Integral partial discharge resistant pressure test device and partial discharge detection method

Abstract

The application provides a kind of integrated partial discharge resistant test device and partial discharge detection method, which integrates programmable high voltage test module, multi-modal sensor assembly containing electrical and acoustic sensors, and signal acquisition and processing module. During testing, the high voltage module applies high voltage to the device under test, and the multi-modal sensor synchronously acquires the electrical and acoustic pulse signals generated by the device. The core is that the signal acquisition and processing module not only analyzes the acquired signals, but also controls the programmable frequency conversion power supply in the high voltage module in reverse according to the analysis results, to change the waveform of the output test high voltage in real time. This closed-loop feedback control mechanism can actively enhance the discharge characteristics or suppress environmental noise. The device significantly improves the accuracy and anti-interference capability of partial discharge detection.

Description

Technical Field

[0001] This invention belongs to the field of partial discharge detection technology, specifically relating to an integrated partial discharge-free withstand voltage test device and a partial discharge detection method. Background Technology

[0002] High-voltage electrical equipment, such as transformers, cables, and switchgear, are core components of power systems. The health of their internal insulation structure directly affects the safety and stability of the entire power grid. Withstand voltage testing and partial discharge testing are two of the most effective factory and field testing methods for evaluating the insulation performance of these devices and detecting early potential defects. Withstand voltage testing aims to verify the equipment's ability to withstand overvoltage, while partial discharge testing can sensitively detect minute defects in the insulation structure, such as air gaps, impurities, or cracks. These defects are the main causes of eventual insulation degradation and breakdown.

[0003] In traditional testing procedures, these two tests are typically performed using a combination of different equipment. Operators first apply a fixed-frequency high voltage to the device under test using a high-voltage generator, and then use a separate partial discharge detector to collect and analyze the discharge signal. This open-loop testing mode has significant limitations. The field environment is often filled with various electromagnetic interferences, and weak partial discharge signals are easily drowned out by strong background noise, leading to missed detections or misjudgments. Operators can only rely on experience, attempting to distinguish between signal and noise through repeated experiments and complex post-processing, a cumbersome and unreliable process. Summary of the Invention

[0004] This invention provides an integrated partial discharge-free withstand voltage test device and a partial discharge detection method to solve the problem of difficulty in distinguishing key signals from noise during the detection process.

[0005] In a first aspect, the present invention provides an integrated partial discharge-free withstand voltage testing device for performing withstand voltage testing and partial discharge detection on the device under test, comprising:

[0006] The high voltage test module is configured to apply a programmable test high voltage to the device under test. The high voltage test module includes a programmable frequency converter and at least one test transformer. The output terminal of the programmable frequency converter is electrically connected to the input terminal of the test transformer.

[0007] A multimodal sensor assembly is used to acquire signals generated by the device under test under high voltage. The multimodal sensor assembly includes at least one electrical sensor for acquiring electrical pulse signals and one acoustic sensor for acquiring acoustic pulse signals.

[0008] The signal acquisition and processing module has its signal input terminal connected to the output terminal of the multimodal sensor assembly, and its control output terminal connected to the control input terminal of the high-voltage test module. The signal acquisition and processing module is used to receive and process electrical pulse signals and acoustic pulse signals, and based on the signal processing results, controls the programmable frequency converter to change the waveform of its output test high voltage.

[0009] Optionally, the high-voltage testing module further includes:

[0010] At least one isolation transformer is connected in series between the programmable frequency converter and the test transformer for electrical isolation of the main circuit of the partial discharge withstand voltage test device;

[0011] A coupling capacitor is connected in parallel in the output circuit of the test transformer. The coupling capacitor is used to couple the electrical pulse signal generated by partial discharge to the electrical sensor.

[0012] A high-voltage measurement component, connected in parallel with the output terminal of the test transformer, is used to measure the amplitude of the output high voltage;

[0013] A safety interlock unit is linked to the shielding door of the test area of ​​the high-voltage test module to prevent high-voltage output when the shielding door is not closed.

[0014] Optionally, the signal acquisition and processing module specifically includes:

[0015] An industrial control computer;

[0016] A partial discharge detection board is installed inside an industrial control computer. The partial discharge detection board has at least two synchronous signal acquisition channels, wherein the first acquisition channel is connected to the output terminal of the electrical sensor and the second acquisition channel is connected to the output terminal of the acoustic sensor.

[0017] A control panel is integrated externally into the signal acquisition and processing module. The control panel is equipped with a power switch and an emergency stop button for controlling the start and stop of the partial discharge withstand voltage test device.

[0018] The industrial control computer is configured to run signal processing algorithms to decompose, correlate, verify, and compare the signals acquired by the partial discharge detection board, and send waveform control commands to the programmable frequency converter through fiber optic or serial communication interfaces based on the comparison results.

[0019] In a second aspect, the present invention also provides a method for detecting partial discharge, the method employing the integrated partial discharge-free withstand voltage test device as described in the first aspect, the method comprising the following steps:

[0020] The high-voltage test module of the control test device applies a test high voltage of the first preset waveform to the device under test, and simultaneously acquires the first mixed electrical pulse signal and the first acoustic pulse signal using the multi-modal sensor component of the test device;

[0021] The first mixed electrical pulse signal is decomposed into a composite signal to obtain the first suspected partial discharge electrical component. The first suspected partial discharge electrical component and the first acoustic pulse signal are then cross-modal time-domain synchronization correlation verification to identify the first group of suspected pulses under the first preset waveform.

[0022] When the first set of suspected pulses meets the preset conditions, the control high voltage test module switches the test high voltage from the first preset waveform to the second preset waveform. The second preset waveform has a voltage change rate distribution characteristic that is different from the first preset waveform.

[0023] Under the excitation of the second preset waveform, the second hybrid electrical pulse signal and the second acoustic pulse signal are synchronously acquired, and the composite signal decomposition and cross-modal time domain synchronization correlation verification are repeated to identify the second set of suspected pulses under the second preset waveform.

[0024] The phase distribution patterns of the first group of suspected pulses and the second group of suspected pulses under their respective excitation waveforms are obtained respectively, and the partial discharge detection results of the device under test are generated according to the change law of the phase distribution pattern when switching from the first preset waveform to the second preset waveform.

[0025] Optionally, the step of performing composite signal decomposition on the first mixed electrical pulse signal to obtain the first suspected partial discharge electrical component includes the following steps:

[0026] Perform wavelet packet transform on the first hybrid electrical pulse signal to decompose the first hybrid electrical pulse signal into multiple sub-frequency band signals;

[0027] Construct the covariance matrix of multiple sub-band signals;

[0028] Eigenvalue decomposition is performed on the covariance matrix to identify the principal component directions representing transient and periodic signal energies;

[0029] Multiple sub-band signals are projected and transformed along the principal component direction;

[0030] Based on the statistical independence of the projection results, the first suspected partial discharge electrical component is obtained by separating and reconstructing it using a blind source separation algorithm.

[0031] Optionally, the step of performing cross-modal time-domain synchronization correlation verification between the first suspected partial discharge electrical component and the first acoustic pulse signal to identify the first group of suspected pulses under the first preset waveform includes the following steps:

[0032] Based on the deployment location of the acoustic sensor in the multimodal sensor assembly and the sound velocity characteristics of the medium of the device under test, the acoustic propagation delay time is calculated.

[0033] The time-domain correlation window is determined based on the timestamp of each pulse in the first suspected partial discharge electrical component and in combination with the acoustic propagation delay time.

[0034] For each acoustic pulse in the first acoustic pulse signal, determine whether the pulse energy of the acoustic pulse exceeds the preset energy threshold;

[0035] If the pulse energy exceeds the energy threshold, the corresponding pulse will be marked as the first group of suspected pulses, until the number of pulses in the first group of suspected pulses exceeds the preset number threshold.

[0036] Optionally, the first preset waveform is a standard sine wave, and the second preset waveform is a flat-topped sine wave.

