An embedded AI edge computing system

Abstract

The application discloses an embedded AI edge computing system and relates to the field of gas leakage detection.The system comprises a dynamic sensing recognition unit, a situation awareness discrimination unit, a spatial missing coverage unit and a visual display storage unit.The dynamic sensing recognition unit is used for preprocessing infrared gas cloud image data of a patrol area, identifying sulfur hexafluoride cloud positioning results according to the preprocessing results and judging gas leakage amount and gas position.The situation awareness discrimination unit is used for simulating the diffusion process of sulfur hexafluoride in the patrol area based on the gas leakage amount and the gas position, discriminating the equipment control failure risk in the patrol area during the diffusion of sulfur hexafluoride.The spatial missing coverage unit is used for generating equipment control failure risk sorting results, combining the leakage position to establish an optimal missing coverage path and adjusting the leakage position of the patrol area.The visual display storage unit can construct the optimal missing coverage path, realize the purpose of ensuring that a patrol personnel can accurately cover the leakage point and the high-risk equipment area during the patrol process.

Description

Technical Field

[0001] This invention relates to the field of gas leak detection, and more particularly to an embedded AI edge computing system. Background Technology

[0002] The gas cloud imaging device is a professional camera designed for large-scale inspections in chemical industrial parks and chemical plants. It utilizes advanced snapshot hyperspectral imaging technology and is specifically designed for large areas and wide-open spaces. It is used for real-time inspection, monitoring, identification, and quantification of gas leaks. The gas cloud imaging device provides leak concentration, leak volume, and leak location, provides real-time alarms and creates actionable alarm information, and records leak video in real time. At the same time, it notifies operators through video, photoelectric, and other means, enabling mechanization to replace manual labor and reducing personnel casualties.

[0003] The AI ​​edge computing system performs data analysis and machine learning tasks on the gas cloud imaging device. In the application of the gas cloud imaging device, the system can achieve more efficient computing, real-time decision-making and autonomous learning, and is especially suitable for industrial environments that require low latency and fast response.

[0004] However, existing AI edge computing systems do not have the function of generating the optimal adjustment path based on the detection results during the gas detection and analysis process. This means that after obtaining the gas leak location, they can only rely on manual experience or preset routes to handle the defect. This can easily lead to repeated inspections, missed inspections, or avoidance of key equipment, which reduces inspection efficiency and safety. As a result, risks may not be intervened in time and may escalate into malfunctions or even safety accidents.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes an embedded AI edge computing system that constructs optimal gap-filling paths, ensuring that inspection personnel can accurately cover leak points and high-risk equipment areas during inspections.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides an embedded AI edge computing system, comprising:

[0009] The dynamic sensing and identification unit is used to preprocess the infrared gas cloud image data in the inspection area, identify the location of the sulfur hexafluoride cloud based on the preprocessing results, and determine the amount of gas leakage and the location of the gas.

[0010] The situational awareness and discrimination unit is used to simulate the diffusion process of sulfur hexafluoride in the inspection area based on the amount and location of gas leakage, and to determine the risk of equipment control failure in the inspection area when sulfur hexafluoride diffuses.

[0011] The space missing coverage unit is configured to generate a device control failure risk ranking result, and establish an optimal missing coverage path in combination with the leakage position, to adjust the leakage coverage of the inspection area;

[0012] The visual display storage unit is configured to store and display the running process of the dynamic awareness identification unit, the situation awareness discrimination unit and the space missing coverage unit, and to perceive the gas change state in the inspection area.

[0013] Preferably, the dynamic awareness identification unit comprises:

[0014] The gas cloud image preprocessing module is configured to monitor the sulfur hexafluoride in the inspection area based on infrared imaging technology to generate raw infrared gas cloud image data, and to perform inverse transform processing on the raw infrared gas cloud image data to obtain infrared gas cloud image data.

[0015] The image reduction processing module is configured to suppress the background state of the infrared gas cloud image data by using a difference morphological filtering technology, to reduce the search range of the sulfur hexafluoride cloud cluster region, and to construct a time-frequency migration tensor graph.

[0016] The positioning result output module is configured to filter the cloud cluster contour target region based on the time-frequency migration tensor graph, to perform clustering filtering and threshold segmentation on the cloud cluster contour target region, and to identify the sulfur hexafluoride cloud cluster result.

[0017] The leakage state judgment module is configured to use the cloud cluster inversion network to perform source analysis on the sulfur hexafluoride cloud cluster result to construct a dynamic volume concentration core, and to output the leakage point position and the leakage amount of the sulfur hexafluoride gas.

[0018] Preferably, the situation awareness discrimination unit comprises:

[0019] The simulation framework creation module is configured to create a multi-dimensional real-time diffusion simulation framework based on the gas leakage amount and the gas position identification result, in combination with the working state of the equipment in the inspection area and the temperature field change.

[0020] The failure analysis module is configured to simulate the concentration change gradient of the sulfur hexafluoride gas in the inspection area by using the multi-dimensional real-time diffusion simulation framework, to obtain the comparison result with the device failure threshold in the target period.

[0021] The failure risk discrimination module is configured to combine the comparison result with the environmental variables, to analyze the sensitivity state of the equipment in the inspection area, and to discriminate the emergency response capability of the equipment in the inspection area when the sulfur hexafluoride diffuses.

[0022] Preferably, the space missing coverage unit comprises:

[0023] The device failure probability acquisition module is configured to arrange the device control failure risks in descending order, generate a device control failure risk ranking result, and acquire the failure probability of each device in the inspection area.

