A multimodal fusion SF6 micro-leakage detection system and method for GIS equipment

Abstract

The application relates to the technical field of power equipment state detection and fault diagnosis, and discloses a GIS device SF6 micro-leakage detection system and method based on multi-modal fusion, which comprises a physical excitation sensing module, a multi-dimensional data fusion module, an intelligent control servo module and a visual interactive early warning module. The physical excitation sensing module is used for performing photoacoustic spectroscopy concentration measurement, infrared thermal image and visible light image synchronous capture, sound emission signal array pickup and environment parameter sensing. The multi-dimensional data fusion module is used for image fusion and heterogeneous data space-time registration, and a unified description vector of leakage state is constructed. The intelligent control servo module is used for coordinating a gimbal scanning coverage and system self-checking calibration. The visual interactive early warning module is used for mapping leakage information to a three-dimensional model, rendering a leakage point position and concentration cloud picture and triggering a hierarchical alarm and the like. Through deep fusion of multi-modal physical responses, accurate concentration measurement, three-dimensional space positioning and diffusion form visualization of SF6 micro-leakage are realized, and the detection sensitivity and intelligent level are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring and fault diagnosis technology, and in particular to a multimodal fusion SF6 micro-leakage detection system and method for GIS equipment. Background Technology

[0002] Sulfur hexafluoride (SF6) gas is widely used as the core insulating medium in gas-insulated gas-insulated switchgear (GIS) due to its excellent insulation and arc-quenching properties. During long-term operation, GIS equipment may experience extremely slow micro-leakage of SF6 gas at its sealing points due to factors such as aging of seals, loosening of flanges, and casting defects. Currently, the power industry commonly uses infrared imaging leak detection, acoustic detection, and offline gas chromatography analysis to detect SF6 leaks. Infrared thermal imaging technology can visually display the leaking gas cloud by observing the absorption characteristics of the gas to specific wavelengths of infrared radiation, making it suitable for large-scale rapid screening; acoustic detection technology uses ultrasonic or audible signals generated by the leak to achieve non-contact location; offline chromatography analysis can provide high-precision quantitative concentration results. These methods have been applied to a certain extent in substation field testing, and corresponding testing procedures and equipment standards have been established.

[0003] The aforementioned and existing related technologies have the following drawbacks: In actual GIS equipment operating environments, single detection methods often have their own physical limitations: Infrared imaging is easily affected by background thermal radiation interference and meteorological conditions, and its sensitivity for identifying trace leaks is insufficient; Acoustic detection is difficult to effectively extract weak leak characteristics against a background of complex mechanical noise; Offline sampling analysis cannot achieve real-time online monitoring and has a long sampling cycle. In addition, existing detection devices mostly use discrete components, lacking the ability to deeply fuse and spatially register multi-source sensor information, making it difficult to simultaneously obtain accurate values ​​of leak concentration, three-dimensional spatial coordinates of the leak source, and the morphological boundaries of leak diffusion. Summary of the Invention

[0004] The technical problem to be solved by the present invention is that the existing technology has the disadvantages of single detection methods, such as infrared imaging being susceptible to interference and having insufficient sensitivity, acoustic detection being difficult to extract weak features under complex noise, and offline analysis being unable to monitor in real time. To this end, we propose a multimodal fusion SF6 micro-leakage detection system and method for GIS equipment.

[0005] To achieve the above objectives, this application adopts the following technical solution: a multimodal fusion SF6 micro-leakage detection system for GIS equipment, comprising:

[0006] Physical excitation sensing module: used to perform photoacoustic spectral measurement of sulfur hexafluoride gas concentration, synchronous capture of infrared thermal radiation and visible light texture of the leak area, pickup of leak acoustic emission signal array, and sensing of environmental temperature, humidity and pressure parameters.

[0007] Multidimensional data fusion module: used to receive concentration values, thermal image sequences, optical images and sound source spatial coordinates output by the physical excitation sensing module, perform infrared and visible light image fusion framework operations, spatiotemporal registration and correlation analysis of multi-source heterogeneous data, and construct a unified description vector of leakage status;

[0008] Intelligent control servo module: used to parse inspection task instructions and coordinate the working sequence of each sensor module, drive the two-degree-of-freedom gimbal to scan and cover the key sealed parts of the GIS equipment, and simultaneously perform light source self-inspection, array phase calibration and detector non-uniformity correction during the system initialization phase.

[0009] Visual interactive early warning module: used to load the 3D digital model of GIS equipment and uniformly describe the leakage status vector to the model surface, render the leakage point markers and concentration distribution cloud map, automatically trigger graded alarms and generate event log messages according to the leakage judgment criteria;

[0010] Basic support and protection module: It is used to provide multiple DC regulated power supplies required by each electronic unit and perform overcurrent protection control, establish a local timing database to store leakage full-cycle detection data and alarm records, and suppress the coupling effect of substation electromagnetic transient interference on weak signal links;

[0011] Integrated structural support module: used to integrate and fasten all hardware components contained in the physical excitation sensing module, intelligent control servo module and basic support guarantee module to form a mobile deployment device.

[0012] Preferably, the physical excitation sensing module includes:

[0013] Laser photoacoustic sensing module: used to generate a wavelength-tunable modulated laser beam and excite gas molecules in a non-resonant photoacoustic cell, receive weak electrical signals generated by photo-induced pressure waves, perform adaptive noise cancellation algorithm and multi-scale time-series spectral feature extraction operation, and output accurate measurement values ​​of sulfur hexafluoride gas concentration.

[0014] Infrared thermal imaging acquisition module: used to capture the difference in thermal radiation distribution between the sulfur hexafluoride leak gas cloud and the background environment within the detection field of view, generate a mid-wave infrared thermal image sequence containing traces of leak diffusion, and transmit the thermal image data to the infrared-visible fusion module;

[0015] Visible light image acquisition module: used to synchronously acquire high-resolution texture images in the visible light band that are completely consistent with the infrared thermal imaging field of view, providing clear geometric information of equipment flanges, valves and pipelines for the fusion framework;

[0016] Acoustic array localization module: used to synchronously collect broadband acoustic emission signals generated by leaking gas through multiple high-sensitivity microphones deployed in space, implement blind source separation and three-dimensional spatial spectrum estimation algorithms based on sparse constraints in complex acoustic environments, and complete the accurate calculation of the azimuth vector and three-dimensional spatial coordinates of weak leaking sound sources.

[0017] Environmental parameter compensation module: used to collect temperature, humidity and atmospheric pressure values ​​at the detection site, and input the environmental parameters into the concentration inversion correction model of the laser photoacoustic sensing module to eliminate the influence of environmental fluctuations on the linearity of the photoacoustic signal response.

[0018] Preferably, the multidimensional data fusion module includes:

[0019] Infrared-visible fusion module: used to receive infrared thermal image and visible light image data streams, perform image fusion framework operations based on multi-scale transformation and salient feature mapping, enhance the contrast of the edge contour and the clarity of the internal texture of the leaking gas cloud, and generate a composite visual image with both thermal features and geometric details.

[0020] Multimodal fusion processing module: Used to receive concentration data from laser photoacoustic sensing module, pixel feature map from infrared visible fusion module and spatial coordinate information from acoustic array positioning module, perform spatiotemporal registration of heterogeneous data and feature layer correlation analysis, and construct a unified description vector of leakage state including leakage intensity value, three-dimensional location of leakage point and diffusion morphology boundary.

[0021] Preferably, the intelligent control servo module includes:

[0022] Central control and scheduling module: used to parse the inspection task instructions from the host computer, coordinate the modulation parameter settings of the laser photoacoustic sensing module, the acquisition triggering of the infrared visible fusion module, the sampling clock synchronization of the acoustic array positioning module, and the generation of motion control instructions for the mechanical scanning module;

[0023] Mechanical scanning servo module: used to drive a two-degree-of-freedom precision gimbal equipped with a laser photoacoustic probe and imaging components to perform closed-loop position control of horizontal azimuth and pitch angle, and achieve full coverage scanning of the flange sealing surface and meter connection of GIS equipment;

[0024] Self-test and calibration module: used to perform light source power stability monitoring, microphone array phase consistency verification and infrared detector non-uniformity correction operations during the system initialization phase, and call the built-in standard concentration gas chamber to perform periodic response calibration of the laser photoacoustic sensing module.

[0025] Preferably, the visual interactive early warning module includes:

[0026] 3D visualization module: used to load the 3D digital model of the detected GIS equipment, map the leakage state description vector output by the multimodal fusion processing module to the node coordinates on the model surface, and render and display the leakage source location markers, concentration distribution pseudo-color cloud map and diffusion impact spatial range;

[0027] Online identification and early warning module: used to perform dynamic threshold comparison analysis on the concentration change slope, cumulative sound source energy value and abnormal area of ​​the fused image in the unified description vector of the leakage status, and automatically trigger audible and visual alarms and generate event log messages according to the leakage severity classification criteria.

