An array acoustic imaging partial discharge detection method and system

Abstract

This invention belongs to the field of acoustic detection technology for partial discharge in power equipment, and relates to an array-type acoustic imaging partial discharge detection method and system. The method includes: establishing the correspondence between channel identifier sequences and array geometric parameters; dividing the array into subgroups according to a fixed subgrid and generating a synchronous sampling time sequence to acquire multi-channel pulse density modulation signals; demodulating and extracting the signals, bandpass filtering and pulse framing; constructing a forward-backward spatial smoothing covariance matrix and using minimum variance distortionless response beamforming to calculate the spatial spectrum to generate an acoustic image; and registering and superimposing the acoustic image with a visible light image to output the positioning result. The technical solution of this application achieves high-contrast imaging and stable positioning output through temporal and geometric consistency constraints, which can improve the consistency of on-site interpretation and meet the requirement of repeatable positioning under weak partial discharge ultrasonic conditions.

Description

Technical Field

[0001] This invention belongs to the field of acoustic detection technology for partial discharge of power equipment, and specifically relates to an array-type acoustic imaging method and system for partial discharge detection. Background Technology

[0002] In existing technologies, acoustic imaging detection of partial discharge in power equipment typically relies on microphone arrays to acquire ultrasonic signals of partial discharge and cameras to acquire visible light images. This is achieved by preprocessing multi-channel acoustic signals, beamforming imaging, and superimposing them onto the visible light images to meet the needs of non-contact inspection and location of equipment such as switchgear, transformers, and reactors. However, existing partial discharge acoustic imaging detection methods have some significant shortcomings in terms of multi-channel acquisition consistency and weak signal imaging stability.

[0003] In practical applications, acoustic imaging systems often need to complete multi-channel synchronous acquisition, filtering and noise reduction, framing, and spatial spectrum imaging in industrial sites. However, due to factors such as acquisition timing consistency, channel phase consistency, and noise spectrum interference, the overall contrast between the main lobe and side lobes of the imaging is low, the consistency of hot spot focusing patterns is weak, and the repeatability and comparability of the positioning results are generally low under conditions of long distance or weak partial discharge signals. This has an adverse effect on the interpretation efficiency and positioning reliability of maintenance personnel.

[0004] Therefore, it is evident that existing technologies often suffer from problems such as low imaging focus contrast and weak stability of positioning results in acoustic imaging detection of partial discharge in power equipment. This is a shortcoming of existing technologies.

[0005] In view of this, it is very necessary to provide an array-type acoustic imaging partial discharge detection method and system to solve the above-mentioned defects in the prior art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of the prior art in acoustic imaging detection of partial discharge in power equipment, namely low imaging focusing contrast and weak stability of positioning results, by providing an array-type acoustic imaging partial discharge detection method and system to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, this application provides an array-type acoustic imaging partial discharge detection method, comprising: Obtain the channel identifier sequence of the microphone array and establish the correspondence between the channel identifier sequence and the array geometric parameters; Based on the channel identifier sequence, the microphone array is divided into subgroups according to a fixed subgrid, and a mapping relationship between the subgroups and the channels is established. Based on the mapping relationship, a synchronous sampling timing sequence is generated and multi-channel synchronous sampling is triggered to obtain multi-channel pulse density modulation signals. Under synchronous sampling timing constraints, low-pass demodulation and decimation operations are performed on multi-channel pulse density modulation signals to obtain multi-channel time-domain signals. Bandpass filtering and pulse framing were performed on the multi-channel time-domain signals to obtain multi-channel partial discharge ultrasonic frames. Based on correspondence and mapping relationships, channel organization is performed on multi-channel partial discharge ultrasound frames and a forward-backward spatial smoothing covariance matrix is ​​constructed. Based on minimum variance distortionless response beamforming, the acoustic spatial spectrum is calculated to generate an acoustic image characterizing the spatial distribution of partial discharge sound sources. The acoustic image and the visible light image are spatially registered and superimposed to output the localization result of the partial discharge sound source.

[0008] By adopting the above technical solution, the microphone array channel identifier sequence and array geometric parameters are uniformly associated, and a full array synchronous sampling organization is formed under the constraint of subgroup and channel mapping relationship. This achieves unified constraints on the multi-channel acquisition and imaging processing link in terms of timing and geometry, which can obtain higher acoustic image focusing contrast and more stable positioning output under complex on-site noise and weak partial discharge ultrasonic signal conditions, thus meeting the needs of improving imaging focusing contrast and enhancing the stability of positioning results in partial discharge acoustic imaging detection.

[0009] Specifically, the correspondence between channel identifier sequences and array geometric parameters establishes a semantic mapping that maintains consistency between channel data and array element spatial positions, providing a deterministic basis for subsequent processing calls to array aperture and element coordinates. Dividing the microphone array into subgroups according to fixed subgrids and establishing a mapping relationship between subgroups and channels provides scalable and reusable structured constraints for multi-channel acquisition organization. Under this mapping relationship, a synchronous sampling timing sequence is generated to trigger multi-channel synchronous sampling, ensuring higher consistency of data across channels under a unified timing reference. Low-pass demodulation and decimation of the pulse density modulation signal are performed under synchronous sampling timing constraints, converting broadband pulse density modulation data into multi-channel time-domain signals suitable for subsequent array processing while also considering computational load. The channel time-domain signal is bandpass filtered and pulsed framed to concentrate the effective frequency band characteristics of partial discharge ultrasound within the frame and enhance the suppression effect on on-site noise, thus providing a more reliable input for robust spatial spectrum estimation. Based on the correspondence and mapping relationship, the partial discharge ultrasound frame is channel-organized and a forward-backward spatial smoothing covariance matrix is ​​constructed to maintain better availability of spatial statistics under weak signal conditions. The acoustic spatial spectrum is calculated using minimum variance distortionless response beamforming to generate an acoustic image, making the focusing effect of the sound source energy distribution clearer and the sidelobe interference more controllable. Finally, the acoustic image and the visible light image are spatially registered and superimposed to output the localization result, making the sound source location more intuitive on the visualization carrier and easier for on-site comparison and interpretation.

[0010] Preferably, the step of obtaining the channel identifier sequence of the microphone array and establishing the correspondence between the channel identifier sequence and the array geometric parameters includes: Read the channel identifier sequence that is associated with each channel, perform a uniqueness check on the channel identifier sequence, and generate the check result; When the verification result indicates that the channel identifier sequence meets the uniqueness constraint, a geometric index table is generated based on the array geometric parameters. The geometric index table is then bound to the channel identifier sequence to form a correspondence between the channel identifier sequence and the array geometric parameters.

[0011] By adopting the above technical solution, the deterministic association between channel and array element positions is achieved by using the uniqueness verification of channel identifier sequence and binding with geometric index table. This can improve the consistency and traceability of channel organization and array geometry call and reduce imaging deviation caused by channel mismatch.

[0012] Preferably, the steps of dividing the microphone array into subgroups according to a fixed subgrid based on the channel identifier sequence, establishing a mapping relationship between the subgroups and channels, generating a synchronous sampling timing sequence based on the mapping relationship, and triggering multi-channel synchronous sampling to obtain a multi-channel pulse density modulation signal include: A subgroup index table is generated based on a fixed subgrid partitioning rule, and the subgroup index table is associated with the channel identifier sequence to obtain the mapping relationship between subgroups and channels; Based on the mapping relationship, a synchronization trigger identifier is assigned to each subgroup and a synchronization sampling timing table is generated. A deterministic delay calibration is performed on the synchronization sampling timing table to generate the calibrated synchronization sampling timing. Multi-channel synchronous sampling is triggered according to the calibrated synchronous sampling timing to form a multi-channel pulse density modulation signal associated with the channel identifier sequence.

[0013] By adopting the above technical solution, cross-subgroup synchronous triggering is achieved by using a synchronous sampling timing sequence generated by a subgroup index table and a synchronous trigger identifier and calibrated with a deterministic delay. This can improve the timing consistency and phase stability of multi-channel sampling and enhance the reliability of spatial spectrum estimation.

[0014] Preferably, the steps of performing low-pass demodulation and decimation operations on the multi-channel pulse density modulated signal to obtain the multi-channel time-domain signal include: Perform demodulation filtering configuration on multi-channel pulse density modulated signals and generate a demodulation parameter set; Low-pass demodulation filtering is performed on the multi-channel pulse density modulated signal based on the demodulation parameter set to obtain the demodulated intermediate signal; The demodulated intermediate signal is configured with a decimation link and a decimation parameter set is generated. Based on the decimation parameter set, the demodulated intermediate signal is subjected to cascaded integral comb filtering to obtain the decimated intermediate signal. A compensated finite-length impulse response filter is applied to the extracted intermediate signal to obtain a multi-channel time-domain signal with the target sampling rate. The multi-channel time-domain signal and the synchronous sampling timing are consistent and then output.