[0024] The gap filling path establishment module is configured to generate a dynamic weight graph by using a graph optimization algorithm to model the devices and the leakage positions in the inspection area in space, and establish an optimal gap filling path by combining a path planning algorithm.

[0025] The gap filling coverage adjustment module is configured to determine a gap filling driving route of the inspection personnel according to the optimal gap filling path, and adjust the gap filling coverage of the leakage positions in the inspection area by using a leakage gap filling measure.

[0026] The present application has the following advantages:

[0027] 1. The present application can quickly detect gas leakage by processing image data in real time, simulate the diffusion process of sulfur hexafluoride in the inspection area based on the amount and position of gas leakage, and evaluate the control failure risk of the devices in the inspection area. By discriminating the device failure risk, the response capability to emergencies is enhanced, and the device control failure risk ranking and the optimal gap filling path are generated to adjust the leakage coverage of the leakage positions in the inspection area, ensuring that device management and risk control can cover all critical positions, reducing the situation that leakage is not repaired in time, and ensuring the safety of device operation.

[0028] 2. The present application arranges the device control failure risks in descending order to help quickly identify high-risk devices, provides data support based on the health status of the devices, so that the inspection personnel can prioritize the processing of high-risk devices, reduce the chain reaction caused by device failure, and efficiently construct the optimal gap filling path by combining the graph optimization algorithm and the path planning algorithm, so that the inspection personnel can accurately cover the leakage points and high-risk device areas during the inspection process, thereby avoiding omission and reducing repeated inspection, and optimizing the allocation of inspection resources. BRIEF DESCRIPTION OF DRAWINGS

[0029] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the description of the exemplary embodiments of the present application and their explanations serve to explain the present application, and do not constitute an improper limitation of the present application.

[0030] Figure 1 is a principle block diagram of an embedded AI edge computing system according to an embodiment of the present application.

[0031] In the drawings:

[0032] 1, dynamic sensing recognition unit; 2, situation awareness discrimination unit; 3, spatial gap filling coverage unit; 4, visual display storage unit. DETAILED DESCRIPTION

[0033] The application will be further described below in connection with the drawings and embodiments.

[0034] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0035] It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0036] The embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0037] Please refer to Figure 1 The application provides an embedded AI edge computing system, which comprises a mainboard, a USB interface, a GMSL expansion card area, an Ethernet expansion card area and a power module, and the mainboard is integrated with a dynamic sensing identification unit 1, a situation awareness discrimination unit 2, a spatial missing coverage unit 3 and a visual display storage unit 4; wherein:

[0038] The dynamic sensing identification unit 1 is used for pre-processing infrared gas cloud image data in a patrol area, identifying a sulfur hexafluoride cloud group positioning result according to a pre-processing result, and judging a gas leakage amount and a gas position.

[0039] In one embodiment, the dynamic sensing identification unit 1 comprises:

[0040] The gas cloud image preprocessing module is used for monitoring sulfur hexafluoride in the patrol area based on infrared imaging technology to generate original infrared gas cloud image data, and performing inverse transform processing on the original infrared gas cloud image data to obtain infrared gas cloud image data.

[0041] It needs to be explained that in the process of realizing the dynamic sensing identification unit 1, mainly based on infrared imaging technology, the gas leakage of sulfur hexafluoride in the inspection area is monitored, and through multi-level data processing and analysis method, efficient gas positioning, leakage judgment and area search are realized, at the same time, the gas cloud image preprocessing module is used to realize real-time monitoring of the inspection area by using infrared imaging technology, and the original infrared gas cloud image data is generated, which reflects the temperature distribution in the inspection area. Especially, the leakage area of sulfur hexafluoride usually shows as a temperature anomaly area, and the infrared imaging technology can capture the position and size of the gas cloud, and usually sulfur hexafluoride gas as a gas with high density, will be in the form of lower or higher temperature in the image, so as to realize real-time gas leakage monitoring and analysis, and ensure early identification of gas leakage.

[0042] After obtaining the original image data, the data is processed by inverse transformation, the key thermal signal area is extracted from the image, the noise and interference are removed, and Fourier transform or wavelet transform can be used to separate the high-frequency noise and background information from the image, highlight the useful temperature difference data, and extract the useful gas leakage information by inverse transformation processing, and remove unnecessary background noise.

[0043] The image reduction processing module is used to suppress the background state of the infrared gas cloud image data by using the difference morphological filtering technology, reduce the search range of the sulfur hexafluoride cloud area, and construct the time-frequency migration tensor graph.

[0044] It needs to be explained that on the image after inverse transformation, the difference morphological filtering technology is used to further suppress the background state and reduce the search range of the sulfur hexafluoride cloud area; the difference morphological filtering is a kind of nonlinear filtering technology, which can remove the small interference area in the background through erosion and expansion operation, so that the sulfur hexafluoride gas cloud in the image is more prominent, the influence of irrelevant area on image analysis is reduced, and the focus is concentrated on the hotspot area of gas cloud, and through the difference morphological filtering, the background interference is removed, the area to be detected is reduced, the gas cloud is highlighted, the identification efficiency of gas leakage area is improved, and the consumption of computing resources is reduced.

[0045] Through the reduced image area, the time-frequency migration tensor graph is used to further analyze and process the image data, the time-frequency migration tensor graph can combine the time domain and frequency domain information of the image, extract the key information of the image in the time-varying and frequency-varying process, and through the transformation into tensor graph form, the diffusion mode, propagation speed and direction of the gas cloud are better captured, and the time and space characteristics are better captured. In the diffusion process of sulfur hexafluoride, the temperature change has the characteristic of time change, and the time-frequency migration tensor graph can identify these change characteristics and help accurately understand the diffusion process of the gas, and then through the construction of the time-frequency migration tensor graph, the dynamic characteristics of the gas leakage are further analyzed from the time and space dimensions.