[0037] Optionally, generating the partial discharge detection result of the device under test based on the change law of the phase distribution pattern when switching from the first preset waveform to the second preset waveform includes the following steps:

[0038] The first derivative curves of the first and second preset waveforms are calculated as a theoretical model.

[0039] The phase distribution pattern of the first group of suspected pulses is matched with the first derivative curve of the first preset waveform to obtain the benchmark compliance.

[0040] The phase distribution pattern of the second group of suspected pulses is matched with the first derivative curve of the second preset waveform to obtain the response compliance.

[0041] Determine whether both the baseline compliance and response compliance are higher than the preset compliance threshold;

[0042] If either or both of the benchmark compliance and response compliance are not higher than the corresponding compliance threshold, the partial discharge detection result of the device under test will be "no partial discharge detected".

[0043] If both the baseline compliance and the response compliance are higher than the compliance threshold, then analyze whether the migration characteristics of the phase distribution pattern are consistent with the morphological change characteristics of the first derivative curve when it changes from the first preset waveform to the second preset waveform.

[0044] If the morphological change characteristics match, the partial discharge detection result of the device under test is generated as "partial discharge detected".

[0045] If the morphological change characteristics do not match, the partial discharge detection result of the device under test will be "no partial discharge detected".

[0046] Optionally, after generating the partial discharge detection result of the device under test indicating that partial discharge has been detected, the following steps are further included:

[0047] The severity level of partial discharge is assessed based on the amplitude information of the first and second groups of suspected pulses;

[0048] By utilizing acoustic propagation delay time and the arrival time difference of signals collected by multiple acoustic sensors, the physical source of partial discharge can be located in three-dimensional space.

[0049] By combining severity level and spatial location results, an equipment status diagnostic report is generated that includes trend analysis and maintenance recommendations.

[0050] Optionally, the method further includes the following steps:

[0051] Before applying the test high voltage with the first preset waveform, the background noise of the device under test under no-load conditions is collected to obtain a background electromagnetic noise sample.

[0052] A background electromagnetic noise model of the device under test is established by learning from a pre-set neural network model of the background electromagnetic noise samples.

[0053] Before performing composite signal decomposition, adaptive noise cancellation processing is performed on the first and second mixed electrical pulse signals using a background noise model.

[0054] The beneficial effects of this invention are:

[0055] This invention integrates electrical and acoustic sensors, enabling a more comprehensive and reliable capture of partial discharge characteristics, effectively reducing potential misjudgments from single signal sources. Based on the real-time collected discharge signal characteristics, the signal acquisition and processing module controls the programmable frequency converter power supply to dynamically adjust the test high-voltage waveform applied to the device under test. This active, interactive testing method can specifically excite specific types of discharges or suppress power frequency interference by changing the frequency, significantly improving the signal-to-noise ratio and detection sensitivity of weak discharge signals. This invention transforms traditional open-loop, static testing into a closed-loop, dynamic diagnostic process, allowing the device not only to determine the presence or absence of discharge but also to delve deeper into the characteristics of the discharge, thereby achieving a more accurate and in-depth assessment of the device's insulation status. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the integrated partial discharge withstand voltage test device according to one embodiment of this application.

[0057] Figure 2 This is a flowchart illustrating a partial discharge detection method in one embodiment of this application.

[0058] Figure 3 This is a schematic diagram of a three-capacitor model for partial discharge in the internal air gap in one embodiment of this application.

[0059] Figure 4 This is one embodiment of the present application. Figure 3 The equivalent circuit diagram.

[0060] Figure 5 This is a schematic diagram of the voltage change in the air gap during partial discharge in one embodiment of this application.

[0061] Figure 6 This is a schematic diagram of the discharge of air bubbles inside the medium in one embodiment of this application.

[0062] Explanation of reference numerals in the attached figures:

[0063] 1. Programmable frequency converter power supply; 2. Grounding terminal; 3. 380V main power switch; 4. High voltage output bushing; 5. Ventilation mesh; 6. Temperature and humidity sensor; 7. Lighting lamp; 8. Shielded glass observation window; 9. Shielded door. Detailed Implementation

[0064] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0065] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0066] The integrated partial discharge-free withstand voltage test device of the present invention will be described in detail below, referring to... Figure 1The device features an integrated design, typically comprising a control console with integrated control and display functions, and a high-voltage test cabinet for generating high voltage and housing the device under test. This integrated structure not only makes the equipment compact and space-saving, but also optimizes the connection between the high-voltage circuit and the control system, helping to reduce external electromagnetic interference and laying the foundation for a partial discharge-free testing environment. The core function of the test device is to perform programmable withstand voltage tests and highly sensitive partial discharge detection on the device under test, such as power transformers, instrument transformers, and cable accessories. The device includes a high-voltage test module, a multi-modal sensor assembly, and a signal acquisition and processing module. The high-voltage test module is configured to apply a programmable test high voltage to the device under test. The core components of this module include a programmable frequency converter and at least one test transformer.

[0067] In this embodiment, the programmable frequency converter (CFD) is the source of the entire high-voltage excitation. It can precisely output AC voltages of specific frequencies (e.g., 50 / 60Hz) and amplitudes (e.g., 0.1V to 420V) according to instructions from the host computer software. Its programmable characteristics are key to the automation and intelligent testing capabilities of this invention, allowing the waveform, frequency, and amplitude of the output voltage to be dynamically adjusted according to test requirements or real-time feedback. The output of this CFD is electrically connected to the input of the test transformer, feeding the adjusted medium voltage to the transformer's input.

[0068] The test transformer is another key component of the high-voltage test module. Its function is to boost the medium voltage output from the programmable frequency converter to the required high-voltage level of the device under test, such as 5kV or higher. The device can be configured with one or more test transformers to accommodate different voltage levels and can simultaneously or time-sharingly test multiple devices under test through a multi-channel design. The device also includes a multimodal sensor assembly for comprehensively acquiring various physical signals that the device under test may generate under high voltage. One innovation of this assembly is its multimodal characteristics, which includes at least one electrical sensor for acquiring electrical pulse signals and one acoustic sensor for acquiring acoustic pulse signals. The electrical sensor is used to capture the transient high-frequency current pulses generated in the circuit by partial discharge events, which is the traditional method for partial discharge detection. The introduction of the acoustic sensor is used to simultaneously capture the ultrasonic signals generated by partial discharge within the insulating medium and propagating to the surface of the device. The two modes of signals are of the same physical origin, but complement each other in signal characteristics and propagation paths, providing richer information dimensions for subsequent signal processing and interference suppression.

[0069] The test apparatus also includes a signal acquisition and processing module. The signal input of this module is connected to the output of the multimodal sensor assembly to receive the raw pulse signals acquired by the electrical and acoustic sensors. Its control output is connected to the control input of the programmable frequency converter in the high-voltage test module. The core task of this module is to receive and process the acquired multimodal signals, such as analyzing signal characteristics through specific algorithms to determine whether a real partial discharge exists. Furthermore, a key technical feature of this invention is that the module can actively and in a closed-loop manner control the programmable frequency converter based on the signal processing results, dynamically changing the waveform of its output test high voltage. For example, when a suspected partial discharge signal is initially identified, the system can automatically switch the standard sine wave to a waveform with different voltage change rates (dV / dt) characteristics (such as a flat-top wave). By observing the response pattern of the discharge behavior under different excitation waveforms, the authenticity of the discharge can be confirmed and its type identified, greatly improving the accuracy of detection and anti-interference capability.