[0028] Preferably, the basic support and guarantee module includes:

[0029] Power management module: used to connect to mobile energy storage power supply or substation AC distribution box, convert input electrical energy into multiple DC regulated outputs required for laser driving constant current source, detector cooling voltage and embedded processor core voltage, and provide overcurrent protection and low power sleep control for each module.

[0030] Data storage and backtracking module: Used to establish a local storage area based on time-series database, record all unified description vectors of leakage status generated by the multimodal fusion processing module, 3D visualization view snapshots and alarm records output by the online identification and early warning module in each detection job, and support trend analysis and report generation of historical leakage events;

[0031] Electromagnetic compatibility protection module: used to apply shielding, grounding and filtering to each signal acquisition link and communication bus of the system, suppressing the coupling crosstalk between transient electromagnetic interference generated by high-voltage switch operation in the substation and weak photoacoustic signals and acoustic array signals, and ensuring the signal-to-noise ratio of the detection data.

[0032] Preferably, the laser photoacoustic sensing module includes:

[0033] Laser modulation drive module: used to generate a drive current signal with a wavelength located at the characteristic absorption peak of sulfur hexafluoride, and to implement sinusoidal frequency modulation and harmonic detection control for distributed feedback semiconductor laser;

[0034] Photoacoustic signal acquisition module: used to convert the weak pressure fluctuations generated by the stimulation of gas molecules in the non-resonant photoacoustic cell into analog voltage signals, and perform pre-amplification and bandpass filtering conditioning;

[0035] Phase-locked demodulation extraction module: Used to receive the modulation frequency reference signal and the signal under test output by the photoacoustic signal pickup module, and extract the second harmonic amplitude that is linearly related to the gas concentration through phase-sensitive detection technology;

[0036] Adaptive noise reduction module: Used to run the minimum mean square error algorithm to filter out random noise and power frequency harmonic interference introduced during phase-locked demodulation, and output clean second harmonic waveform data to the multi-scale feature extraction unit.

[0037] Preferably, the environmental parameter compensation module includes:

[0038] Temperature and humidity sensing and acquisition module: used to acquire the ambient temperature and relative humidity values ​​at the detection site through a digital temperature and humidity composite sensor;

[0039] Atmospheric pressure measurement module: used to acquire the ambient atmospheric pressure value at the detection site through a silicon piezoresistive absolute pressure sensor;

[0040] Concentration correction calculation module: This module substitutes temperature, humidity and atmospheric pressure values ​​into the empirical correction polynomial of photoacoustic signal intensity and environmental parameters to perform linear compensation calculations on the original concentration value output by the laser photoacoustic sensing module.

[0041] Preferably, the self-test and calibration module includes:

[0042] Light source status monitoring module: used to monitor the fluctuation range of laser output optical power through a built-in photodetector during the system startup phase;

[0043] Array phase verification module: used to inject standard sinusoidal test signals into each channel of the microphone array, calculate the phase consistency deviation between each channel and generate a correction coefficient table;

[0044] Infrared non-uniformity correction module: used to acquire images of uniform blackbody radiation sources and calculate the gain coefficient and bias coefficient of each pixel of the infrared detector to eliminate fixed pattern noise.

[0045] Calibration chamber switching module: Used to control the miniature solenoid valve to switch the gas path between external sampling state and built-in standard concentration sulfur hexafluoride gas chamber.

[0046] Preferably, the present invention provides another technical solution: a multimodal fusion SF6 micro-leakage detection method for GIS equipment, applied to a multimodal fusion SF6 micro-leakage detection system for GIS equipment, comprising the following implementation steps:

[0047] S1: The self-test and calibration module completes the system initialization calibration, and the central control and scheduling module receives the inspection task instruction and configures the working parameters of each sensor module.

[0048] S2: The mechanical scanning servo module drives the gimbal to scan and cover the target area, while the physical excitation sensor module simultaneously collects photoacoustic concentration data, infrared thermal image sequence, visible light image and acoustic emission signal.

[0049] S3: The infrared and visible fusion module and the multimodal fusion processing module perform fusion calculations on the collected data to construct a unified description vector of the leakage state;

[0050] S4: The acoustic array localization module performs blind source separation and spatial spectrum estimation, and calculates the three-dimensional spatial coordinates of the leaking sound source;

[0051] S5: The 3D visualization module renders cloud maps of leak locations and concentration distribution, while the online identification and early warning module performs threshold judgment and graded alarms based on a unified description vector of the leak status.

[0052] S6: The data storage and backtracking module saves the data for the entire testing cycle and integrates the structural support module's moving device to the next preset point.

[0053] The technical effects and advantages of this invention are as follows:

[0054] This invention effectively overcomes the limitations of traditional single detection methods, such as the susceptibility of infrared imaging to interference, the difficulty in extracting weak features through acoustic detection, and the inability to monitor in real time through offline analysis, by organically integrating multiple physical excitation sensing methods, including laser photoacoustic spectroscopy, infrared thermal imaging, visible light imaging, and acoustic array positioning, and combining multi-dimensional data deep fusion and intelligent control servo technology. The system utilizes photoacoustic spectroscopy to achieve ultra-high sensitivity and accurate measurement of SF6 concentration, and eliminates the influence of temperature, humidity, and pressure fluctuations after environmental parameter compensation, ensuring the quantitative accuracy of trace leaks. Simultaneously, the multi-scale fusion of infrared and visible light images significantly enhances the edge contour and texture clarity of the leaking gas cloud, while blind source separation and spatial spectral estimation... The acoustic positioning technology can accurately calculate the three-dimensional spatial coordinates of leakage sources in complex substation noise backgrounds. Based on this, the multi-modal fusion processing module constructs a unified description vector of the leakage state, including leakage intensity, spatial location, and diffusion pattern. It also uses a three-dimensional digital model to realize intuitive rendering of leakage point markings and concentration distribution cloud maps. Combined with a graded early warning mechanism, it automatically triggers audible and visual alarms and generates event log messages. The system also has complete self-testing calibration, electromagnetic compatibility protection, and data storage and traceability capabilities to ensure long-term stable operation in harsh electromagnetic environments. It can perform full-coverage scanning of key sealing parts of GIS equipment, significantly improving the sensitivity, positioning accuracy, and intelligence level of micro-leakage detection. Attached Figure Description

[0055] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0056] Figure 1 This is a system architecture diagram of a multimodal fusion GIS equipment SF6 micro-leakage detection system and method according to the present invention;

[0057] Figure 2 This is a system logic diagram of a multimodal fusion SF6 micro-leakage detection system and method for GIS equipment according to the present invention;

[0058] Figure 3 This is a diagram of the laser photoacoustic sensing module architecture of a multimodal fusion GIS equipment SF6 micro-leakage detection system and method according to the present invention.

[0059] Figure 4 This is an architecture diagram of the environmental parameter compensation module of a multimodal fusion GIS equipment SF6 micro-leakage detection system and method according to the present invention;

[0060] Figure 5 This is a diagram illustrating the self-testing and calibration module architecture of a multimodal fusion GIS equipment SF6 micro-leakage detection system and method according to the present invention.

[0061] Figure 6 This is an overall flowchart of the SF6 micro-leakage detection system and method for GIS equipment based on the present invention. Detailed Implementation

[0062] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0063] Reference Figures 1-5 As shown, the present invention provides a technical solution: a multimodal fusion SF6 micro-leakage detection system for GIS equipment, which mainly consists of the following six modular units: physical excitation sensing module, multidimensional data fusion module, intelligent control servo module, visual interactive early warning module, basic support and guarantee module, and integrated structural bearing module.

[0064] The physical excitation sensing module is the core sensing unit of this system, responsible for the synchronous acquisition of data from multiple physical fields. This module comprises five sub-modules: a laser photoacoustic sensing module, an infrared thermal imaging acquisition module, a visible light image acquisition module, an acoustic array positioning module, and an environmental parameter compensation module. Each sub-module is connected to the central control and scheduling module in the intelligent control servo module via a high-speed data bus, enabling unified configuration of operating parameters and the aggregation and transmission of acquired data.

[0065] The multi-dimensional data fusion module is responsible for deep fusion processing of the multi-source heterogeneous data output from the physical excitation sensing module. This module includes two sub-modules: an infrared-visible fusion module and a multi-modal fusion processing module. The infrared-visible fusion module receives thermal image data streams from the infrared thermal imaging acquisition module and visible light image data streams from the visible light image acquisition module, and performs image fusion framework operations based on multi-scale transformation and saliency feature mapping. The multi-modal fusion processing module further receives concentration data from the laser photoacoustic sensing module, pixel feature maps from the infrared-visible fusion module, and spatial coordinate information from the acoustic array positioning module, performs spatiotemporal registration of heterogeneous data and feature layer correlation analysis, and constructs a unified description vector of the leakage state.

