Robot dog platform traffic tunnel inspection system based on AI technology

Abstract

The invention discloses a robotic dog platform traffic tunnel inspection system based on an AI technology, relates to the technical field of data analysis, and solves the technical problems that it is difficult to adapt to cross-sea channels with a long distance, it is difficult to ensure timely analysis and processing of data, and it is difficult to predict possible potential faults or degradation. The data from the robot dog can be received and processed in real time through the edge data management node, the delay of data transmission is reduced, the response speed is improved, the data processing pressure of the cloud is relieved through data processing and analysis at the edge, and the whole system is more efficient; the edge data management node can perform security analysis according to the data collected by the robot dog, and timely discover and warn potential security risks; through the safety degree analysis model, the safety condition of each grid area can be intuitively known, and a manager can quickly identify a potential risk area so as to reasonably allocate resources and energy.

Description

Technical Field

[0001] The present invention belongs to the field of data analysis, and specifically relates to a traffic tunnel inspection system for a robot dog platform based on AI technology. Background Art

[0002] With the acceleration of urbanization and the increasing complexity of the traffic network, the safety and maintenance of traffic channels have become increasingly important, such as the inspection of channels like tunnels, bridges, and roads. Traditional inspection methods mostly rely on manual labor, suffering from problems such as low efficiency, easy errors, and lagging data processing. In addition, due to the complex environment and narrow space of many traffic channels, manual inspection faces many challenges, such as the inability to timely detect potential safety hazards and maintenance requirements. It is particularly difficult to efficiently inspect cross-sea channels with complex regional structures, such as the Shenzhen-Zhongshan Channel.

[0003] For a general robot dog platform channel inspection system, by analyzing and calculating data on a unified server, it is difficult to adapt to long cross-sea channels, difficult to ensure timely analysis and processing of data, and at the same time, lack of targeted analysis of each area of the channel, making it difficult to predict potential faults or deteriorations that may exist. Summary of the Invention

[0004] The present invention aims to at least solve one of the technical problems existing in the prior art; for this purpose, the present invention proposes a traffic tunnel inspection system for a robot dog platform based on AI technology, which is used to solve the technical problems of being difficult to adapt to long cross-sea channels, difficult to ensure timely analysis and processing of data, and at the same time, difficult to predict potential faults or deteriorations that may exist.

[0005] To solve the above problems, the first aspect of the present invention provides a traffic tunnel inspection system for a robot dog platform based on AI technology, including: an edge terminal and a cloud management center;

[0006] The edge terminal includes:

[0007] A region allocation module: obtaining channel data of the cross-sea channel region, dividing the target region into a wind tower channel region, a pipe gallery channel region, and an external channel region, and setting robot dogs that conform to the channel management rules of different regions in different regions, and dividing grid areas of different detection types in different regions;

[0008] A data acquisition module: setting different data acquisition strategies for different grids, and the robot dogs in each grid area collect data according to the data acquisition strategies of different grid areas and send them to the edge data management node;

[0009] Edge data management node: Edge data management nodes are set at intervals of a safe communication distance in the cross-sea channel area; the edge data management nodes communicate with the robotic dogs within the safe communication distance for scheduling and control, and perform security analysis on different types of grids based on the data collected by the robotic dogs in the grid area, and issue security alerts for grids with security analysis lower than the threshold.

[0010] The cloud management center includes:

[0011] Model construction module: Based on the security analysis results of grids in different regions, establish security level analysis models for different grid areas and perform security level analysis on each grid area.

[0012] Robotic dog scheduling module: According to the fault degree analysis results of the grid area and the detection progress of the grids in the grid area, support and schedule the robotic dogs in different grid areas through the edge data management nodes.

[0013] Optionally, in an example of the above aspect, the area allocation module includes:

[0014] Area division unit: Obtain the channel data of the cross-sea channel area, including: channel detection area data, channel accident records, channel types, and channel map data. According to the channel detection area data and the channel map data, divide the channel connected to the wind tower into the wind tower channel area, divide the channel accommodating various pipelines into the pipe gallery channel area, and divide the channel accommodating the vehicle and pedestrian passage area into the external channel area.