[0046] The positioning result output module is configured to filter a cloud cluster contour target region based on the time-frequency migration tensor diagram, perform clustering filtering and threshold segmentation on the cloud cluster contour target region, and identify a sulfur hexafluoride cloud cluster result.

[0047] Specifically, in the process of filtering the cloud cluster contour target region based on the time-frequency migration tensor diagram, performing clustering filtering and threshold segmentation on the cloud cluster contour target region, and identifying the sulfur hexafluoride cloud cluster result, the neural diffuse reflection network can be used to perform physical enhancement on a low signal-to-noise area of the time-frequency migration tensor diagram, and the area point set of the target cloud cluster contour in the time-frequency migration tensor diagram can be estimated according to the similarity theorem; the temperature difference between the background and the target cloud cluster contour of the time-frequency migration tensor diagram is randomly defined as a feature vector of the area point from the area point set, the Euclidean distance between the feature vector and any point of the area point set is calculated; the probability of any point being selected as a clustering center is judged according to the Euclidean distance, the clustering center is selected, and the minimum average distance from any point in the area point set to the clustering center is obtained as a cloud cluster contour target point; a threshold segmentation value is generated by setting a window with the cloud cluster contour target point as the center, the shape of the sulfur hexafluoride cloud cluster is estimated, and the sulfur hexafluoride cloud cluster result is identified according to the area ratio weighted average result of the cloud cluster shape.

[0048] It needs to be explained that in the process of identifying the sulfur hexafluoride cloud cluster, the time-frequency migration tensor diagram is constructed by image data to obtain the thermal field feature distribution of the gas changing with time, but since the original signal-to-noise ratio of the image is low, the neural diffuse reflection network is introduced to enhance the low signal-to-noise area. This physical enhancement simulates the light diffusion and boundary gradual change mechanism in nature to realize local contrast enhancement and boundary contour clarification of the image, so that the gas contour is no longer submerged by the background interference; according to the similarity theorem, the thermal diffusion contour area of the cloud cluster in the time-frequency distribution is derived in the diagram to obtain an initial area point set representing the possible edge or center area of the sulfur hexafluoride cloud cluster.

[0049] To further screen and classify these regional point sets, a random feature vector definition method is adopted, that is, representative points are randomly selected in the regional point set, and the difference between the representative points and the surrounding background in the temperature field is calculated to construct a temperature difference feature vector. Each feature vector is usually an n-dimensional vector (for example, n=8 represents the gradient difference value of 8 frames of time-frequency continuous graphs), which is used to express the independence or mutation degree of the region point relative to the surrounding time-frequency background. At the same time, the Euclidean distance between these feature vectors and any point in the set is further calculated, which reflects the similarity of the point to the possible contour center. The smaller the distance, the more likely it belongs to the same heat diffusion path or the same cloud structure. Based on the distance distribution, the probability of each point being selected as a clustering center is estimated using a probability density function, and an adaptive density clustering method (such as improved DBSCAN or Gaussian kernel K-Means) is used to determine a set of clustering centers, which represent the high-confidence boundary points or center points of the cloud. In order to further refine the boundary, the minimum average distance from each point to the corresponding clustering center is calculated, and the "cloud contour target points" that meet the convergence threshold are screened out. A local analysis window (such as a 5 or 7 pixel region) is constructed around each target point, an adaptive temperature gradient histogram is constructed in the window, and local extreme values and turning points are extracted. A dynamic threshold for image segmentation is calculated, which not only considers the temperature difference around the target point, but also integrates the hot spot distribution structure inside the window, thereby dynamically adapting to the thermal feature changes of different regions and avoiding the misidentification caused by traditional fixed threshold methods. All segmented contour points are constructed into a candidate cloud contour set, and further weighted averaging is performed according to the cloud shape area ratio (such as area / perimeter, boundary integrity coefficient, etc.) to remove irregular, non-continuous, and small-area interference regions. The final output cloud contour not only has accurate spatial boundaries, but also has physical diffusion consistency, ensuring that the identified sulfur hexafluoride cloud result meets both the image thermal field characteristics and the gas physical diffusion model.

[0050] Further, accurate segmentation and identification of sulfur hexafluoride gas clouds can still be achieved under complex image background and time-varying interference, thereby greatly improving the accuracy and stability of micro-gas leakage detection in industrial scenarios.

[0051] The leakage state judgment module is used to trace the analysis of the cloud inversion network to construct a dynamic volume concentration core for the sulfur hexafluoride cloud result, and output the leakage point position and leakage amount of the sulfur hexafluoride gas.

[0052] Specifically, when using a cloud inversion network to perform source tracing analysis on sulfur hexafluoride (SF6) cloud results to construct a dynamic volume concentration kernel and output the location and amount of SF6 gas leakage, the SF6 cloud results can be overlaid in the time dimension to construct a gas cloud deformation time distribution tensor. Motion field sensing technology is then introduced to extract the expansion rate vector field and gradient tension tensor of the cloud edge. Based on the expansion rate vector field and gradient tension tensor, the cloud inversion network is trained to reverse-trace the initial expansion boundary of the SF6 cloud results, identifying the corresponding potential source point probability hotspots. The integral of the potential source point probability hotspot is determined based on the concentration distribution field, and combined with diffusion rate correction technology to construct a dynamic volume concentration kernel. A diffusion factor is introduced to estimate the total leakage amount of SF6 gas. Based on the equipment layout diagram within the inspection area, spatial constraints and location confirmation of the leakage point are performed, and confidence is verified by combining temporal evolution consistency, outputting the leakage point location and total leakage amount.

