potential well spacing and damping coefficient parameters. Simultaneously, it uses fourth-order cumulant calculation to suppress Gaussian noise components. After establishing the weak feature vector, it fuses it with the transient feature vector, achieving effective extraction and amplification of weak discharge signals. Furthermore, this invention also solves the technical problem of low positioning accuracy caused by differences in acoustic wave propagation characteristics among different discharge types in multi-medium environments. This invention adaptively adjusts the propagation velocity parameters of the multipath delay correction matrix according to the determined discharge type. For tip discharge, surface discharge, and internal discharge, the weights of gas sound velocity, solid sound velocity, and liquid sound velocity are increased to the corresponding sound velocity weight values, respectively. After updating the propagation path analysis results, the spatial location coordinates of the discharge source are recalculated using the arrival time difference matrix and the maximum likelihood estimation method. This effectively compensates for the influence of the differences in sound wave propagation paths for different discharge types on the positioning results and significantly improves the accuracy of discharge source positioning.

[0069] Specifically, the principle of this invention is as follows: The invention solves this technical problem by combining a physical propagation mechanism with a feature enhancement mechanism. The multi-path delay correction matrix calculates the medium type and propagation distance of each propagation path using a ray tracing algorithm. Combined with the reflection and transmission coefficients of the medium interface, it can accurately identify the propagation characteristics of direct waves and multiple reflected waves, providing a physical basis for subsequent feature separation. The hybrid architecture design of the waveform propagation enhancement model conforms to the physical laws of ultrasonic multi-medium propagation. The three propagation layers of the waveform propagation network structure correspond to the acoustic characteristics of the three media. The longitudinal wave sub-network and the transverse wave sub-network achieve accurate modeling of wave mode conversion through a phase modulation mechanism. The multi-hop graph propagation network structure captures the internal correlation of time-domain parameters, the internal correlation of frequency-domain parameters, and cross-domain cross-correlation by constructing graph node relationships between feature parameters. This allows weak discharge characteristics to be gradually amplified and identified even under strong interference backgrounds through the multi-hop propagation mechanism. Furthermore, this invention adaptively adjusts the propagation velocity parameters of the multipath delay correction matrix according to the discharge type, and increases the sound velocity weight of the corresponding medium for tip discharge, surface discharge, and internal discharge respectively, thereby realizing dynamic optimization of propagation path analysis and further improving the adaptability of feature extraction under different discharge scenarios, thus theoretically ensuring the effectiveness of the method.

[0070] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0071] In this embodiment, the specific implementation of step S01 is the same as described above, and will not be repeated in detail here.

[0072] The specific implementation of step S02 involves inputting the acquired raw ultrasound signal into a pre-established multipath delay correction matrix for propagation path analysis and identification. matrix elements Indicates from the discharge point to the... The ultrasonic sensor's first The comprehensive parameters of the propagation path are calculated using the following formula:

[0073] ;

[0074] In the formula, The multipath delay correction matrix is ​​the first one. Line number Column matrix elements; The ultrasonic sensor is numbered, with a value ranging from 1 to... ; This is the propagation path number, with a value ranging from 1 to... ; This represents the total number of ultrasonic sensors. To reach the The total number of propagation paths for each sensor; From the discharge point to the first The ultrasonic sensor's first The theoretical propagation delay of each propagation path, in units of ; For the first The energy attenuation factor of each propagation path, dimensionless; For the first The medium type weight of each propagation path is dimensionless.

[0075] Among them, theoretical propagation delay The calculation formula is expressed as follows:

[0076] ;

[0077] In the formula, For the first The propagation distance of a propagation path in a gaseous medium, in units of It is obtained through ray tracing algorithm; The speed of sound in a gaseous medium is empirically valued at 340. ; For the first The propagation distance of a propagation path in a liquid medium, in units of It is obtained through ray tracing algorithm; The velocity of sound in a liquid medium is empirically valued at 1400. ; For the first The propagation distance of a propagation path in a solid medium, in units of It is obtained through ray tracing algorithm; The velocity of sound in a solid medium is empirically valued at 5000. .