[0015] Robotic dog allocation unit: Set robotic dogs that conform to the channel management rules of different regions according to the detected area in the channel detection area data of the wind tower channel area, pipe gallery channel area, and external channel area.

[0016] Grid division unit: According to the channel data of each region, divide equipment grid areas, channel grid areas, and wind tower monitoring grid areas in the wind tower channel area, divide channel grid areas, equipment grid areas, and pipeline network detection grid areas in the pipe gallery channel area, and divide channel grid areas and equipment grid areas in the external channel area.

[0017] Optionally, in an example of the above aspect, the data acquisition module sets different data acquisition strategies for different grids, including:

[0018] For all grid areas in the wind tower channel area, pipe gallery channel area, and external channel area, detect basic operation data, including: wind speed, temperature, humidity, high-definition image data, and infrared image data of the channel.

[0019] For the equipment grid areas in the wind tower channel area, pipe gallery channel area, and external channel area, the data acquisition strategy is:

[0020] Collect high-definition images and infrared images of the devices in the additional collection device grid area, as well as infrared images and electrical parameters at the device connection points;

[0021] For the channel grid areas in the wind tower channel area and the pipe gallery channel area, the data collection strategy is:

[0022] Additionallly collect the smoke condition of the channels in the channel grid area;

[0023] For the pipe network detection grid area in the pipe gallery channel area, the data collection strategy is:

[0024] Additionallly collect high-definition images and polarization images at the cables and pipes in the pipe network detection grid area;

[0025] For the channel grid area in the external channel area, the data collection strategy is:

[0026] Additionallly collect the salt fog concentration of the channels in the channel grid area.

[0027] Optionally, in an example of the above aspect, the edge data management node performs scheduling control by communicating with the robotic dog within the secure communication distance, including the following steps:

[0028] The edge data management node establishes a communication connection with the robotic dog within the secure communication distance;

[0029] After receiving the robotic dog scheduling control instruction between different grid areas in the same region sent by the robotic dog scheduling module, according to the scheduling control instruction, send the data collection strategy of the corresponding block to the corresponding robotic dog, and send an instruction to go to the corresponding block for data collection.

[0030] Optionally, in an example of the above aspect, the edge data management node performs security analysis on different types of grids based on the data collected by the robotic dog in the grid area, and issues a security alert for grids with a security analysis lower than the threshold, including the following steps:

[0031] Obtain the high-definition image data, infrared image data, and polarization images with surface anomalies, as well as the high-definition image data, infrared image data, and polarization images without surface anomalies in the historical detection data of the cross-sea channel, and mark the positions of the surface anomalies;

[0032] Train a deep learning model through the image data of the historical detection data to identify and mark the abnormal positions of the high-definition image data, infrared image data, and polarization images, and calculate the area of the abnormal area;

[0033] When obtaining historical data, for the high-definition image data, infrared image data, and polarization images when each area is operating normally, use the trained deep learning model to detect the area of the abnormal area and take the average value as the maximum abnormal fault area threshold for the high-definition image data, infrared image data, and polarization images;

[0034] For all grid areas in the wind tower channel area, pipe gallery channel area, and external channel area, during the preset detection time interval, detect the basic operation data, and perform grid basic safety analysis through the following formula:

[0035]

[0036] Where S is the grid basic safety analysis value, Sv is the grid wind speed safety analysis value, Sth is the grid temperature and humidity safety analysis value, Sp is the grid image abnormality safety analysis value, w1, w2, and w3 are the corresponding weights of the grid wind speed safety analysis value, grid temperature and humidity safety analysis value, and grid image abnormality safety analysis value respectively; v is the average wind speed of the corresponding grid during the preset detection time interval, v0 is the abnormal wind speed threshold of the corresponding grid area, T0 is the average temperature of the corresponding grid during the preset detection time interval, H0 is the average humidity of the corresponding grid during the preset detection time interval, σt is the standard deviation of the temperature of the corresponding grid during the preset detection time interval, σh is the standard deviation of the humidity of the corresponding grid during the preset detection time interval, fai, fbi, and fci are the abnormal areas detected by the high-definition image data, infrared image data, and polarization images in the grid during the preset detection time interval. If no abnormality is detected or there is no corresponding detection data for the corresponding grid, the corresponding abnormal area is set to zero, and famax, fbmax, and fcmax are the maximum abnormal fault area thresholds detected by the high-definition image data, infrared image data, and polarization images;

