Multifunctional comprehensive detection device for partial discharge of power equipment

Abstract

The invention discloses a multifunctional power equipment partial discharge comprehensive detection device, and relates to the technical field of power equipment state monitoring and fault diagnosis. The system comprises a multi-mode sensor interface (TEV / HFCT / UHF / air coupling ultrasound), a phase reference and cross-channel alignment unit, a physical scene adaptive window estimation module, an event sparse state fusion unit and a joint criterion and control unit. The device adapts the time difference of arrival and a phase window and generates a frequency band weight according to equipment geometry / medium information, event level judgment is completed under the common constraint of delta t / delta phi consistency, spectral kurtosis and fusion score in combination with an event purification mask and an evidence score, PRPD / PRPS is output, and graded alarming is performed; a coaxial composite ultrasonic wave concentrator improves the acoustic signal-to-noise ratio, and a calibration and traceability unit (a detachable UHF calibration cavity, a fast edge source and a probe identification memory) realizes link sensitivity and group delay field compensation. Power-off-free and cross-scene high-precision detection is realized, false alarm and missing detection are reduced, and consistency and traceability are enhanced.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, specifically a multifunctional power equipment partial discharge comprehensive detection device. Background Technology

[0002] Partial discharge is one of the most sensitive early signs of insulation degradation in high-voltage power equipment, typically occurring in switchgear (including ring main units), GIS, transformer leads, cables and their terminals, and overhead line fittings. Due to the influence of operating environment, electromagnetic interference, and structural differences, single-physical-quantity detection (such as UHF only, TEV only, or ultrasonic only) often has detection blind spots and potential for misjudgment. To avoid power outages, the industry generally uses live-line testing; however, limited on-site space, complex interference sources, and diverse equipment types necessitate portable, robust, and cross-scenario consistent testing methods.

[0003] Existing technologies include: metal casing coupling based on surface voltage (TEV), electromagnetic radiation based on ultra-high frequency (UHF), acoustic methods based on ultrasound (AE / air coupling), and conductor current methods based on high-frequency current transformers (HFCT). Handheld instruments combining two or three types of sensors are also available. Some solutions introduce phase references to generate PRPD maps, or utilize empirical thresholds, simple features, or classifiers for discrimination. While these technologies can be effective in specific scenarios, they reveal common problems under conditions of multiple devices, strong interference, and long-term online inspection: portable multimodal solutions often involve "parallel acquisition and independent discrimination," lacking a unified phase scale and high-precision time alignment across channels, resulting in insufficient cross-modal comparability of PRPD / PRPS; thresholds and time difference of arrival (Δt) windows often rely on experience or fixed configurations, making it difficult to adapt to device geometry and medium characteristics, resulting in poor cross-scenario mobility; and under strong RFI and on-site noise, there is a lack of event-level cleanup and evidence accumulation mechanisms, making it difficult to balance false alarm and false negative rates.

[0004] Ultrasonic telemetry suffers from limited signal-to-noise ratio, insufficient directivity, and inadequate sidelobe suppression.

[0005] Link sensitivity and group delay rely heavily on factory calibration, and there is a lack of miniaturized and detachable calibration traceability systems in the field, resulting in insufficient long-term data consistency and traceability. Alarm strategies are mostly triggered by single thresholds, lacking the coordination of frequency band weighting and hysteresis / debouncing, which easily leads to jitter alarms in the field. Data management and probe parameter management are scattered, and the transmission of calibration parameters after probe replacement is not standardized, affecting the reproducibility of results.

[0006] Existing technologies improve applicability by using multiple sensors in parallel, adding phase references, or adopting empirical rules, but they still have certain limitations: such as inconsistent criteria due to cross-modal time / phase asynchrony, difficulty in matching different equipment structures with fixed Δt / phase windows, lack of event-level purification and evidence scoring for strong interference, insufficient ultrasonic directivity and SNR, and lack of field-demountable UHF calibration cavities and automatic compensation links.

[0007] Therefore, there is an urgent need for a multifunctional power equipment partial discharge comprehensive detection device that integrates multimodal acquisition, non-contact phase reference and high-precision cross-channel alignment, physical scene adaptive window estimation based on device geometry / medium, event sparse state fusion and joint criteria and calibration traceability, so as to achieve cross-scenario, high consistency and low false alarm / low missed detection engineering applications under power outage conditions. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and propose a multifunctional power equipment partial discharge comprehensive detection device to solve the above-mentioned problems.

[0009] The objective of this invention is achieved through the following technical solution: a multifunctional power equipment partial discharge comprehensive detection device, characterized in that it comprises:

[0010] A multimodal sensor interface and acquisition unit for parallel acquisition of pulse signals from at least two of the following probes: surface voltage probe, high-frequency current transformer, ultra-high frequency near-field antenna, and air-coupled ultrasonic probe;

[0011] The phase reference and cross-channel alignment unit includes a non-contact phase reference module (which picks up the power frequency phase through an electric field or magnetic flux and obtains it through a digital phase-locked loop) and a hardware alignment module that works in conjunction with a time-to-digital converter and a processor to map different mode pulses to a unified power frequency phase scale, with a cross-channel alignment resolution of no more than 100ps.

[0012] The physical scene adaptive window estimation module outputs the arrival time difference window Δt between different modes based on the type of the measured object and its geometric / medium parameters (including equipment category, typical size, thickness of metal or insulating material, and relative permittivity). min ,Δt max With phase correction It generates frequency band weight vectors for spectral feature calculation and spectrum drawing; the physical scene adaptive window estimation module adopts a local multi-scale window modeling strategy that separates the representation of each modal channel and interacts only on the channel axis to improve cross-scene generalization ability.