[0066] The intelligent control servo module is responsible for the system's timing coordination and motion control. This module includes three sub-modules: a central control and scheduling module, a mechanical scanning servo module, and a self-test and calibration module. The central control and scheduling module, as the core control unit of the system, runs an embedded real-time operating system, parses the inspection task instructions issued by the host computer, and sends working parameter configuration and acquisition trigger commands to each sensor module according to the preset timing logic. The mechanical scanning servo module drives a two-degree-of-freedom precision gimbal equipped with a laser photoacoustic probe and imaging components to perform closed-loop position control of horizontal azimuth and pitch angle, achieving full coverage scanning of key sealed parts of the GIS equipment. The self-test and calibration module performs initialization calibration operations such as light source status monitoring, microphone array phase consistency verification, and infrared detector non-uniformity correction every time the system is powered on.

[0067] The visual interactive early warning module is responsible for the 3D visualization of detection results and the dissemination of early warning information. This module comprises two sub-modules: a 3D visualization display module and an online identification and early warning module. The 3D visualization display module loads the 3D digital model of the detected GIS equipment, maps the unified description vector of the leakage state output by the multimodal fusion processing module to the coordinates of node nodes on the model surface, and renders and displays the leakage source location markers, a pseudo-color cloud map of concentration distribution, and the spatial range of diffusion impact. The online identification and early warning module performs dynamic threshold comparison analysis on the slope of concentration change, the cumulative value of sound source energy, and the abnormal area of ​​the fused image in the unified description vector of the leakage state. Based on the leakage severity grading criteria, it automatically triggers audible and visual alarms and generates event log messages.

[0068] The basic support module is responsible for the system's power supply, data storage, and electromagnetic compatibility (EMC) protection. This module comprises three sub-modules: a power management module, a data storage and backtracking module, and an EMC protection module. The power management module connects to a mobile energy storage power source or a substation AC distribution box, converting the input power into a 28V DC regulated output required for the laser driver's constant current source, a 5V DC regulated output required for the infrared detector's cooling, and a 3.3V DC regulated output required for the embedded processor core. It also provides overcurrent protection and low-power sleep control for each module. The data storage and backtracking module establishes a local storage area based on a time-series database, using an SQLite embedded database management system. After each detection operation, it records a unified description vector of all leakage states, a 3D visualization snapshot, and alarm records, supporting trend analysis and report generation for historical leakage events. The EMC protection module applies shielding, grounding, and filtering to all signal acquisition links and communication buses in the system. It sets a low-pass filter at the analog signal front-end to suppress high-frequency electromagnetic interference and uses transient suppression diodes at the digital communication interface to protect against electrostatic discharge.

[0069] The integrated structural support module integrates and securely assembles all the hardware components contained in the five modules mentioned above. This module uses an aluminum alloy frame structure, with the circuit boards of each functional unit installed in a standard 19-inch 3U height enclosure, allowing for quick loading and unloading via guide rails. Sensor components such as laser photoacoustic probes, infrared thermal imagers, visible light cameras, and acoustic arrays are fixed to the pan-tilt mounting panel via standardized mechanical interfaces. The entire unit is equipped with casters and a telescopic lifting column, allowing for flexible deployment around the GIS equipment as needed for inspection.

[0070] The laser photoacoustic sensing module is the core sensing unit for achieving ultra-high sensitivity SF6 concentration detection. (Refer to...) Figure 1 and Figure 3 The specific implementation method is as follows:

[0071] The laser modulation drive module adopts a technology solution based on a distributed feedback DFB laser. The nominal center wavelength of the distributed feedback semiconductor laser is 10.6 μm, located at the strong characteristic absorption peak of SF6 gas. The laser modulation drive module integrates a digital signal synthesizer and a laser drive constant current source circuit, capable of generating a sinusoidal frequency modulation signal. The modulation frequency is set to 1.6 kHz with a duty cycle of 50%. The current modulation of the laser drive constant current source is set to 100 mA, superimposed on a 200 mA DC bias current. Under sinusoidal modulation, the output optical power of the laser changes periodically, exciting the stimulated absorption of SF6 gas molecules within the photoacoustic cell, generating periodic pressure waves proportional to the gas concentration.

[0072] The non-resonant photoacoustic cell serves as the medium for generating and propagating photoacoustic signals. Employing a non-resonant photoacoustic cell design, the cell is cylindrical with an internal volume of 50 mL. The inlet and outlet are connected to the system's gas path via PTFE tubing. ZnSe lens windows are installed at both ends of the cell, allowing the laser beam to pass through and excite the gas molecules within. A high-sensitivity electret condenser microphone is installed inside the cell to convert the weak pressure fluctuations generated by the excitation into an analog voltage signal.

[0073] The photoacoustic signal acquisition module consists of two parts: a preamplifier circuit and a bandpass filter circuit. The preamplifier circuit employs a low-noise instrumentation amplifier structure with an input impedance set to 10MΩ and an equivalent input noise voltage density of less than 5nV / √Hz. The amplifier's signal gain is set to 60dB, which can amplify the weak voltage signal output from the microphone to the input range of the analog-to-digital converter. The bandpass filter circuit uses a fourth-order Butterworth filter structure with a passband frequency range of 1.2kHz to 2.0kHz and a center frequency of 1.6kHz, used to filter out environmental noise and high-frequency interference components.

[0074] The phase-locked demodulation extraction module employs phase-sensitive detection technology to extract the second harmonic amplitude, which exhibits a linear relationship with the gas concentration. The module integrates a digital lock-in amplifier, using the modulation frequency reference signal output from the laser modulation drive module as the phase-locked reference to perform quadrature demodulation on the bandpass-filtered test signal. The demodulation algorithm uses a digital phase detection method, multiplying the in-phase and quadrature components of the test signal and the reference signal respectively, and then integrating the results through a low-pass filter to obtain the in-phase output component representing the second harmonic amplitude. This component exhibits a strictly linear proportional relationship with the SF6 gas concentration in the photoacoustic cell and can be directly converted to a concentration value using calibration coefficients.

[0075] The adaptive noise reduction module is used to filter out random noise and power frequency harmonic interference introduced during phase-locked demodulation by running the Least Mean Square Error (LMS) algorithm. The module integrates an adaptive digital filter with an order of 256, and the adaptive update algorithm uses the normalized LMS method. The reference noise signal is acquired through an independently configured background noise acquisition channel. The filter adjusts its weight coefficients in real time based on the statistical correlation between the reference signal and the signal to be processed. After iterative convergence, it outputs clean second harmonic waveform data to the multi-scale feature extraction unit.

[0076] The infrared thermal imaging acquisition module uses a mid-wave infrared thermal imager to capture the thermal radiation of the leaking gas cloud. The imager's detector is an MCT-type photovoltaic detector with a spectral response range of 3μm–5μm and a thermal sensitivity (NETD) better than 25mK. The imager is equipped with a dedicated germanium lens group, a field of view of 15° × 12°, and a spatial resolution better than 0.5mrad. The imager operates at a frame rate of 30fps and can generate a sequence of mid-wave infrared thermal images containing traces of leak diffusion. Infrared image data is transmitted to the multi-dimensional data fusion module via a GigEVision gigabit Ethernet interface.

[0077] The visible light image acquisition module employs a high-resolution visible light camera to capture the geometric texture information of the detection scene. The image sensor of the visible light camera is a CMOS global shutter sensor with an effective pixel count of 2448×2048 and a pixel size of 3.45μm×3.45μm. The lens uses a fixed-focus structure with a focal length set to 25mm, and its field of view is precisely matched with that of the infrared thermal imager, ensuring complete consistency between the visible light image and the infrared thermal image's field of view. Visible light image data is transmitted to the multi-dimensional data fusion module via the CoaXPress high-speed interface.

[0078] The acoustic array localization module is a key sensing unit for realizing the three-dimensional spatial localization of leakage sound sources, referring to... Figure 1 The specific implementation method is as follows:

[0079] The microphone array deployment module is used to spatially arrange multiple MEMS microphones in a three-dimensional spiral configuration to form an acoustic sampling aperture with spatial directional gain. In this embodiment, the microphone array adopts an 8-channel three-dimensional spiral configuration design, with eight MEMS microphones installed at the eight vertices of a cube, and the spacing between adjacent microphones set to 15cm. This three-dimensional spiral configuration design enables the array to have uniform spatial sampling characteristics in three-dimensional space, facilitating subsequent three-dimensional spatial spectrum estimation calculations. The microphone sensitivity is -26dBFS (1Pa@1kHz), the frequency response range is 20Hz~20kHz, and the noise floor is 26dBA.