[0070] In one embodiment, the high-voltage test module also includes at least one isolation transformer connected in series between the programmable frequency converter and the test transformer. Its main function is to electrically isolate the main power supply circuit of the device, effectively filtering conducted interference from the power grid, such as surges, spikes, and noise, ensuring the generation of a clean, partial discharge-free test power supply—a prerequisite for achieving high-sensitivity partial discharge measurement. It also improves operational safety. The device is also equipped with a coupling capacitor connected in parallel to the output circuit of the test transformer, i.e., between the high-voltage output terminal and ground. Its function is to couple the nanosecond-level high-frequency electrical pulse signal generated by partial discharge down from the high-voltage bus, forming a low-voltage signal path for subsequent electrical sensors to detect. The selection of the coupling capacitor directly affects the measurement sensitivity and bandwidth; the specifications in this embodiment ensure effective capture of weak discharge signals.

[0071] To accurately monitor the voltage applied to the device under test, the high-voltage test module also includes a high-voltage measurement component. This component, which can be in the form of a high-voltage measuring resistor, forms a high-voltage divider connected in parallel with the output of the test transformer. It precisely reduces the high voltage to a safe voltage range that the measuring instrument can handle, thereby achieving real-time and accurate measurement of the output high-voltage amplitude. The measurement results are not only displayed but also serve as important feedback for the entire closed-loop control system. Finally, to ensure the safety of operators and equipment, the device integrates a safety interlock unit. This safety interlock unit is linked to the shielding door of the high-voltage test module's test area. When the shielding door is not fully closed, the interlock device will issue a prohibition signal, forcibly preventing the high-voltage test module from starting or outputting high voltage. High-voltage output is only permitted after confirming that the shielding door is reliably closed. This design is a mandatory safety standard for high-voltage testing equipment and effectively prevents personal injury accidents caused by misoperation.

[0072] In one implementation, the core of the signal acquisition and processing module is an industrial control computer (IPC) with high reliability and powerful computing capabilities, adaptable to industrial environments. All data analysis, user interaction, and equipment control commands are handled by its internal software system. Inside the IPC, one or more partial discharge detection boards are installed. These boards are dedicated high-performance data acquisition hardware, such as the NI PCI-6220. Each partial discharge detection board has at least two channels capable of strictly synchronous signal acquisition. The first acquisition channel is connected to the output of the electrical sensor in the aforementioned multimodal sensor assembly via a coaxial cable or similar means, for high-speed acquisition of coupled electrical pulse signals. Simultaneously, the second acquisition channel is connected to the output of the acoustic sensor for synchronous acquisition of acoustic pulse signals. This synchronous acquisition capability is fundamental for subsequent cross-modal signal time-domain correlation verification.

[0073] To facilitate on-site control by operators, a control panel is integrated externally into the signal acquisition and processing module. This panel is equipped with necessary physical control elements, such as a power switch to control the start and stop of the entire device and an emergency stop button to immediately cut off the main power supply to the high-voltage circuit in case of an emergency, ensuring that the operator has the highest priority control under any circumstances. The industrial control computer is configured to run a complex signal processing algorithm. This algorithm can perform in-depth processing on the multimodal signals acquired by the partial discharge detection board, such as performing composite signal decomposition on electrical signals mixed with various interferences to extract suspected partial discharge components; then, it performs cross-modal correlation verification and time-domain synchronization analysis with synchronously acquired acoustic signals; finally, it can also perform phase distribution (PRPD) mode comparison on the confirmed discharge signals. Based on these complex analysis and comparison results, the industrial control computer generates corresponding waveform control commands and sends the commands to the programmable frequency converter through a stable and reliable communication interface, such as a fiber optic or RS232 serial communication interface, thereby realizing intelligent and dynamic control of the test voltage waveform.

[0074] Figure 2 This is a flowchart illustrating a partial discharge detection method in one embodiment. The partial discharge detection method employs the integrated partial discharge-free withstand voltage test device described in any of the above embodiments. It should be understood that, although... Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in any other order. Furthermore, Figure 2 At least some steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps. For example Figure 2 As shown, the partial discharge detection method disclosed in this invention specifically includes the following steps:

[0075] S101. The high-voltage test module of the control test device applies a test high voltage of the first preset waveform to the device under test, and uses the multi-modal sensor component of the test device to synchronously acquire the first mixed electrical pulse signal and the first acoustic pulse signal.

[0076] In the initial stage of the partial discharge detection method, a test high voltage with a first preset waveform is applied to the device under test placed inside the high-voltage test cabinet by controlling the high-voltage test module of the test device. The occurrence of partial discharge is closely related to the electric field strength and voltage change rate applied to the insulation system. By applying a sufficiently high voltage, the electrical stress that the device experiences during actual operation can be simulated or exceeded, thereby exciting discharge at potential insulation defects within the device. The first preset waveform is usually selected as a standard power frequency sine wave, i.e. ,in This represents the peak amplitude of the test voltage, while This represents the power frequency angular frequency. A standard sine wave was chosen as the initial excitation because its characteristics are well-known and a large amount of historical data and diagnostic spectra are available for comparison. Simultaneously with the application of high voltage, a multimodal sensor assembly deployed on or near the device under test is activated to acquire signals synchronously. Electrical sensors, such as a detection impedance connected in series with a coupling capacitor, capture the instantaneous high-frequency current pulses propagating in the main circuit caused by partial discharge, forming the first hybrid electrical pulse signal. Simultaneously, acoustic sensors, such as piezoelectric ultrasonic probes, are used to pick up the mechanical stress waves (i.e., sound waves) generated within the insulating medium by the discharge event, forming the first acoustic pulse signal.

[0077] S102. Perform composite signal decomposition on the first mixed electrical pulse signal to obtain the first suspected partial discharge electrical component, and perform cross-modal time-domain synchronization correlation verification on the first suspected partial discharge electrical component and the first acoustic pulse signal to identify the first group of suspected pulses under the first preset waveform.

[0078] After obtaining the raw data, the crucial signal purification and preliminary identification stage begins. The aim is to separate suspected partial discharge signals from the noise-overwhelmed first mixed electrical pulse signal and perform preliminary verification using acoustic signals. The principle is that genuine partial discharge pulses possess unique transient characteristics in both the time and frequency domains, significantly different from common periodic noise (such as power electronic switch interference) or white noise. Partial discharge, as a physical phenomenon, is inevitably accompanied by electromagnetic radiation and acoustic wave release; these two different modal signals share a common origin. In this step, a composite signal decomposition algorithm, such as wavelet packet transform, can be applied to the first mixed electrical pulse signal to decompose it into multiple signal components of different frequency bands. Then, through techniques such as principal component analysis or blind source separation, the signal representing the transient pulse energy—the first suspected partial discharge electrical component—is identified and reconstructed. Subsequently, this electrical component is compared with the synchronously acquired first acoustic pulse signal for cross-modal time-domain synchronization verification. The verification is implemented as follows: for each suspected electrical pulse occurrence time, a theoretical acoustic propagation delay time is calculated based on the acoustic sensor location and the sound velocity in the medium, and an extremely narrow correlation time window is set around this time. If an acoustic pulse with energy exceeding a threshold is detected within the time window, then this pair of electrical-acoustic pulses is confirmed as a valid suspected discharge event.

[0079] S103. When the first group of suspected pulses meets the preset conditions, the control high voltage test module switches the test high voltage from the first preset waveform to the second preset waveform. The second preset waveform has a voltage change rate distribution characteristic that is different from the first preset waveform.

[0080] In this method, when the first group of suspected pulses identified meets a preset activation condition in terms of quantity or amplitude, such as the pulse repetition rate exceeding a set threshold within a unit time, indicating significant discharge activity in the device under test and requiring further diagnosis, the method enters the active detection phase. During this phase, the high-voltage test module is controlled to seamlessly switch its output test high voltage from a first preset waveform (standard sine wave) to a second preset waveform. The selection of the second preset waveform is a key innovation of this method; it must have a voltage change rate (i.e., dV / dt) distribution characteristic significantly different from the first preset waveform. The physical principle is based on the fact that the initial discharge voltage of different types of partial discharge (such as internal hole discharge, surface discharge, or corona discharge) is not only related to the voltage amplitude but also closely related to the instantaneous voltage change rate. For example, internal hole discharge tends to occur in the region of maximum voltage change rate. As one implementation method, the second preset waveform can be selected as a flat-topped sine wave. The voltage change rate of a standard sine wave reaches its maximum at the voltage zero-crossing point and is zero at the peaks and troughs. In contrast, the flat-topped sine wave exhibits extremely high voltage change rates during the rapid rise and fall edges of the waveform, while the voltage change rate approaches zero in the flat top region. By actively altering the characteristics of the excitation source in this way, it is equivalent to performing a "perturbation-response" test on the physical source of the discharge.