[0053] Specifically, in this leakage status judgment module, the core task is to perform source tracing analysis on the results of sulfur hexafluoride (SF6) clouds through a cloud inversion network, further construct a dynamic volume concentration kernel, and finally output the location and leakage amount of the gas leak point. This is achieved by superimposing SF6 cloud results from different time points along a time dimension to construct a gas cloud deformation time distribution tensor. This tensor integrates the expansion behavior of SF6 gas clouds at different times, forming a multi-dimensional spatiotemporal data structure. Information such as the cloud location, shape, and concentration at each time point is embedded in this tensor, forming the time-varying characteristics of gas diffusion. Furthermore, by superimposing image data from different times, a composite tensor containing time-varying information can be created to capture the dynamic changes in cloud expansion. This comprehensively describes the historical evolution of cloud expansion, providing rich temporal information for subsequent motion field perception and source tracing analysis.

[0054] Furthermore, motion field sensing technology is used to conduct in-depth analysis of the dynamic expansion process of the cloud cluster, extracting the expansion rate vector field and gradient tension tensor. The expansion rate vector field can capture the expansion speed and direction of the cloud cluster edge in the time dimension, while the gradient tension tensor reflects the rate of change of the cloud cluster shape and the irregularity of the morphological changes. By extracting the expansion rate vector field and gradient tension tensor, the dynamic expansion behavior and shape changes of the cloud cluster can be analyzed, providing key data for reverse tracing.

[0055] Based on the data extracted from the time-frequency migration tensor map and the motion field perception technology, the cloud cluster inversion network is trained to perform backtracking on the initial boundary of the cloud cluster expansion, i.e., to trace back from the expansion boundary of the cloud cluster along the air flow direction to the possible leakage source, which uses a reverse expansion algorithm combined with gradient tension information to reversely calculate the starting position of the gas leakage and identify the probability hot zone of the potential source point. The gas concentration in the hot zone is higher, indicating that it may be a leakage source area, so that through the backtracking of the cloud cluster inversion network, the potential gas leakage source area is identified, and the leakage source probability of each area point is given.

[0056] Based on the cloud cluster inversion result, a concentration distribution field is constructed to reflect the gas concentration value at each location point. On this basis, a dynamic volume concentration kernel is constructed by combining diffusion rate correction technology. The diffusion rate correction technology is based on environmental factors (such as wind speed, temperature, air flow direction, etc.) and gas diffusion model to correct the deviation in the gas diffusion process, thereby improving the accuracy of the leakage amount estimation. By correcting the diffusion rate, the actual amount of gas leakage can be more accurately estimated, and the concentration change during the gas diffusion process is dynamically tracked to improve the estimation accuracy of the leakage amount.

[0057] Combined with the equipment layout map in the inspection area, the spatial constraint and positioning confirmation of the leakage position point are performed. The equipment layout map provides the location, structure and activity range of all equipment in the inspection area, which is compared with the spatial position of the leakage source point to determine the specific position of the leakage point in the actual equipment. At the same time, the confidence check of time evolution consistency is performed to ensure the high reliability of the confirmation result of the leakage position.

[0058] Further, by combining time-frequency tensor analysis, motion field perception, cloud cluster inversion, diffusion rate correction and equipment spatial constraint, accurate detection, positioning and quantification of sulfur hexafluoride gas leakage are ensured.

[0059] The situation awareness discrimination unit 2 is used to simulate the diffusion process of sulfur hexafluoride in the inspection area based on the gas leakage amount and the gas position, and to discriminate the risk of equipment control failure in the inspection area during the diffusion of sulfur hexafluoride.

[0060] In one embodiment, the situation awareness discrimination unit 2 includes:

[0061] The simulation framework creation module is used to create a multi-dimensional real-time diffusion simulation framework based on the gas leakage amount and the gas position identification result, combined with the working state of the equipment in the inspection area and the temperature field change.

[0062] The multi-dimensional real-time diffusion simulation framework is created based on the gas leakage amount and the gas position recognition result, in combination with the working state and temperature field change of the equipment in the inspection area. The multi-dimensional real-time diffusion simulation framework includes: based on the leakage amount and the leakage position recognition result of sulfur hexafluoride, a space-time correlation model is used to finely divide the inspection area in the spatial coordinate system, and an initial gas concentration field of sulfur hexafluoride is generated; using the dynamic air mass tracking mechanism of the particle filtering algorithm, sulfur hexafluoride is set as a diffusion cloud composed of multiple particles in the gas concentration field, and each particle represents a volume unit of sulfur hexafluoride; according to the set result and the gas characteristics, the position is updated and the concentration is adjusted, a multi-granularity diffusion simulation framework is constructed, and the working state of the equipment in the inspection area and the temperature field change under the working state are obtained; based on the working state of the equipment and the temperature field change, a thermal effect coupling mechanism is introduced to redivide the grid of the multi-granularity diffusion simulation framework according to the dynamic scene change, and a multi-dimensional real-time diffusion simulation framework is created.