[0080] The multi-channel synchronous acquisition module provides strictly in-phase sampling clock signals to each element in the microphone array deployment module, enabling parallel, time-biased acquisition of multi-channel sound pressure waveform data. The module integrates an 8-channel synchronous sampling ADC, with each channel set to a sampling rate of 48kHz and a sampling bit depth of 24 bits. A high-performance field-programmable gate array (FPGA) chip is integrated within the module to control the phase synchronization of the sampling clock for each channel, ensuring that the sampling time deviation of each channel is better than 1μs. The acquired data is transmitted via a high-speed data bus to the subsequent blind source separation and filtering module for processing.

[0081] The blind source separation and filtering module is used to run an iterative optimization algorithm based on independent component analysis (ICA) to separate the background mechanical vibration and corona discharge interference components from the multi-channel mixed acoustic signal. In the substation operating environment, the signal received by the acoustic array is typically a mixture of leakage sound and background noise. The background noise mainly includes mechanical vibration noise from transformer operation and broadband noise generated by corona discharge from high-voltage equipment. The blind source separation and filtering module employs an independent component analysis method based on the FastICA algorithm, gradually achieving source signal separation through iterative optimization. The module first pre-filters the 8-channel acquired data to remove DC offset and power frequency interference components; then, it constructs a joint probability density function model of the observed signals, maximizing the non-Gaussianity index to estimate the source signal separation; the separated independent components are then used to identify the leakage sound signal component through posterior power spectrum analysis.

[0082] The direction-of-arrival (DOA) estimation module receives the clean leakage acoustic signal after blind source separation and applies the MUSIC spatial spectrum estimation algorithm to calculate the elevation and azimuth parameters of the leakage source relative to the array center. The basic principle of the MUSIC algorithm is to construct signal and noise subspaces by calculating the eigenvalue decomposition of the array covariance matrix, and then construct a spatial spectrum function using the orthogonality of the two subspaces. The DOA estimation module first calculates the covariance matrix of the 8-channel leakage acoustic signal and performs eigenvalue decomposition; then, it distinguishes between the signal and noise subspaces based on the magnitude of the eigenvalues; next, it constructs the MUSIC spatial spectrum function, performing peak searches within the azimuth search range of 0°–360° and the elevation search range of 0°–90°; finally, it calculates the three-dimensional spatial coordinates of the leakage source using a peak refinement search algorithm.

[0083] The environmental parameter compensation module is used to collect environmental parameters at the testing site and correct and compensate for the concentration detection results, referring to... Figure 1 and Figure 4 The specific implementation method is as follows:

[0084] The temperature and humidity sensing module uses a digital temperature and humidity composite sensor to acquire the ambient temperature and relative humidity values ​​at the detection site. The sensor employs a MEMS capacitive sensing element, with a temperature measurement range of -40℃ to +85℃ and an accuracy of ±0.3℃, and a humidity measurement range of 0%RH to 100%RH and an accuracy of ±2%RH. The sensor outputs measurement data via an I²C digital interface, with a sampling rate set to 1Hz.

[0085] The atmospheric pressure measurement module uses a silicon piezoresistive absolute pressure sensor to acquire the ambient atmospheric pressure values ​​at the testing site. The sensor employs a MEMS piezoresistive sensing element, with a measurement range of 50 kPa to 150 kPa and a measurement accuracy of ±0.1 kPa. The sensor also outputs measurement data via an I²C digital interface.

[0086] The concentration correction calculation module is used to substitute temperature, humidity, and atmospheric pressure values ​​into the empirical correction polynomial of photoacoustic signal intensity and environmental parameters to perform linear compensation calculations on the original concentration value output by the laser photoacoustic sensing module. The mathematical expression of the empirical correction polynomial is:

[0087]

[0088] Among them, C corrected For the corrected concentration value, C raw P represents the raw concentration value output by the laser photoacoustic sensing module. meas P represents the measured atmospheric pressure value. ref For reference atmospheric pressure (101.325 kPa), T meas The measured temperature value is T. ref For reference temperature (25℃), H meas H represents the measured relative humidity value. ref For reference relative humidity (50%RH), α H The humidity correction factor is set to 0.001 / %RH. The corrected concentration value output by the concentration correction calculation module is transmitted to the multidimensional data fusion module as the final detection result.

[0089] The infrared-visible fusion module is responsible for performing deep fusion processing of infrared thermal images and visible light images, referring to... Figure 1 The specific implementation method is as follows:

[0090] The infrared-visible fusion module first preprocesses the input infrared thermal image and visible light image. Infrared thermal image preprocessing includes three steps: non-uniformity correction, temperature calibration, and pseudo-color mapping. Non-uniformity correction uses correction coefficients generated by the infrared non-uniformity correction module in the self-test and calibration module to correct the response consistency of each pixel. Temperature calibration converts the grayscale values ​​of the corrected infrared image into temperature values ​​corresponding to the scene. Pseudo-color mapping maps the temperature values ​​into color codes for subsequent synthesis with the visible light image to create a visual composite image. Visible light image preprocessing includes three steps: distortion correction, brightness equalization, and contrast enhancement. Distortion correction uses distortion correction coefficients generated by camera calibration to perform geometric correction on the image. Brightness and contrast equalization uses a histogram specification method to match the brightness distribution characteristics of the visible light image and the infrared thermal image.

[0091] After preprocessing, the infrared-visible fusion module performs image fusion framework operations based on multi-scale transformation. This embodiment employs an image fusion method based on stationary wavelet transform (SWT), with the following specific steps: First, stationary wavelet transforms are performed on both the infrared thermal image and the visible light image, decomposing them into low-frequency approximate components and high-frequency detail components. Then, a weighted average method is used to fuse the low-frequency components of each decomposition layer, with the weight coefficients adaptively determined based on the local information entropy of each image. A fusion strategy based on saliency feature mapping is used for the high-frequency components of each decomposition layer, selecting the component with the largest absolute value of its coefficient as the fusion result. Finally, an inverse stationary wavelet transform is performed on the fused wavelet coefficients to generate a composite visual image that combines thermal features and geometric details. In this composite image, the edge contour contrast of the leaking gas cloud is significantly enhanced compared to using the infrared thermal image alone, and the clarity of the internal texture is also significantly improved.

[0092] The multimodal fusion processing module is responsible for receiving concentration data from the laser photoacoustic sensing module, pixel feature maps from the infrared visible fusion module, and spatial coordinate information from the acoustic array positioning module. It performs spatiotemporal registration of heterogeneous data and feature layer correlation analysis to construct a unified description vector of the leakage state, which includes leakage intensity values, three-dimensional location of leakage points, and diffusion morphology boundaries.

[0093] Spatiotemporal registration is a crucial step in multimodal data fusion. Because the spatial sampling positions and temporal sampling times differ among the four sensing methods—laser photoacoustic sensing, infrared imaging, visible light imaging, and acoustic positioning—spatiotemporal registration is necessary for data correlation analysis. This embodiment employs a timestamp-based soft synchronization method: each sensing module labels its data with a precise timestamp (microsecond accuracy) while acquiring data. The multimodal fusion processing module aligns the time axes of each modality's data based on these timestamps, with a time window width set to 100ms. For spatial registration, the multimodal fusion processing module receives position feedback information from the mechanical scanning servo module, calculates the instantaneous attitude of each sensor, and maps the image coordinates to the geodetic coordinate system based on the camera's intrinsic and extrinsic parameter models, thus achieving spatial position registration and correlation.

[0094] Feature-layer correlation analysis, based on spatiotemporal registration, extracts and correlates leakage features from various modal data. Specifically, the multimodal fusion processing module extracts concentration peaks and gradient change features from laser photoacoustic concentration data; extracts pixel area and edge contour features of the leakage region from the infrared-visible fused image; and extracts the azimuth vector and energy features of the leakage sound source from the acoustic localization data. These features are fused and judged using a Bayesian inference framework to output a unified leakage state description vector. This unified leakage state description vector includes the following components: leakage intensity (calculated based on concentration data, unit μL / L), three-dimensional coordinates of the leakage point (X, Y, Z, calculated based on acoustic localization data, unit m), leakage diffusion area (calculated based on the fused image, unit m²), and leakage timestamp (based on soft synchronization timestamp, unit ms). This unified leakage state description vector, as the core output of the system, is transmitted to the visualization and interactive early warning module for further processing.