[0015] By adopting the above technical solution, the target sampling rate reconstruction is achieved by using demodulation parameter set and decimation parameter set to drive low-pass demodulation and cascaded integral comb decimation, combined with compensated finite impulse response filtering. This can maintain the consistency of amplitude and phase characteristics of the effective frequency band while controlling the computational load and improving the stability of subsequent framing and imaging input.

[0016] Preferably, the step of performing bandpass filtering and pulse framing on the multi-channel time-domain signal to obtain a multi-channel partial discharge ultrasonic frame includes: A bandpass filter configuration is generated based on the partial discharge ultrasonic frequency band constraint, and bandpass filtering is performed on multi-channel time-domain signals; Based on the multi-channel signal after bandpass filtering, a pulse detection sequence is generated and the start and end intervals of the pulse events are determined. The multi-channel signal is aligned and truncated according to the start and end intervals of the pulse events and a frame index table is generated. The multi-channel partial discharge ultrasound frames are output based on the frame index table and associated with the channel identifier sequence.

[0017] By adopting the above technical solution, bandpass filtering is performed according to the partial discharge ultrasonic frequency band constraint, and the start and end intervals of the event are determined by pulse detection sequence to achieve multi-channel aligned interception. This can improve the intra-frame energy concentration and cross-channel alignment consistency, and enhance the repeatability of sound source localization results.

[0018] Preferably, the steps of organizing multi-channel partial discharge ultrasound frames based on correspondence and mapping relationships and constructing a forward-backward spatially smoothed covariance matrix, and calculating the acoustic spatial spectrum based on minimum variance distortionless response beamforming to generate an acoustic image characterizing the spatial distribution of partial discharge sound sources include: Array guidance information is generated based on the correspondence, and subarray index sequences are generated based on the mapping relationship; The multi-channel partial discharge ultrasound frames are organized into channels according to the subarray index sequence, and a forward covariance matrix is ​​constructed. Perform channel inversion mapping based on subarray index sequence and construct backward covariance matrix; The forward and backward covariance matrices are fused to obtain the forward-backward spatially smoothed covariance matrix; Based on the forward-backward spatial smoothing covariance matrix and array guidance information, the minimum variance distortionless response beamforming weights are solved and the acoustic spatial spectrum is generated, outputting an acoustic image.

[0019] By employing the above technical solution, the forward covariance matrix and the backward covariance matrix are constructed using array guidance information and subarray index sequence, and then fused to obtain the forward-backward spatial smooth covariance matrix. The minimum variance distortionless response weight is then solved to achieve spatial spectrum imaging, which can suppress sidelobe interference and improve the focusing contrast and positioning reliability of weak partial discharge sound sources.

[0020] Preferably, the step of fusing the forward covariance matrix and the backward covariance matrix to obtain the forward-backward spatially smoothed covariance matrix includes: Generate a subarray sliding window sequence based on the subarray index sequence, and generate a window weight sequence for each sliding window; Calculate the forward covariance submatrix and backward covariance submatrix corresponding to each sliding window according to the submatrix sliding window sequence; The forward and backward covariance submatrices are weighted and fused based on the window weight sequence to output a forward-backward spatially smoothed covariance matrix.

[0021] By employing the above technical solution, a weighted fusion of the forward and backward covariance submatrices is achieved using a subarray sliding window sequence and a window weight sequence to construct a robust spatial smoothing statistic. This can improve the stability of covariance estimation and reduce the impact of single-pulse fluctuations on the consistency of imaging morphology.

[0022] Preferably, the steps of spatially registering the acoustic image and the visible light image and superimposing them to output the localization result of the partial discharge sound source include: Establish an acoustic imaging coordinate system and generate an acoustic imaging correspondence table between acoustic image pixel coordinates and the acoustic imaging coordinate system; Establish an optical imaging coordinate system and generate an optical imaging correspondence table between visible light image pixel coordinates and the optical imaging coordinate system; The mapping relationship between the acoustic imaging coordinate system and the optical imaging coordinate system is calculated based on the acoustic imaging correspondence table and the optical imaging correspondence table, and a registration transformation table is generated. Based on the registration transformation table, the acoustic image is geometrically transformed and superimposed with the visible light image to output the localization result of the partial discharge sound source.

[0023] By adopting the above technical solution, the mapping relationship between the two coordinate systems is calculated using the acoustic imaging correspondence table and the optical imaging correspondence table, and a registration transformation table is generated to drive the geometric transformation and superposition output of the acoustic image. This can improve the consistency and intuitiveness of the acoustic image and optical image superposition and enhance the positioning usability of on-site interpretation.

[0024] Secondly, this application also provides an array-type acoustic imaging partial discharge detection system, comprising: The channel binding module is used to obtain the channel identifier sequence of the microphone array and establish the correspondence between the channel identifier sequence and the array geometric parameters; The group synchronization module is used to divide the microphone array into subgroups according to a fixed subgrid based on the channel identifier sequence, establish a mapping relationship between the subgroups and the channels, generate a synchronous sampling timing sequence based on the mapping relationship and trigger multi-channel synchronous sampling to obtain multi-channel pulse density modulation signals. The demodulation and decimation module is used to perform low-pass demodulation and decimation operations on multi-channel pulse density modulation signals under synchronous sampling timing constraints to obtain multi-channel time-domain signals. The frame preprocessing module is used to perform bandpass filtering and pulse framing on the multi-channel time-domain signal to obtain multi-channel partial discharge ultrasonic frames. The smoothing imaging module is used to organize the channels of multi-channel partial discharge ultrasound frames based on correspondence and mapping relationships and construct the forward-backward spatial smoothing covariance matrix. It also calculates the acoustic spatial spectrum based on minimum variance distortionless response beamforming to generate an acoustic image characterizing the spatial distribution of partial discharge sound sources. The registration and overlay module is used to spatially register acoustic images and visible light images and overlay them to output the localization results of partial discharge sound sources.

[0025] Preferably, the smoothing imaging module includes: The index organization submodule is used to generate array guidance information based on the correspondence and generate subarray index sequences based on the mapping relationship, and to organize the multi-channel partial discharge ultrasound frames according to the subarray index sequences; The forward-backward fusion submodule is used to construct the forward covariance matrix and the backward covariance matrix, and fuse the forward covariance matrix and the backward covariance matrix to obtain the forward-backward spatial smoothing covariance matrix. The weighting submodule is used to solve for the minimum variance distortionless response beamforming weights based on the forward-backward spatial smoothing covariance matrix and array guidance information, and to generate the acoustic spatial spectrum and output the acoustic image.

[0026] As can be seen from the above technical solutions, the present invention has the following advantages: This application provides an array-based acoustic imaging partial discharge detection method and system. By uniformly associating the microphone array channel identifier sequence with the array geometric parameters and forming a full-array synchronous sampling organization under the constraint of the subgroup and channel mapping relationship, the multi-channel acquisition and imaging processing link achieves unified constraints in terms of timing and geometry. This enables higher acoustic image focusing contrast and more stable positioning output under complex on-site noise and weak partial discharge ultrasonic signal conditions, meeting the needs of improving imaging focusing contrast and enhancing the stability of positioning results in partial discharge acoustic imaging detection. Attached Figure Description

[0027] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of an array-type acoustic imaging partial discharge detection method provided by the present invention; Figure 2 This is a block diagram illustrating the principle of an array-type acoustic imaging partial discharge detection system provided by the present invention.

[0029] The module consists of: 1. Channel binding module; 2. Group synchronization module; 3. Demodulation extraction module; 4. Frame preprocessing module; 5. Smoothing imaging module; and 6. Registration and overlay module. Detailed Implementation

[0030] Various embodiments of this disclosure are described more fully below with reference to the accompanying drawings. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0031] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0032] It should be noted in advance that, in order to facilitate a clear and accurate description of the technical solutions in the embodiments of this application, the following is a brief explanation of some terms and related technologies involved in the embodiments of this application: 1. Pulse Density Modulation (PDM): This is a digital encoding method that uses a high-frequency 1-bit pulse sequence to represent the amplitude of an analog signal. The signal amplitude is represented by the density of "1" pulses per unit time. It is often used for digital microphone output. Afterward, it usually needs to be low-pass demodulated and decimated to recover the conventional multi-bit time-domain sampled signal.

[0033] 2. Minimum Variance Distortionless Response Beamforming: Often abbreviated as MVDR (Minimum Variance Distortionless Response), it is a type of adaptive array beamforming method. Its idea is to minimize the output power while maintaining the gain in the target direction without distortion, thereby suppressing interference and noise in non-target directions and improving the spatial spectrum focusing capability. It is often used for high-resolution spatial spectrum estimation in sound source localization and acoustic imaging.