[0078] Energy decay factor The calculation formula is expressed as follows:

[0079] ;

[0080] In the formula, For the first The number of medium interfaces traversed by each propagation path; This refers to the media interface number; For the first The reflection coefficient of each medium interface is dimensionless and obtained through experimental measurement. The empirical value of the reflection coefficient of the gas-solid interface is 0.85, and the empirical value of the reflection coefficient of the liquid-solid interface is 0.72. For the first The transmittance coefficient of each medium interface is dimensionless and obtained through experimental measurement. The empirical value for the transmittance coefficient of the gas-solid interface is 0.15, and the empirical value for the transmittance coefficient of the liquid-solid interface is 0.28.

[0081] Media type weight The calculation formula is expressed as follows:

[0082] ;

[0083] In the formula, For the first The total propagation distance of the propagation paths, in units of The calculation method is as follows .

[0084] The specific implementation of step S03 involves performing time-frequency domain joint analysis on the separated ultrasonic signal components using synchronous compressed wavelet transform, and the transient eigenvectors. The construction formula is expressed as follows:

[0085] ;

[0086] In the formula, These are transient feature vectors; The pulse rise time is expressed in units of 1 / 2000. It is obtained by measuring the duration of the rising edge of the ultrasound signal waveform from the baseline to the peak. For reference, the rise time is set at 10. ; Peak amplitude, in units of It is obtained by measuring the maximum absolute value of the ultrasonic signal waveform; For reference peak amplitude, the empirical value is 1. ; Pulse width, in units of It is obtained by measuring the full width at half maximum (FWHM) of the ultrasonic signal waveform; For reference pulse width, the empirical value is 50. ; The decay time constant is expressed in units of 1. The time required for the ultrasonic signal amplitude to decay to 36.8% of its initial value was measured. For reference decay time constant, the empirical value is 100. ; Instantaneous frequency, unit: It is obtained through synchronous compressed wavelet transform extraction; For reference frequency, the empirical value is 100. ; Main frequency bandwidth, in units of It is obtained by calculating the energy distribution on the statistical time-frequency plane; For reference bandwidth, the empirical value is 50. ; The centroid of the frequency, in units of It is obtained by calculating the position of the frequency centroid of the energy distribution on the statistical time-frequency plane; This is a reference value for the frequency centroid; the empirical value is 100. .

[0087] The specific implementation of step S04 is the same as described above, and will not be repeated in detail here.

[0088] The specific implementation of step S05 involves performing random resonance processing on signal components in the enhanced feature vector whose peak amplitude is less than the weak signal threshold. The weak signal threshold... The calculation formula is expressed as follows:

[0089] ;

[0090] In the formula, The threshold for weak signals, in units of ; The 90th percentile of the peak amplitude of normal operating noise in the training set, in units of The value was obtained by calculating the peak amplitude distribution of all normal operating noise samples in the statistical training set, with an empirical value of 0.23. ; The peak amplitude of the normal operating noise in the training set is given in units of By statistically analyzing the 90th percentile of the peak amplitude of normal operating noise and multiplying it by 1.5, the discrimination threshold is determined, which ensures the effective identification of weak signals while avoiding misjudging normal noise as a discharge signal.

[0091] Potential function of a bistable system The statement is as follows:

[0092] ;

[0093] In the formula, Let be the potential function of the bistable system, in units of . ; For system state variables, the unit is ; The potential well spacing parameter is expressed in units of 1000 m / s. The signal-to-noise ratio is obtained by adjusting the settings to reach its maximum value; an empirical value of 1.2 is used. ; This is the barrier height parameter, in units of The value is adaptively adjusted based on the collected background noise intensity, with an empirical value of 0.8. A double potential well structure is used to achieve a stochastic resonance effect. By adjusting the potential well spacing parameter and the barrier height parameter, weak signals can be amplified with the aid of noise. The fourth-order cumulant has zero value characteristics for Gaussian noise but remains sensitive to non-Gaussian discharge signals, thereby achieving noise suppression.