[0013] The event sparse state fusion unit performs a fusion of arrival time (t) and amplitude of a multimodal pulse event sequence. Power frequency phase Frequency band indexing performs event-driven sparse state updates and outputs an event cleanup mask (clean). mask With evidence scoring score This unit uses a parallelizable alternative dynamics approximation for state updates during the training phase and maintains event-driven, low-power computation during the inference phase.

[0014] The joint criterion and control unit, under the adaptive window constraints given by the physical scene adaptive window estimation module, performs event-level multimodal joint determination on events purified by the event sparse state fusion unit: only when pulses from at least two different modes are within the Δt window and phase window... When internal coordination occurs, and its spectral kurtosis is not lower than the threshold κ, and the fusion score σ meets the threshold condition, it is determined to be a partial discharge event; the fusion score σ is composed of the consistency term Δt, The repetition rate, amplitude normalization, and morphological similarity are obtained by linear weighting α, β, γ, and δ, where θ is 5°–15°, κ is 2.5–4.0, and the threshold σ is... thr The range is 0.5 to 0.9, and α, β, γ, δ ∈ (0,1) and their sum is 1;

[0015] The joint criterion and control unit uses the judgment results to output phase-resolved partial discharge maps and / or phase-resolved pulse sequence maps and graded alarms, and uses the frequency band weights of the physical scene adaptive window estimation module for spectral kurtosis calculation and adaptive setting of the frequency band grid of the phase-resolved pulse sequence map.

[0016] The physical scene adaptive window estimation module provides typical Δt window ranges for different scenarios, including: 0.3–5 μs for UHF near-field antenna surface voltage probe, 0.5–8 μs for UHF near-field antenna high-frequency current transformer, and 0.2–6 μs for surface voltage probe high-frequency current transformer. When the field configuration changes, the delay for completing the sub-window inference does not exceed 20 ms.

[0017] The event-driven sparse state fusion unit performs sparse computation during the inference phase based on event triggers. The hidden state dimension is 64–128, the event time window is 2–10 ms, the edge processing latency does not exceed 5 ms, and the sparse computation duty cycle is no higher than 15%. Its clean... mask Used to eliminate spurious pulses, evidence score Weights used for the repetition rate and morphological similarity in a given fusion score σ.

[0018] The phase reference and cross-channel alignment unit ensures that the phase resolution of each mode is no less than 256 bins and that the equivalent phase error of different modes at 50Hz is no greater than 1°.

[0019] The alarms of the joint criterion and control unit are hierarchical alarms with hysteresis and debouncing logic, and the alarm response time is no more than 200ms.

[0020] The bandwidth and sampling specifications of the multimodal sensor interface and acquisition unit are as follows: surface voltage probe 3~100MHz, ultra-high frequency near-field antenna 300MHz~2.0GHz, high frequency current transformer 30kHz~20MHz, and air-coupled ultrasonic probe 20~120kHz; the sampling rate of the ultra-high frequency near-field antenna / surface voltage probe channel is not less than 500MS / s, the high frequency current transformer channel is not less than 20MS / s, the air-coupled ultrasonic probe channel is not less than 500kS / s, and the analog-to-digital conversion bit width is not less than 12bit.

[0021] It further includes a coaxial composite ultrasonic focusing device, which consists of a coaxially arranged horn mouth and a parabolic reflector, with a diameter of Φ60~120mm, a focal length of 25~60mm, a half-power angle of no more than 12° at 40kHz and an acoustic gain of no less than 8dB relative to the bare probe, and is connected to the air-coupled ultrasonic probe through a quick-release mechanism.

[0022] It further includes a calibration and traceability unit, which contains a fast-edge pulse source with a rising edge of no more than 3ns and a detachable small UHF calibration cavity; during calibration, the UHF link sensitivity and group delay are measured and written into the probe identification memory, and the device automatically compensates for the Δt window, phase correction amount and alarm threshold accordingly.

[0023] The multimodal sensor interface and acquisition unit adopt a 50Ω SMA or BNC physical interface, and are equipped with a probe identification memory to record the probe serial number, calibration coefficient and the most recent calibration date; the device integrates and archives event waveforms, phase-resolved partial discharge patterns / phase-resolved pulse sequence patterns, alarm levels and time / location tags.

[0024] The device is portable, weighs no more than 2.8kg, has an IP54 or IP65 protection rating, a battery life of no less than 8 hours, and an operating temperature range of -20℃ to 55℃. The communication interface includes at least one of USB-C or Ethernet, and includes wireless communication for data export and device management.

[0025] The beneficial effects of this invention are:

[0026] 1. By leveraging a multi-modal sensor interface and acquisition unit, complementary acquisition of electrical, radio frequency, and acoustic signals is achieved, enhancing detectability and anti-interference capabilities under weak discharge and complex interference environments. It is suitable for live-line inspection of various types of power equipment.

[0027] 2. By using phase reference and cross-channel alignment units to unify the pulses of each mode to the same power frequency phase scale, the consistency and comparability of cross-modal PRPD / PRPS spectra are ensured, significantly reducing misjudgments and missed detections caused by channel asynchrony.

[0028] 3. A physical scene adaptive window estimation module is introduced, which explicitly maps device type and geometric / medium constraints to arrival time difference window and phase correction, and outputs frequency band weight vector to achieve physical adaptation of threshold and frequency band self-adaptation, thereby enhancing the generalization ability across devices and scenarios.

[0029] 4. An event-driven sparse state fusion unit is used to perform event-driven sparse updates on pulse events, generating event cleanup masks and evidence scores. This enables pseudo-pulse suppression and evidence accumulation under low power conditions, improving real-time performance and stability.