[0034] 3. Forward-backward spatial smoothing covariance matrix: Forward-backward spatial smoothing is a robust covariance matrix processing method. It forms multiple sub-array covariances by sliding windows of array data into sub-arrays and then performs weighted fusion. At the same time, it introduces the backward statistics obtained by channel inversion to enhance the rank and stability of covariance estimation, thereby improving the reliability of spatial spectrum estimation under conditions of strong correlation signals or insufficient snapshots and reducing imaging pseudo-peak interference.

[0035] To address the limitations of multi-channel acquisition consistency control and low imaging focusing contrast and weak positioning stability under weak partial discharge ultrasonic conditions during partial discharge acoustic imaging detection of power equipment, which make it difficult to meet the actual needs of on-site inspections for high-contrast imaging and repeatable positioning output, this application discloses an array-type acoustic imaging partial discharge detection method and system. By establishing the association between channel identifiers and array geometric parameters and organizing synchronous sampling and signal processing under fixed subgrid subgroup mapping constraints, combined with demodulation extraction and bandpass framing processing for partial discharge ultrasound, as well as a spatial spectrum imaging strategy of forward-backward spatial smoothing and minimum variance distortionless response beamforming, it can achieve clear characterization of the spatial distribution of sound sources and stable output of positioning results under complex noise backgrounds and weak signal conditions. This effectively improves the focusing contrast and positioning reliability of partial discharge acoustic images, further enhances on-site interpretation efficiency and inspection quality, reduces the impact of human experience differences on positioning results, and ensures the timeliness and consistency of power equipment condition assessment and maintenance decisions.

[0036] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] like Figure 1 As shown, this embodiment provides an array-type acoustic imaging partial discharge detection method, including: Step S1: Obtain the channel identifier sequence of the microphone array and establish the correspondence between the channel identifier sequence and the array geometric parameters; Step S2: Based on the channel identifier sequence, the microphone array is divided into subgroups according to a fixed subgrid, and a mapping relationship between the subgroups and the channels is established. Based on the mapping relationship, a synchronous sampling timing sequence is generated and multi-channel synchronous sampling is triggered to obtain multi-channel pulse density modulation signals. Step S3: Under the constraint of synchronous sampling timing, perform low-pass demodulation and decimation operations on the multi-channel pulse density modulation signal to obtain the multi-channel time domain signal; Step S4: Perform bandpass filtering on the multi-channel time-domain signal and pulse framing to obtain multi-channel partial discharge ultrasonic frames; Step S5: Based on the correspondence and mapping relationship, the multi-channel partial discharge ultrasound frames are organized into channels and a forward-backward spatial smoothing covariance matrix is ​​constructed. Based on the minimum variance distortionless response beamforming, the acoustic spatial spectrum is calculated to generate an acoustic image characterizing the spatial distribution of the partial discharge sound source. Step S6: Spatial registration of the acoustic image and the visible light image is performed and the results of localization of the partial discharge sound source are superimposed and output.

[0038] This embodiment employs a channel organization method based on the consistent association between channel identifier sequences and array geometric parameters, ensuring a definite mapping of multi-channel acoustic signals in terms of array element spatial location, thus providing a controllable geometric reference for subsequent spatial spectrum estimation. By dividing the microphone array into subgroups according to a fixed subgrid and generating synchronous sampling timing based on the mapping relationship between subgroups and channels, the multi-channel acquisition process achieves higher consistency under a unified timing reference, reducing the impact of inter-channel timing and phase offsets on imaging results. By performing low-pass demodulation and decimation on the pulse density modulation signal under synchronous sampling timing constraints, the multi-channel data is converted from the pulse density domain into a time-domain signal suitable for array processing, balancing data volume and computational load, thereby improving the feasibility of real-time on-site imaging processing. By performing bandpass filtering and pulse framing on the multi-channel time-domain signals... This approach concentrates the effective frequency band characteristics of partial discharge ultrasound within the frame and more effectively suppresses background noise, improving the input quality of subsequent covariance statistics and spatial spectrum estimation. By constructing a forward-backward spatially smoothed covariance matrix and employing minimum variance distortionless response beamforming to calculate the acoustic spatial spectrum, the spatial focusing capability and sidelobe suppression capability under weak partial discharge ultrasound conditions are enhanced, improving the focusing contrast and hotspot presentation consistency of the acoustic image. By spatially registering the acoustic image and the visible light image and superimposing the output positioning result, the sound source location is presented more intuitively in the visualization carrier and is easier for on-site interpretation and comparison. Overall, it achieves high-contrast imaging presentation and stable positioning output in partial discharge acoustic imaging detection under complex noise background and weak signal conditions, improving the efficiency and consistency of inspection interpretation and supporting the timeliness and reliability of maintenance decisions.

[0039] The above steps will be specifically described below based on the embodiments of this application.

[0040] In step S1, the core task is to solidify the channel identifier sequence of each acquisition channel into a unique and traceable index entry, and establish a definite correspondence between the channel identifier sequence and the array geometric parameters on this entry, so that subsequent array guidance information calculations can directly retrieve the two-dimensional coordinates of the array elements and the array element spacing aperture from the channel identifier and maintain the geometric aperture consistency.

[0041] In this embodiment, the channel identifier sequence of the microphone array can be obtained and a correspondence between the channel identifier sequence and the array geometric parameters can be established. The channel identifier sequence is composed of hardware identifier codes output by each acquisition channel after power-on, arranged in the access order, and is written into the channel registration table to freeze the index aperture. The array geometric parameters are used to characterize the array element row index, array element column index, array element two-dimensional coordinates, and array element spacing aperture, and are written into the geometric parameter table to freeze the geometric aperture.

[0042] In some embodiments of this application, to ensure that any subsequent frame of data can be stably traced back to the geometric position of the array element using the hardware identifier code as the primary key, hard constraint verification can be performed to output a machine-readable verification result, and only after the verification passes can the data enter the geometric table and be bound. Specifically, the channel identifier sequence associated with each channel can be read. The engineering meaning of reading is to retrieve the hardware identifier code from the identifier register or configuration area of ​​each channel and write the hardware identifier code, along with the acquisition board number and the board channel number, into the channel registration table to form a traceable index link. Based on this, a uniqueness verification is performed on the channel identifier sequence and a verification result is generated. The uniqueness verification simultaneously covers duplicate identifier detection and missing identifier detection to avoid channel mismatch caused by code reuse or missing codes. The verification result can be calculated using:

[0043] in, Indicates the verification result. Indicates an indicator function, This represents the deduplication set operator. Indicates the channel identifier sequence. Represents the cardinality of a set. Indicates the total number of channels.

[0044] When the verification result is valid, meaning the verification result indicates that the channel identifier sequence meets the uniqueness constraint, the process proceeds to array coordinate system construction and geometric index table generation. At this point, the geometric index table can be generated based on the array geometric parameters. The array coordinate system has its origin at the array geometric center, and the two orthogonal directions within the array plane are defined as... shaft and The axis is defined as the normal to the array plane. An axis is used to ensure that the two-dimensional coordinates of array elements are expressed under a unified spatial scale. The geometric index table uses the hardware identifier as the primary key and writes the array element row index, array element column index, array element two-dimensional coordinates, and array element spacing scale into it, so that the array element two-dimensional coordinate vector can be retrieved by simply providing the hardware identifier. The calculation of the array element two-dimensional coordinates can be done as follows:

[0045] in, Indicates the first The coordinate vector of the corresponding array element of the path channel. Indicates that the array element is in Axial coordinates, Indicates that the array element is in Axial coordinates, Indicates the row index of the array element. Indicates the column index of the array element. Indicates the center row index of the array. Indicates the index of the center column of the array. Indicates the aperture of the array element spacing. This indicates the transpose operation.

[0046] Furthermore, to ensure that the element spacing aperture meets spatial sampling requirements and suppresses spatial aliasing, the wavelength corresponding to the center frequency and the element spacing aperture need to meet sampling constraints and be fixed in the geometric parameter table during the design phase. For example, the wavelength and element spacing aperture constraints adopt the following:

[0047] in, Indicates the wavelength corresponding to the center frequency. Indicates the speed of sound. Indicates the center frequency of partial discharge ultrasound. This indicates the aperture of the element spacing. For example, when the air speed of sound is 340 m / s and the center frequency is 60 kHz, the wavelength is approximately 5.7 mm, and an element spacing aperture on the order of 2 mm can suppress spatial aliasing and improve angular resolution.