[0094] Weak eigenvectors The construction formula is expressed as follows:

[0095] ;

[0096] In the formula, It is a weak feature vector; The peak amplitude of the output of the bistable system is given in units of 1 / 2. It is obtained by measuring the maximum absolute value of the output signal of the bistable system; The output pulse width, in units of The full width at half maximum (FWHM) of the output signal of the bistable system is obtained by measuring the full width at half maximum (FWHM). The output pulse width is a reference value, with an empirical value of 50. ; Output spectral energy, unit: The spectral energy is obtained by performing spectral analysis on the output signal of the bistable system and calculating the spectral energy. For reference spectral energy, the empirical value is 0.01. ; This is the magnitude of the fourth-order cumulant, in units of... The value is obtained by calculating the fourth-order cumulant of the output signal of the bistable system. For reference, the fourth-order cumulative quantity is empirically set at 0.001. .

[0097] The specific implementation of step S06 involves fusing the seven parameters of the transient feature vector with the four parameters of the weak feature vector to form an eleven-dimensional feature space, and then processing the fused feature vector... Perform principal component analysis, principal components The calculation formula is expressed as follows:

[0098] ;

[0099] In the formula, For the first One principal component, dimensionless; Principal component number, with a value range from 1 to 5; The first covariance matrix is ​​the first... The th eigenvector of the th feature vector Each component is dimensionless and is obtained by eigenvalue decomposition of the covariance matrix of the fused feature vectors. is the feature vector dimension index, with a value ranging from 1 to 11; For the fused feature vector One element, dimensionless.

[0100] Euclidean distance between the enhanced eigenvector and the standard discharge eigenvector The calculation formula is expressed as follows:

[0101] ;

[0102] In the formula, To enhance the feature vector and the first The Euclidean distance of the eigenvectors of the standard discharge, in units of It is obtained by taking the square root of the sum of squares of the differences between each dimension in the feature space; This is the discharge type number, with a value of 1 representing tip discharge, a value of 2 representing surface discharge, and a value of 3 representing internal discharge. The feature vector dimension index, with values ​​ranging from 1 to... ; For the enhanced feature vector, the first... One element, dimensionless; For the first The first of the standard discharge eigenvectors One element, dimensionless; The dimension of the feature vector is 11 in this scheme.

[0103] Equivalent discharge intensity matrix The construction formula is expressed as follows:

[0104] ;

[0105] In the formula, This is the equivalent discharge intensity matrix, a two-row, five-column matrix; The first five principal components are dimensionless. The contribution rate corresponding to the first five principal components is dimensionless and is obtained through principal component analysis. The Euclidean distance between the enhanced eigenvector and the standard tip discharge eigenvector is expressed in units of 1. ; The Euclidean distance between the enhanced eigenvector and the standard surface discharge eigenvector is expressed in units of 1. ; The Euclidean distance between the enhanced eigenvector and the standard internal discharge eigenvector is expressed in units of 1. .

[0106] Discharge type discrimination index The calculation formula is expressed as follows:

[0107] ;

[0108] In the formula, This is a dimensionless index for determining the discharge type. The standard discharge energy threshold has an empirical value of 0.65. The standard frequency threshold is empirically set at 110. The weighting coefficients of 0.5, 0.3, and 0.2 in this formula reflect the dominant role of energy distribution characteristics in the discrimination result.

[0109] The specific implementation method of step S07 is the same as described above, and will not be repeated in detail here.

[0110] The specific implementation of step S08 involves adjusting the propagation velocity parameter in the multipath delay correction matrix according to the determined discharge type, thus updating the theoretical propagation delay. The calculation formula is expressed as follows:

[0111] ;

[0112] In the formula, The updated theoretical propagation delay is expressed in units of... ; This is the gas sound velocity weighting adjustment coefficient, which is dimensionless. For tip discharge, the first sound velocity weighting value is taken, with an empirical value of 0.85. In other cases, it is taken as 1. is the liquid sound velocity weighting adjustment coefficient, which is dimensionless. For internal discharge, the third sound velocity weighting value is taken, with an empirical value of 0.80. In other cases, it is taken as 1. This is the solid-state sound velocity weighting adjustment coefficient, which is dimensionless. For surface discharge, the second sound velocity weighting value is taken, with an empirical value of 0.75. In other cases, it is taken as 1.

[0113] The specific implementation of step S09 is to calculate the spatial location coordinates of the discharge source and the arrival time difference vector based on the updated propagation path analysis results. The construction formula is expressed as follows:

[0114] ;

[0115] In the formula, This is the arrival time difference vector; The reference sensor receives the raw ultrasonic signal at the time specified in units. ; For the first The time at which each sensor receives the raw ultrasonic signal, in units of... ; This is the sensor serial number, with a value ranging from 2 to... ; For the first The propagation path correction delay of each sensor, in units of It is calculated based on the updated multipath delay correction matrix.