[0063] Specifically, the thermal effect coupling mechanism is introduced based on the working state of the equipment and the temperature field change to redivide the grid of the multi-granularity diffusion simulation framework according to the dynamic scene change, and a multi-dimensional real-time diffusion simulation framework is created, which includes: based on the working state of the equipment and the temperature field change, the thermal effect simulation model is used to link the thermal influence of the equipment with the environmental airflow field for simulation, and the potential influence of the equipment temperature change on gas diffusion is analyzed; using multi-physical field coupling and fluid dynamics, the diffusion direction and speed offset of sulfur hexafluoride caused by the temperature change of the equipment are dynamically simulated, and the resolution of the multi-granularity diffusion simulation framework is adjusted; based on the adaptive subdivision grid technology, the grid density of the multi-granularity diffusion simulation framework is optimized to ensure the stability of the airflow disturbance of the equipment in the inspection area, and a multi-dimensional real-time diffusion simulation framework is created.

[0064] It should be explained that the simulation framework creation module is based on the identified sulfur hexafluoride leakage amount (such as 200ppm·m) and the spatial coordinate position, uses a space-time correlation model to finely divide the space of the inspection area, such as dividing it into 10cm×10cm grid units, generates an initial gas concentration field in these grids, gives an initial sulfur hexafluoride concentration value in each grid unit, uses a particle filtering algorithm for dynamic air mass tracking, models sulfur hexafluoride as a diffusion cloud composed of hundreds of particles in the concentration field, and each particle represents a specific volume of gas unit, for example, 0.01m 3and the concentration correction is made according to the diffusion equation and the wind field simulation; at the same time, the working state parameters (such as working current, load, rotating speed, etc.) of each device in the inspection area and the temperature field change data (such as thermal infrared imaging results) matched therewith are synchronously obtained, which are used to establish the temperature field distribution diagram under the influence of the device, and the thermal effect coupling mechanism is introduced, that is, the temperature change of the device is coupled and simulated with the environmental airflow field through a heat source influence model, the offset effect of the heat generated by the device on the diffusion trajectory of the gas is simulated, for example, the airflow disturbance vertically upward or outward around the hot spot device can cause the diffusion direction of sulfur hexafluoride to deviate by 10°-30°, and the speed to deviate by 0.2-0.5 m / s.

[0065] Based on the thermal disturbance information, the granularity and precision of the diffusion simulation framework are dynamically adjusted, the adaptive subdivision grid technology (such as octree subdivision and Voronoi re-partitioning) is used to re-construct the diffusion field grid, higher resolution (such as 5 cm x 5 cm) is set for the area near the device or the area with severe airflow disturbance, and lower resolution is maintained for the area far from the device, so as to reduce the calculation amount while ensuring the precision, and the whole multi-granularity diffusion simulation framework thus forms a dynamic updating, thermal-flow coupling three-dimensional gas diffusion field, finally, the framework can predict the distribution of sulfur hexafluoride at each moment, and mark the area where the concentration exceeds the device control threshold (such as 150 ppm) as a high-risk failure area, combine the device function table to judge whether there is a potential threat of control failure, and generate a risk level map.

[0066] Further, multi-dimensional perception of the leakage situation can be realized, so that not only the risk can be identified under static conditions, but also the diffusion path can be perceived in advance under dynamic and thermal disturbance background, the risk position and time point can be accurately judged, and high credible decision basis can be provided for inspection scheduling and emergency response.

[0067] The failure analysis module is used to simulate the concentration change gradient of sulfur hexafluoride gas in the inspection area by using the multi-dimensional real-time diffusion simulation framework, and obtain the comparison result with the device failure threshold in the target period.

[0068] The failure risk judgment module is used to combine the comparison result with the environmental variables, analyze the sensitivity state of the device in the inspection area, and judge the emergency response capability of the device in the inspection area when sulfur hexafluoride diffuses.

[0069] Specifically, the core function of the failure analysis module is to simulate the diffusion process of sulfur hexafluoride gas in the inspection area through a multi-dimensional real-time diffusion simulation framework, especially the concentration gradient of the gas. During the simulation process, the gas concentration value of each grid unit (e.g., 10 cm x 10 cm) in the inspection area is continuously calculated and updated. At the same time, the change of the gas concentration field is corrected based on the device location and thermal effect. Then, the diffusion simulation framework is used to calculate and output the concentration gradient of the gas, reflecting the diffusion speed, diffusion direction, and spatial relationship with the device. Meanwhile, according to factors such as the device's location, type, working state, and gas concentration changes near the device, the gas concentration at each device location is calculated and compared with the device's failure threshold (e.g., when the sulfur hexafluoride concentration exceeds 100 ppm, the device fails).

[0070] The failure risk discrimination module combines the comparison results obtained from the failure analysis module (i.e., the comparison of sulfur hexafluoride gas concentration and device failure threshold) with environmental variables (such as wind speed, temperature, humidity, etc.) to comprehensively evaluate the sensitivity state and emergency response capability of the devices in the inspection area. The sensitivity state of the device refers to whether the device is prone to failure under specific gas concentration and environmental conditions, while the emergency response capability refers to whether the device can respond in time to avoid failure or damage in the case of excessive gas concentration.

[0071] The failure risk discrimination module analyzes the working state of the device (such as load, voltage, current, etc.) and external environmental factors (such as temperature, humidity, wind speed, etc.) to determine whether the device is in a high sensitivity state. For example, excessive wind speed (more than 5 m / s) may cause the gas diffusion speed to increase, thereby increasing the risk of device failure; high temperature (more than 70°C) may cause the device to respond to gas leakage in time, increasing the probability of failure.