[0081] In one implementation, determining whether the first group of suspected pulses meets the preset conditions specifically includes the following steps:

[0082] The temporal entropy reflecting the disorder of the temporal distribution of the first group of suspected pulses is calculated based on the timestamp and power frequency period of the first group of suspected pulses.

[0083] The first group of suspected pulses is spatially located based on the time difference of arrival of signals collected by multiple acoustic sensors in the multimodal sensor assembly, and the spatial entropy reflecting the concentration of the physical source of the first group of suspected pulses is calculated based on the spatial location results.

[0084] The spatiotemporal entropy index characterizing the development trend of partial discharge is obtained by weighted calculation of temporal entropy and spatial entropy.

[0085] Determine whether the spatiotemporal entropy index is lower than a preset trend deterioration threshold. When the spatiotemporal entropy index is lower than the trend deterioration threshold, the first group of suspected pulses is deemed to meet the preset conditions.

[0086] In this embodiment, after obtaining the first group of suspected pulses through preliminary screening, in order to quantitatively assess their potential risk from a time dimension, it is necessary to calculate the disorder of the temporal distribution of these pulses, i.e., time entropy. A mature and stable partial discharge defect source often exhibits discharge behavior tightly locked to a specific phase of the AC voltage, displaying a high degree of periodicity and regularity; its pulses' phase distribution becomes concentrated rather than random. Conversely, background noise composed of multiple sources or initially unstable discharges exhibit greater randomness in their occurrence. Information entropy is a physical quantity that measures the degree of uncertainty or disorder in a system. Therefore, a time distribution with a low entropy value precisely reflects the high orderliness and predictability of the discharge behavior, which is usually associated with a serious defect that is continuously developing. Specifically, a power frequency cycle (e.g., 20 milliseconds) is first divided into segments according to a set resolution. A discrete phase window is defined. Then, all pulses in the first group of suspected pulses are traversed, the number of pulses falling within the i-th phase window is counted, and the probability of their occurrence is calculated. Finally, the time entropy is calculated using the Shannon entropy formula. Its expression is This allows the temporal distribution characteristics of the pulse sequence to be condensed into a single quantitative index. .

[0087] To pinpoint the physical location of defects in space and assess their concentration, multiple acoustic sensors within a multimodal sensor assembly are used for spatial localization, and spatial entropy is calculated. The physical principle is based on the property that sound waves propagate at a finite speed in a medium. When a partial discharge occurs at a point inside the device, the generated ultrasonic signal propagates outwards and is received by acoustic sensors at different locations at different times. The time difference between these signal arrivals contains geometric information about the location of the discharge source. A single, fixed defect source will produce a highly concentrated spatial location, while dispersed noise sources or multiple unrelated weak discharges will result in a diffuse spatial distribution. Spatial entropy is the metric used to quantify the concentration of this spatial distribution. Specifically, this is achieved by solving a set of nonlinear equations based on the signal arrival time difference to locate each valid event in the first set of suspected pulses in three-dimensional space. After obtaining a large number of location points, the physical space of the entire device under test is divided into... Given a set of 3D mesh cells (or voxels), count the number of locations falling within the j-th mesh cell and calculate their probability. Subsequently, the spatial entropy was calculated using the Shannon entropy formula. Its expression is This step not only presents the physical source of the discharge in a three-dimensional visualization, but more importantly, it produces a quantitative indicator. This characterizes the physical concentration of the discharge head. A low spatial entropy value strongly suggests the existence of a locally active hotspot within the device, providing crucial spatial dimensional information for risk assessment.

[0088] After obtaining the temporal entropy and spatial entropy separately, to form a comprehensive index that can accurately characterize the development trend of partial discharge, it is necessary to fuse these two different dimensions of information, that is, to calculate the spatiotemporal entropy index by weighting the temporal and spatial entropy. A discharge source that is temporally regular and spatially concentrated is far more dangerous than a temporally random or spatially diffuse source. Considering either dimension alone may lead to misjudgment. For example, a spatially highly concentrated source may only be a transient, unstable initial discharge with a high temporal entropy; conversely, a temporally highly regular signal may originate from spatially dispersed external periodic interference. Therefore, it is essential to combine the two to form a more robust diagnostic criterion. The specific implementation method is to construct the spatiotemporal entropy index by weighted summation. Its calculation formula is In this formula, and In particular, the weighting coefficients of temporal entropy and spatial entropy reflect the relative importance of temporal regularity and spatial concentration in the comprehensive evaluation, and usually satisfy the following conditions: These weights can be preset based on the type of equipment under test, its insulation structure, and historical operating experience.

[0089] After obtaining the spatiotemporal entropy index, it is determined whether the first group of suspected pulses meets the conditions for initiating further advanced diagnostics. Specifically, it can be determined whether the spatiotemporal entropy index is lower than a pre-set trend deterioration threshold. When the spatiotemporal entropy index is lower than the trend deterioration threshold, it is determined that the discharge activity represented by the first group of suspected pulses has reached a severity level requiring in-depth analysis, thus meeting the preset conditions. This trend deterioration threshold is scientifically set based on extensive experimental data from similar equipment, historical fault case analysis, and industry standards. This implementation method can intelligently determine the severity of the current discharge, avoiding the need to initiate complex diagnostic procedures for every weak signal, thereby significantly improving detection efficiency. At the same time, it can ensure timely and accurate response to dangerous defects with a real deterioration trend, triggering subsequent more refined diagnostic procedures.

[0090] S104. Under the excitation of the second preset waveform, the second mixed electrical pulse signal and the second acoustic pulse signal are synchronously acquired, and the composite signal decomposition and cross-modal time-domain synchronization correlation verification are repeated to identify the second set of suspected pulses under the second preset waveform.

[0091] Under continuous excitation by a second preset waveform (such as a flat-top sine wave), the detection process repeats the previous signal acquisition and processing steps. The high-voltage test module stably outputs the high voltage of the second preset waveform, and the multi-modal sensor component continues to synchronously acquire the second mixed electrical pulse signal and the second acoustic pulse signal. The principle is to ensure the fairness and consistency of the comparative experiment, that is, using the exact same measurement and analysis methods when only a single variable (voltage waveform) is changed, thereby ensuring that any subsequent signal changes are caused by the change of that variable. In specific implementation, the newly acquired second mixed electrical pulse signal is also decomposed into composite signals, and the second suspected partial discharge electrical component under the current excitation is separated and reconstructed using wavelet analysis and blind source separation techniques. Then, cross-modal time-domain synchronization correlation verification is performed again, comparing the electrical component with the synchronously acquired second acoustic pulse signal, and eliminating pseudo-pulses without accompanying acoustic signals. This verification process is completely consistent with the previous one, ensuring the consistency of noise filtering standards.

[0092] S105. Obtain the phase distribution patterns of the first group of suspected pulses and the second group of suspected pulses under their respective excitation waveforms, and generate the partial discharge detection results of the device under test according to the change law of the phase distribution patterns when switching from the first preset waveform to the second preset waveform.