[0037] According to the grid basic safety analysis results, perform additional safety analysis for different areas;

[0038] Perform a weighted average of the grid basic safety analysis results and the additional safety analysis to obtain the final safety analysis result of the grid, and issue a safety alarm for grids with a safety analysis lower than the threshold.

[0039] Optionally, in an example of the above aspect, performing additional safety analysis for different areas includes the following steps:

[0040] For the equipment grid areas in each area, obtain the high-definition images and infrared images of the equipment in the additional acquisition equipment grid area, as well as the infrared images and electrical parameters at the equipment connection points, and perform additional safety analysis of the equipment through the following formula:

[0041]

[0042] Among them, Sep is the additional security analysis value of the device, fej is the device grid area of each region, the abnormal area of the j-th image data collected additionally, femaxj is the maximum abnormal fault area threshold of the corresponding image type of the j-th image data collected additionally, σek is the standard deviation of the k-th electrical parameter collected additionally in the device grid area of each region during the preset detection time interval, and Tek is the mean value of the k-th electrical parameter collected additionally in the device grid area of each region during the preset detection time interval;

[0043] For the channel grid areas of the wind tower channel area and the pipe gallery channel area, obtain the smoke condition of the channels in the additionally collected channel grid area. If smoke is detected, immediately send a safety alarm signal for the corresponding grid; otherwise, do not send a safety alarm signal.

[0044] For the pipe network detection grid area of the pipe gallery channel area, obtain the high-definition image and polarized image at the cable and pipeline in the additionally collected pipe network detection grid area, and calculate the additional security analysis value of the cable and pipeline according to the same method as the calculation method of the image abnormal security analysis value of the basic operation data grid.

[0045] For the channel grid area of the external channel area, obtain the salt fog concentration of the channels in the additionally collected channel grid area, and calculate the ratio of the mean value of the detected salt fog concentration within the preset detection time interval to the abnormal salt fog concentration threshold as the additional security analysis value of the salt fog.

[0046] Add up all the additionally calculated security analysis values in each grid. If the sum is zero, the additional security analysis result of the corresponding grid is zero, and the additional security analysis value of the grid does not participate in the subsequent calculation of the final security analysis result.

[0047] If the sum is greater than zero, take the reciprocal after adding 1 to the sum as the additional security analysis result of the corresponding grid, and participate in the subsequent calculation of the final security analysis result.

[0048] Optionally, in an example of the above aspect, the model construction module establishes a safety level analysis model for different grid areas according to the safety analysis results of the grids in different regions, and conducts a safety level analysis on each grid area, including the following steps:

[0049] Obtain the historical detection data of different grid areas before failure and the historical detection data of different grid areas with long-term normal operation in different grid areas of different regions, and label the data before failure and normal data in the historical detection data.

[0050] The deep learning model trained by the edge data management node detects the area of the abnormal region of the image data, adds it to the corresponding historical data to replace the original image data, and calculates the safety analysis result of the grid of the historical data;

[0051] According to the historical data of different grid areas, the LSTM-full connection hybrid model is trained respectively to predict whether the safety analysis result of the grid is pre-fault data or normal data.