[0095] The central control and scheduling module, as the core control unit of the system, runs an embedded real-time operating system and is responsible for parsing the host computer's inspection task instructions and coordinating the working timing of each sensor module. The hardware of the central control and scheduling module uses an industrial-grade ARM processor, runs a Linux real-time operating system, and has a built-in multi-threaded task scheduler.

[0096] The core functions of the central control and dispatch module include:

[0097] (1) Inspection task parsing: Parse the inspection task instructions sent by the host computer via Ethernet. The task instructions include information such as the target GIS equipment number, the list of inspection areas, and the configuration of inspection parameters;

[0098] (2) Parameter configuration distribution: The detection parameter configuration is distributed to each sensing module, including the modulation frequency and modulation depth of the laser photoacoustic sensing module, the acquisition frame rate and integration time of the infrared thermal imaging acquisition module, and the sampling rate and acquisition duration of the acoustic array positioning module.

[0099] (3) Timing coordination control: Send acquisition trigger commands to each sensing module according to the preset timing logic to ensure synchronous acquisition of multimodal data;

[0100] (4) Data aggregation and feedback: Receive the collected data uploaded by each sensor module, perform preprocessing, and then forward it to the multidimensional data fusion module for fusion processing;

[0101] (5) Gimbal motion control: Send scanning motion control commands to the mechanical scanning servo module to control the gimbal to scan and cover the target area according to the preset trajectory.

[0102] The mechanical scanning servo module is responsible for driving the gimbal to perform scanning motion, as shown in the reference module. Figure 1 The specific implementation method is as follows:

[0103] The precision gimbal actuator module provides mechanical rotational freedom for the horizontal and pitch axes through a combination of stepper motors and harmonic reducers, supporting the laser photoacoustic probe and imaging components for attitude adjustment. The horizontal axis stepper motor has a step angle of 1.8°, with 16 microsteps and an angular resolution better than 0.0125°; the pitch axis uses a stepper motor of the same specifications. The harmonic reducer has a reduction ratio of 50:1 and a transmission efficiency higher than 90%. The gimbal has a maximum load capacity of 5kg, a horizontal rotation range of 0° to 360°, and a pitch rotation range of -30° to +60°.

[0104] The angle position feedback module reads the current azimuth and pitch angle values ​​of the gimbal in real time through an optical encoder, forming a position closed-loop negative feedback loop in the servo control circuit. The optical encoder is an incremental optical encoder with a pulse count of 2500 PPR, and its angular resolution is better than 0.0144° after quadruple frequency subdivision. The angle position feedback module reads the encoder values ​​at a sampling rate of 100Hz, compares them with the target angle sent by the central control and scheduling module, and generates a position error signal. The position error signal is processed by a PID control algorithm and then output to the stepper motor driver to achieve closed-loop position control.

[0105] The motion trajectory planning module calculates the gimbal's velocity curve and acceleration feedforward compensation during the start-stop phase based on the scanning area instructions issued by the central control and scheduling module. The module employs an S-shaped velocity curve planning algorithm, using a parabolic acceleration variation during acceleration and deceleration to avoid gimbal jitter caused by abrupt changes in the velocity curve. The acceleration feedforward compensation is calculated based on the rotational inertia of the gimbal load and is used to compensate for inertia errors during the start-stop phase, improving position control accuracy.

[0106] Reference Figure 1 and Figure 5 The self-test and calibration module is used to perform self-test and calibration operations during the system initialization phase to ensure the measurement accuracy and operational reliability of the system.

[0107] During system startup, the light source status monitoring module monitors the fluctuation range of the laser's output optical power using a built-in photodetector. The photodetector is installed in the laser's output optical path, with a sampling rate of 10kHz, to collect real-time output optical power data. If the optical power fluctuation exceeds a threshold (set to ±5%), the system automatically issues a laser performance malfunction alarm.

[0108] The array phase calibration module injects a standard sinusoidal test signal into each channel of the microphone array, calculates the phase consistency deviation between channels, and generates a correction coefficient table. The frequency of the standard sinusoidal test signal is set to 1kHz, and the amplitude is set to 94dBSPL. Test signals are injected into each channel sequentially, the response signals of each channel are acquired, and the phase deviation of each channel relative to the reference channel is calculated. The phase deviation calculation formula is as follows: ,in: φ represents the phase deviation of the i-th channel, where i represents the microphone channel number; i φ represents the measured phase of the i-th channel; ref This indicates the phase of the reference channel. Phase deviation data is stored in non-volatile memory in the form of a correction coefficient table. The subsequent blind source separation and filtering module calls this correction coefficient table for phase compensation during signal processing.

[0109] The infrared non-uniformity correction module acquires images of a uniform blackbody radiation source and calculates the gain coefficient and bias coefficient of each pixel in the infrared detector to eliminate fixed pattern noise. The temperature of the uniform blackbody radiation source is set to 40℃, and the radiation brightness is uniform. The module first acquires multiple frames of blackbody images, averages them, and uses the average as a reference image; then it calculates the gain coefficient (reference value / pixel value) and bias coefficient (pixel value - reference value) of each pixel. The specific calculation formula is as follows:

[0110] , where: G p,q This represents the gain coefficient, p represents the cell row index, and q represents the cell column index; O p,q V represents the bias coefficient; ideal Indicates the reference response value; V p,q This represents the measured response value of the (p,q)th pixel. During actual measurement, the corrected output value Y... p,q Calculate using the following formula: In the formula: Y p,q Indicates the corrected output value; X p,q This represents the original pixel response value. Finally, the gain coefficient and bias coefficient are stored in non-volatile memory, and the infrared thermal imaging acquisition module calls these coefficients for real-time correction during image acquisition.

[0111] The calibration chamber switching module controls a miniature solenoid valve to switch the gas path between external sampling mode and the built-in standard concentration SF6 chamber. The nominal SF6 gas concentration in the standard concentration chamber is 10 μL / L, with an accuracy of ±1%. The system periodically (recommended once a week) switches to the calibration chamber mode to calibrate the laser photoacoustic sensor module. The updated calibration coefficients are stored in non-volatile memory.

[0112] The 3D visualization module is responsible for presenting the detection results to users in a 3D visualization format. This module loads the 3D digital model of the detected GIS equipment, maps the unified description vector of the leakage state output by the multimodal fusion processing module to the coordinates of the nodes on the model surface, and renders and displays the leakage source location markers, the concentration distribution pseudo-color cloud map, and the spatial range of the diffusion impact.

[0113] The 3D digital model is constructed based on the engineering design drawings of GIS equipment and is modeled using 3D CAD software, with a model accuracy better than 1mm. The model file is stored in a common 3D format (such as OBJ or FBX format) and contains all geometric structural information of the equipment shell, flange sealing surfaces, valves, pipelines, etc. When the 3D visualization module runs, it first parses the 3D digital model file and establishes the rendering pipeline of the scene; then, according to the unified description vector of the leakage state, it finds the nearest node position on the model surface as the leakage point marker position; next, according to the unified description vector of the leakage state, it renders the coverage area of ​​the pseudo-color cloud map, and the higher the leakage intensity, the more red the cloud map color. Finally, it renders the spatial influence range of the leakage sound source, displayed as a hemispherical transparent volume.

[0114] The online identification and early warning module is responsible for real-time analysis of detection data and triggering early warnings. This module performs dynamic threshold comparison analysis on the concentration change slope, cumulative sound source energy value, and abnormal area in the fused image in the unified description vector of the leakage status. Based on the leakage severity classification criteria, it automatically triggers audible and visual alarms and generates event log messages.

[0115] The severity of leaks is classified into four levels:

[0116] Normal (green): The slope of concentration change is less than 0.1 μL / L / min, the cumulative value of sound source energy is less than 100, and the abnormal area of ​​the fused image is less than 0.01 m². The system displays a green status indicator and no alarm is required.

[0117] Note (yellow): If the slope of the concentration change is 0.1μL / L / min to 1μL / L / min, the cumulative value of the sound source energy is 100 to 1000, and the abnormal area of ​​the fused image is 0.01m² to 0.1m², a note-level alarm will be triggered, a yellow status indicator will be displayed, and a note-level event log message will be generated.

[0118] Warning (Orange): If the slope of the concentration change is 1μL / L / min to 10μL / L / min, the cumulative value of the sound source energy is 1000 to 10000, and the abnormal area of ​​the fused image is 0.1m² to 1m², a warning-level alarm will be triggered, an orange status indicator will be displayed, a warning-level event log message will be generated, and an intermittent audible and visual alarm will be activated.

[0119] Danger (Red): If the slope of the concentration change is greater than 10 μL / L / min, the cumulative value of the sound source energy is greater than 10000, or the abnormal area of ​​the fused image is greater than 1m², a danger level alarm will be triggered, a red status indicator will be displayed, a danger level event record message will be generated, and the audible and visual alarm will be activated to issue a continuous audible and visual alarm. An emergency alarm SMS will also be pushed to the host computer via the wireless communication module.