[0116] Place the power supply space coordinates The objective function is solved using the maximum likelihood estimation method. The statement is as follows:

[0117] ;

[0118] In the formula, The objective function is... Let be the spatial coordinates of the discharge source to be solved, in units of . ; For the first The spatial coordinates of each ultrasonic sensor, in units of ; The spatial position coordinates of the reference sensor are in units of ; Effective speed of sound, unit: It is obtained by calculating based on the updated media type weight.

[0119] Effective speed of sound The calculation formula is expressed as follows:

[0120] .

[0121] To better understand and implement this invention, a specific application scenario is provided below as Example 2: A technical team needs to detect and locate discharges in a 220kV transformer. The transformer operates in a complex environment, containing transformer oil (liquid medium), gaseous insulation areas, and solid insulation materials. The presence of multiple media makes the ultrasonic wave propagation path complex, making it difficult for traditional detection methods based on a single-medium assumption to accurately identify the discharge type and location. The technical team decided to use the ultrasonic feature extraction method generated by typical discharge phenomena under complex operating conditions, as described in this invention, for detection.

[0122] The technical team evenly distributed eight ultrasonic sensors on the surface of the transformer casing. The spatial coordinates of the sensors are shown in Table 1. The sensor sampling frequency was set to 2MHz to synchronously acquire the raw ultrasonic signals generated by the discharge. During one acquisition process, all eight sensors synchronously recorded obvious ultrasonic pulse signals. Sensor 1 received the signal earliest at 0μs, while the other sensors received the signals at 12μs, 18μs, 25μs, 31μs, 39μs, 45μs, and 52μs, respectively.

[0123] Table 1 Spatial coordinates of ultrasonic sensor

[0124]

[0125] The technical team first established a three-dimensional acoustic propagation model of the transformer. This model sets the sound velocity in the gaseous medium region to 340 m / s, the sound velocity in the transformer oil liquid medium region to 1380 m / s, and the sound velocity in the solid insulating material region to 2850 m / s. A ray tracing algorithm was used to simulate the propagation path of ultrasonic waves from possible discharge points to each sensor, considering both direct waves and multipath propagation with up to three reflections. By calculating the reflection and transmission coefficients at different medium interfaces, a multipath delay correction matrix was established. This matrix is ​​8 rows and 20 columns, with each row corresponding to one sensor and each column corresponding to a possible propagation path. The matrix elements include three parameters: theoretical propagation delay, energy attenuation factor, and medium type weight.

[0126] Next, the propagation path of the acquired raw ultrasonic signal was analyzed. Using a multipath delay correction matrix, it was identified that the signal received by sensor 1 contained one direct wave path and two reflected wave paths. The direct wave primarily propagated through the gas medium (73%), followed by the liquid medium (18%) and the solid medium (9%). After separating the direct wave signal component, it underwent synchronous compressed wavelet transform processing. The wavelet transform decomposed the signal into 16 scales, and the instantaneous frequencies corresponding to the three scales with the largest amplitudes were extracted as 85kHz, 132kHz, and 178kHz, with instantaneous amplitudes of 0.68V, 0.52V, and 0.31V, respectively. Further calculations revealed a pulse rise time of 3.2μs, a peak amplitude of 0.82V, a pulse width of 15.6μs, and a decay time constant of 28.4μs. Statistical analysis of the energy distribution in the time-frequency plane showed a dominant frequency bandwidth of 95kHz and a frequency centroid of 126kHz. These seven parameters are normalized to form a transient feature vector with normalized values ​​of 0.64, 0.78, 0.52, 0.71, 0.68, 0.63, and 0.75.