[0072] In this process, a device risk state diagram is generated, and each device is assigned a risk level based on the combination of its gas concentration exposure and environmental variables. For example, if the gas concentration exceeds the failure threshold and the device is in a high load state, the device's risk level will be rated as "high"; if the environmental conditions are stable (e.g., low wind speed, normal temperature), the risk level will be lower. By combining gas concentration data and environmental variables, the risk state of the device is analyzed, and the sensitivity state and emergency response capability of each device are evaluated, generating a risk level for each device. This allows multiple factors affecting device failure to be considered, making the risk assessment not only limited to gas concentration, but also taking into account environmental factors and the working state of the device itself.

[0073] The spatial missing coverage unit 3 is used to generate a device control failure risk ranking result and establish an optimal missing coverage path in combination with the leakage location to adjust the leakage coverage of the inspection area.

[0074] In one embodiment, the spatial missing coverage unit 3 comprises:

[0075] The device failure probability acquisition module is configured to arrange the device control failure risks in descending order, generate a device control failure risk ranking result, and acquire the failure probabilities of the devices in the inspection area.

[0076] The missing path establishment module is configured to use a graph optimization algorithm to perform spatial modeling on the devices and the leakage position in the inspection area to generate a dynamic weight graph, and combine a path planning algorithm to establish an optimal missing path.

[0077] Specifically, the use of a graph optimization algorithm to perform spatial modeling on the devices and the leakage position in the inspection area to generate a dynamic weight graph, and combining a path planning algorithm to establish an optimal missing path comprises: performing rasterization processing on the two-dimensional map of the inspection area, discretizing the device positions into graph nodes, simultaneously converting the leakage points into concentration influence nodes, and establishing edge connections at the concentration influence nodes to generate edges; adjusting the concentration influence node state and edge weight based on the temperature disturbance intensity and the operation accessibility obstacle factor to form a dynamic weight graph that can be adaptively updated over time; performing sectioning on the dynamic weight graph to establish a topological graph, combining a path planning algorithm to generate an initial missing path, simultaneously using an elastic band algorithm to optimize the initial missing path to obtain a locally optimal path; constructing an initial tree based on the topological graph, combining a sectioning line constraint and an informed set constraint to construct a dynamic sampling domain, and randomly sampling in the dynamic sampling domain to optimize the initial tree to establish an optimal missing path.

[0078] The construction of the initial tree based on the topological graph, the combination of the sectioning line constraint and the informed set constraint to construct the dynamic sampling domain, the random sampling in the dynamic sampling domain to optimize the initial tree, and the establishment of the optimal missing path comprise: constructing the informed set with the length of the locally optimal path as the long axis, and defining the informed set as the limit to define the sectioning line, and taking the midpoint of the sectioning line as the node under the action of the topological graph; starting from the initial state to construct the initial tree and randomly generating a sampling point in the dynamic sampling domain, expanding the sampling point to the initial tree, and judging in the initial tree whether the parent node of the sampling point is the starting node; if it is the starting node, the sampling point is directly updated, and if it is not the starting node, the node with the minimum cost is selected from the adjacent sectioning line nodes of the sectioning line where the sampling point is located; adjusting all child nodes on the sectioning line where the sampling point is located through the reconnection of the initial tree, and sequentially updating the cost of the subsequent nodes, and establishing the optimal missing path based on the update result.

[0079] The missing coverage adjustment module is configured to determine the missing driving route of the inspection personnel according to the optimal missing path, and adopt a leakage missing measure to adjust the missing coverage of the leakage position in the inspection area.

[0080] It should be explained that in the spatial missing coverage unit 3, by integrating technologies such as equipment control failure risk ranking, leak location modeling, and dynamic path planning, a comprehensive path optimization and leak coverage scheme is generated. Through the planning of the optimal path, it is ensured that in the event of a gas leak, the inspection personnel can cover all high-risk equipment areas and respond to the possible leakage impact in a timely and effective manner.

[0081] The core task of the equipment failure probability acquisition module is to rank the failure probabilities of each piece of equipment in the inspection area based on the previous risk analysis results. It uses the gas exposure and emergency response capabilities of each piece of equipment obtained from the failure analysis module to determine the failure probability. For example, assuming equipment A has a gas concentration of 100 ppm in the leak area, its failure probability at this concentration is 0.85; equipment B is located in an area with high wind speeds but a concentration of 70 ppm, its failure probability might be 0.5. These probability values ​​are sorted in descending order to generate the equipment failure risk ranking result. The ranked equipment list serves as the priority target for path planning and leak coverage. Through accurate failure risk ranking, high-risk equipment can be identified, ensuring that the equipment most likely to fail is prioritized during inspections, thereby improving the efficiency and safety of inspections.

[0082] The core function of the gap-filling path creation module is to generate an optimized inspection path. This path not only covers high-risk equipment but also adjusts according to changes in leak location and gas concentration. The two-dimensional map of the inspection area is rasterized, transforming the location of each device and leak point into nodes in the graph. Device nodes and leak point nodes are connected by edges, with edge weights related to concentration impact. For example, the closer the leak point is to the device, the greater the concentration impact and the higher the weight. Simultaneously, a dynamic weight adjustment mechanism is introduced based on environmental changes. For instance, factors such as the intensity of temperature disturbances, wind speed, and humidity around the equipment affect gas diffusion and concentration distribution. Environmental factors influence the state of concentration-affected nodes and edge weights, forming a dynamically updated weighted graph. This graph is continuously updated over time, reflecting real-time changes in gas concentration and equipment risk. This allows path planning to respond promptly to environmental changes, ensuring the real-time nature and adaptability of the inspection path and improving the effective coverage of leak sources by inspection personnel.