[0048] Furthermore, after the geometric index table is generated, it undergoes a binding and solidification process to make the channel identifier sequence the unique geometric entry point. During this process, the geometric index table can be bound to the channel identifier sequence, forming a correspondence between the channel identifier sequence and the array geometric parameters. The engineering implementation of this binding involves a one-to-one association between hardware identifier codes and geometric index table row numbers, and this association is written into the binding table. Simultaneously, the version number and verification result are recorded for subsequent tracing of the geometric aperture version corresponding to the positioning result.

[0049] Thus, step S1 completes the uniqueness verification of the channel identifier sequence, the placement and binding of the geometric index, and forms a hardware identifier code-driven geometric entry system. This ensures that subsequent guidance information calculations are always based on a unified geometric caliber, providing a reliable geometric input foundation for array imaging stability.

[0050] In step S2, the core task is to divide the array into several subgroups according to the fixed subgrid based on the channel identifier sequence and form a mapping relationship between the subgroups and the channels. Then, a synchronous sampling timing sequence is generated on the mapping relationship and deterministic delay calibration is completed, so that the high channel scale can still maintain the consistency of cross-subgroup frame boundaries and cross-subgroup phase under the conditions of multiple boards and multiple links, and output the multi-channel pulse density modulation signal associated with the channel identifier.

[0051] In this embodiment, the microphone array can be divided into subgroups according to a fixed subgrid based on the channel identifier sequence, and a mapping relationship between the subgroups and channels can be established. Based on this mapping relationship, a synchronous sampling timing sequence is generated and multi-channel synchronous sampling is triggered, thereby obtaining a multi-channel pulse density modulation signal. The fixed subgrid defines the sub-window size in the array row and column index domain. and Furthermore, coverage rules are defined to ensure that each subgroup has stable boundaries and can be reused. For example, a 32×32 array can be divided into 4 subgroups by taking 16×16 subgrids to reduce triggering and caching pressure, and an 8×8 subgrid can be divided into 16 subgroups to increase the number of subsequent spatial smoothing windows.

[0052] In some embodiments of this application, to solidify the subgroup structure into a directly callable data structure and avoid on-site interpretation differences, a subgroup index table can be generated based on fixed subgrid partitioning rules, and the subgroup index table can be associated with the channel identifier sequence to obtain the mapping relationship between subgroups and channels. The association process uses the correspondence in step S1 as a bridge and converts the subgroup boundary into a hardware identifier code set, while preserving the local row and column order of the subgroups for subsequent subarray sliding and channel reversal mapping. For example, the mapping relationship from subgroup to channel can be expressed using a set for subsequent data retrieval and organization. The mapping relationship can adopt the following:

[0053] in, Indicates the starting point of the line. and column start point Defined subgroup channel index set, Indicates row index, Indicates column index, Indicates the row dimension of the subgrid. Indicates the column dimension of the subgrid. This represents logical AND. Subgroup The mapping relationship to the channel identifier set. Indicates the first Road access identification code.

[0054] Furthermore, after the mapping relationship is established, a synchronous sampling timing sequence can be generated and deterministic delay calibration can be performed to eliminate fixed delay differences between subgroups. Specifically, a synchronous trigger identifier can be assigned to each subgroup based on the mapping relationship, and a synchronous sampling timing table can be generated. The synchronous sampling timing table includes the frame number, frame start time, frame length, and synchronous trigger identifier, and uses the frame boundary as the unified time calibrator for all subsequent processing. Based on this, deterministic delay calibration is performed on the synchronous sampling timing table to generate the calibrated synchronous sampling timing sequence. The deterministic delay calibration uses a unified reference edge as a benchmark, obtains the equivalent delay sample number for each subgroup by locating the cross-correlation peak, and writes it back to the synchronous sampling timing table. Delay estimation can be performed using:

[0055] in, Indicates the equivalent delayed sample number of the subgroup. Represents a delayed search set. Indicates the cross-correlation window length. Represents the discrete sequence of reference edges. This represents the discrete sequence of reference edges acquired by the subgroup. This indicates the independent variable that takes the maximum value.

[0056] Furthermore, the equivalent delayed sample count is used for sample domain alignment to form a calibrated synchronous sampling timing, and to align the cross-subgroup frame start points to the same master clock boundary. Sample domain compensation employs:

[0057] in, This represents the discrete sequence after delay calibration. This represents the discrete sequence before delayed calibration. Indicates the equivalent delayed sample number of the subgroup. Indicates a subgroup index. Indicates the channel index. This indicates the sequence number of the discrete sampling point.

[0058] In some embodiments of this application, multi-channel synchronous sampling can be triggered according to the calibrated synchronous sampling timing to form a multi-channel pulse density modulation signal associated with the channel identifier sequence. The output organization uses the frame sequence number as the primary key and the hardware identifier code as the channel primary key, and links each pulse density modulation signal to the channel registration table record to ensure that each subsequent frame can be traced back to the two-dimensional coordinates of the array element. For example, the pulse density modulation clock is on the order of 1MHz, the frame length is on the order of 4096 points, and the delay search set is from -32 to 32 to cover the fixed delay range of the subgroup link.

[0059] Thus, step S2 forms a closed-loop system of fixed subgrid subgrouping, subgroup-channel mapping, synchronous sampling timing and deterministic delay calibration, and outputs a multi-channel pulse density modulation signal associated with channel identifiers, providing a unified frame boundary and phase alignment basis for subsequent demodulation and extraction.

[0060] In step S3, the core task is to convert the multi-channel pulse density modulation signal into a multi-channel time-domain signal with the target sampling rate through an achievable low-pass demodulation and decimation link, and to map the frame boundary from the synchronous sampling timing to the target sampling rate sample domain, so that subsequent frequency band limiting and pulse framing can be performed in the target sampling rate sample domain, thereby maintaining consistent alignment across subgroups.

[0061] In this embodiment, under synchronous sampling timing constraints, low-pass demodulation and decimation operations can be performed on multi-channel pulse density modulation signals to obtain multi-channel time-domain signals. During this process, to ensure the link is configurable and reusable, the demodulation parameter set and decimation parameter set can be fixed, and low-pass demodulation filtering, cascaded integral comb filtering, and compensated finite-length impulse response filtering can be performed sequentially. Simultaneously, the output and frame boundary consistency label are output together.

[0062] In some embodiments of this application, demodulation filtering configuration can be performed on multi-channel pulse density modulation signals to generate a demodulation parameter set. The demodulation parameter set includes a low-pass cutoff frequency, filter length, coefficient table index, output bit width, and saturation strategy to adapt to the 30kHz to 100kHz frequency band of partial discharge ultrasound and suppress the PDM quantization noise rise band. Based on this, low-pass demodulation filtering can be performed on the multi-channel pulse density modulation signals based on the demodulation parameter set to obtain a demodulated intermediate signal. Low-pass demodulation employs a symmetrical finite-length impulse response structure and reuses the same coefficient table across all channels to ensure cross-channel amplitude and phase consistency, and avoids frame edge distortion through overlapping storage. For example, the convolutional representation of low-pass demodulation uses:

[0063] in, Indicates the first The demodulated time-domain signal, This represents the time-domain response of the low-pass demodulation filter. Indicates the first Pulse density modulated signal, This represents the convolution operation.

[0064] Furthermore, decimation link configuration is performed on the demodulated intermediate signal, and a decimation parameter set is generated. The decimation parameter set includes the decimation ratio. The cascaded integrator-comb filter stages, the length of the compensated finite impulse response filter, and the coefficient table index are adjustable between computational complexity and bandwidth specifications. For example, when the pulse density modulation clock is 1MHz and the target sampling rate is 250kHz, the decimation factor is 4 and the cascaded integrator-comb filter stages are 3, while the length of the compensated finite impulse response filter is on the order of 64 points. Further, cascaded integrator-comb filtering decimation is performed on the demodulated intermediate signal based on the decimation parameter set to obtain the decimated intermediate signal. The cascaded integrator-comb filtering decimation uses:

[0065] in, This represents the discrete sequence output by the integral stage. This represents the discrete output sequence of the comb stage. Indicates the extraction multiplier. Indicates the original sampling point number. This indicates the sequence number of the sampling point after extraction.

[0066] Furthermore, a compensated finite impulse response (FOR) filter is applied to the decimated intermediate signal to obtain a multi-channel time-domain signal with the target sampling rate. The compensated FOR filter compensates for passband droop and suppresses decimation aliasing, while ensuring cross-channel consistency by multiplexing the same coefficient table. The compensation filter employs:

[0067] in, The target sampling rate is represented by the first... Multi-channel time-domain discrete signal sequence This represents the coefficients of the compensated finite impulse response filter. Indicates the filter length. This indicates the extraction of an intermediate discrete sequence.