[0093] The final judgment is made by analyzing the phase distribution patterns of two suspected pulses under their respective excitation waveforms. The principle behind this step is that the phase distribution of a partial discharge physical source dependent on electric field changes predictably and regularly shifts with the change in the dV / dt distribution of the voltage waveform. Random noise or fixed-frequency interference does not exhibit this correlation response characteristic. In practice, the phases of the first and second suspected pulses under their respective excitation waveforms are first acquired, generating two phase distribution pattern diagrams (PRPD diagrams). Then, the phase distribution pattern of the first pulse group is matched with the voltage rate of change curve of a first preset waveform (standard sine wave), and the degree of agreement is calculated. Similarly, the phase distribution pattern of the second pulse group is matched with the voltage rate of change curve of a second preset waveform (flat-top sine wave). Only when the phase distributions of both sets of pulses closely match the peak regions of the voltage change rate of their corresponding excitation waveforms, and the migration characteristics of the phase distribution pattern (e.g., migration from near the zero-crossing point of a sine wave to near the steep edge of a flat-topped wave) perfectly match the theoretical migration characteristics of the peak position of the voltage change rate, can it be definitively determined that the device under test has a true partial discharge. By utilizing the concept of phase-locked amplification—actively changing the excitation conditions and observing the response—it fundamentally distinguishes between true partial discharge and complex interference, achieving extremely high detection accuracy and anti-interference capability, and generating reliable partial discharge detection results.

[0094] In one embodiment, decomposing the first mixed electrical pulse signal into a composite signal to obtain the first suspected partial discharge electrical component includes the following steps:

[0095] Perform wavelet packet transform on the first hybrid electrical pulse signal to decompose the first hybrid electrical pulse signal into multiple sub-frequency band signals;

[0096] Construct the covariance matrix of multiple sub-band signals;

[0097] Eigenvalue decomposition is performed on the covariance matrix to identify the principal component directions representing transient and periodic signal energies;

[0098] Multiple sub-band signals are projected and transformed along the principal component direction;

[0099] Based on the statistical independence of the projection results, the first suspected partial discharge electrical component is obtained by separating and reconstructing it using a blind source separation algorithm.

[0100] In this embodiment, the partial discharge signal is a typical non-stationary transient signal, with its energy highly concentrated in both time and frequency, while the background noise typically manifests as broadband white noise or periodic interference at specific frequencies. Traditional Fourier transforms can only provide global frequency information and cannot simultaneously pinpoint the time of signal occurrence. Wavelet packet transform, as a sophisticated time-frequency analysis tool, can logarithmically or linearly divide the signal across the entire frequency domain, providing finer frequency resolution for the high-frequency components, thus perfectly matching the need for separating transient pulses. By convolving the first mixed electrical pulse signal with a set of wavelet basis functions generated by scaling and translating the mother wavelet function, multi-level decomposition of the signal is achieved. This process continuously bisects the signal's frequency band, forming a complete wavelet packet decomposition tree, where each leaf node corresponds to a signal component of a specific frequency band. This process transforms the original one-dimensional time series signal into a two-dimensional time-frequency matrix composed of dozens or even hundreds of parallel sub-band signals. This step initially separates the different components hidden in the complex mixed signal into different sub-band channels based on their frequency characteristics. This does not directly remove noise, but rather maps the original signal from a one-dimensional space to a high-dimensional feature space, creating the necessary conditions for subsequent pattern recognition and signal separation using statistical methods.

[0101] After decomposing the signal into multiple sub-bands, the covariance matrix of these sub-band signals is constructed to quantify their statistical correlation. The energy of a partial discharge pulse is simultaneously reflected in multiple adjacent frequency bands, causing these sub-band signals to exhibit a strong positive correlation. Signals from different sources, such as partial discharge and independent periodic noise, are statistically uncorrelated or weakly correlated. The covariance matrix is ​​the mathematical tool for describing this internal correlation structure of multidimensional data. In practice, if wavelet packet transform generates N sub-band signals, then the covariance matrix C is... A square matrix. The element located in the i-th row and j-th column of the matrix. Defined as the covariance between the i-th and j-th sub-band signals. Elements on the diagonal. This represents the variance of the i-th sub-band signal itself, reflecting the signal energy within that band. By calculating the pairwise covariances of all sub-band signals, a complete covariance matrix can be constructed. This step transforms dynamic time-domain signal information into a covariance matrix describing the second-order statistical characteristics of the signal. This matrix captures the inherent patterns of energy interaction between frequency components within the mixed signal.

[0102] Any symmetric covariance matrix can be decomposed into a set of orthogonal eigenvectors and their corresponding eigenvalues. Each eigenvector represents a principal direction (or principal component) of data variation, while its corresponding eigenvalue represents the magnitude of the variance of the data in that direction, i.e., the degree of energy concentration. In mixed signals, the energy source with the strongest energy, whether it is periodic strong interference or transient partial discharge, will inevitably dominate the structure of the covariance matrix, forming one or more directions with maximal eigenvalues. Therefore, by finding the largest eigenvalue, we can find the pattern representing the main signal energy distribution. One specific implementation method is to solve the characteristic equation. Where C is the covariance matrix, These are the eigenvectors to be solved, and These are the corresponding eigenvalues. Solving this equation yields N eigenvalues ​​and their corresponding eigenvectors. By sorting the eigenvalues, the eigenvectors corresponding to the largest eigenvalues ​​represent the dominant directions of the strongest transient signals (such as partial discharge) and periodic signals (such as noise), respectively. This step rotates the original frequency band-based coordinate system to a new coordinate system based on the principal direction of signal energy change. This new coordinate system can more effectively reveal the intrinsic structure of the data, extracting the main energy of the signal from multiple related frequency bands and concentrating it on a few unrelated principal component directions.

[0103] After identifying the principal component directions, the original multiple sub-band signals need to be projected along these new directions. Each projection calculates the magnitude of the original signal component along a specific principal component direction. Since these principal component directions are selected based on the principle of maximizing signal energy, the projected new signal sequence (i.e., principal component scores) can retain the information of the original signal to the greatest extent, while simultaneously achieving dimensionality reduction and decorrelation. For example, if the first principal component direction mainly represents some kind of periodic noise, then the projection of all sub-band signals along that direction will form a new time series that mainly contains the waveform of this periodic noise. In practice, this involves performing matrix multiplication on the data matrix containing all sub-band signals and a transformation matrix composed of selected eigenvectors. The result of the operation is a new set of time series signals, each corresponding to a principal component. These new time series are linear combinations of the original sub-band signals and are statistically orthogonal and uncorrelated with each other.

[0104] Signals from different physical sources (such as real partial discharges, power supply noise, antenna effects, etc.) are statistically independent. Even after linear mixing, this inherent independence can still be revealed through higher-order statistical analysis. The goal of blind source separation is to find an unmixing matrix that, when used with the observed mixed signal, maximizes the statistical independence between the output signal components. In practice, the algorithm takes the principal component signal obtained in the previous step as input and assumes that it satisfies a linear mixing model. ,in It is the observed principal component signal vector. Let A be an unknown independent source signal vector, and let A be an unknown mixing matrix. The algorithm iteratively optimizes to find a demixing matrix W such that the estimated source signals... The components are most independent of each other (usually measured by maximizing non-Gaussianity). After separation, the waveform characteristics of each independent component are analyzed, such as the steepness, duration, and kurtosis of the pulse, to identify the component that best matches the physical characteristics of partial discharge (instantaneous, non-Gaussian, high steepness). Finally, this component is reconstructed separately to obtain the pure first suspected partial discharge electrical component.

[0105] In one embodiment, performing cross-modal time-domain synchronization correlation verification on the first suspected partial discharge electrical component and the first acoustic pulse signal to identify the first set of suspected pulses under the first preset waveform includes the following steps:

[0106] Based on the deployment location of the acoustic sensor in the multimodal sensor assembly and the sound velocity characteristics of the medium of the device under test, the acoustic propagation delay time is calculated.

[0107] The time-domain correlation window is determined based on the timestamp of each pulse in the first suspected partial discharge electrical component and in combination with the acoustic propagation delay time.

[0108] For each acoustic pulse in the first acoustic pulse signal, determine whether the pulse energy of the acoustic pulse exceeds the preset energy threshold;

[0109] If the pulse energy exceeds the energy threshold, the corresponding pulse will be marked as the first group of suspected pulses, until the number of pulses in the first group of suspected pulses exceeds the preset number threshold.