[0127] like Figure 2 As shown, the transient feature vector is processed by the input waveform propagation enhancement model. This model employs a hybrid architecture of waveform propagation network and multi-hop graph propagation network. The waveform propagation network contains three propagation layers, simulating sound wave propagation in gas, liquid, and solid media, respectively. Each propagation layer contains longitudinal wave sub-networks and transverse wave sub-networks, achieving energy conversion between wave modes through an interference module. Since the multipath delay correction matrix shows that the weight for the gas medium type is at most 0.73, the interference intensity parameter is set to the first interference intensity value of 0.42. The multi-hop graph propagation network uses seven feature parameters as graph nodes. Based on the fact that the peak amplitude normalization value of 0.78 is greater than the first peak normalization threshold of 0.75, a two-hop propagation mode is adopted. The first hop propagation captures the correlation between time-domain parameters, and the second hop propagation captures the correlation between frequency-domain parameters. After model processing, the enhanced feature vector is output, with its dimension expanded to 14.

[0128] like Figure 3As shown, the technical team extracted weak signals from the enhanced feature vector. Since the peak amplitude of 0.82V was higher than the weak signal threshold of 0.35V, some weak signal components still needed processing. A bistable system was constructed with a double-well potential function, where both wells were 1.8V deep, and the barrier height was adaptively adjusted to 0.95V based on the background noise intensity. The weak signal components were input into the bistable system and superimposed with the acquired background noise, which had a root mean square value of 0.028V. By adjusting the well spacing parameter to 2.4 and the damping coefficient to 0.36, the signal-to-noise ratio of the system output signal reached its maximum value of 18.6dB. The adjusted bistable system output signal had a peak amplitude of 0.23V, an output pulse width of 8.7μs, and an output spectral energy of 0.16J. Fourth-order cumulant calculations were performed on the output signal, yielding a fourth-order cumulant amplitude of 0.048. These four parameters are normalized to form a weak feature vector with normalized values ​​of 0.46, 0.38, 0.42, and 0.51, respectively.

[0129] The technical team fused the seven parameters of the transient eigenvector and the four parameters of the weak eigenvector, and performed principal component analysis on the 11 parameters after fusion. The first five principal components were extracted, with contribution rates of 0.32, 0.24, 0.18, 0.15, and 0.11, respectively. Multiplying these five principal components by their corresponding weight coefficients formed the first row of the equivalent discharge intensity matrix. Simultaneously, the Euclidean distances between the enhanced eigenvector and the standard tip discharge eigenvector were calculated to be 2.35, 4.87, and 5.62, respectively. These three Euclidean distances, after normalization, were 0.18, 0.38, and 0.44, forming the second row of the equivalent discharge intensity matrix. This matrix clearly reflects the energy distribution characteristics and type similarity characteristics of the discharge signal.

[0130] The discharge type discrimination index is calculated based on the equivalent discharge intensity matrix. The weighted sum of the five elements in the first row of the equivalent discharge intensity matrix is ​​0.87. Normalizing this sum by dividing it by the sum of the weighted sum and the standard discharge energy threshold of 0.65 yields 0.572. The minimum distance value of 0.18 is normalized by dividing it by the sum of the three Euclidean distance values, yielding 0.180. The instantaneous frequency parameter of 126kHz is normalized by dividing it by the sum of the instantaneous frequency parameter and the standard frequency threshold of 110kHz, yielding 0.534. The three normalized values ​​are summed using weighting coefficients of 0.5, 0.3, and 0.2, resulting in a discharge type discrimination index of 0.497. This index is greater than the first discrimination threshold of 0.45, therefore it is determined to be a tip discharge.

[0131] After determining the discharge type, the technical team adjusted the propagation velocity parameters of the multipath delay correction matrix. For tip discharges, the weight of the gas sound velocity was increased to the first sound velocity weight value of 0.85, the weight of the liquid sound velocity was decreased to 0.10, and the weight of the solid sound velocity was decreased to 0.05. Based on the updated propagation path analysis results, the spatial location of the discharge source was calculated using the arrival time difference matrix and the maximum likelihood estimation method. Sensor 1 was selected as the reference sensor, and the arrival time differences of other sensors were calculated to be 12μs, 18μs, 25μs, 31μs, 39μs, 45μs, and 52μs, respectively. The propagation path was corrected according to the updated multipath delay correction matrix. After deducting the additional time delay introduced by multiple reflection paths, the corrected time differences were 10.8μs, 16.2μs, 22.5μs, 27.9μs, 35.1μs, 40.5μs, and 46.8μs, respectively. The objective function was optimized using the gradient descent algorithm, and converged after 58 iterations. The spatial coordinates of the discharge source were calculated to be X=0.78m, Y=0.62m, and Z=0.95m.