[0052] Optionally, in an example of the above aspect, according to the historical data of different grid areas, the LSTM-full connection hybrid model is trained respectively to predict whether the safety analysis result of the grid is pre-fault data or normal data, including the following steps:

[0053] Using the data of any time period in the historical data of different grid areas as input and the data of the corresponding next time period as output, the LSTM multi-modal time series model is trained respectively to predict the future detection data and the safety analysis result of the grid according to the historical data;

[0054] A feature processing layer and a connection layer are set at the output end of the LSTM multi-modal time series model. The edge data management node is called through the feature processing layer to calculate the safety analysis result of the future grid according to the prediction result of the future detection data, which is used as a feature value, and the labels of pre-fault data and normal data are reserved on the predicted value;

[0055] The safety analysis result feature value of the future grid and the safety analysis result predicted value output by the LSTM multi-modal time series model are input into the full connection classifier through the connection layer to predict whether the safety analysis result of the grid is pre-fault data or normal data of different grid areas.

[0056] Optionally, in an example of the above aspect, the machine dog scheduling module performs support scheduling on the machine dogs in different grid areas through the edge data management node according to the fault degree analysis result of the grid area and the detection progress of the grid in the grid area, including the following steps:

[0057] Obtain the fault degree analysis result of the grid area and the detection progress of the grid in the grid area;

[0058] If the fault degree analysis result of the grid area is normal data, no scheduling is performed on the machine dogs in different grid areas;

[0059] If the fault degree analysis result of the grid area is pre-fault data, it is detected whether the detection progress of the grid in the grid area is greater than 50%. If so, no scheduling is performed;

[0060] Otherwise, check the detection progress of the grid areas in the same region, and select the grid areas with normal analysis results and a detection progress greater than 50% as alternative grid areas. If there are no alternative grid areas, no scheduling will be performed;

[0061] If there are alternative grid areas, select the machine dog of the grid area closest to the grid area before the fault, and send a support scheduling command and the machine dog number to the edge data management node.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] The present invention can receive and process data from machine dogs in real time through the edge data management node, reduce the latency of data transmission, improve the response speed, and relieve the data processing pressure on the cloud by performing data processing and analysis at the edge, making the entire system more efficient; the edge data management node can perform security analysis based on the data collected by the machine dog, and timely discover and warn of potential security risks; since the data is processed on the edge side, the risk of data leakage during transmission is reduced, improving the security of the data.

[0064] Through the security level analysis model of the present invention, the security status of each grid area can be intuitively understood, which helps managers quickly identify potential risk areas, thereby reasonably allocating resources and energy. Through the edge data management node, support scheduling for machine dogs is performed according to the degree of failure and detection progress to ensure timely and effective monitoring of key areas. Through model analysis and scheduling of data nodes, the problem can be quickly located, and nearby machine dogs can be mobilized for support. According to the analysis results of the security level and failure degree of the grid area, the deployment and scheduling strategies of machine dogs are dynamically adjusted to avoid resource idleness or over-concentration. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0066] Figure 1 It is a schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0068] Please refer to Figure 1 , an embodiment of the first aspect of the present invention provides a traffic tunnel inspection system for a robotic dog platform based on AI technology, including: an edge end and a cloud management center;

[0069] The edge end includes:

[0070] Area allocation module: Obtain the channel data of the cross-sea channel area, divide the target area into a wind tower channel area, a pipe gallery channel area, and an external channel area, set robotic dogs that comply with the channel management rules of different areas in different areas, and divide grid areas with different detection types in different areas;

[0071] Data acquisition module: Set different data acquisition strategies for different grids. The robotic dogs in each grid area collect data according to the data acquisition strategies of different grid areas and send it to the edge data management node;

[0072] Edge data management node: Edge data management nodes are set at intervals of a safe communication distance in the cross-sea channel area; the edge data management node performs scheduling control by communicating with the robotic dogs within the safe communication distance, and analyzes the security of different types of grids based on the data collected by the robotic dogs in the grid area, and issues a security alarm for grids with a security analysis lower than the threshold;

[0073] The cloud management center includes:

[0074] Model construction module: Establish a security level analysis model for different grid areas based on the security analysis results of the grids in different areas, and analyze the security level of each grid area;

[0075] Robotic dog scheduling module: Support and schedule the robotic dogs in different grid areas through the edge data management node according to the fault degree analysis results of the grid areas and the detection progress of the grids in the grid areas.