[0083] In the establishment process of the supplement path, the graph optimization algorithm and the path planning algorithm play a key role. Through rasterization, the two-dimensional map of the inspection area is converted into a graph structure, and the connection edges between each device node and the leakage node are weighted. The weight of the edge is dynamically adjusted according to the distance between the device and the leakage source, the concentration influence, and the environmental disturbance (such as temperature, wind speed, etc.). For example, the device node near the leakage point may have an increased weight due to the high concentration. The path planning algorithm is used to solve the shortest path from the starting point to each device node, which takes into account the priority of each device. In order to optimize the efficiency of the path, the elastic band algorithm is used to optimize the preliminary path, adjusting the curvature of the path and avoiding unnecessary redundancy. The topology graph divides the graph of the inspection area into multiple sub-areas and generates a sampling domain in each sub-area. The initial tree is continuously optimized through random sampling points, and the optimal supplement path is formed on the topology graph. The combination of the section line constraint and the informed set constraint makes the path planning more efficient. The section line constraint can optimize the path locally according to the optimization direction of the current path. The informed set constraint limits the range of sampling points to speed up the path search, thereby obtaining a locally optimal path. The dynamic sampling domain is combined to determine the inspection path, effectively reducing the complexity of path planning and ensuring efficient and accurate inspection path generation, greatly improving the inspection efficiency and safety.

[0084] The leak coverage adjustment module determines the specific driving route of the inspection personnel based on the optimal supplement path and takes effective leak coverage adjustment measures to cover the missed areas. The process of leak coverage adjustment includes the generation of dynamic paths and environmental adaptation, adjustment of the inspection personnel's travel route, and timely inspection of all high-risk devices and missed areas to improve the integrity and flexibility of path planning and avoid missing any high-risk devices, thereby reducing the potential risks caused by missed inspection.

[0085] Therefore, through the graph optimization algorithm, path planning algorithm, and dynamic leak adjustment, the optimal inspection path can be ensured at each moment, thereby effectively reducing the risk of device failure and improving the efficiency and accuracy of inspection response.

[0086] The visual display storage unit 4 is used to store and display the running process of the dynamic sensing recognition unit 1, the situation awareness discrimination unit 2, and the spatial missing coverage unit 3, and to perceive the gas change state in the inspection area.

[0087] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the present embodiment can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0088] Although the specific embodiments of the present application are described above in combination with the drawings, it is not a limitation on the scope of protection of the present application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. An embedded AI edge computing system, characterized in that, include: The dynamic sensing and identification unit is used to preprocess the infrared gas cloud image data in the inspection area, identify the location of the sulfur hexafluoride cloud based on the preprocessing results, and determine the amount of gas leakage and the location of the gas. The situational awareness and discrimination unit is used to simulate the diffusion process of sulfur hexafluoride in the inspection area based on the amount and location of gas leakage, and to determine the risk of equipment control failure in the inspection area when sulfur hexafluoride diffuses. Spatial gap coverage unit is used to generate equipment control failure risk ranking results and combine them with the leak location to establish the optimal gap filling path, and adjust the leak coverage of the leak location in the inspection area. The visual display storage unit is used to store and display the operation process of the dynamic perception and recognition unit, the situational perception and discrimination unit and the spatial missing coverage unit, and to perceive the gas change status in the inspection area. The situational awareness and discrimination unit includes: The simulation framework creation module is used to finely divide the inspection area within the spatial coordinate system based on the identification results of sulfur hexafluoride (SF6) leakage amount and location, and generate an initial SF6 gas concentration field. Utilizing the dynamic gas cloud tracking mechanism of the particle filter algorithm, SF6 is represented as a diffusion cloud composed of multiple particles within the gas concentration field, with each particle representing a volume unit of SF6. The module updates the location and adjusts the concentration based on the set results and gas characteristics, constructing a multi-granularity diffusion simulation framework and acquiring the equipment operating status and temperature field changes within the inspection area. Based on the equipment operating status and temperature field changes, a thermal effect simulation model is used to simulate the linkage between the equipment's thermal impact and the ambient airflow field, analyzing the potential impact of equipment temperature changes on gas diffusion. Using multiphysics coupling and fluid dynamics, the module dynamically simulates the diffusion direction and velocity shift of SF6 caused by equipment temperature changes and adjusts the resolution of the multi-granularity diffusion simulation framework. Based on adaptive subdivision mesh technology, the grid density of the multi-granularity diffusion simulation framework is optimized to ensure stability at the equipment airflow disturbance points within the inspection area, creating a multi-dimensional real-time diffusion simulation framework.

2. The embedded AI edge computing system according to claim 1, characterized in that, The dynamic sensing and recognition unit includes: The gas cloud image preprocessing module is used to monitor sulfur hexafluoride in the inspection area based on infrared imaging technology, generate raw infrared gas cloud image data, and perform inverse transformation processing on the raw infrared gas cloud image data to obtain infrared gas cloud image data. The image downsizing module is used to suppress the background state of infrared gas cloud image data using differential morphological filtering technology, reduce the search range of sulfur hexafluoride cloud region, and construct a time-frequency transition tensor map. The localization result output module is used to filter the target region of cloud contour based on the time-frequency migration tensor map, and to perform clustering filtering and threshold segmentation on the target region of cloud contour to identify the sulfur hexafluoride cloud results. The leakage status judgment module is used to perform source analysis on the sulfur hexafluoride cloud results using the cloud inversion network to construct a dynamic volume concentration kernel and output the location and leakage amount of sulfur hexafluoride gas leakage point.