[0120] The event log message generated by the online identification and early warning module includes the following fields: alarm timestamp, GIS device number, 3D coordinates of the leak point, leak intensity, leak diffusion area, leak severity level, and handling recommendations. The event log message is simultaneously stored in the data storage and backtracking module and reported to the host computer management system via a communication interface.

[0121] The power management module is responsible for providing a stable and reliable DC power supply to all electronic units in the system, and for performing overcurrent protection and low-power management functions.

[0122] The power management module can be powered by either a mobile energy storage power supply or an AC distribution box in a substation. In AC power mode, the power management module first suppresses conducted interference from the grid side through an EMI filter, and then electrically isolates the input voltage from the system through an isolation transformer. The isolated AC power is then converted to DC power by a rectifier and filter circuit, and then converted to the different voltage levels required by each power-consuming unit by a multi-channel DC-DC switching power supply module.

[0123] Specifically, the power management module outputs the following multiple DC regulated power supplies:

[0124] +28V DC regulated output: This voltage powers the constant current source driving the laser and is also the power supply for the laser modulation drive module.

[0125] +5V DC regulated output: Powers the stepper motor driver for the infrared detector cooling circuit and the precision gimbal execution module;

[0126] +3.3V DC regulated output: Powers the ARM processor, FPGA chip, and various digital circuit modules of the central control and scheduling module;

[0127] ±5V dual-channel DC regulated output: powering analog signal conditioning circuits and instrumentation amplifiers;

[0128] +12V DC regulated output: Powers the microphone pre-amplifier circuit of the acoustic array positioning module.

[0129] The power management module also integrates overcurrent protection circuitry, implemented using an electronic current limiting circuit. The protection current threshold can be set according to the rated current value of each power supply. When the output current of a power supply exceeds the threshold, the electronic current limiting circuit quickly shuts off that output, protecting the downstream circuitry. Furthermore, the power management module supports low-power sleep mode control. When the system is in standby mode, it can shut down high-power units such as the laser power supply and cooling circuit power supply, maintaining power only for the central control scheduling module and communication interface. The sleep mode power consumption is less than 5W.

[0130] The data storage and backtracking module is responsible for establishing a local database storage area, recording various types of data throughout the entire detection cycle, and supporting the query, backtracking, and analysis report generation of historical data.

[0131] The data storage and backtracking module uses an embedded SQLite time-series database as its storage engine. The database file is stored in an industrial-grade solid-state drive (SSD) with a capacity of 128GB, which can store the raw data and processing results of no less than 10,000 complete test jobs.

[0132] The database table structure is designed as follows:

[0133] Inspection Task Table: Records task information for each inspection job, including task number, task creation time, GIS equipment number, list of inspection areas, operator number, etc.

[0134] Concentration data table: Records the concentration measurement data output by the laser photoacoustic sensing module, including timestamp, concentration value (μL / L), environmental parameters (temperature, humidity, air pressure), etc.

[0135] Infrared Image Table: Records thermal image data acquired by the infrared thermal imaging acquisition module, and stores the compressed image data in BLOB field format;

[0136] Visible light image table: records texture image data acquired by the visible light image acquisition module;

[0137] Acoustic localization table: Records the three-dimensional coordinate data of the leaking sound source output by the acoustic array localization module, including timestamp, X coordinate, Y coordinate, Z coordinate, confidence level, etc.

[0138] Fusion Result Table: Records the unified description vector data of leakage status generated by the multimodal fusion processing module;

[0139] Alarm Log: Records various alarm information triggered by the online identification and warning module, including alarm time, alarm level, alarm reason, trigger threshold, actual measurement value, etc.

[0140] System log table: Records various log information during system operation, including self-test log, calibration log, communication log, exception log, etc.

[0141] The data storage and traceability module supports trend analysis and report generation for historical leak events. Users can set a time range to query historical detection data, and the system automatically generates analysis reports such as concentration change trend curves and leak frequency statistics. Reports can be exported in PDF format for easy archiving and presentation.

[0142] The electromagnetic compatibility protection module is responsible for suppressing the interference of the complex electromagnetic environment in the substation on the weak signal acquisition link of the system, and ensuring the signal-to-noise ratio of the test data.

[0143] Substations contain strong sources of electromagnetic interference, primarily including transient electromagnetic pulses generated by high-voltage switch operation, power frequency magnetic field radiation during transformer operation, and broadband electromagnetic noise generated by corona discharge from high-voltage equipment. These interference signals may enter the system's signal acquisition link through spatial coupling or conductive coupling, causing severe interference to weak photoacoustic signals and acoustic array signals.

[0144] The electromagnetic compatibility protection module employs multi-layered comprehensive protection measures:

[0145] (1) Shielding and grounding design: The system casing is made of conductive aluminum alloy. Each circuit board is installed in a metal shielding box, which is reliably connected to the ground through a low-impedance grounding wire. The signal cable is a double-shielded cable, with the shielding layer grounded at one end on the system side to suppress coupling interference from external electromagnetic fields.

[0146] (2) Power line filtering: A power filter is installed at the input of the power management module to suppress electromagnetic interference conducted along the power line. The power filter adopts a π-type filter structure with a common-mode insertion loss greater than 60dB and a differential-mode insertion loss greater than 40dB.

[0147] (3) Signal link filtering: An active low-pass filter is set at the front end of the analog signal, with the cutoff frequency set to 1.5 times the bandwidth of the signal under test, effectively filtering out high-frequency electromagnetic interference and radio frequency interference. The acoustic signal acquisition channel is also equipped with a dedicated 50Hz notch filter to suppress power frequency and harmonic interference.

[0148] (4) Transient suppression protection: Transient suppression diodes (TVS) and surge protectors are installed at each communication interface and external connection terminal to prevent electrostatic discharge and surge voltage from damaging the system circuit.

[0149] (5) Reasonable wiring and partitioning: Digital signal lines and analog signal lines inside the system are arranged in partitions, and digital ground and analog ground are connected at a single point at the power input to avoid digital switch noise coupling to the analog signal link.

[0150] Through the above comprehensive protection measures, the electromagnetic compatibility protection module can improve the system's anti-disturbance capability to the EMCB level standard requirements, enabling it to work reliably in the complex electromagnetic environment of substation sites.

[0151] Reference Figure 1 , Figure 2 and Figure 6 This embodiment details a multimodal fusion method for detecting SF6 micro-leakage in GIS equipment, applied to a multimodal fusion SF6 micro-leakage detection system for GIS equipment, including the following specific implementation steps:

[0152] Step 1: System initialization and self-test phase;

[0153] The self-test and calibration module performs a complete initialization calibration operation after the system is powered on. The specific process is as follows:

[0154] (1) The central control and scheduling module performs a power-on self-test to check the operating status of key hardware such as processor memory, storage media, and communication interfaces;

[0155] (2) The light source status monitoring module is activated, collects the output optical power data of the laser, and evaluates the working stability of the laser. If the optical power fluctuation exceeds the threshold, the system issues a laser alarm and enters fault protection state;

[0156] (3) The array phase verification module injects standard sinusoidal test signals into each channel of the microphone array, calculates the phase consistency deviation between each channel, and generates and stores the correction coefficient table.

[0157] (4) The infrared non-uniformity correction module acquires images of a uniform blackbody radiation source, calculates the gain coefficient and bias coefficient of each pixel of the infrared detector, and generates and stores the non-uniformity correction parameters.

[0158] (5) The calibration chamber switching module controls the micro solenoid valve to switch to the calibration chamber state, and the laser photoacoustic sensing module performs response measurement on the standard concentration chamber and updates the calibration coefficient.

[0159] (6) The environmental parameter compensation module reads the current ambient temperature, humidity and atmospheric pressure values ​​as the reference parameters for subsequent concentration correction.

[0160] After the self-test and calibration module completes the system initialization calibration, the central control and scheduling module receives the inspection task instruction from the host computer. The inspection task instruction includes information such as the target GIS equipment number, the list of inspection areas, and the configuration of inspection parameters. After parsing the task instruction, the central control and scheduling module sends the corresponding working parameter configuration to each sensor module, including the modulation frequency and modulation depth of the laser photoacoustic sensor module, the acquisition frame rate and integration time of the infrared thermal imaging acquisition module, and the sampling rate and acquisition duration of the acoustic array positioning module.

[0161] Step Two: Scanning and Acquisition Phase;

[0162] The mechanical scanning servo module drives a two-degree-of-freedom gimbal to scan and cover the target detection area according to the inspection task instructions.