[0030] 5. Under adaptive window constraints, the joint criterion and control unit uses arrival time difference consistency, phase consistency, frequency domain spectral kurtosis and evidence score as parallel necessary conditions for event-level determination, which significantly reduces the false alarm rate and maintains a high recall rate.

[0031] 6. The coaxial composite ultrasonic focusing device improves the acoustic signal-to-noise ratio in the background of long-distance noise by enhancing directivity and suppressing sidelobe, thus expanding the effective coverage of ultrasonic detection.

[0032] 7. The calibration and traceability unit, in conjunction with a detachable UHF calibration cavity and a fast-edge signal source, completes on-site quantization and automatic compensation of link sensitivity and group delay. Combined with the probe identification memory, it enables probe-level parameter write-back and full traceability, ensuring consistency and reproducibility in long-term operation.

[0033] 8. The frequency band weight vector simultaneously drives the frequency band grid adaptation of spectral kurtosis calculation and phase-resolved pulse sequence map, making feature extraction and map presentation more sensitive to the target frequency band and improving the intuitiveness and accuracy of diagnosis and interpretation.

[0034] 9. Tiered alarms, combined with hysteresis and debouncing strategies, reduce alarm jitter under boundary conditions and improve the availability and safety margin of operation and maintenance decisions.

[0035] 10. The device structure and algorithm are integrated into a single design, taking into account portability, rapid deployment and data management in the field, reducing detection costs and maintenance complexity, and meeting the engineering requirements of power systems for uninterrupted and routine inspections. Attached Figure Description

[0036] Figure 1 This is a system architecture diagram of the present invention;

[0037] Figure 2 This is a flowchart of the signal processing of the present invention;

[0038] Figure 3 This is a flowchart of the fusion score calculation process of the present invention;

[0039] Figure 4 This is a comparison chart of the missed detection rates for different equipment types according to the present invention;

[0040] Figure 5 This is a comparison chart of the false negative rates of various configurations of the present invention;

[0041] Figure 6 This is a phase error comparison diagram of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] It should be noted that the directional concepts of "left", "right", "up", "down", "front", "back", "inner", and "outer" in the following scheme are all relative directions, and will not be listed one by one here.

[0044] Example 1: Multimodal Acquisition and Phase Alignment

[0045] like Figures 1 to 3 As shown, this embodiment relates to a multi-modal acquisition and phase alignment technology solution for a multifunctional power equipment partial discharge comprehensive detection device. This solution utilizes multiple sensor modules working in parallel to acquire pulse signals from different sensors. Through high-precision phase reference and cross-channel alignment technology, it ensures that all modal signals can be analyzed under a unified time domain and phase scale, thereby achieving more accurate partial discharge detection.

[0046] The device in this embodiment includes multiple sensor interfaces for parallel acquisition of at least two pulse signals. The acquired signal sources include a surface voltage probe, an ultra-high frequency near-field antenna, a high-frequency current transformer, and an air-coupled ultrasonic probe. These signals operate in different frequency bands to capture partial discharge phenomena in electrical equipment. Each sensor is independently designed to adapt to different frequency ranges: the surface voltage probe operates in the 3–100 MHz band, the ultra-high frequency near-field antenna in the 300 MHz–2.0 GHz band, the high-frequency current transformer in the 30 kHz–20 MHz band, and the ultrasonic probe in the 20–120 kHz band. To satisfy the Nyquist sampling theorem and ensure signal integrity, the sampling rate of each sensor is set according to its frequency band range: the sampling rate of the UHF near-field antenna channel is no less than 500 MS / s (to meet the 2.0 GHz bandwidth requirement), the sampling rate of the surface voltage probe channel is no less than 250 MS / s (to meet the 100 MHz bandwidth requirement), the sampling rate of the high-frequency current transformer channel is no less than 50 MS / s (to meet the 20 MHz bandwidth requirement), the sampling rate of the ultrasonic channel is no less than 500 kS / s (to meet the 120 kHz bandwidth requirement), and the analog-to-digital conversion bit width is no less than 12 bits to ensure dynamic range.

[0047] The phase reference and cross-channel alignment unit ensures the synchronous alignment of signals acquired by the multi-mode sensor. This unit includes a non-contact phase reference module and a time-to-digital converter module. The non-contact phase reference module picks up signals via electric field or magnetic flux and acquires power frequency phase information using a digital phase-locked loop. Signals from all modes are time-domain aligned through the collaborative operation of the time-to-digital converter and the processor, thus ensuring power frequency phase synchronization between different modes. To achieve higher accuracy, the resolution of cross-channel alignment is limited to no more than 100 ps, ​​corresponding to (3 × 10⁻⁶ ps). -2 The electromagnetic wave propagation distance of 1000 mm ensures sub-millimeter spatial positioning accuracy. The phase error of each channel is controlled within 1°, corresponding to a time error of approximately 55.6 μs at a 50 Hz power frequency, guaranteeing high signal precision and consistency.

[0048] After time-domain and phase alignment of the multimodal signals, the device performs further signal processing through a physical scene adaptive window estimation module. This module generates an arrival time difference window (Δt) based on the type, size, and dielectric constant parameters of the material of the electrical equipment under test. min ),(Δt max and phase correction amount Based on this information, the time and phase windows of the signal are adaptively adjusted to improve detection accuracy under different equipment or power system environments. Simultaneously, the physical scene adaptive window estimation module outputs a frequency band weight vector for spectrum plotting, ensuring comprehensive processing of signals from different frequency bands. This module's design employs separate representations of each modal channel and interacts only along the channel axes, thereby enhancing cross-scene generalization capabilities.