[0132] like Figure 4 As shown, the technical team output a complete set of characteristic parameters, including the discharge type as tip discharge, the coordinates of the discharge source location, the discharge intensity rating as medium, the main frequency as 126kHz, and the energy level as 0.87. These results provide a reliable basis for transformer condition assessment and maintenance decisions. Field verification revealed electric field concentration at the end of the high-voltage lead near the calculated discharge source location, confirming the accuracy of the detection results.

[0133] It should be explained that the various threshold parameters involved in this embodiment are determined based on experimental data statistics and physical model calculations, and the specific numerical ranges are as follows.

[0134] The first discrimination threshold ranges from 0.42 to 0.48. This threshold is used to distinguish between tip discharge and surface discharge, and is obtained by statistically analyzing the discharge type discrimination index of 100 tip discharge samples and 100 surface discharge samples. The discrimination index of tip discharge samples typically ranges from 0.45 to 0.72, while the discrimination index of surface discharge samples typically ranges from 0.28 to 0.45. The dividing threshold between the two is taken as the arithmetic mean, with a typical value of 0.45.

[0135] The second discrimination threshold ranges from 0.25 to 0.32. This threshold is used to distinguish between surface discharge and internal discharge, and is obtained by statistically analyzing the discharge type discrimination index of 100 surface discharge samples and 100 internal discharge samples. The discrimination index of surface discharge samples typically ranges from 0.28 to 0.45, while the discrimination index of internal discharge samples typically ranges from 0.12 to 0.28. The dividing threshold between the two is taken as the arithmetic mean, with a typical value of 0.28.

[0136] The first sound velocity weight value ranges from 0.75 to 0.90. This weight value corresponds to the proportion of the gas medium propagation path to the total propagation path during tip discharge. Since tip discharge mainly occurs in a gas medium, ultrasonic waves preferentially propagate through gas. According to the mathematical model of three-dimensional acoustic propagation of power equipment, the proportion of the gas medium propagation path is usually between 75% and 90%, with a typical value of 0.85.

[0137] The second sound velocity weight value ranges from 0.65 to 0.82. This weight value corresponds to the proportion of the solid medium propagation path to the total propagation path during surface discharge. Surface discharge occurs on the surface of a solid medium, and ultrasonic waves mainly propagate along the solid structure. According to the mathematical model of three-dimensional acoustic propagation of power equipment, the proportion of the solid medium propagation path is usually between 65% and 82%, with a typical value of 0.75.

[0138] The third sound velocity weight value ranges from 0.70 to 0.88. This weight value corresponds to the proportion of the propagation path through the liquid medium to the total propagation path during internal discharge. Internal discharge occurs inside the liquid medium, and ultrasonic waves preferentially propagate through the liquid. According to the mathematical model of three-dimensional acoustic propagation of power equipment, the proportion of the propagation path through the liquid medium is usually between 70% and 88%, with a typical value of 0.80.

[0139] The first peak normalization threshold ranges from 0.72 to 0.82. This threshold is used to determine the propagation steps of a multi-hop graph propagation network structure and is obtained by statistically analyzing the distribution of peak amplitude normalized values ​​in the training set. The 80th percentile of the peak amplitude normalized values ​​is calculated, indicating that 80% of the samples have peak amplitude normalized values ​​below this threshold; a typical value is 0.75.

[0140] The second peak normalization threshold ranges from 0.35 to 0.45. This threshold is also used to determine the propagation steps of a multi-hop graph propagation network structure, and is obtained by statistically analyzing the distribution of peak amplitude normalization values ​​in the training set. The 40th percentile of the peak amplitude normalization values ​​is calculated, indicating that 40% of the samples have peak amplitude normalization values ​​below this threshold, with a typical value of 0.40.

[0141] The first interference intensity value ranges from 0.35 to 0.52. This value corresponds to the energy conversion ratio between the longitudinal wave propagation mode and the transverse wave propagation mode in the gas medium. In the gas medium, due to the lower density of the medium, the energy conversion between wave modes is relatively weak. Through laboratory measurements, the energy conversion ratio in the gas medium is usually between 35% and 52%, with a typical value of 0.42.