[0110] In this embodiment, partial discharge events are the common physical source of both electromagnetic pulses and acoustic pulses. Electromagnetic waves propagate at near the speed of light, and their arrival time at the electrical sensor can be approximated as the instantaneous moment of the discharge. However, acoustic waves propagate much slower in insulating media (such as transformer oil, epoxy resin, etc.), typically on the order of thousands to several kilometers per second. Therefore, there must be a calculable propagation delay from the occurrence of the discharge to the capture of the acoustic wave by the acoustic sensor on the device surface. This delay is the physical basis for establishing the time-domain correlation between electro-acoustic signals. First, a three-dimensional model of the device under test is created, and the precise spatial coordinates of each acoustic sensor are calibrated in the model. Then, it is necessary to know in advance or measure the sound velocity of the main insulating medium inside the device under test. Since the initial location of the discharge source is unknown, a possible discharge region can be assumed, and the geometric center point of that region can be taken. As a starting point for estimation. Acoustic propagation delay time. It can be obtained by calculating the straight-line distance between two points and dividing by the speed of sound. The formula is as follows: .

[0111] In a real partial discharge event, the electrical and acoustic pulses generated are synchronous at the physical source. Therefore, on the time axis, the acoustic pulse must lag behind the electrical pulse, and this lag time is theoretically equal to the sound wave propagation time. By setting a dynamic search window around this theoretical delay point, it is possible to selectively search for acoustic events associated with specific electrical pulses from a massive amount of acoustic signals. In practice, for any pulse in the first suspected partial discharge electrical component, its precise occurrence timestamp is recorded. Based on the acoustic propagation delay time calculated in the previous step, the theoretically corresponding arrival time of the acoustic pulse can be determined as follows: Considering the uncertainty of the power supply location and measurement errors, it is necessary to set a width of [missing information] on both sides of the theoretical arrival time. The tolerance range is used to form a time-domain correlation window. The starting time of this window is... The end time is Once the time-domain correlation window is determined, for the elements within the time-domain correlation window... Acoustic signal segment acquired internally Calculate its pulse energy. Pulse energy It can be obtained by integrating the square of the signal amplitude over time, i.e. Then, the calculated energy value is compared with a pre-set energy threshold. This threshold is usually determined based on statistical analysis of the noise level (e.g., taking the mean of the noise energy plus three standard deviations) when collecting background noise data from the device.

[0112] Only when a suspected electrical pulse, within its dedicated time-domain correlation window, contains an acoustic pulse with energy exceeding a threshold, are these two electro-acoustic pulses jointly certified as a high-confidence partial discharge event. This cross-modal cross-validation is the most powerful means of eliminating various single-mode interferences (such as external electromagnetic interference with only electrical signals or mechanical vibration with only acoustic signals). In practice, the program performs a loop check: if the acoustic pulse energy calculated in the previous step is greater than the energy threshold, the electrical pulse that triggered this correlation window is marked as valid. This marked electrical pulse, along with its amplitude, phase, and timestamp information, is formally stored in a set called "First Group of Suspected Pulses." Simultaneously, the system accumulates and counts the number of pulses in this set. This process continues until the number of pulses in the set exceeds a preset threshold. Once the threshold is reached, it is considered that enough statistically significant discharge samples have been collected, and the next step of mode analysis can proceed.

[0113] In one embodiment, generating the partial discharge detection result of the device under test based on the change pattern of the phase distribution mode when switching from a first preset waveform to a second preset waveform includes the following steps:

[0114] The first derivative curves of the first and second preset waveforms are calculated as a theoretical model.

[0115] The phase distribution pattern of the first group of suspected pulses is matched with the first derivative curve of the first preset waveform to obtain the benchmark compliance.

[0116] The phase distribution pattern of the second group of suspected pulses is matched with the first derivative curve of the second preset waveform to obtain the response compliance.

[0117] Determine whether both the baseline compliance and response compliance are higher than the preset compliance threshold;

[0118] If either or both of the benchmark compliance and response compliance are not higher than the corresponding compliance threshold, the partial discharge detection result of the device under test will be "no partial discharge detected".

[0119] If both the baseline compliance and the response compliance are higher than the compliance threshold, then analyze whether the migration characteristics of the phase distribution pattern are consistent with the morphological change characteristics of the first derivative curve when it changes from the first preset waveform to the second preset waveform.

[0120] If the morphological change characteristics match, the partial discharge detection result of the device under test is generated as "partial discharge detected".

[0121] If the morphological change characteristics do not match, the partial discharge detection result of the device under test will be "no partial discharge detected".

[0122] In this embodiment, taking internal air gap partial discharge as an example, refer to... Figure 3 and Figure 4 Cg represents the capacitance of the air gap, Cb (which is Cb1 and Cb2 connected in series) represents the capacitance of the dielectric in the section connected in series with Cg, and represents the capacitance of the remaining insulating portion. If an AC voltage is applied between the electrodes, the voltage appearing on Cg is ug, i.e.

[0123]

[0124] Because the air gap is very small, Cg is much larger than Cb, therefore g is much smaller. The voltage and current changes in the air gap during partial discharge are shown in the figure below. As g increases, when it rises to the initial discharge voltage (g reaches the discharge voltage of Cg), the air gap in Cg discharges. Therefore, the voltage on Cg quickly drops from [value missing] to [value missing], and the discharge is extinguished. In the formula, The residual voltage on Cg , This represents the corresponding applied voltage value. The voltage reconstructed on Cg after discharge will differ from ug, but its upward trend with increasing applied voltage will resemble that of ug. Rise, when it rises to That is, the applied voltage difference has increased. At that time, Cg discharged again, the discharge extinguished again, and the voltage dropped again. The voltage variation on Cg is to The time interval, that is, the time during which a partial discharge pulse is generated, during which a pulse current i flows through Cg in the external circuit, such as... Figure 5 As shown. The discharge diagram of the bubbles inside the dielectric is as follows. Figure 6As shown. Discharge will only occur when the voltage changes rapidly enough that the voltage across the defect accumulates within a short time and reaches its breakdown threshold. Therefore, the amplitude distribution of the first derivative curve of the voltage (i.e., the voltage rate of change curve) theoretically determines the phase position where partial discharge is most likely to occur during an AC cycle.

[0125] In this embodiment, the standard sine wave, which is the first preset waveform, is... Its first derivative curve is The curve reaches its peak near the voltage zero-crossing points (0 degrees and 180 degrees). For the flat-topped sine wave, which serves as the second preset waveform, its mathematical expression is more complex, but a significant characteristic of its first derivative curve is the appearance of sharp, narrow, and high peaks at the edges of rapid rises and falls in the waveform. By differentiating the mathematical functions of these two preset waveforms, two theoretical discharge probability distribution curves can be obtained. After obtaining the theoretical model, the first group of suspected pulses is statistically analyzed according to their phases to generate a discrete phase distribution histogram, i.e., the PRPD spectrum. Then, this actually observed distribution histogram is morphologically compared with the first derivative curve of the first preset waveform (standard sine wave) calculated in the previous step. Statistical tools such as the Pearson correlation coefficient or cross-correlation function can be used to quantify the similarity between the two. For example, by calculating the correlation coefficient between the two curves, a value ranging from -1 to 1 can be obtained, which is the baseline consistency. A correlation coefficient value close to 1 indicates that the actual discharge phase distribution is highly consistent with the theoretical prediction.