[0049] After the signal is processed by the physical scene adaptive window estimation module, it enters the event sparse state fusion unit. This unit performs event-driven sparse state updates on the multimodal signal. The event sparse state fusion unit outputs an event cleanup mask (clean) by updating the state of each event. mask and evidence scoring score This information is used for subsequent joint decision-making. During the training phase, the module employs a parallelizable alternative dynamics approximation to optimize state update efficiency, while maintaining an event-driven, low-power computational approach during inference. mask Used to eliminate spurious pulses, evidence score This is used to quantify the morphological consistency and repeatability of events, thereby improving the accuracy and reliability of detection.

[0050] Finally, the processed signal enters the joint criterion and control unit, which performs event-level multimodal joint determination based on the adaptive window provided by the physical scene adaptive window estimation module and the purification result generated by the event sparse state fusion unit. Only when pulse signals from at least two modes fall within the specified (Δt) window and phase window... A partial discharge event is determined only when internal synergy occurs and its spectral kurtosis (κ) is not lower than a set threshold. During the determination process, the fusion score (σ) is calculated using the following formula:

[0051] Among them: (α, β, γ, δ∈(0,1)) and (α+β+γ+δ=1);

[0052] For consistency terms, calculate (Δt) and The time and phase differences reflect the consistency of different modal signals; (f rep (f) represents the repetition rate term, indicating the frequency of signal occurrence and measuring its stability in continuous measurements; amp (f) is the amplitude normalization term, used to eliminate the influence of signal amplitude, ensuring that signals of different amplitudes are treated fairly during processing; morph () is the morphological similarity term, used to compare the waveform morphology of different events to ensure that only signals with similar morphology are identified as partial discharges;

[0053] Specifically, the consistency item is calculated as follows:

[0054] Where (σ t )and These are the standard deviation parameters for time and phase, respectively.

[0055] After the above comprehensive judgment, the device will output a phase-resolved partial discharge spectrum or a phase-resolved pulse sequence spectrum, and issue a graded alarm based on a set threshold. The alarm level is set according to the fusion score (σ): (0.5≤σ<0.7) is a low-level alarm, (0.7≤σ<0.9) is a medium-level alarm, and (σ≥0.9) is a high-level alarm. In addition, the frequency band weight vector is also used for spectral kurtosis calculation and adaptive setting of the frequency band grid of the phase-resolved pulse sequence spectrum, thereby ensuring that the discharge characteristics of different signal sources are accurately captured and distinguished.

[0056] Through the above process, this embodiment provides an efficient and high-precision partial discharge detection device that can provide flexible detection solutions across devices and scenarios with the support of multi-mode signals, and can maintain high accuracy and low false alarm rate even in strong interference environments.

[0057] Example 2: Physical Scene Adaptive Window Estimation and Event Sparse State Fusion

[0058] like Figures 1 to 3 As shown, this embodiment details the technical solution of a multifunctional power equipment partial discharge comprehensive detection device based on Embodiment 1, particularly the combined application of the physical scene adaptive window estimation module and the event sparse state fusion unit. In this embodiment, through precise capture and adaptive processing of multimodal signals, this device can achieve efficient and accurate partial discharge detection in complex power equipment environments.

[0059] The physical scene adaptive window estimation module calculates and generates an adaptive arrival time difference window (Δt) based on the geometric and dielectric parameters of the measured power equipment (including equipment type, typical size, thickness of metal or insulation material, and relative permittivity). min ),(Δt max and phase correction amount These parameters directly affect the accuracy of signal processing, ensuring accurate capture of partial discharge signals under different devices and operating environments. To cope with variations in devices and environments, the physical scene adaptive window estimation module automatically adjusts the frequency band weight vector according to the specific application scenario. These frequency band weights play a crucial role in spectral feature calculation and map rendering, ensuring thorough analysis of signals in different frequency bands.

[0060] The core algorithm of the physical scene adaptive window estimation module is based on the propagation characteristics of electromagnetic waves in different media. For a given device configuration, the time difference of arrival window is determined by the following formula:

[0061] Where: d is the propagation path length; v is the propagation speed of electromagnetic waves in the medium; c is the speed of light in vacuum (3×10⁻⁶). 8 (m / s); (ε) r () represents the relative permittivity;

[0062] The physical scene adaptive window estimation module employs a local multi-scale window modeling strategy, where signals of different modes are represented separately. This design avoids unnecessary interference between channels by interacting only on the channel axis, thus improving generalization ability in complex scenarios. Specifically, the module can effectively adjust the arrival time difference window between UHF near-field antenna surface voltage probes to 0.3–5 μs, the window for UHF near-field antenna high-frequency current transformers to 0.5–8 μs, and the window between surface voltage probes and high-frequency current transformers to 0.2–6 μs, allowing for fine-tuning to meet the needs of different devices. When field configurations change, the physical scene adaptive window estimation module can complete window inference within 20 ms, ensuring the device can quickly adapt to changes.

[0063] After the signal is processed by the physical scene adaptive window estimation module, it enters the event sparse state fusion unit. The main function of this unit is to perform event-driven sparse state updates on pulse event signals collected by multiple sensors (surface voltage probe, high-frequency current transformer, ultra-high frequency near-field antenna, and ultrasonic probe). The event sparse state fusion unit processes the multimodal signal by updating the state of each event, and simultaneously outputs an event cleanup mask. mask and evidence scoring score clean mask Used to filter out spurious pulses, while evidence score This quantifies the credibility and consistency of each event.