[0142] The second interference intensity value ranges from 0.55 to 0.72. This value corresponds to the energy conversion ratio between the longitudinal wave propagation mode and the transverse wave propagation mode in the liquid medium. The density and viscosity characteristics of the liquid medium result in a moderate energy conversion between the wave modes. Through laboratory measurements, the energy conversion ratio of the liquid medium is typically between 55% and 72%, with a typical value of 0.63.

[0143] The third interference intensity value ranges from 0.68 to 0.85. This value corresponds to the energy conversion ratio between the longitudinal wave propagation mode and the transverse wave propagation mode in the solid medium. The rigid structure of the solid medium results in a strong energy conversion between the longitudinal and transverse waves. Through laboratory measurements, the energy conversion ratio of the solid medium is typically between 68% and 85%, with a typical value of 0.76.

[0144] The range of the initial learning rate value is: to This learning rate, used for training and optimization of the waveform propagation enhancement model, is determined through a learning rate range test method. The learning rate ranges from... Gradually increase to During the testing process, observe the rate of decrease of the loss function and training stability. The optimal learning rate is usually located at... to Between, the typical value is .

[0145] The weak signal threshold ranges from 0.30V to 0.42V. This threshold is used to determine whether random resonance processing of the signal component is required. By statistically analyzing the peak amplitude distribution of normal operating noise in the training set, the 90th percentile is calculated, indicating that 90% of the noise peak amplitude is below this value. 1.5 times this value is taken as the weak signal threshold, with a typical value of 0.35V.

[0146] The standard discharge energy threshold ranges from 0.58 to 0.75. This threshold is used for the normalization calculation of the discharge type discrimination index. Typical discharge type samples are created in a laboratory environment, and the energy of the acquired ultrasonic signals is calculated. The median of the energy distribution is statistically analyzed; typical discharge energies are usually distributed between 0.58 and 0.75, with a typical value of 0.65.

[0147] The standard frequency threshold ranges from 95 kHz to 125 kHz. This threshold is used for the normalization calculation of the discharge type discrimination index. Typical discharge type samples are created in a laboratory environment, and the acquired ultrasonic signals are subjected to spectral analysis to statistically analyze the median of the dominant frequency distribution. The typical dominant discharge frequency is usually distributed between 95 kHz and 125 kHz, with a typical value of 110 kHz.

[0148] The numerical ranges of the aforementioned threshold parameters are determined based on statistical analysis of extensive experimental data and theoretical model calculations. In practical applications, they can be appropriately adjusted according to the specific type of power equipment, operating conditions, and detection accuracy requirements. The thresholds are interconnected, collectively forming a complete discharge detection and identification system, ensuring the accuracy and robustness of the method under complex operating conditions.

[0149] It should be noted that the variables involved in this invention are explained in detail in Tables 2 and 3.

[0150] Table 2. Variable Explanation Table (Part 1)

[0151]

[0152] Table 3. Variable Explanation Table (Part Two)

[0153]

[0154] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting ultrasonic features generated by typical discharge phenomena under complex working conditions, characterized in that, Multiple ultrasonic sensors are deployed on the surface of the power equipment casing to collect the raw ultrasonic signals generated by the discharge. The spatial coordinates of each ultrasonic sensor and the arrival time of the raw ultrasonic signal are recorded synchronously. The raw ultrasonic signals are input into a multipath delay correction matrix for propagation path analysis to identify the propagation paths of direct waves and multiple reflected waves, separating ultrasonic signal components of different propagation modes. Synchronous compressed wavelet transform processing is performed on the ultrasonic signal components to extract the time-frequency characteristic parameters of transient pulses and establish transient feature vectors. These transient feature vectors are input into a waveform propagation enhancement model, which simulates the propagation interference process of sound waves in gaseous, liquid, and solid media through a waveform propagation network structure, outputting... The enhanced eigenvector is used to extract weak signals. Stochastic resonance amplification is achieved by adjusting the bistable system parameters to establish a weak eigenvector. The weak eigenvector is then fused with the transient eigenvector to establish an equivalent discharge intensity matrix, and the discharge type discrimination index is calculated. The discharge type is determined based on the numerical range of the discharge type discrimination index. The propagation speed parameters of the multipath delay correction matrix are adjusted according to the determined discharge type to update the propagation path analysis results. Based on the updated propagation path analysis results, the spatial coordinates of the discharge source are calculated using the arrival time difference matrix and the maximum likelihood estimation method. The spatial coordinates of the discharge source and the set of characteristic parameters are then output.