[0126] Similar to the previous step, a second pattern matching is needed to compare the phase distribution pattern of the second set of suspected pulses with the first derivative curve of the second preset waveform to obtain the response compliance. The principle behind this step is to verify the response of the discharge behavior after being disturbed (i.e., a change in voltage waveform). A real partial discharge physical source must not only behave normally under standard conditions, but more importantly, its response must also change synchronously and predictably when the excitation conditions change. If the discharge phase distribution can closely follow the unique voltage change rate curve of the second preset waveform, then it becomes extremely strong evidence of a real discharge. The specific implementation is exactly the same as the previous step: first, plot the phase distribution histogram of the second set of suspected pulses, then compare it with the first derivative curve of the second preset waveform, and calculate the correlation coefficient between the two; this value is the response compliance. A high response compliance indicates that the discharge behavior can accurately track drastic changes in the applied voltage change rate, which is crucial for distinguishing real discharges from fixed-phase interferences unaffected by the voltage change rate (such as switching noise synchronized with the grid's fixed phase).

[0127] After calculating the baseline compliance and response compliance, a threshold judgment needs to be applied to these two indicators. Insufficient compliance under any condition may mean that the observed signal is not from a voltage-driven discharge, but rather some kind of noise or spurious signal. Therefore, a compliance threshold (e.g., 0.8) needs to be preset. This threshold is based on extensive experiments and experience and represents the minimum acceptable level of compliance. Then, it is checked whether the baseline compliance and response compliance simultaneously exceed the compliance threshold. If either or both of the baseline compliance and response compliance fail to exceed the preset compliance threshold, the detection process will directly reach a negative conclusion, the program will immediately interrupt subsequent analysis processes, and directly generate the final detection result. The report will clearly state that "no (credible) partial discharge was detected" in this test.

[0128] Once both the baseline compliance and response compliance successfully pass the threshold judgment, the detection process will proceed to the final confirmation stage: analyzing the migration characteristics of the phase distribution pattern. Since the phase distribution of the partial discharge physical source must not only statically match the dV / dt curve, its dynamic migration law must also conform to the migration law of the dV / dt curve. For example, when switching from a standard sine wave to a flat-top wave, the peak region of dV / dt will significantly migrate from near the voltage zero-crossing point to the steep edge of the flat-top wave. Therefore, the actual discharge pulse must also undergo the same regular and predictable phase drift. In practice, it is necessary to compare and analyze the PRPD spectra of the first and second groups of pulses to observe whether the central phase of their main discharge clusters has migrated in accordance with theoretical predictions. The change in the centroid phase of the two distributions can be calculated and compared with the change in the peak phase of the theoretical first derivative curve. This process can be automatically completed by image processing algorithms or pattern recognition algorithms. If, after final migration feature analysis, it is confirmed that the migration characteristics of the phase distribution pattern completely match the morphological change characteristics of the first derivative curve as it changes from the first preset waveform to the second preset waveform, the system will generate a final positive detection result. Conversely, if the final migration feature analysis finds that the migration characteristics of the phase distribution pattern do not match the theoretical prediction, even if its static conformity is high, the system will generate a negative detection result.

[0129] In one embodiment, after generating a partial discharge detection result for the device under test indicating that partial discharge has been detected, the following steps are further included:

[0130] The severity level of partial discharge is assessed based on the amplitude information of the first and second groups of suspected pulses;

[0131] By utilizing acoustic propagation delay time and the arrival time difference of signals collected by multiple acoustic sensors, the physical source of partial discharge can be located in three-dimensional space.

[0132] By combining severity level and spatial location results, an equipment status diagnostic report is generated that includes trend analysis and maintenance recommendations.

[0133] In this embodiment, the amplitude of the partial discharge pulse (typically measured in picocoos per 100 pC) directly reflects the amount of energy released in each discharge event. A larger amplitude indicates a stronger destructive effect of a single discharge on the insulating material, and the faster the defect may develop. By analyzing the amplitude distribution of a large number of pulses, a more comprehensive severity assessment can be obtained than a single measurement. One specific implementation involves merging the partial discharge pulses identified in the first and second groups to form a complete discharge dataset. Then, statistical analysis is performed on all pulse amplitudes in this dataset to calculate key statistics, such as the maximum discharge quantity and the average discharge quantity. The maximum discharge quantity represents the discharge intensity under the worst-case scenario, while the average discharge quantity reflects the overall level of discharge. Furthermore, a discharge quantity-repetition rate (qn) graph can be plotted to analyze the discharge frequency in different amplitude ranges. Based on industry standards or reference values ​​provided by equipment manufacturers, the calculated maximum and average discharge quantities can be classified into different severity level ranges.

[0134] When a partial discharge event occurs, the resulting sound waves spread outwards and are received sequentially by at least four (for three-dimensional positioning) non-coplanar acoustic sensors deployed at different locations within the device. Since sound waves propagate at a uniform speed in the medium, the time difference between any two sensors receiving the same sound wave signal corresponds to the distance difference between the discharge source and those two sensors. By establishing a set of nonlinear hyperboloid equations with the discharge source coordinates as unknowns and solving these equations, the three-dimensional spatial coordinates of the discharge source can be derived. One specific implementation involves recording the time difference between ... One acoustic sensor ( Received timestamp Using the first sensor that receives the signal as a reference, we can obtain... Time difference Let the coordinates of the power source be... The coordinates of the i-th sensor are The speed of sound in the medium is Then we can establish a system of equations:

[0135]

[0136] The coordinates of the discharge source can be obtained by solving this system of equations using iterative optimization algorithms such as Taylor series expansion. Finally, after obtaining the severity level and precise spatial location of the localized discharge, this information needs to be organically combined to generate a comprehensive equipment condition diagnostic report that includes trend analysis and maintenance recommendations. The report will first clearly list the key results of this inspection: severity level indicators such as maximum discharge quantity, average discharge quantity, and discharge repetition rate, along with a PRPD (Pressure, Discharge, and Damage) map. Then, the 3D location results will be visually overlaid on the structural diagram of the tested equipment, clearly marking the physical location of the defect. In the trend analysis section, if historical data is available, the report will plot the trend curve of discharge quantity over time, and determine whether the defect is in a stable, slowly developing, or rapidly deteriorating phase. In the maintenance recommendations section, specific suggestions will be given based on the severity level, spatial location, and development trend.

[0137] In one embodiment, the partial discharge detection method further includes the following steps:

[0138] Before applying the test high voltage with the first preset waveform, the background noise of the device under test under no-load conditions is collected to obtain a background electromagnetic noise sample.

[0139] A background electromagnetic noise model of the device under test is established by learning from a pre-set neural network model of the background electromagnetic noise samples.

[0140] Before performing composite signal decomposition, adaptive noise cancellation processing is performed on the first and second mixed electrical pulse signals using a background noise model.

[0141] In this implementation, any partial discharge measurement is inevitably subject to electromagnetic interference from the environment and the test system itself. This noise, like the actual partial discharge signal, is an electrical pulse signal. If its characteristics are not understood beforehand, it will be superimposed on the actual signal during subsequent boost tests, severely affecting the sensitivity and accuracy of the detection. In practice, all test circuits need to be connected, and the control system and measuring instruments need to be started, but the programmable frequency converter should not output voltage. In this state, using the same electrical sensors and data acquisition boards as in the formal test, electromagnetic signals are collected for a period of time. This signal duration needs to be long enough to cover various periodic changes and random fluctuations that may occur in the noise, thereby forming a statistically representative sample of background electromagnetic noise.

[0142] After obtaining background electromagnetic noise samples, an LSTM network is used as the pre-set model. The collected background electromagnetic noise sample sequence is used as input data, and the network is trained through supervised learning. The training goal is to enable the network to accurately predict the noise signal at the next moment based on the noise signal at the previous moment, continuously adjusting the weights and biases within the network by minimizing the mean square error between the predicted and true values. After sufficient training, a nonlinear dynamic model capable of real-time noise prediction is formed. The collected mixed signal is input into the trained neural network noise model in real time, and the model predicts the noise component in the signal at the next moment based on its learned patterns. Subtracting this accurately predicted noise component from the original mixed signal leaves mainly unpredictable and unrelated partial discharge signals. Due to the adaptability of the neural network, even if the characteristics of the background noise change slowly, the model can track and adapt to a certain extent. The final effect of this step is to accurately and adaptively remove most of the background noise from the original mixed signal without damaging the real partial discharge signal, greatly improving the signal-to-noise ratio. This allows subsequent composite signal decomposition and pattern recognition algorithms to be performed on a cleaner data basis, thereby significantly improving the sensitivity and reliability of the entire detection method.