[0076] Specifically, in this embodiment, the wind tower channel area: includes wind turbines, transmission tower bases, and connecting corridors, and deploy high-wind-speed-resistant robotic dogs;

[0077] Pipe gallery channel area: Inside the pipe gallery for accommodating pipelines or cables and the maintenance channel under the sea, deploy explosion-proof and waterproof robotic dogs;

[0078] External channel area: And the emergency channel, deploy all-terrain patrol robotic dogs;

[0079] Edge data management node, hardware configuration

[0080] Computing unit: NVIDIA Jetson AGX Xavier, 32 TOPS AI computing power;

[0081] Storage system: RAID 1 dual solid-state drives, 1TB×2, read / write speed 550MB / s,

[0082] Communication module: Supports 5G / satellite dual-link redundancy, uplink and downlink rate ≥1Gbps.

[0083] Edge data management nodes are set at every safe communication distance in the cross-sea channel area; the edge data management nodes communicate with the watchdog within the safe communication distance for scheduling and control, and based on the data collected by the watchdog in the grid area, perform security analysis on different types of grids, and issue security alerts for grids with security analysis below the threshold;

[0084] The edge data management node can receive and process data from the watchdog in real time, reducing data transmission latency and improving response speed. By performing data processing and analysis at the edge, the data processing pressure on the cloud is reduced, making the entire system more efficient. The edge data management node can perform security analysis based on the data collected by the watchdog, and promptly discover and warn of potential security risks. Since the data is processed at the edge side, the risk of data leakage during transmission is reduced, improving data security.

[0085] The setting of the edge data management node facilitates flexible adjustment according to the specific situation of the cross-sea channel to adapt to different safe communication distances and grid areas. With the development of technology and the growth of demand, it is convenient to add new edge data management nodes and watchdogs to expand the scale and functions of the system.

[0086] The model construction module establishes a security level analysis model for different grid areas based on the security analysis results of the grids in different regions, and performs security level analysis on each grid area; the watchdog scheduling module performs support scheduling on the watchdogs in different grid areas through the edge data management node according to the fault level analysis results of the grid area and the detection progress of the grids in the grid area.

[0087] Through the security level analysis model, the security status of each grid area can be intuitively understood, which helps managers quickly identify potential risk areas, so as to reasonably allocate resources and energy. By the edge data management node performing support scheduling on the watchdog according to the fault level and detection progress, it ensures that key areas are monitored in a timely and effective manner, improving the overall security management efficiency.

[0088] In case of a failure or a safety incident, through model analysis and the scheduling of data nodes, the location of the problem can be quickly identified, and nearby robotic dogs can be mobilized for support, shortening the emergency response time. Based on the analysis results of the safety level and failure degree of the grid area, the deployment and scheduling strategies of the robotic dogs are dynamically adjusted to avoid resource idleness or over-concentration. Through reasonable resource allocation, appropriate monitoring and support are ensured for each grid area, improving resource utilization efficiency.

[0089] Through continuous monitoring and scheduling, potential safety hazards can be detected and handled in a timely manner, reducing the probability of accidents and enhancing overall safety.

[0090] In one embodiment of the present invention, the area allocation module includes:

[0091] Area division unit: Obtain the channel data of the cross-sea channel area, including: channel detection area data, channel accident records, channel types, and channel map data. According to the channel detection area data and the channel map data, the channel connected to the wind tower is divided into the wind tower channel area, the channel accommodating various pipelines is divided into the pipe gallery channel area, and the channel accommodating the vehicle and pedestrian passage area is divided into the external channel area;

[0092] Robotic dog allocation unit: Set robotic dogs that comply with the channel management rules of different areas according to the detection area areas in the channel detection area data of the wind tower channel area, the pipe gallery channel area, and the external channel area;

[0093] Grid division unit: According to the channel data of each area, divide the equipment grid area, the channel grid area, and the wind tower monitoring grid area in the wind tower channel area, divide the channel grid area, the equipment grid area, and the pipe network detection grid area in the pipe gallery channel area, and divide the channel grid area and the equipment grid area in the external channel area.