[0068] It should be noted that, in order to ensure that subsequent pulse event windows can be directly located in the target sampling rate sample domain and consistent with the synchronous sampling timing, in this embodiment, the multi-channel time-domain signal and the synchronous sampling timing can be consistently labeled and output. During this process, the frame start point and frame sequence number can be mapped from the synchronous sampling timing to the target sampling rate sample domain and output as a consistency label. The consistency label includes the frame sequence number, the frame start sample, and the decimation rate, and serves as a unified time anchor point for subsequent framing.

[0069] Thus, step S3 forms an executable link for low-pass demodulation and decimation and outputs a multi-channel time-domain signal at the target sampling rate, transferring frame boundary consistency to the target sampling rate sample domain, and providing a time-scope closure basis for subsequent bandpass filtering and pulse framing.

[0070] In step S4, the core task is to limit the multi-channel time-domain signal to the partial discharge ultrasonic frequency band, form a stable pulse detection sequence to determine the start and end interval of the pulse event, and perform multi-channel aligned truncation to output multi-channel partial discharge ultrasonic frames associated with the channel identification sequence, providing observation data of the same event and the same window for subsequent covariance matrix construction.

[0071] In this embodiment, bandpass filtering and pulse framing can be performed on multi-channel time-domain signals to obtain multi-channel partial discharge ultrasonic frames. Specifically, a bandpass filtering configuration can be generated based on partial discharge ultrasonic frequency band constraints, and bandpass filtering can be performed on the multi-channel time-domain signals. The bandpass filtering configuration includes passband upper and lower limits, filter length, coefficient table index, and group delay compensation strategy, and ensures that the passband covers the partial discharge ultrasonic frequency band, thereby suppressing low-frequency mechanical noise and high-frequency quantization noise. The bandpass filtering adopts:

[0072] in, Indicates the first Partial discharge ultrasonic signal after bandpass filtering This represents the time-domain response of the bandpass filter. This indicates the output of step S3. Time-domain signal of the path, This represents the convolution operation.

[0073] Based on this, a pulse detection sequence is generated from the bandpass-filtered multi-channel signal, and the start and end intervals of the pulse events are determined. For example, to suppress false detections caused by single-channel spike noise and enhance the detection stability of weak signals at long distances, in this embodiment, channel aggregation energy and cross-channel consistency are used as core criteria, and the detection process is divided into three stages: short-time energy aggregation, normalization, and consistency scoring. Specifically, short-time energy aggregation is calculated using a sliding window and summed along the channel dimension to form a channel aggregation energy sequence, while normalization using the moving mean and moving standard deviation forms the pulse detection sequence. For example, the channel aggregation energy uses:

[0074] in, Indicates the first Energy is aggregated through the channels of a sliding window. Indicates the total number of channels. Indicates the window length. Indicates the step size. Indicates the first Discrete sequence after the band is turned on.

[0075] Furthermore, the normalized pulse detection sequence uses:

[0076] in, Indicates a pulse detection sequence. This represents the moving average of the channel pooling energy. The sliding standard deviation of the channel pooling energy. To represent a stable term that prevents the denominator from being zero, it can be taken as a very small positive number, such as... .

[0077] In some embodiments of this application, after the candidate window is obtained, a consistency score can be further introduced to enhance the partial discharge characteristics of "multi-channel common visibility" and suppress isolated noise windows. This consistency score considers both energy concentration and arrival time consistency, and together with a preset consistency threshold, determines the start and end interval of the pulse event. For example, the window length is 256 points, the step size is 128 points, the preset consistency threshold is 3.0, and the minimum number of consecutive windows is 2 to avoid false detections from a single window.

[0078] To further enhance stability in weak partial discharge scenarios at distances of 10m to 50m, this embodiment adds a lightweight discriminant model to the consistency score for secondary discrimination of candidate windows, and uses the model output as a correction term for the pulse detection sequence. The model input includes multi-channel aggregated energy, the variance of the energy peak time, and a short-time spectral amplitude vector near the center frequency; the output is the pulse existence probability. The model structure employs two layers of one-dimensional convolutions and one layer of gated recurrent units, with a fully connected output layer to output probabilities. The number of convolutional kernels is set to 16 and 32, the kernel length to 7 and 5, and the hidden layer dimension of the gated recurrent unit is set to 64. During training, manually labeled partial discharge pulse windows are used as positive samples, and no-pulse windows are used as negative samples. Random amplitude scaling, random additive noise, and random temporal jitter are used for data augmentation to cover propagation attenuation and noise variations. The training loss function is cross-entropy, which can be expressed as:

[0079] in, Indicates training loss, Indicates the batch sample size. Indicates the first Each sample label Indicates the first Predict the probability of each sample.

[0080] Furthermore, after determining the start and end intervals of the pulse events, the process proceeds to framed output. Multi-channel signals are aligned and truncated according to the pulse event start and end intervals, generating a frame index table. The frame index table contains frame number, start sample, end sample, and corresponding hardware identifier code range references, ensuring that all channel truncation windows under the same frame number are completely consistent. Based on this, multi-channel partial discharge ultrasound frames associated with the channel identifier sequence are output based on the frame index table. During output, the frame number is used as the primary key, and the hardware identifier code is used as the channel primary key to organize the frame data, ensuring that subsequent channel organization and geometric index retrieval can be directly closed.

[0081] Thus far, step S4 has completed the partial discharge ultrasonic frequency band limitation, pulse detection sequence generation, pulse event start and end interval determination, and multi-channel aligned framed output. It also improves the detection stability in weak signal environments through consistency scoring and lightweight discrimination model, providing an observation input basis aligned with the same event for subsequent covariance matrix construction.

[0082] In step S5, the core task is to organize the multi-channel partial discharge ultrasound frames based on the correspondence in step S1 and the mapping relationship in step S2, construct the forward-backward spatial smoothing covariance matrix, calculate the acoustic spatial spectrum based on the minimum variance distortionless response beamforming, and generate an acoustic image to characterize the spatial distribution of the partial discharge sound source and maintain imaging stability under conditions of insufficient single-event snapshots and noise interference.

[0083] In some embodiments of this application, array guidance information can be generated based on correspondence, and subarray index sequences can be generated based on mapping relationships. The array guidance information is determined by the two-dimensional coordinates of the array elements, the sound velocity, and the center frequency, and is used to describe the array manifold vectors under different spatial directions. To bind the directional scan to the array element geometry, the propagation delay and manifold vector elements are:

[0084] in, Indicates the first One microphone in the direction The propagation delay below Indicates the first Each microphone coordinate vector Represents the direction of the unit vector. Indicates the speed of sound. Indicates the center frequency. Represents the elements of a manifold vector. It represents the imaginary unit.

[0085] Furthermore, to enhance the distinction between partial discharge pulses and noise and improve the stability of frequency domain observations, matched filtering can be performed on the multi-channel partial discharge ultrasonic frames before covariance construction, and the center frequency component can be extracted to form the observation vector. The matched filtering uses:

[0086] in, Indicates the first The enhanced partial discharge time-domain signal, This represents the time-domain response of the matched filter. This indicates the bandpass signal output in step S4.

[0087] Based on this, a short-time Fourier transform is performed on the enhanced signal, and the center frequency component is extracted to construct an observation vector. This observation vector is then used as the input for covariance estimation. The center frequency component observation vector is obtained using:

[0088] in, Indicates the first Road time-frequency signal, Represents the short-time Fourier transform. Indicates the center frequency. Represents the observation vector at the center frequency. Indicates the total number of channels.

[0089] In some embodiments of this application, during the channel organization and covariance construction stage, multi-channel partial discharge ultrasound frames can be organized according to a subarray index sequence to construct a forward covariance matrix. The subarray index sequence divides the full array observation vector into multiple subarray observation vectors and forms a virtual snapshot to alleviate the covariance rank deficiency caused by insufficient single-event snapshots. The forward covariance matrix is ​​calculated based on the virtual snapshot and can be expressed as:

[0090] in, Indicates the first The forward covariance matrix corresponding to each virtual snapshot Indicates the first A virtual snapshot signal vector, This indicates the conjugate transpose.

[0091] Building upon this, a backward covariance is constructed to enhance resolution and improve estimation stability in scenarios with related sound sources. Specifically, channel inversion mapping can be performed based on the subarray index sequence to construct the backward covariance matrix. The backward covariance matrix is ​​obtained through the inversion matrix. The forward covariance matrix is ​​obtained by performing complex conjugate and inverse transformations, and expressed as follows:

[0092] in, Indicates the first The backward covariance matrix corresponding to each virtual snapshot Represents the reversed matrix. Indicates complex conjugation.

[0093] In the fusion phase, the forward and backward covariance matrices are fused to obtain the forward-backward spatially smoothed covariance matrix. The fusion process simultaneously performs the averaging function for spatial smoothing and aggregates multiple submatrix virtual snapshots into a stable estimate, which can be expressed as:

[0094] in, Represents the forward-backward spatial smoothing covariance matrix. This indicates the total number of virtual snapshots.