[0143] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0144] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. An integrated no-partial-discharge withstand voltage test device, characterized by comprising: The application relates to a device for voltage withstand test and partial discharge detection. The high-voltage test module comprises a programmable variable-frequency power supply and at least one test transformer, and the output end of the programmable variable-frequency power supply is electrically connected with the input end of the test transformer. The multi-modal sensor assembly comprises at least one electrical sensor for collecting electrical pulse signals and one acoustic sensor for collecting acoustic pulse signals. The signal acquisition and processing module is connected with the output end of the multi-modal sensor assembly, and the control output end of the signal acquisition and processing module is connected with the control input end of the high-voltage test module. The high-voltage test module is used for applying test high voltage of a first preset waveform to the device under test, and the multi-modal sensor assembly is used for synchronously collecting first mixed electrical pulse signals and first acoustic pulse signals. The signal acquisition and processing module is used for performing composite signal decomposition on the first mixed electrical pulse signals to obtain first suspected partial discharge electrical components, and performing cross-modal time-domain synchronism correlation verification on the first suspected partial discharge electrical components and the first acoustic pulse signals to identify a first group of suspected pulses under the first preset waveform. The high-voltage test module is further used for switching the test high voltage from the first preset waveform to a second preset waveform when the first group of suspected pulses satisfies a preset condition, and the second preset waveform has a different voltage change rate distribution characteristic from the first preset waveform. The multi-modal sensor assembly is further used for synchronously collecting second mixed electrical pulse signals and second acoustic pulse signals under excitation of the second preset waveform.

2. The integrated no-partial-discharge withstand test device according to claim 1, wherein The signal acquisition and processing module is further used for repeatedly performing the composite signal decomposition and the cross-modal time-domain synchronism correlation verification to identify a second group of suspected pulses under the second preset waveform, obtaining phase distribution patterns of the first group of suspected pulses and the second group of suspected pulses under respective excitation waveforms, and generating a partial discharge detection result of the device under test according to a variation rule of the phase distribution patterns from the first preset waveform to the second preset waveform. The high-voltage test module further comprises: At least one isolation transformer connected in series between the programmable variable-frequency power supply and the test transformer, used for electrical isolation of the main circuit of the partial discharge-free voltage withstand test device; A coupling capacitor connected in parallel with the output circuit of the test transformer, used for coupling electrical pulse signals generated by the partial discharge to the electrical sensor; A high-voltage measurement assembly connected in parallel with the output end of the test transformer, used for measuring the amplitude of the output high voltage; 3. The integrated no-partial-discharge pressure-test device of claim 1, wherein, A safety interlocking unit linked with a shielding door of the test area of the high-voltage test module, used for prohibiting high-voltage output when the shielding door is not closed. The signal acquisition and processing module specifically comprises: An industrial control computer; A partial discharge detection board card installed in the industrial control computer, the partial discharge detection board card having at least two synchronous signal acquisition channels, wherein the first acquisition channel is connected with the output end of the electrical sensor, and the second acquisition channel is connected with the output end of the acoustic sensor; A control panel integrated on the outside of the signal acquisition and processing module, the control panel being provided with a power switch and an emergency stop button used for controlling the start and stop of the partial discharge-free voltage withstand test device. The industrial control computer is configured to run a signal processing algorithm for decomposing, correlation checking and pattern comparison of signals collected by the partial discharge detection board card, and send waveform control instructions to the program-controlled variable frequency power supply through an optical fiber or a serial communication interface according to the comparison result.

4. The integrated no-partial-discharge pressure-test device of claim 1, wherein, The composite signal decomposition on the first mixed electrical pulse signal to obtain the first suspected partial discharge electrical component includes the following steps: Performing wavelet packet transform on the first mixed electrical pulse signal to decompose the first mixed electrical pulse signal into a plurality of sub-band signals; Constructing a covariance matrix of the plurality of sub-band signals; Performing eigenvalue decomposition on the covariance matrix to identify the principal component direction representing the transient signal energy and the periodic signal energy; Projecting the plurality of sub-band signals along the principal component direction; Based on the statistical independence of the projection result, the first suspected partial discharge electrical component is separated and reconstructed by using a blind source separation algorithm to obtain the first suspected partial discharge electrical component.

5. The integrated no-partial-discharge pressure-test device of claim 1, wherein, The cross-modal time-domain synchronicity correlation check on the first suspected partial discharge electrical component and the first acoustic pulse signal to identify the first group of suspected pulses under the first preset waveform includes the following steps: Based on the deployment position of the acoustic sensor in the multi-modal sensor assembly and the medium sound velocity characteristics of the device under test, the acoustic propagation delay time is calculated; Taking the timestamp of each pulse in the first suspected partial discharge electrical component as a reference, and combining the acoustic propagation delay time to determine the time-domain correlation window; For each acoustic pulse in the first acoustic pulse signal, it is judged whether the pulse energy of the acoustic pulse exceeds a preset energy threshold; If the pulse energy exceeds the energy threshold, the corresponding mark is marked as the first group of suspected pulses, until the number of pulses of the first group of suspected pulses exceeds a preset number threshold.

6. The integrated no-partial-discharge pressure-test device of claim 1, wherein, The first preset waveform is a standard sine wave, and the second preset waveform is a flat-top sine wave.

7. The integrated no-partial-discharge pressure-test device of claim 1, wherein, The local discharge detection result of the device under test is generated according to the variation rule when the phase distribution pattern is switched from the first preset waveform to the second preset waveform, including the following steps: Calculate the first derivative curve of the first preset waveform and the second preset waveform as a theoretical model; Perform first pattern matching on the phase distribution pattern of the first group of suspected pulses and the first derivative curve of the first preset waveform to obtain a reference compliance degree; Perform second pattern matching on the phase distribution pattern of the second group of suspected pulses and the first derivative curve of the second preset waveform to obtain a response compliance degree; Determine whether the reference compliance degree and the response compliance degree are higher than a preset compliance threshold; If any one or both of the reference compliance degree and the response compliance degree is not higher than the corresponding compliance threshold, the local discharge detection result of the device under test is generated as no local discharge is detected; If the reference compliance degree and the response compliance degree are both higher than the compliance threshold, analyze whether the migration characteristics of the phase distribution pattern are consistent with the morphological change characteristics of the first derivative curve when the first derivative curve changes from the first preset waveform to the second preset waveform; If the morphological change characteristics are consistent, the local discharge detection result of the device under test is generated as local discharge is detected; If the morphological change characteristics are not consistent, the local discharge detection result of the device under test is generated as no local discharge is detected.

8. The integrated no-partial-discharge pressure-test device of claim 7, wherein, After the partial discharge detection result of the device under test is generated as detecting partial discharge, the method further comprises the following steps: Based on the amplitude information of the first group of suspected pulses and the second group of suspected pulses, the severity level of the partial discharge is evaluated; Using the acoustic propagation delay time and the time difference of the signals collected by the plurality of acoustic sensors, the physical source of the partial discharge is three-dimensionally positioned in space; Combining the severity level and the spatial positioning result, a device state diagnosis report containing trend analysis and maintenance suggestions is generated.

9. The integrated no-partial-discharge pressure-test device of claim 1, wherein, Further comprising: Before the test high voltage of the first preset waveform is applied, the background noise of the device under test in the no-load state is collected to obtain a background electromagnetic noise sample; Using a preset neural network model, the background electromagnetic noise sample is learned to establish a background noise model of the device under test; Before the composite signal decomposition is performed, the first mixed electrical pulse signal and the second mixed electrical pulse signal are subjected to adaptive noise cancellation processing using the background noise model.

Images