[0094] Specifically, in this embodiment, robotic dogs that comply with the channel management rules of different areas are set according to the detection area areas in the channel detection area data of the wind tower channel area, the pipe gallery channel area, and the external channel area;

[0095] The wind tower channel area is usually relatively narrow and may have high wind speeds and wind direction changes.

[0096] Channel management rules for the wind tower channel area:

[0097] It is necessary to conduct inspections, maintenance, and fault troubleshooting on the wind tower equipment.

[0098] The robotic dog should have good stability and wind resistance.

[0099] The robotic dog is set as:

[0100] Select a machine dog with a compact structure and light weight to reduce the impact on wind resistance.

[0101] The machine dog is equipped with a high-definition camera, an infrared thermal imager sensor, a Doppler radar (wind speed), a temperature and humidity sensor, as well as a smoke sensor and a current and voltage detector for the equipment grid area, in order to conduct a detailed inspection of the wind tower equipment.

[0102] The pipe gallery passage area is usually located underground, with a long, narrow and enclosed space. There may be problems such as electromagnetic interference and difficulties in wireless signal transmission.

[0103] Pipe gallery passage area channel management rules:

[0104] It is necessary to conduct inspections and fault troubleshooting on facilities such as pipelines and cables.

[0105] The machine dog should have good wireless communication capabilities and anti-interference capabilities.

[0106] The machine dog is set to:

[0107] Select a machine dog with strong wireless communication capabilities and good anti-interference performance.

[0108] Equip with a high-performance wireless base station and a roaming-type airborne terminal to ensure wireless signal coverage and stable transmission in the pipe gallery passage.

[0109] The machine dog is equipped with a high-definition camera, an infrared thermal imager sensor, a temperature and humidity sensor, as well as a smoke sensor, a polarized light camera, and a current and voltage detector for the equipment grid area, in order to conduct real-time monitoring and data analysis on facilities such as pipelines and cables.

[0110] The external passage area is usually relatively open, but may face different climate conditions, as well as terrain conditions such as sandy, rocky, and snowy terrains.

[0111] External passage area channel management rules:

[0112] It is necessary to conduct inspections, monitoring, and emergency handling on external facilities.

[0113] The machine dog should have good driving capabilities and all-weather working capabilities.

[0114] The machine dog is set to:

[0115] Select a machine dog with tires or tracks to adapt to various road surfaces; it has waterproof, dustproof, and temperature-resistant properties to ensure normal operation under various conditions.

[0116] The machine dog is equipped with a high-definition camera, an infrared thermal imager sensor, a temperature and humidity sensor, as well as a Doppler radar for detecting wind speed, a salt fog concentration sensor, and a current and voltage detector for the equipment grid area, in order to improve the machine dog's environmental perception ability and positioning accuracy.

[0117] In one embodiment of the present invention, the data acquisition module sets different data acquisition strategies for different grids, including:

[0118] For all grid areas in the wind tower passage area, pipe gallery passage area, and external passage area, detect basic operation data, including: wind speed, temperature, humidity, high-definition image data, and infrared image data of the passage;

[0119] For the equipment grid areas in the wind tower passage area, pipe gallery passage area, and external passage area, the data acquisition strategy is:

[0120] Extraordinarily collect high-definition images and infrared images of the equipment in the equipment grid area, as well as infrared images and electrical parameters at the equipment connection points;

[0121] For the passage grid areas in the wind tower passage area and pipe gallery passage area, the data acquisition strategy is:

[0122] Extraordinarily collect the smoke condition of the passage in the passage grid area;

[0123] For the pipe network detection grid areas in the pipe gallery passage area, the data acquisition strategy is:

[0124] Extraordinarily collect high-definition images and polarization images at the cables and pipes in the pipe network detection grid area;

[0125] For the passage grid areas in the external passage area, the data acquisition strategy is:

[0126] Extraordinarily collect the salt fog concentration of the passage in the passage grid area.