[0095] Furthermore, to ensure the spatial smoothing process is controllable and can be stably reused across different array sizes, the subarray sliding window and window weights can be explicitly written into the implementation chain. Specifically, a subarray sliding window sequence can be generated based on the subarray index sequence, and a window weight sequence can be generated for each sliding window. Simultaneously, the forward and backward covariance submatrices corresponding to each sliding window are calculated according to the subarray sliding window sequence. The forward and backward covariance submatrices are then weighted and fused based on the window weight sequence to output the forward-backward spatial smoothing covariance matrix. For example, the subarray sliding window uses 7×7 elements, and the window weights use Hamming weights to suppress the bias of edge windows on the estimation.

[0096] During the beamforming stage, the minimum variance distortionless response beamforming weights are calculated based on the forward-backward spatial smoothing covariance matrix and array guidance information, and an acoustic spatial spectrum is generated, outputting an acoustic image. The MVDR weights and spatial spectrum are calculated using:

[0097] in, Indicates direction beam weight vector, Indicates direction array manifold vectors, Denotes the inverse of the smoothed covariance matrix. Indicates the energy value of the spatial spectrum. Represents the observation vector at the center frequency. It represents the square of the modulus.

[0098] It should be noted that the spatial spectral energy is calculated on the scanning grid and mapped to the pixel brightness of the acoustic image, with the peak position used as the localization output of the direction of the partial discharge sound source. For example, the azimuth scanning range is -60° to 60° with a step of 1°, the elevation scanning range is -30° to 30° with a step of 1°, and the total number of virtual snapshots is on the order of 64 to ensure stable covariance estimation.

[0099] Thus far, step S5 forms a virtual snapshot through the subarray index sequence and forms a spatial smooth covariance matrix through forward-backward fusion. At the same time, it combines MVDR weights to generate spatial spectrum and acoustic image and improves the main lobe-side lobe ratio under weak signal and insufficient snapshot conditions, providing a visual positioning input basis for subsequent acoustic-optical superposition.

[0100] In step S6, the core task is to align the acoustic image and the visible light image to the same spatial reference and perform superposition output. This allows the spatial distribution of the partial discharge sound source to be intuitively presented against the background of the visible light scene, and outputs a result that can be used for on-site localization. This ensures that the registration transformation is reusable and the error is quantifiable. In this embodiment, the acoustic image and the visible light image can be spatially registered and superimposed to output the localization result of the partial discharge sound source.

[0101] In some embodiments of this application, to ensure the registration process has a clear coordinate system and is easy to implement, an acoustic imaging coordinate system and an optical imaging coordinate system can be established separately, and a correspondence table between pixel coordinates and their respective coordinate systems can be generated. Specifically, an acoustic imaging coordinate system can be established, and an acoustic imaging correspondence table between acoustic image pixel coordinates and the acoustic imaging coordinate system can be generated. The acoustic imaging correspondence table maps the acoustic image pixel coordinates to directional unit vectors or spatial rays and is consistent with the scanning grid in step S5; simultaneously, an optical imaging coordinate system is established, and an optical imaging correspondence table between visible light image pixel coordinates and the optical imaging coordinate system can be generated. The optical imaging correspondence table maps the visible light pixel coordinates to optical rays and is consistent with the camera's intrinsic and extrinsic parameter apertures.

[0102] After the two correspondence tables are formed, the mapping relationship between the acoustic imaging coordinate system and the optical imaging coordinate system is calculated based on the acoustic imaging correspondence table and the optical imaging correspondence table, and a registration transformation table is generated for reuse. The registration transformation table can be represented by a two-dimensional homography matrix, solved by least-squares fitting of multiple sets of corresponding points, and the registration accuracy is quantified by reprojection error. The two-dimensional homography transformation adopts:

[0103] in, Represents the pixel coordinates of the acoustic image. Represents the pixel coordinates of a visible light image. Represents the registration transformation matrix. This represents the normalization scaling factor.

[0104] Furthermore, after the registration transformation table is generated, a geometric transformation is performed on the acoustic image based on the table, and it is then superimposed on the visible light image to finally output the localization result of the partial discharge sound source. During superposition, the spatial spectrum energy is logarithmically compressed to improve the visibility of weak energy areas, and the peak position is marked and superimposed on the visible light image. Simultaneously, the azimuth and elevation angles corresponding to the peak directions are output for on-site verification. For example, the visible light resolution is 1920×1080, the acoustic image pixel grid is 200×200, the number of corresponding points is more than 12, and the reprojection error is controlled to the order of 3 pixels to meet the on-site localization display requirements.

[0105] At this point, step S6 completes the spatial registration and overlay output of the acoustic image and the visible light image, and presents the localization result of the partial discharge sound source in a visual way against the background of the visible light scene, providing a visual output basis for on-site localization decision-making.

[0106] In summary, this method establishes a definite correspondence between channel identifier sequences and array geometric parameters, completes synchronous sampling timing calibration under a fixed subgrid subgrouping organization, and combines frequency band limitation and consistency enhancement discrimination of pulse events to achieve stable framed observation of partial discharge ultrasonic signals. It also generates a high-resolution acoustic spatial spectrum using minimum variance distortion-free beamforming with spatial smooth covariance constraints. Furthermore, it performs spatial registration and superposition output with visible light images, maintaining stable positioning peaks and sidelobe suppression under conditions of high channel scale, strong noise, and weak signal propagation attenuation. This reduces errors introduced by channel mismatch and timing drift, improves the positioning accuracy of partial discharge sound sources, imaging interpretability, and on-site handling efficiency, thereby reducing reliance on inspection downtime and manual troubleshooting, and improving the operational reliability and maintenance safety of electrical equipment.

[0107] It should be noted that, although the embodiments in this application are based on... Figure 1 The steps are described sequentially, but this does not mean that the steps must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which each step is described is intended to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, the step numbers are only used to distinguish different steps and do not constitute a limitation on the execution order of the steps; the specific execution order of each step can be appropriately adjusted according to actual needs, functional requirements, and the inherent logic in actual application scenarios.

[0108] In some embodiments of this application, an array-type acoustic imaging partial discharge detection method is applied to the on-site movement detection of a 35kV switchgear bay. The distance between the array and the cabinet is 20m. During the acquisition, the ambient noise is mainly composed of fans and human voices. The goal is to complete the localization of a partial discharge sound source and output the sound-light superposition result within 30s.

[0109] The complete implementation process may include the following steps: Step 1: Site setup and parameter loading.

[0110] For example, a uniform rectangular array is used, with an element size of 32×32 and a total number of channels. The spacing between array elements is taken The visible light camera has a resolution of 1920×1080, and the acoustic image raster is 200×200. The center frequency of the partial discharge ultrasound is taken as... sound speed Pulse density modulation clock take Target sampling rate is taken Extraction multiplier A single frame is 4096 points long, and 64 frames are continuously collected as one detection batch.

[0111] Step 2: Channel primary key solidification and geometric table placement.

[0112] Write the hardware identifier code returned by each acquisition link into the channel registration table to form a channel identifier sequence. .right Perform deduplication counting and generate a checksum. To exclude the reuse of the same code and missing codes:

[0113] in, Represents the uniqueness check value. Indicates an indicator function, This represents the deduplication set operator. Indicates the channel identifier sequence. Represents the cardinality of a set. This represents the total number of channels. For example, if the base number is still 1024 after deduplication, then... The value is 1. Then, the array row and column indices and the two-dimensional coordinates are written into the geometric index table. The two-dimensional coordinate vector of the path array element is denoted as Its calculation uses:

[0114] in, Represents the coordinate vector of the array element. Indicates the horizontal coordinate of the array element. Represents the vertical coordinate of the array element. Indicates row index, Indicates column index, Indicates the center row index. Indicates the index of the center column. Indicates the spacing between array elements. This indicates transpose.

[0115] For example, the center frequency corresponds to the wavelength. ,Pick satisfy This involves engineering sampling constraints and reducing the risk of spatial aliasing. Finally, the channel identifier code is linked with... Perform one-to-one binding and write to the binding table so that any subsequent data can retrieve geometric information using the channel identifier as the unique entry point.

[0116] Step 3: Subgroup organization, synchronization triggering, and delay alignment.