[0064] The state update mechanism of the event sparse state fusion unit is based on a sparse neurodynamics model:

[0065]

[0066] Where: h(t) is the hidden state vector; τ is the time constant; w i The weight of the i-th event; δ(tt) i ) represents the event i at time (t) i The impulse function of ).

[0067] During the training phase, the event-sparse state fusion unit employs a parallelizable alternative dynamics approximation for state updates to improve processing efficiency; during the inference phase, the module maintains an event-driven, low-power computation mode. In this way, the event-sparse state fusion unit not only achieves low-latency processing but also operates efficiently on large-scale data, adapting to various power equipment and field applications. The hidden state dimension used in the state update process is 64–128, the event time window is 2–10 ms, the edge processing latency is controlled to no more than 5 ms, and the duty cycle of sparse computation is no higher than 15%. These design parameters ensure the system's high efficiency and low power consumption in real-time detection.

[0068] The output of the event sparse state fusion unit is clean. mask and evidence score The signal is then further transmitted to the joint criterion and control unit for final event determination. During this process, the system converts pulse signals from different modes into signals with varying Δt and phase windows. The occurrence of synergy within the signal is analyzed. If the signal meets the set threshold conditions and the kurtosis (κ) of the frequency domain spectrum is not lower than the set threshold, the event is determined to be a partial discharge. The specific fusion score (σ) is calculated by the following formula:

[0069]

[0070] Among them: (α, β, γ, δ∈(0,1)) and (α+β+γ+δ=1);

[0071] This is a consistency term that measures the degree of matching between events in time and phase.

[0072] (f rep The repetition rate term reflects the stability and persistence of the event, and is calculated as (f rep =n rep / n total ), where (n rep ) represents the number of repeated events, (n) total (f) represents the total number of events; amp ) is the amplitude normalization term, calculated as (f amp =A / A max ), used to eliminate the influence of signal amplitude on the judgment; (f morph The morphological similarity term is used to calculate waveform similarity using cross-correlation coefficients;

[0073] Finally, based on the fused judgment results, the device outputs a phase-resolved partial discharge spectrum and / or a phase-resolved pulse sequence spectrum, triggering a tiered alarm. During the alarm process, the system sets different alarm levels according to the severity and reliability of the event. Specifically, no alarm is triggered when the fusion score is (σ<0.5), a low-level alarm is triggered when (0.5≤σ<0.7), a medium-level alarm is triggered when (0.7≤σ<0.9), and a high-level alarm is triggered when (σ≥0.9). Simultaneously, the system adaptively adjusts the frequency band grid of the phase-resolved pulse sequence spectrum according to the frequency band weight vector. This adjustment of the frequency band weights ensures that signals in different frequency bands receive appropriate attention and effectively avoids the influence of interference signals, thereby improving detection accuracy.

[0074] In summary, this embodiment provides an efficient and flexible partial discharge detection scheme by combining a physical scene adaptive window estimation module with an event sparse state fusion unit. This scheme maintains good detection performance under different power equipment and operating environments, and exhibits excellent cross-scene adaptability. In environments with strong interference, the system's low-power computing and fast response capabilities make it an ideal tool for partial discharge detection in power equipment.

[0075] Example 3: Focusing Unit and Calibration Unit

[0076] like Figures 1 to 3 As shown, this embodiment describes the design of the coaxial composite ultrasonic concentrator and the calibration and traceability unit in the multifunctional power equipment partial discharge integrated detection device, based on embodiments 1 and 2. The application of these technical solutions aims to improve the measurement accuracy and stability of the detection device, ensuring that it can provide high-quality partial discharge detection under different power equipment and operating conditions.

[0077] First, the coaxial composite ultrasonic concentrator in this embodiment employs a coaxial structure combining a horn-shaped aperture and a parabolic reflector. This structure features high gain and low sidelobe characteristics, making it particularly suitable for long-distance detection applications. At a frequency of 40kHz, the concentrator's half-power angle is no greater than 12°, and it has an acoustic gain of no less than 8dB relative to the bare probe. This allows the concentrator to effectively concentrate ultrasonic energy during detection, increasing signal strength and thus enhancing the detection capability of partial discharge signals. The design also incorporates a quick-release mechanism, allowing users to easily and quickly replace the ultrasonic probe and concentrator as needed, ensuring the flexibility and operability of the device.

[0078] The acoustic gain of a wave concentrater is calculated based on the following formula:

[0079] Where: G is the acoustic gain (dB); (A eff (λ) represents the effective aperture area; (λ) represents the wavelength of the sound wave.

[0080] For a 40kHz ultrasound, the wavelength in air is approximately:

[0081] The focuser employs a coaxial structure, which effectively reduces ultrasonic signal dispersion and provides a smaller half-power angle, ensuring precise focusing of the sound waves and minimizing signal loss. The focuser's aperture ranges from 60 to 120 mm (Φ), and its focal length from 25 to 60 mm, allowing for flexible selection of appropriate parameters based on specific testing requirements. This design enables more concentrated ultrasonic signals, reduces interference, and effectively enhances signal strength in long-distance measurements.

[0082] The relationship between the half-power angle and the aperture is determined by the following formula:

[0083] Where: (θ HPBW ) represents the half-power angle (degrees); D represents the focusing aperture;

[0084] For an 80mm aperture screen, the theoretical half-power angle at 40kHz is approximately:

[0085]

[0086] In actual design, the shape of the parabolic surface is optimized to ensure that the half-power angle is no greater than 12°.

[0087] Secondly, the calibration and traceability unit in this embodiment includes a high-performance UHF calibration cavity. The calibration process is achieved by measuring the UHF link sensitivity and group delay. The UHF calibration cavity ensures the accuracy of the signal transmission link. The detection system can use this device to accurately compensate for delays and signal strength during signal transmission, ensuring that subsequent signal processing will not suffer from reduced detection accuracy due to transmission errors.