[0127] In one embodiment of the present invention, the edge data management node performs scheduling control by communicating with the watchdog within the safe communication distance, including the following steps:

[0128] The edge data management node establishes a communication connection with the watchdog within the safe communication distance;

[0129] After receiving the watchdog scheduling control instruction between different grid areas in the same area sent by the watchdog scheduling module, send the data acquisition strategy of the corresponding block to the corresponding watchdog according to the scheduling control instruction, and send the instruction to go to the corresponding block for data acquisition.

[0130] In one embodiment of the present invention, the edge data management node performs security analysis on different types of grids according to the data collected by the watchdog in the grid area, and issues a security alarm for the grids with security analysis lower than the threshold, including the following steps:

[0131] Obtain the high-definition image data, infrared image data, and polarization image with surface anomalies, as well as the high-definition image data, infrared image data, and polarization image without surface anomalies in the historical detection data of the cross-sea channel, and mark the positions of surface anomalies;

[0132] Train a deep learning model with the image data of the historical detection data to identify and mark the abnormal positions of the high-definition image data, infrared image data, and polarization image, and calculate the area of the abnormal region;

[0133] Obtain the high-definition image data, infrared image data, and polarization image when each region is operating normally in the historical data. Through the trained deep learning model, detect the area of the abnormal region and take the average value as the maximum abnormal fault area threshold of the high-definition image data, infrared image data, and polarization image;

[0134] For all grid areas in the wind tower channel area, pipe gallery channel area, and external channel area, detect the basic operation data in the preset detection time interval, and conduct grid basic safety analysis through the following formula:

[0135]

[0136] Where S is the grid basic safety analysis value, Sv is the grid wind speed safety analysis value, Sth is the grid temperature and humidity safety analysis value, Sp is the grid image anomaly safety analysis value, w1, w2, and w3 are the corresponding weights of the grid wind speed safety analysis value, grid temperature and humidity safety analysis value, and grid image anomaly safety analysis value respectively; v is the average wind speed of the corresponding grid in the preset detection time interval, v0 is the abnormal wind speed threshold of the corresponding grid area, T0 is the average temperature of the corresponding grid in the preset detection time interval, H0 is the average humidity of the corresponding grid in the preset detection time interval, σt is the standard deviation of the temperature of the corresponding grid in the preset detection time interval, σh is the standard deviation of the humidity of the corresponding grid in the preset detection time interval, fai, fbi, and fci are the abnormal areas detected by the high-definition image data, infrared image data, and polarization image in the grid in the preset detection time interval. If no anomaly is detected or the corresponding grid does not have the corresponding detection data, the corresponding abnormal area is set to zero, and famax, fbmax, and fcmax are the maximum abnormal fault area thresholds detected by the high-definition image data, infrared image data, and polarization image;

[0137] According to the grid basic safety analysis results, conduct additional safety analysis for different regions for different regions;

[0138] Perform a weighted average of the grid basic safety analysis results and the additional safety analysis to obtain the final safety analysis result of the grid, and issue a safety alert for grids with a safety analysis lower than the threshold.

[0139] Specifically, in this embodiment, for the grids that only detect basic operation data, the basic grid security analysis result is used as the final security analysis result of the grid. For the grids that conduct additional security analysis, the basic grid security analysis result and the additional security analysis are weighted and averaged. In this embodiment, by statistically analyzing the data of a large number of well - operating grid areas and faulty grid areas, basic security analysis and additional security analysis are carried out. The weight of the basic grid security analysis result is set to 0.75, and the weight of the additional security analysis result is set to 0.25. At the same time, through experiments, the final security analysis result of the grid is obtained. The security analysis results of well - operating grid areas are all greater than 0.75. Therefore, the threshold for security analysis is set to 0.75.

[0140] At the same time, in this embodiment, for the grids in the wind tower passage area, w1, w2, and w3 are set to 0.4, 0.3, and 0.3 respectively; for the grids in the pipe gallery passage area, w1, w2, and w3 are set to 0.2, 0.5, and 0.3 respectively; for the grids in the external passage area, w1, w2, and w3 are set to 0.2, 0.3, and 0.5 respectively.