3. The embedded AI edge computing system according to claim 2, characterized in that, The method of filtering cloud contour target regions based on time-frequency migration tensor maps, and performing clustering filtering and threshold segmentation on the cloud contour target regions to identify sulfur hexafluoride clouds includes: We use a neural diffuse reflection network to perform object-like enhancement on the low signal-to-noise region of the time-frequency migration tensor, and estimate the set of region points of the target cloud outline in the time-frequency migration tensor based on the similarity theorem. The temperature difference between the background of the time-frequency migration tensor and the outline of the target cloud is randomly defined from the set of regional points as the feature vector of the regional points, and the Euclidean distance between the feature vector and any point in the set of regional points is calculated. The probability of any point being selected as a cluster center is determined based on Euclidean distance. Cluster centers are selected, and the minimum average distance from any point in the region point set to the cluster center is obtained as the target point of the cloud outline. A threshold segmentation value is generated by setting a window centered on the target point of the cloud outline to estimate the shape of the sulfur hexafluoride cloud. Based on the weighted average result of the area ratio of the cloud shape, the sulfur hexafluoride cloud result is identified.

4. The embedded AI edge computing system according to claim 3, characterized in that, The method of using a cloud inversion network to perform source tracing analysis on sulfur hexafluoride (SF6) cloud results to construct a dynamic volume concentration kernel, and outputting the location and leakage amount of SF6 gas leakage points, including: The results of sulfur hexafluoride clouds were superimposed in the time dimension to construct the gas cloud deformation time distribution tensor, and motion field sensing technology was introduced to extract the expansion rate vector field and gradient tension tensor of the cloud edge. Based on the training of cloud inversion network using the extended rate vector field and gradient tension tensor, the initial extended boundary of sulfur hexafluoride cloud results is traced back to identify the corresponding potential source point probability hotspots. Based on the integral of the probability hot zone of potential source points determined by the concentration distribution field, and combined with the diffusion rate correction technique, a dynamic volume concentration kernel is constructed, and a diffusion factor is introduced to estimate the total leakage of sulfur hexafluoride gas. Based on the equipment layout diagram within the inspection area, spatial constraints and location confirmation are performed on the leak location, and confidence is verified by combining the consistency of time evolution. The location of the leak point and the total leakage amount are then output.

5. An embedded AI edge computing system according to claim 1, characterized in that, The situational awareness and discrimination unit also includes: The failure analysis module is used to simulate the concentration change gradient of sulfur hexafluoride gas in the inspection area using a multi-dimensional real-time diffusion simulation framework, and obtain the comparison results with the equipment failure threshold within the target time period. The failure risk assessment module is used to combine the comparison results with environmental variables to analyze the sensitivity status of equipment in the inspection area and determine the emergency response capability of equipment in the inspection area when sulfur hexafluoride diffuses.

6. An embedded AI edge computing system according to claim 1, characterized in that, The spatial missing coverage unit includes: The equipment failure probability acquisition module is used to sort the equipment control failure risk in descending order, generate the equipment control failure risk ranking result, and obtain the failure probability of each piece of equipment in the inspection area. The gap filling path establishment module is used to use graph optimization algorithms to spatially model the equipment and leakage locations in the inspection area to generate a dynamic weighted graph, and combine it with path planning algorithms to establish the optimal gap filling path. The leak coverage adjustment module is used to determine the inspection personnel's travel route for leak coverage based on the optimal leak coverage path, and to adjust the leak coverage of the inspection area by adopting leak coverage measures.

7. An embedded AI edge computing system according to claim 6, characterized in that, The process of using graph optimization algorithms to spatially model equipment and leak locations within the inspection area to generate a dynamic weighted graph, and combining this with path planning algorithms to establish the optimal gap-filling path, includes: The two-dimensional map of the inspection area is rasterized, the equipment location is discretized into graph nodes, the leak point is transformed into concentration-affected nodes, and edge connections are established between the concentration-affected nodes to generate edges. Based on the influence of temperature perturbation intensity and operational accessibility barrier factor on the concentration of the influence on node state and edge weight, a dynamic weight graph that can be adaptively updated over time is formed. The dynamic weighted graph is partitioned to create a topology graph. An initial missing path is generated by combining a path planning algorithm. At the same time, the elastic band algorithm is used to optimize the initial missing path to obtain the local optimal path. An initial tree is constructed based on the topology graph, and a dynamic sampling domain is constructed by combining the partition line constraint and the informed set constraint. Random sampling is performed in the dynamic sampling domain to optimize the initial tree and establish the optimal missing path.

8. An embedded AI edge computing system according to claim 7, characterized in that, The process of constructing an initial tree based on a topological graph, and then constructing a dynamic sampling domain by combining partition line constraints and informed set constraints, followed by random sampling within the dynamic sampling domain to optimize the initial tree and establish the optimal missing path includes: Construct an information set with the length of the local optimal path as the major axis, and use the information set as a constraint to define the partition line. Under the action of the topology graph, the midpoint of the partition line is taken as a node. Starting from the initial state, construct an initial tree and randomly generate sampling points within the dynamic sampling domain. Extend the sampling points to the initial tree and determine whether the parent node of the sampling point is the starting node within the initial tree. If it is the starting node, the sampling point is updated directly; if it is not the starting node, the node with the lowest cost is selected from the adjacent nodes of the subdivision line. By reconnecting the initial tree, all child nodes on the partition line where the sampling point is located are adjusted, and the cost of subsequent nodes is updated in turn. The optimal missing path is established based on the update results.

Images