[0088] The calibration unit is also equipped with a fast-edge pulse source with a rise time of no more than 3ns, used to provide a high-precision pulse signal during calibration. By measuring the response of the pulse signal, the system can accurately calculate the sensitivity and group delay of the link, and automatically compensate for the delay and intensity of the acquired signal based on these data.

[0089] The formula for calculating group delay is:

[0090] Where: (τ g (φ) represents the group delay; (ω) represents the phase response; and (ω) represents the angular frequency.

[0091] The calibration and traceability unit operates as follows: First, the system uses a fast-edge pulse source to emit a fast pulse to the probe. This pulse has a wide spectral range, covering DC to approximately 1 GHz (calculated based on a 3 ns rise time). BW ≈0.35, t r =0.35, 3ns≈117MHz), considering higher harmonics up to 1GHz). The propagation characteristics of the pulse signal, such as signal delay and attenuation, were then measured using an ultra-high frequency calibration cavity. The calibration cavity employs a standard TEM cell structure with known transmission characteristics; its transfer function is:

[0092] [H(f)=A(f)·e -jφ(f) ]; where A(f) is the amplitude frequency response and (φ(f)) is the phase frequency response.

[0093] Then, based on the measured group delay and sensitivity data, the calibration module automatically calculates and adjusts the acquired signal to ensure the accuracy of the measurement results. Through this process, the device can guarantee that the detection data maintains high accuracy even in the complex environment of power equipment.

[0094] This embodiment also includes a probe identification memory, which uses an EEPROM chip to record the probe's serial number, calibration coefficients, and most recent calibration date. The probe identification memory is designed to enable management and traceability of each probe. The stored calibration coefficients include:

[0095] Sensitivity coefficient: (S=V) out / E in (mV / (pC)); Frequency response correction factor: (K(f));

[0096] Temperature compensation coefficient: (α) T Group delay compensation value: (τ) comp );

[0097] By storing this information, the system can quickly calibrate and trace the probe on-site, and ensure that each test result can be traced back to the specific probe and calibration status.

[0098] During use, when the user replaces the probe, the system automatically reads the information from the probe identification memory, loads the calibration coefficient and the latest calibration date corresponding to that probe, and automatically performs compensation based on this data. The compensation algorithm is: [S corrected (f,T)=S measured (f)·K(f)·(1+α T ·(TT0))];

[0099] Among them: (S) corrected (S) represents the corrected signal; measured(T0) represents the measurement signal; T represents the current temperature; (T0) represents the calibration temperature (usually 20℃).

[0100] In this way, the system can respond in real time to the replacement of different probes, avoiding errors that may be caused by manual calibration, and improving the convenience of operation and the stability of detection.

[0101] By combining the coaxial composite ultrasonic focusing device with the calibration unit in this embodiment, the device can provide high-gain, low-sidelobe, accurate measurements. Furthermore, automatic calibration and traceability technologies ensure high-precision detection under various equipment and environmental conditions. These designs significantly improve the accuracy, flexibility, and stability of partial discharge detection, making it particularly suitable for real-time, reliable partial discharge monitoring in complex power equipment environments.

[0102] Example 4:

[0103] To verify the technological advancement and practicality of this invention, this embodiment systematically evaluated the performance contribution of each key module of the device through control group experiments and ablation experiments. The experimental design followed the principle of scientific control to ensure the reliability and repeatability of the data.

[0104] Comparison table of control group and experimental group configurations

[0105]

[0106]

[0107] Experiment 1: Comparison of False Detection Rates

[0108] Experimental conditions

[0109] Test environment: Actual operating environment of a 220kV substation

[0110] Testing equipment: gas-insulated switchgear, ring main units, cable terminals, switch cabinets, overhead lines

[0111] Test duration: 72 hours of continuous monitoring for each device.

[0112] Partial discharge power supply: A standard pulse generator injects a calibration pulse with known amplitude and frequency.

[0113] False alarm rate control: fixed at 5%

[0114] Experimental data

[0115] Equipment type Missed detection rate (%) in the control group False negative rate in the experimental group (%) The rate of false negatives decreased by (%) Relative improvement rate (%) Gas-insulated switchgear 12.5±0.8 9.0±0.6 3.5 28.0 Ring main unit 15.2±1.0 11.2±0.7 4.0 26.3 Cable termination 14.0±0.9 10.5±0.7 3.5 25.0 switch cabinet 13.0±0.8 9.5±0.6 3.5 26.9 Overhead lines 16.0±1.1 11.0±0.8 5.0 31.3

[0116] like Figure 4As shown, the false negative rate of the experimental group was significantly lower than that of the control group on various types of power equipment (p<0.01, two-tailed t-test). Under the same false alarm rate, the false negative rate was reduced by an average of 27.5%, especially in the strong interference environment of overhead lines, where the improvement was more significant (31.3%). This verifies the effectiveness of multimodal fusion and adaptive windowing technology.

[0117] Experiment 2: Multi-scenario Adaptability (Ablation Experiment)

[0118] Experimental Design

[0119] By gradually removing critical modules, the contribution of each module to system performance is evaluated:

[0120] Configuration A: Complete system (physical scene adaptive window estimation module + event sparse state fusion unit)

[0121] Configuration B: Remove the physical scene adaptive window estimation module, retain the event sparse state fusion unit. Configuration C: Remove the event sparse state fusion unit, retain the physical scene adaptive window estimation module. Configuration D: Remove both modules (baseline system).