2. The method according to claim 1, characterized in that, The establishment of the multipath delay correction matrix specifically involves establishing a three-dimensional acoustic propagation model of the power equipment, using a ray tracing algorithm to simulate the propagation path of ultrasonic waves from the discharge point to each ultrasonic sensor, calculating the medium type and propagation distance of each propagation path, calculating the theoretical propagation delay of each propagation path based on the sound velocity parameters of different medium types, and calculating the energy attenuation factor of multiple reflection paths considering the reflection coefficient and transmission coefficient of the medium interface.

3. The method according to claim 2, characterized in that, The three-dimensional acoustic propagation model of the power equipment includes a gas medium region, a liquid medium region, and a solid structure region. The multipath delay correction matrix establishes a two-dimensional matrix for all propagation paths according to the ultrasonic sensor number and path sequence number. The matrix elements include three parameters: theoretical propagation delay, energy attenuation factor, and medium type weight.

4. The method according to claim 3, characterized in that, The establishment of the transient feature vector involves performing synchronous compressed wavelet transform on the ultrasonic signal component to decompose it into wavelet coefficients of multiple scales, extracting the instantaneous frequency and instantaneous amplitude corresponding to the three scales with the largest amplitude among the wavelet coefficients, calculating the pulse rise time, peak amplitude, pulse width and decay time constant of the ultrasonic signal component, and statistically analyzing the energy distribution of the ultrasonic signal component on the time-frequency plane to calculate the main frequency bandwidth and frequency centroid.

5. The method according to claim 4, characterized in that, The transient feature vector is composed of seven parameters after normalization: pulse rise time, peak amplitude, pulse width, decay time constant, instantaneous frequency, main frequency bandwidth, and frequency centroid.

6. The method according to claim 5, characterized in that, The waveform propagation enhancement model adopts a hybrid architecture of waveform propagation network structure and multi-hop graph propagation network structure. The waveform propagation network structure contains three propagation layers, each of which simulates the propagation of sound waves in a medium. The first propagation layer simulates propagation in a gaseous medium, the second propagation layer simulates propagation in a liquid medium, and the third propagation layer simulates propagation in a solid medium.

7. The method according to claim 6, characterized in that, Each propagation layer contains a longitudinal wave sub-network and a transverse wave sub-network. The longitudinal wave sub-network and the transverse wave sub-network exchange information through an interference module. The interference module uses a phase modulation mechanism to calculate the interference mode of the longitudinal wave propagation mode and the transverse wave propagation mode.

8. The method according to claim 7, characterized in that, The multi-hop graph propagation network structure uses the seven parameters in the transient feature vector as graph nodes, establishes edge connections based on the physical correlation between the parameters, and propagates information in multiple hops on the graph node structure. The first hop propagation captures the correlation between time-domain parameters, the second hop propagation captures the correlation between frequency-domain parameters, and the third hop propagation captures the cross-correlation between time-domain parameters and frequency-domain parameters.

9. The method according to claim 8, characterized in that, The number of propagation steps in the multi-hop graph propagation network structure is determined based on the normalized value of the peak amplitude in the transient feature vector. When the normalized value of the peak amplitude is greater than the first peak normalization threshold, two-hop propagation is used. When the normalized value of the peak amplitude is between the second peak normalization threshold and the first peak normalization threshold, three-hop propagation is used. When the normalized value of the peak amplitude is less than the second peak normalization threshold, four-hop propagation is used.

10. The method according to claim 9, characterized in that, The interference intensity parameter of the waveform propagation network structure is determined based on the maximum value of the medium type weight in the multipath delay correction matrix. When the weight of the gas medium type is the largest, the interference intensity parameter is set to the first interference intensity value; when the weight of the liquid medium type is the largest, the interference intensity parameter is set to the second interference intensity value; and when the weight of the solid medium type is the largest, the interference intensity parameter is set to the third interference intensity value.

Images