[0117] To reduce the complexity of parallel organization with 1024 channels, the array is divided into 32 subgroups according to a fixed subgrid, with each subgroup containing 32 channels. The subgroup index is denoted as... The set of channel identifier codes within a subgroup is denoted as A synchronization trigger identifier is assigned to each subgroup, and a synchronization sampling timing table is generated, containing the frame number, frame start point, and frame length. To eliminate fixed delay differences in subgroup links, the equivalent delay sample number for each subgroup is obtained by cross-correlation peak localization using a unified reference edge. :

[0118] in, Indicates the equivalent delayed sample number of the subgroup. This indicates the independent variable that corresponds to the maximum value. Represents a delayed search set. Indicates the cross-correlation window length. Represents the discrete sequence of reference edges. This represents the discrete sequence of reference edges acquired by the subgroup.

[0119] For example, Taking values ​​from -32 to 32, a certain subgroup is obtained. Then perform sample domain alignment on all channels of that subgroup:

[0120] in, This represents a discrete sequence after delay alignment. This represents the discrete sequence before alignment. Indicates the number of delayed samples in the subgroup. Indicates a subgroup index. Indicates the channel index. Indicates the sampling point number.

[0121] After alignment is completed, synchronous sampling is triggered according to the timing table and 1024 pulse density modulation sequences are buffered and output. All outputs are indexed into the database according to the channel identifier code and carry the frame sequence number.

[0122] Step four: PDM demodulation and extraction obtains the target sampling rate waveform.

[0123] Each PDM sequence First, it passes through a low-pass demodulation filter. Obtain the demodulated signal :

[0124] in, This represents the demodulated time-domain signal. This represents the response of the low-pass demodulation filter. Indicates PDM signal, This represents convolution.

[0125] For example, the demodulation filter length is 64 points, the cutoff frequency is 120kHz, and the output bit width is 16 bits. Then... The sampling process involves decimation and passband compensation using a 64-point compensated finite-length impulse response filter, ultimately yielding the discrete sequence at the target sampling rate. The "frame number - starting sample - extraction rate" is written as a consistency marker into the frame index record, so that subsequent pulse windows can be... Direct location within the sample domain.

[0126] Step 5: Frequency band limitation, candidate pulse discovery, and secondary model discrimination.

[0127] right Perform bandpass filtering to obtain For example, the passband ranges from 30kHz to 100kHz, the filter length is 128 points, and the same coefficient table is reused to ensure cross-channel amplitude and phase consistency. To detect candidate pulses, the channel pooling energy is calculated. The detection sequence was obtained by normalization. :

[0128]

[0129] in, This indicates that the window aggregates energy. Indicates a pulse detection sequence. Represents the moving average. Indicates the sliding standard deviation. Indicates a stable term. Indicates the window length. Indicates the step size.

[0130] For example, , The preset consistency threshold is 3.0. When two consecutive windows are greater than 3.0, the candidate start and end intervals are determined and 128 points are extended on both sides of the interval as protection boundaries to form candidate event segments.

[0131] To suppress false triggering caused by spike noise and occasional knocking sounds, a lightweight discriminant model is introduced to the candidate events, outputting the pulse existence probability for secondary screening. The model is denoted as... The input consists of three types of features concatenated together: 1) Aggregation energy fragments within the candidate interval ; 2) Variance of peak arrival time across channels ; 3) Short-time spectral amplitude vector near the center frequency .

[0132] The model structure consists of two 1D convolutional layers, one gated recurrent unit layer, and a fully connected output layer. The number of convolutional kernels is 16 and 32, and the kernel length is 7 and 5. The hidden layer dimension of the gated recurrent unit is 64, and the output is a probability. For example, during the training phase, 5000 positive samples and 5000 negative samples are selected. Data augmentation includes random amplitude scaling from 0.7 to 1.3, random additive noise signal-to-noise ratio from -5dB to 15dB, and random time jitter of ±8 points. Training is performed using cross-entropy loss.

[0133] in, Indicates training loss, Indicates the batch sample size. Indicates a label, This represents the predicted probability.

[0134] For example, a threshold of 0.6 is taken during online runtime. The candidate event is retained and the start and end samples are written to the frame index table. Then, the 1024 channels are processed according to the start and end intervals. Uniformly aligned and captured, outputting multi-channel partial discharge ultrasonic frames.

[0135] Step six: Spatial spectrum imaging and acousto-optic superposition output.

[0136] Perform a short-time Fourier transform on each frame of partial discharge ultrasound and extract the center frequency. The complex spectral components at the location form the observation vector. Then, based on the array element coordinates... Generate guidance information. Direction Propagation delay With array manifold elements use:

[0137] in, Indicates the propagation delay. Represents the coordinates of the array elements. Represents the direction of the unit vector. Represents a manifold element. Indicates the speed of sound. Indicates the center frequency.

[0138] Virtual snapshots are generated by subarray sliding, and forward and backward covariances are constructed and fused to obtain a forward-backward spatially smoothed covariance matrix. The MVDR weights are solved and the spatial spectrum is calculated on a scan grid with azimuth angles from -60° to 60° and elevation angles from -30° to 30°. ,Will The image is mapped to an acoustic image, and the peak position is taken as the result of the sound source direction localization.

[0139] Finally, the acoustic image and the visible light image are registered and superimposed. For example, 12 sets of corresponding points are selected to fit a two-dimensional homography matrix. The acoustic image is geometrically transformed and superimposed onto the visible light image, outputting a positioning result of "visible light background + acoustic heat + peak marker" and simultaneously outputting the peak azimuth and elevation angles for on-site verification.

[0140] Through the complete implementation process described above, this method maintains the consistency of geometric and temporal apertures in high-channel observations based on channel identification and array geometry binding. It improves the reliability of pulse event screening through frequency band limitation and model secondary discrimination. Furthermore, it obtains high-resolution positioning results by combining MVDR spatial spectrum imaging with spatial smoothing covariance constraints and completes acousto-optic registration and superposition output. This method can maintain stable positioning peaks and sidelobe suppression effects under long-distance attenuation and strong noise environments, reduce errors introduced by channel mismatch and timing drift, improve the positioning accuracy of partial discharge sound sources and the efficiency of on-site handling, thereby reducing reliance on manual investigation and downtime and improving equipment operation reliability and maintenance safety.

[0141] It should be understood that the step numbers identified by "Step 1, Step 2" and other similar forms in the above embodiments are only used to distinguish different steps and do not limit the steps to be executed in the order of these numbers. The specific execution order of each step can be adjusted according to its functional requirements and the inherent logic in the actual application scenario. The above step numbers should not be interpreted as a limitation on the implementation process of the embodiments of this application.

[0142] like Figure 2 As shown, the following is an embodiment of an array-type acoustic imaging partial discharge detection system provided by this disclosure. This array-type acoustic imaging partial discharge detection system and the array-type acoustic imaging partial discharge detection methods of the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the array-type acoustic imaging partial discharge detection system, please refer to the embodiments of the above-described array-type acoustic imaging partial discharge detection methods.

[0143] Based on the same concept, another embodiment of this application provides an array-type acoustic imaging partial discharge detection system, comprising: Channel binding module 1 is used to obtain the channel identifier sequence of the microphone array and establish the correspondence between the channel identifier sequence and the array geometric parameters; Group synchronization module 2 is used to divide the microphone array into subgroups according to a fixed subgrid based on the channel identifier sequence, establish a mapping relationship between the subgroups and the channels, generate a synchronous sampling timing sequence based on the mapping relationship and trigger multi-channel synchronous sampling to obtain multi-channel pulse density modulation signals; Demodulation and decimation module 3 is used to perform low-pass demodulation and decimation operations on multi-channel pulse density modulation signals under synchronous sampling timing constraints to obtain multi-channel time-domain signals; The frame preprocessing module 4 is used to perform bandpass filtering and pulse framing on the multi-channel time-domain signal to obtain multi-channel partial discharge ultrasonic frames. The smoothing imaging module 5 is used to organize the multi-channel partial discharge ultrasound frames based on correspondence and mapping relationships and construct the forward-backward spatial smoothing covariance matrix. It also calculates the acoustic spatial spectrum based on the minimum variance distortionless response beamforming to generate an acoustic image characterizing the spatial distribution of the partial discharge sound source. The registration and overlay module 6 is used to spatially register the acoustic image and the visible light image and overlay them to output the localization result of the partial discharge sound source.

[0144] In some embodiments of this application, the smoothing imaging module 5 includes: The index organization submodule is used to generate array guidance information based on the correspondence and generate subarray index sequences based on the mapping relationship, and to organize the multi-channel partial discharge ultrasound frames according to the subarray index sequences; The forward-backward fusion submodule is used to construct the forward covariance matrix and the backward covariance matrix, and fuse the forward covariance matrix and the backward covariance matrix to obtain the forward-backward spatial smoothing covariance matrix. The weighting submodule is used to solve for the minimum variance distortionless response beamforming weights based on the forward-backward spatial smoothing covariance matrix and array guidance information, and to generate the acoustic spatial spectrum and output the acoustic image.