[0122] Experimental data

[0123] Equipment type Configure the false negative rate (%) Configure C to detect false negative rate (%) Configuration B: False negative rate (%) Configure A for the false negative rate (%) Gas-insulated switchgear 18.5±1.2 13.0±0.9 17.5±1.1 9.0±0.6 Ring main unit 19.2±1.3 14.5±1.0 18.0±1.2 11.2±0.7 Cable termination 21.0±1.4 15.0±1.0 20.0±1.3 10.5±0.7 switch cabinet 17.8±1.2 12.0±0.8 16.5±1.1 9.5±0.6 Overhead lines 20.5±1.4 15.0±1.1 19.0±1.3 11.0±0.8

[0124] like Figure 5 As shown, the contribution rate of the physical scene adaptive window estimation module is:

[0125] Contribution rate of the event sparse state fusion unit:

[0126] Among them (R) i ) represents the false negative rate of configuration i.

[0127] The calculation results show that the physical scene adaptive window estimation module has an average contribution rate of 62.3%, and the event sparse state fusion unit has an average contribution rate of 37.7%. The two have a synergistic effect, and the performance improvement when used together exceeds the sum of their individual contributions.

[0128] Experiment 3: Phase Consistency Test

[0129] Experimental methods

[0130] The phase error of the system output was measured by injecting a pulse signal at a known phase using a standard pulse generator. The test was conducted at a 50Hz power frequency, with 1000 measurements taken at each phase point, and the mean and standard deviation were calculated.

[0131] Experimental data

[0132] Equipment type Phase error of the control group (°) Phase error of the experimental group (°) Phase accuracy improvement factor Gas-insulated switchgear 5.6±0.4 0.8±0.1 7.0 Ring main unit 6.2±0.5 1.0±0.1 6.2 Cable termination 5.0±0.3 0.7±0.1 7.1 switch cabinet 5.8±0.4 1.1±0.1 5.3 Overhead lines 6.5±0.5 1.3±0.2 5.0

[0133] Phase resolution calculation

[0134] like Figure 6 As shown, at a 50Hz power frequency, a 360° phase corresponds to a 20ms period. The time error corresponding to a 1° phase error in the experimental group is:

[0135] This is still a large margin compared to the 100ps resolution of the time-to-digital converter, with the main error stemming from the uncertainty of signal transmission and processing delays.

[0136] Experiment 4: Calibration and Traceability Verification

[0137] Experimental Design

[0138] Option 1: No calibration (using default parameters)

[0139] Option 2: Manual calibration (traditional method, calibrated every 24 hours)

[0140] Option 3: Automatic calibration and traceability (the method of this invention, real-time compensation)

[0141] Long-term stability test (running continuously for 30 days)

[0142] Equipment type Accuracy of Option 1 (%) Accuracy of Option 2 (%) Accuracy of Option 3 (%) Initial (Day 1) 82.0±2.0 88.5±1.5 91.5±1.0 Mid-term (Day 15) 75.5±3.0 86.0±2.0 90.8±1.1 Final stage (day 30) 68.0±4.0 84.0±2.5 90.0±1.2 Accuracy degradation rate (%) / day 0.47 0.15 0.05

[0143] Temperature effect test (-20℃ to 55℃)

[0144] Temperature (°C) Uncompensated error (%) Manual error compensation (%) Automatic error compensation (%) -20 12.5±1.5 5.2±0.8 2.1±0.3 0 8.3±1.0 3.5±0.5 1.5±0.2 20 0 (Reference point) 0 (Reference point) 0 (Reference point) 40 6.8±0.8 3.0±0.4 1.3±0.2 55 10.2±1.2 4.5±0.6 1.8±0.3

[0145] Verification of the effectiveness of the temperature compensation coefficient:

[0146] Statistical Analysis and Conclusions

[0147] Analysis of variance (ANOVA)

[0148] One-way ANOVA was performed on the data from each experimental group, and the results showed that:

[0149] Comparison of false negative rates: F(1,8)=45.3, p<0.001

[0150] Phase coherence: F(1,8)=126.7, p<0.001

[0151] Calibration result: F(2,12)=38.9, p<0.001

[0152] All key performance indicators showed significant differences between the experimental group and the control group.

[0153] Technical Summary

[0154] Improved detection accuracy: Through multimodal fusion and adaptive window technology, the false negative rate is reduced by an average of 27.5%, with more significant improvement in environments with strong interference.

[0155] Phase accuracy improvement: Cross-channel alignment technology reduces phase error from more than 5° to less than 1°, an improvement of 5-7 times.

[0156] Long-term stability: The automatic calibration and traceability functions enable the system to maintain more than 90% accuracy after 30 days of continuous operation, with an accuracy decline rate of only 0.05% / day.

[0157] Temperature adaptability: Within the range of -20℃ to 55℃, automatic temperature compensation keeps the error within 2.1%, which is significantly better than traditional methods.

[0158] Module synergy: The combined use of the physical scene adaptive window estimation module and the event sparse state fusion unit produced a performance improvement that exceeded the sum of their individual contributions, proving the rationality of the system design.

[0159] The systematic verification through the above control group experiment and ablation experiment demonstrates the technological advancement of this invention in the field of partial discharge detection, providing a more reliable technical guarantee for the safe operation of power equipment.