[0163] First, the motion trajectory planning module calculates the scanning path of the gimbal based on the detection area list. The detection area list contains several rectangular sub-regions, each defined by the geodetic coordinates of its four corner points. The motion trajectory planning module discretizes each sub-region into a sequence of grid points, calculates the horizontal azimuth and pitch angles of the gimbal as it passes through each grid point, and generates S-shaped velocity curve planning data.

[0164] Then, the precision gimbal execution module drives the gimbal to move according to the path data generated by the motion trajectory planning module. The angle position feedback module reads the position feedback signal from the photoelectric encoder in real time, forming a position closed-loop control loop to ensure that the gimbal accurately reaches the target position at each path point.

[0165] During the gimbal motion scanning process, each sub-module of the physical excitation sensing module synchronously performs data acquisition:

[0166] (1) The laser photoacoustic sensing module continuously samples and detects the SF6 gas concentration in the optical path at a sampling rate of 10Hz. The concentration data is then output to the multidimensional data fusion module after being corrected by the environmental parameter compensation module.

[0167] (2) The infrared thermal image acquisition module acquires a sequence of thermal images of the detection field at a frame rate of 30fps, and the thermal image data is transmitted to the multi-dimensional data fusion module in real time;

[0168] (3) The visible light image acquisition module synchronously acquires high-resolution visible light texture images that are consistent with the field of view of the infrared thermal image, and the acquisition frame rate is matched with the infrared thermal image to ensure the consistency of the field of view;

[0169] (4) The acoustic array positioning module synchronously acquires 8 channels of sound pressure waveform data at a sampling rate of 48kHz. The multi-channel synchronous acquisition module ensures that each channel is strictly in phase.

[0170] Each sensing module labels the data with a precise timestamp while collecting it, which facilitates subsequent spatiotemporal registration processing.

[0171] Step 3: Multimodal data fusion processing stage;

[0172] The multidimensional data fusion module performs fusion processing on the multi-source heterogeneous data collected by the physical excitation sensing module.

[0173] First, the infrared-visible fusion module performs image fusion framework operations based on multi-scale transformation on the infrared thermal image and the visible light image. The fusion algorithm uses the stationary wavelet transform method detailed in the previous embodiments to generate a composite visual image that combines thermal features and geometric details. The fused image enhances the contrast of the leaking gas cloud's edge contours and the clarity of its internal texture.

[0174] Then, the multimodal fusion processing module receives concentration data from the laser photoacoustic sensing module, pixel feature maps from the infrared-visible fusion module, and spatial coordinate information from the acoustic array positioning module, and performs spatiotemporal registration and feature layer correlation analysis of heterogeneous data. The multimodal fusion processing module aligns the time axis of each modal data according to the timestamp, performs spatial registration based on the gimbal position feedback information, and then performs feature layer fusion of each modal data using a Bayesian inference framework to construct a unified description vector of the leakage state, including leakage intensity, three-dimensional coordinates of the leakage point, leakage diffusion area, and timestamp.

[0175] Step 4: Sound source localization calculation stage;

[0176] The acoustic array positioning module processes the acquired multi-channel sound pressure waveform data and calculates the three-dimensional spatial coordinates of the leaking sound source.

[0177] First, the blind source separation and filtering module runs an iterative optimization algorithm based on Independent Component Analysis (ICA). After preprocessing the 8-channel acquired data, the module constructs a joint probability density function model of the observed signals and achieves source signal separation through iterative convergence using the FastICA algorithm. The separated independent components are then subjected to power spectrum analysis to identify the independent components that match the acoustic emission characteristics of SF6 gas leakage as the pure leakage acoustic signal.

[0178] Then, the direction-of-arrival (DOA) estimation module receives the clean leakage sound signal after blind source separation and applies the MUSIC spatial spectrum estimation algorithm to calculate the elevation and azimuth parameters of the leakage sound source relative to the array center. The module performs spectral peak search within the azimuth search range of 0° to 360° and the elevation search range of 0° to 90°, and calculates the azimuth vector of the leakage sound source after a fine search.

[0179] Finally, combining the gimbal position feedback information from the mechanical scanning servo module, the azimuth vector of the leakage sound source is converted into three-dimensional spatial coordinates (X, Y, Z) in the geodetic coordinate system. The positioning result is transmitted to the multi-dimensional data fusion module as an important component of the unified description vector of the leakage state.

[0180] Step 5: Visualization and Early Warning Stage;

[0181] The visual interactive early warning module provides a three-dimensional visual display and early warning judgment of the fusion processing results.

[0182] First, the 3D visualization module loads the 3D digital model of the detected GIS equipment. The module reads the 3D model file, establishes a rendering pipeline, and maps the 3D coordinates of the leak point in the unified description vector of the leak status to the coordinates of the nodes on the model surface. The module renders and displays the leak source location marker (represented by a red diamond icon), the concentration distribution pseudo-color cloud map (rendered with gradient colors according to the leak intensity, the higher the intensity, the more red the color), and the spatial range of the diffusion impact (displayed as a hemispherical transparent volume).

[0183] Simultaneously, real-time captured composite visual images are overlaid on the 3D view interface, achieving an augmented reality (AR) effect for visualizing leaks. Users can perform interactive operations such as zooming, rotating, and panning in the 3D view to observe the location and shape of the leak point from any angle.

[0184] Then, the online identification and early warning module performs dynamic threshold comparison analysis on the concentration change slope, cumulative sound source energy value, and abnormal area in the fused image within the unified description vector of the leakage status. The module compares the real-time calculated indicators with preset thresholds and determines the current leakage level based on the leakage severity grading criteria.

[0185] Based on the judgment results, the online identification and early warning module executes the corresponding alarm operations: for alert, warning, and danger level alarms, the module automatically triggers the audible and visual alarms (a single sound for alert level, an intermittent sound for warning level, and a continuous sound for danger level), and simultaneously generates an event log message for the corresponding level. The event log message includes fields such as alarm timestamp, GIS device number, three-dimensional coordinates of the leak point, leak intensity, leak diffusion area, leak severity level, and handling recommendations. The message data is stored in the data storage and backtracking module and reported to the host computer management system via the communication interface.

[0186] Step Six: Data Storage and Transition Stage;

[0187] After the detection operation is completed, the data storage backtracking module performs data storage operations.

[0188] The data storage and backtracking module writes all the data from this testing operation into a local SQLite database, including: testing task information, concentration data sequences, infrared and visible light image data, acoustic positioning data, fusion result data, alarm records, system logs, etc. The data is stored in a compressed format; image data is compressed using JPEG, and raw acquisition data is compressed using GZIP.

[0189] After the inspection is completed, the central control and scheduling module controls the mechanical scanning servo module to reset the gimbal to its initial position, and at the same time controls the power management module to enter a low-power standby state.

[0190] The integrated structural support module carries the entire system device to the next preset inspection point, and repeats the inspection process from step one to step six. Multiple preset points are inspected sequentially to achieve full coverage inspection of the key sealing parts of the target GIS equipment.

[0191] After all inspection points are inspected, the central control and dispatch module sends an inspection completion report to the host computer, summarizing the inspection results, discovered leaks, and triggered alarm records. Users can query inspection data for any time period through the historical data storage and backtracking module to generate trend analysis reports.