[0145] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. An array acoustic imaging partial discharge detection method, characterized in that, include: Obtain the channel identifier sequence of the microphone array and establish the correspondence between the channel identifier sequence and the array geometric parameters; Based on the channel identifier sequence, the microphone array is divided into subgroups according to a fixed subgrid, and a mapping relationship between the subgroups and the channels is established. Based on the mapping relationship, a synchronous sampling timing sequence is generated and multi-channel synchronous sampling is triggered to obtain multi-channel pulse density modulation signals. Under synchronous sampling timing constraints, low-pass demodulation and decimation operations are performed on multi-channel pulse density modulation signals to obtain multi-channel time-domain signals. Bandpass filtering and pulse framing were performed on the multi-channel time-domain signals to obtain multi-channel partial discharge ultrasonic frames. Based on correspondence and mapping relationships, channel organization is performed on multi-channel partial discharge ultrasound frames and a forward-backward spatial smoothing covariance matrix is ​​constructed. Based on minimum variance distortionless response beamforming, the acoustic spatial spectrum is calculated to generate an acoustic image characterizing the spatial distribution of partial discharge sound sources. The acoustic image and the visible light image are spatially registered and superimposed to output the localization result of the partial discharge sound source.

2. The array-type acoustic imaging partial discharge detection method as described in claim 1, characterized in that, The steps of obtaining the channel identifier sequence of the microphone array and establishing the correspondence between the channel identifier sequence and the array geometric parameters include: Read the channel identifier sequence that is associated with each channel, perform a uniqueness check on the channel identifier sequence, and generate the check result; When the verification result indicates that the channel identifier sequence meets the uniqueness constraint, a geometric index table is generated based on the array geometric parameters. The geometric index table is then bound to the channel identifier sequence to form a correspondence between the channel identifier sequence and the array geometric parameters.

3. The array-type acoustic imaging partial discharge detection method as described in claim 2, characterized in that, Based on the channel identifier sequence, the microphone array is divided into subgroups according to a fixed subgrid, a mapping relationship between the subgroups and channels is established, a synchronous sampling timing sequence is generated based on the mapping relationship, and multi-channel synchronous sampling is triggered to obtain a multi-channel pulse density modulation signal. The steps include: A subgroup index table is generated based on a fixed subgrid partitioning rule, and the subgroup index table is associated with the channel identifier sequence to obtain the mapping relationship between subgroups and channels; Based on the mapping relationship, a synchronization trigger identifier is assigned to each subgroup and a synchronization sampling timing table is generated. A deterministic delay calibration is performed on the synchronization sampling timing table to generate the calibrated synchronization sampling timing. Multi-channel synchronous sampling is triggered according to the calibrated synchronous sampling timing to form a multi-channel pulse density modulation signal associated with the channel identifier sequence.

4. The array-type acoustic imaging partial discharge detection method as described in claim 3, characterized in that, The steps of performing low-pass demodulation and decimation operations on a multi-channel pulse density modulated signal to obtain a multi-channel time-domain signal include: Perform demodulation filtering configuration on multi-channel pulse density modulated signals and generate a demodulation parameter set; Low-pass demodulation filtering is performed on the multi-channel pulse density modulated signal based on the demodulation parameter set to obtain the demodulated intermediate signal; The demodulated intermediate signal is configured with a decimation link and a decimation parameter set is generated. Based on the decimation parameter set, the demodulated intermediate signal is subjected to cascaded integral comb filtering to obtain the decimated intermediate signal. A compensated finite-length impulse response filter is applied to the extracted intermediate signal to obtain a multi-channel time-domain signal with the target sampling rate. The multi-channel time-domain signal and the synchronous sampling timing are consistent and then output.

5. The array-type acoustic imaging partial discharge detection method as described in claim 4, characterized in that, The steps of performing bandpass filtering and pulse framing on multi-channel time-domain signals to obtain multi-channel partial discharge ultrasonic frames include: A bandpass filter configuration is generated based on the partial discharge ultrasonic frequency band constraint, and bandpass filtering is performed on multi-channel time-domain signals; Based on the multi-channel signal after bandpass filtering, a pulse detection sequence is generated and the start and end intervals of the pulse events are determined. The multi-channel signal is aligned and truncated according to the start and end intervals of the pulse events and a frame index table is generated. The multi-channel partial discharge ultrasound frames are output based on the frame index table and associated with the channel identifier sequence.

6. The array-type acoustic imaging partial discharge detection method as described in claim 1, characterized in that, The steps include: organizing multi-channel partial discharge ultrasound frames based on correspondence and mapping relationships, constructing a forward-backward spatially smoothed covariance matrix, and calculating the acoustic spatial spectrum based on minimum variance distortionless response beamforming to generate an acoustic image characterizing the spatial distribution of partial discharge sound sources. Array guidance information is generated based on the correspondence, and subarray index sequences are generated based on the mapping relationship; The multi-channel partial discharge ultrasound frames are organized into channels according to the subarray index sequence, and a forward covariance matrix is ​​constructed. Perform channel inversion mapping based on subarray index sequence and construct backward covariance matrix; The forward and backward covariance matrices are fused to obtain the forward-backward spatially smoothed covariance matrix; Based on the forward-backward spatial smoothing covariance matrix and array guidance information, the minimum variance distortionless response beamforming weights are solved and the acoustic spatial spectrum is generated, outputting an acoustic image.

7. The array-type acoustic imaging partial discharge detection method as described in claim 6, characterized in that, The steps for fusing the forward and backward covariance matrices to obtain the forward-backward spatially smoothed covariance matrix include: Generate a subarray sliding window sequence based on the subarray index sequence, and generate a window weight sequence for each sliding window; Calculate the forward covariance submatrix and backward covariance submatrix corresponding to each sliding window according to the submatrix sliding window sequence; The forward and backward covariance submatrices are weighted and fused based on the window weight sequence to output a forward-backward spatially smoothed covariance matrix.

8. The array-type acoustic imaging partial discharge detection method as described in claim 7, characterized in that, The steps for spatially registering acoustic and visible light images and then superimposing them to output the localization result of the partial discharge sound source include: Establish an acoustic imaging coordinate system and generate an acoustic imaging correspondence table between acoustic image pixel coordinates and the acoustic imaging coordinate system; Establish an optical imaging coordinate system and generate an optical imaging correspondence table between visible light image pixel coordinates and the optical imaging coordinate system; The mapping relationship between the acoustic imaging coordinate system and the optical imaging coordinate system is calculated based on the acoustic imaging correspondence table and the optical imaging correspondence table, and a registration transformation table is generated. Based on the registration transformation table, the acoustic image is geometrically transformed and superimposed with the visible light image to output the localization result of the partial discharge sound source.

9. An array-type acoustic imaging partial discharge detection system, characterized in that, include: The channel binding module is used to obtain the channel identifier sequence of the microphone array and establish the correspondence between the channel identifier sequence and the array geometric parameters; The group synchronization module is used to divide the microphone array into subgroups according to a fixed subgrid based on the channel identifier sequence, establish a mapping relationship between the subgroups and the channels, generate a synchronous sampling timing sequence based on the mapping relationship and trigger multi-channel synchronous sampling to obtain multi-channel pulse density modulation signals. The demodulation and decimation module is used to perform low-pass demodulation and decimation operations on multi-channel pulse density modulation signals under synchronous sampling timing constraints to obtain multi-channel time-domain signals. The frame preprocessing module is used to perform bandpass filtering and pulse framing on the multi-channel time-domain signal to obtain multi-channel partial discharge ultrasonic frames. The smoothing imaging module is used to organize the channels of multi-channel partial discharge ultrasound frames based on correspondence and mapping relationships and construct the forward-backward spatial smoothing covariance matrix. It also calculates the acoustic spatial spectrum based on minimum variance distortionless response beamforming to generate an acoustic image characterizing the spatial distribution of partial discharge sound sources. The registration and overlay module is used to spatially register acoustic images and visible light images and overlay them to output the localization results of partial discharge sound sources.

10. The array-type acoustic imaging partial discharge detection system as described in claim 9, characterized in that, The smoothing imaging module includes: The index organization submodule is used to generate array guidance information based on the correspondence and generate subarray index sequences based on the mapping relationship, and to organize the multi-channel partial discharge ultrasound frames according to the subarray index sequences; The forward-backward fusion submodule is used to construct the forward covariance matrix and the backward covariance matrix, and fuse the forward covariance matrix and the backward covariance matrix to obtain the forward-backward spatial smoothing covariance matrix. The weighting submodule is used to solve for the minimum variance distortionless response beamforming weights based on the forward-backward spatial smoothing covariance matrix and array guidance information, and to generate the acoustic spatial spectrum and output the acoustic image.

Images