[0160] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A multifunctional power equipment partial discharge comprehensive detection device, characterized in that, include: A multimodal sensor interface and acquisition unit for parallel acquisition of pulse signals from at least two of the following probes: surface voltage probe, high-frequency current transformer, ultra-high frequency near-field antenna, and air-coupled ultrasonic probe; The phase reference and cross-channel alignment unit includes a non-contact phase reference module (which picks up the power frequency phase through an electric field or magnetic flux and obtains it through a digital phase-locked loop) and a hardware alignment module that works in conjunction with a time-to-digital converter and a processor to map different mode pulses to a unified power frequency phase scale, with a cross-channel alignment resolution of no more than 100ps. The physical scene adaptive window estimation module outputs the arrival time difference window Δt between different modes based on the type of the measured object and its geometric / medium parameters (including equipment category, typical size, thickness of metal or insulating material, and relative permittivity). min ,Δt max With phase correction It generates frequency band weight vectors for spectral feature calculation and spectrum drawing; the physical scene adaptive window estimation module adopts a local multi-scale window modeling strategy that separates the representation of each modal channel and interacts only on the channel axis to improve cross-scene generalization ability. The event sparse state fusion unit performs a fusion of arrival time (t) and amplitude of a multimodal pulse event sequence. Power frequency phase Frequency band indexing performs event-driven sparse state updates and outputs an event cleanup mask (clean). mask With evidence scoring score This unit uses a parallelizable alternative dynamics approximation for state updates during the training phase and maintains event-driven, low-power computation during the inference phase. The joint criterion and control unit, under the adaptive window constraints given by the physical scene adaptive window estimation module, performs event-level multimodal joint determination on events purified by the event sparse state fusion unit: only when pulses from at least two different modes are within the Δt window and phase window... When internal coordination occurs, and its frequency domain spectral kurtosis is not lower than the threshold κ, and the fusion score σ meets the threshold condition, it is determined to be a partial discharge event; the fusion score σ is composed of the consistency term Δt, The repetition rate, amplitude normalization, and morphological similarity are obtained by linear weighting α, β, γ, and δ, where θ is 5°–15°, κ is 2.5–4.0, and the threshold σ is... thr The range is 0.5 to 0.9, and α, β, γ, δ ∈ (0,1) and their sum is 1; The joint criterion and control unit uses the judgment result to output phase-resolved partial discharge maps and / or phase-resolved pulse sequence maps and graded alarms, and uses the frequency band weight of the physical scene adaptive window estimation module for spectral kurtosis calculation and adaptive setting of the frequency band grid of the phase-resolved pulse sequence map.

2. The apparatus according to claim 1, characterized in that, The physical scene adaptive window estimation module provides typical Δt window ranges for different scenarios, including: 0.3–5 μs for UHF near-field antenna surface voltage probe, 0.5–8 μs for UHF near-field antenna high-frequency current transformer, and 0.2–6 μs for surface voltage probe high-frequency current transformer; when the field configuration changes, the delay for completing the sub-window inference does not exceed 20 ms.

3. The apparatus according to claim 1, characterized in that, The event sparse state fusion unit described herein performs sparse computation based on event triggering during the inference phase. The hidden state dimension is 64–128, the event time window is 2–10 ms, the edge processing latency does not exceed 5 ms, and the sparse computation duty cycle is no higher than 15%. Its clean mask Used to eliminate spurious pulses, evidence score Weights used for the repetition rate and morphological similarity in a given fusion score σ.

4. The apparatus according to claim 1, characterized in that, The phase reference and cross-channel alignment unit ensures that the phase resolution of each mode is not less than 256 bins and that the equivalent phase error of different modes at 50Hz is not greater than 1°.

5. The apparatus according to claim 1, characterized in that, The alarm of the joint criterion and control unit is a hierarchical alarm with hysteresis and debouncing logic, and the alarm response time is no more than 200ms.

6. The apparatus according to claim 1, characterized in that, The bandwidth and sampling specifications of the multimodal sensor interface and acquisition unit are as follows: surface voltage probe 3-100MHz, ultra-high frequency near-field antenna 300MHz-2.0GHz, high frequency current transformer 30kHz-20MHz, and air-coupled ultrasonic probe 20-120kHz; the sampling rate of the ultra-high frequency near-field antenna / surface voltage probe channel is not less than 500MS / s, the sampling rate of the high frequency current transformer channel is not less than 20MS / s, the sampling rate of the air-coupled ultrasonic probe channel is not less than 500kS / s, and the analog-to-digital conversion bit width is not less than 12bit.

7. The apparatus according to claim 1, characterized in that, It further includes a coaxial composite ultrasonic focusing device, which consists of a coaxially arranged horn mouth and a parabolic reflective surface, with a diameter of Φ60~120mm, a focal length of 25~60mm, a half-power angle of no more than 12° at 40kHz and an acoustic gain of no less than 8dB relative to the bare probe, and is connected to the air-coupled ultrasonic probe through a quick-release mechanism.

8. The apparatus according to claim 1, characterized in that, It further includes a calibration and traceability unit, which contains a fast-edge pulse source with a rising edge of no more than 3ns and a detachable small UHF calibration cavity; during calibration, the UHF link sensitivity and group delay are measured and written into the probe identification memory, and the device automatically compensates for the Δt window, phase correction amount and alarm threshold accordingly.

9. The apparatus according to claim 1, characterized in that, The multimodal sensor interface and acquisition unit adopt a 50Ω SMA or BNC physical interface and are equipped with a probe identification memory to record the probe serial number, calibration coefficient and the most recent calibration date; the device integrates and archives event waveforms, phase-resolved partial discharge patterns / phase-resolved pulse sequence patterns, alarm levels and time / location tags.

10. The apparatus according to claim 1, characterized in that, The device is portable, weighs no more than 2.8kg, has an IP54 or IP65 protection rating, a battery life of no less than 8 hours, and an operating temperature range of -20℃ to 55℃. The communication interface includes at least one of USB-C or Ethernet, and includes wireless communication for data export and device management.

Images