[0192] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A multimodal fusion SF6 micro-leakage detection system for GIS equipment, characterized in that, include: Physical excitation sensing module: used to perform photoacoustic spectral measurement of sulfur hexafluoride gas concentration, synchronous capture of infrared thermal radiation and visible light texture of the leak area, pickup of leak acoustic emission signal array, and sensing of environmental temperature, humidity and pressure parameters. Multidimensional data fusion module: used to receive concentration values, thermal image sequences, optical images and sound source spatial coordinates output by the physical excitation sensing module, perform infrared and visible light image fusion framework operations, spatiotemporal registration and correlation analysis of multi-source heterogeneous data, and construct a unified description vector of leakage status; Intelligent control servo module: used to parse inspection task instructions and coordinate the working sequence of each sensor module, drive the two-degree-of-freedom gimbal to scan and cover the key sealed parts of the GIS equipment, and simultaneously perform light source self-inspection, array phase calibration and detector non-uniformity correction during the system initialization phase. Visual interactive early warning module: used to load the 3D digital model of GIS equipment and uniformly describe the leakage status vector to the model surface, render the leakage point markers and concentration distribution cloud map, automatically trigger graded alarms and generate event log messages according to the leakage judgment criteria; Basic support and protection module: It is used to provide multiple DC regulated power supplies required by each electronic unit and perform overcurrent protection control, establish a local timing database to store leakage full-cycle detection data and alarm records, and suppress the coupling effect of substation electromagnetic transient interference on weak signal links; Integrated structural support module: used to integrate and fasten all hardware components contained in the physical excitation sensing module, intelligent control servo module and basic support guarantee module to form a mobile deployment device. The physical excitation sensing module includes: Laser photoacoustic sensing module: used to generate a wavelength-tunable modulated laser beam and excite gas molecules in a non-resonant photoacoustic cell, receive weak electrical signals generated by photo-induced pressure waves, perform adaptive noise cancellation algorithm and multi-scale time-series spectral feature extraction operation, and output accurate measurement values ​​of sulfur hexafluoride gas concentration. Infrared thermal imaging acquisition module: used to capture the difference in thermal radiation distribution between the sulfur hexafluoride leak gas cloud and the background environment within the detection field of view, generate a mid-wave infrared thermal image sequence containing traces of leak diffusion, and transmit the thermal image data to the infrared-visible fusion module; Visible light image acquisition module: used to synchronously acquire high-resolution texture images in the visible light band that are completely consistent with the infrared thermal imaging field of view, providing clear geometric information of equipment flanges, valves and pipelines for the fusion framework; Acoustic array localization module: used to synchronously collect broadband acoustic emission signals generated by leaking gas through multiple high-sensitivity microphones deployed in space, implement blind source separation and three-dimensional spatial spectrum estimation algorithms based on sparse constraints in complex acoustic environments, and complete the accurate calculation of the azimuth vector and three-dimensional spatial coordinates of weak leaking sound sources. Environmental parameter compensation module: used to collect temperature, humidity and atmospheric pressure values ​​at the detection site, and input the environmental parameters into the concentration inversion correction model of the laser photoacoustic sensing module to eliminate the influence of environmental fluctuations on the linearity of the photoacoustic signal response.

2. The SF6 micro-leakage detection system for GIS equipment based on multimodal fusion according to claim 1, characterized in that: The multidimensional data fusion module includes: Infrared-visible fusion module: used to receive infrared thermal image and visible light image data streams, perform image fusion framework operations based on multi-scale transformation and salient feature mapping, enhance the contrast of the edge contour and the clarity of the internal texture of the leaking gas cloud, and generate a composite visual image with both thermal features and geometric details. Multimodal fusion processing module: Used to receive concentration data from laser photoacoustic sensing module, pixel feature map from infrared visible fusion module and spatial coordinate information from acoustic array positioning module, perform spatiotemporal registration of heterogeneous data and feature layer correlation analysis, and construct a unified description vector of leakage state including leakage intensity value, three-dimensional location of leakage point and diffusion morphology boundary.

3. The SF6 micro-leakage detection system for GIS equipment based on multimodal fusion according to claim 2, characterized in that: The intelligent control servo module includes: Central control and scheduling module: used to parse the inspection task instructions from the host computer, coordinate the modulation parameter settings of the laser photoacoustic sensing module, the acquisition triggering of the infrared visible fusion module, the sampling clock synchronization of the acoustic array positioning module, and the generation of motion control instructions for the mechanical scanning module; Mechanical scanning servo module: used to drive a two-degree-of-freedom precision gimbal equipped with a laser photoacoustic probe and imaging components to perform closed-loop position control of horizontal azimuth and pitch angle, and achieve full coverage scanning of the flange sealing surface and meter connection of GIS equipment; Self-test and calibration module: used to perform light source power stability monitoring, microphone array phase consistency verification and infrared detector non-uniformity correction operations during the system initialization phase, and call the built-in standard concentration gas chamber to perform periodic response calibration of the laser photoacoustic sensing module.

4. The SF6 micro-leakage detection system for GIS equipment with multimodal fusion according to claim 3, characterized in that: The visual interactive early warning module includes: 3D visualization module: used to load the 3D digital model of the detected GIS equipment, map the leakage state description vector output by the multimodal fusion processing module to the node coordinates on the model surface, and render and display the leakage source location markers, concentration distribution pseudo-color cloud map and diffusion impact spatial range; Online identification and early warning module: used to perform dynamic threshold comparison analysis on the concentration change slope, cumulative sound source energy value and abnormal area of ​​the fused image in the unified description vector of the leakage status, and automatically trigger audible and visual alarms and generate event log messages according to the leakage severity classification criteria.

5. The SF6 micro-leakage detection system for GIS equipment based on multimodal fusion according to claim 4, characterized in that: The basic support and guarantee module includes: Power management module: used to connect to mobile energy storage power supply or substation AC distribution box, convert input electrical energy into multiple DC regulated outputs required for laser driving constant current source, detector cooling voltage and embedded processor core voltage, and provide overcurrent protection and low power sleep control for each module. Data storage and backtracking module: Used to establish a local storage area based on time-series database, record all unified description vectors of leakage status generated by the multimodal fusion processing module, 3D visualization view snapshots and alarm records output by the online identification and early warning module in each detection job, and support trend analysis and report generation of historical leakage events; Electromagnetic compatibility protection module: used to apply shielding, grounding and filtering to each signal acquisition link and communication bus of the system, suppressing the coupling crosstalk between transient electromagnetic interference generated by high-voltage switch operation in the substation and weak photoacoustic signals and acoustic array signals, and ensuring the signal-to-noise ratio of the detection data.

6. The SF6 micro-leakage detection system for GIS equipment with multimodal fusion according to claim 1, characterized in that: The laser photoacoustic sensing module includes: Laser modulation drive module: used to generate a drive current signal with a wavelength located at the characteristic absorption peak of sulfur hexafluoride, and to implement sinusoidal frequency modulation and harmonic detection control for distributed feedback semiconductor laser; Photoacoustic signal acquisition module: used to convert the weak pressure fluctuations generated by the stimulation of gas molecules in the non-resonant photoacoustic cell into analog voltage signals, and perform pre-amplification and bandpass filtering conditioning; Phase-locked demodulation extraction module: Used to receive the modulation frequency reference signal and the signal under test output by the photoacoustic signal pickup module, and extract the second harmonic amplitude that is linearly related to the gas concentration through phase-sensitive detection technology; Adaptive noise reduction module: Used to run the minimum mean square error algorithm to filter out random noise and power frequency harmonic interference introduced during phase-locked demodulation, and output clean second harmonic waveform data to the multi-scale feature extraction unit.

7. The SF6 micro-leakage detection system for GIS equipment with multimodal fusion according to claim 6, characterized in that: The environmental parameter compensation module includes: Temperature and humidity sensing and acquisition module: used to acquire the ambient temperature and relative humidity values ​​at the detection site through a digital temperature and humidity composite sensor; Atmospheric pressure measurement module: used to acquire the ambient atmospheric pressure value at the detection site through a silicon piezoresistive absolute pressure sensor; Concentration correction calculation module: This module substitutes temperature, humidity and atmospheric pressure values ​​into the empirical correction polynomial of photoacoustic signal intensity and environmental parameters to perform linear compensation calculations on the original concentration value output by the laser photoacoustic sensing module.

8. The SF6 micro-leakage detection system for GIS equipment with multimodal fusion according to claim 3, characterized in that: The self-testing and calibration module includes: Light source status monitoring module: used to monitor the fluctuation range of laser output optical power through a built-in photodetector during the system startup phase; Array phase verification module: used to inject standard sinusoidal test signals into each channel of the microphone array, calculate the phase consistency deviation between each channel and generate a correction coefficient table; Infrared non-uniformity correction module: used to acquire images of uniform blackbody radiation sources and calculate the gain coefficient and bias coefficient of each pixel of the infrared detector to eliminate fixed pattern noise. Calibration chamber switching module: Used to control the miniature solenoid valve to switch the gas path between external sampling state and built-in standard concentration sulfur hexafluoride gas chamber.

9. A multimodal fusion method for detecting SF6 micro-leakage in GIS equipment, applied to the multimodal fusion SF6 micro-leakage detection system for GIS equipment as described in claim 5, characterized in that: The implementation steps include the following: S1: The self-test and calibration module completes the system initialization calibration, and the central control and scheduling module receives the inspection task instruction and configures the working parameters of each sensor module. S2: The mechanical scanning servo module drives the gimbal to scan and cover the target area, while the physical excitation sensor module simultaneously collects photoacoustic concentration data, infrared thermal image sequence, visible light image and acoustic emission signal. S3: The infrared and visible fusion module and the multimodal fusion processing module perform fusion calculations on the collected data to construct a unified description vector of the leakage state; S4: The acoustic array localization module performs blind source separation and spatial spectrum estimation, and calculates the three-dimensional spatial coordinates of the leaking sound source; S5: The 3D visualization module renders cloud maps of leak locations and concentration distribution, while the online identification and early warning module performs threshold judgment and graded alarms based on a unified description vector of the leak status. S6: The data storage and backtracking module saves the data for the entire testing cycle and integrates the structural support module's moving device to the next preset point.

Images