[0141] In one embodiment of the present invention, for different regions, different - region additional security analysis is carried out, including the following steps:

[0142] For the equipment grid areas in each region, obtain the high - definition images and infrared images of the equipment in the additionally collected equipment grid areas, as well as the infrared images and electrical parameters at the equipment connection points, and conduct additional security analysis of the equipment through the following formula:

[0143]

[0144] Among them, Sep is the additional security analysis value of the equipment, fej is the area of the abnormal region in the j - th image data additionally collected in the equipment grid area of each region, femaxj is the maximum abnormal fault area threshold corresponding to the j - th image type of the additionally collected image data, σek is the standard deviation of the k - th electrical parameter additionally collected in the equipment grid area of each region during the preset detection time interval, and Tek is the mean value of the k - th electrical parameter additionally collected in the equipment grid area of each region during the preset detection time interval;

[0145] For the passage grid areas in the wind tower passage area and the pipe gallery passage area, obtain the smoke condition in the additionally collected passage grid areas. If smoke is detected, immediately send a safety alarm signal to the corresponding grid; otherwise, do not send a safety alarm signal.

[0146] For the pipe network detection grid area in the utility tunnel passage area, obtain high-definition images and polarization images at the cables and pipelines in the pipe network detection grid area collected additionally, and calculate the additional safety analysis values of the cables and pipelines according to the same method as the calculation method of the image anomaly safety analysis value of the basic operation data grid;

[0147] For the passage grid area in the external passage area, obtain the salt mist concentration in the passage in the passage grid area collected additionally, and calculate the ratio of the average value of the detected salt mist concentration within the preset detection time interval to the abnormal salt mist concentration threshold as the additional safety analysis value of the salt mist;

[0148] Add up all the additional safety analysis values calculated in each grid. If the sum is zero, the additional safety analysis result of the corresponding grid is zero, and the additional safety analysis value of the grid does not participate in the subsequent calculation of the final safety analysis result;

[0149] If the sum is greater than zero, take the reciprocal after adding 1 to the sum as the additional safety analysis result of the corresponding grid, and participate in the subsequent calculation of the final safety analysis result.

[0150] In one embodiment of the present invention, the model construction module establishes a safety degree analysis model for different grid areas according to the safety analysis results of the grids in different areas, and conducts safety degree analysis on each grid area, including the following steps:

[0151] Obtain the historical detection data of different grid areas before failure and the historical detection data of different grid areas operating normally for a long time in different grid areas of different regions, and label the pre-failure data and normal data in the historical detection data;

[0152] Detect the area of the abnormal area of the image data through the deep learning model trained by the edge data management node, add it to the corresponding historical data to replace the original image data, and calculate the safety analysis result of the grid of the historical data;

[0153] Train the LSTM-full connection hybrid model respectively according to the historical data of different grid areas to predict whether the safety analysis result of the grid is pre-failure data or normal data.

[0154] In one embodiment of the present invention, training the LSTM-full connection hybrid model respectively according to the historical data of different grid areas to predict whether the safety analysis result of the grid is pre-failure data or normal data includes the following steps:

[0155] Using the data for any time period in the historical data of different grid areas as input and the corresponding data for the subsequent time period as output, train the LSTM multi-modal time series model respectively, and predict the future detection data and the safety analysis results of the grid based on the historical data;

[0156] A feature processing layer and a connection layer are provided at the output end of the LSTM multi-modal time series model. Through the feature processing layer, the edge data management node is called to calculate the safety analysis results of the future grid based on the prediction results of the future detection data, which are used as feature values, and the labels of the pre-fault data and the normal data are retained on the estimated values;

[0157] The feature values of the safety analysis results of the future grid and the predicted values of the safety analysis results output by the LSTM multi-modal time series model are input into the fully connected classifier through the connection layer to predict whether the safety analysis results of the grid are pre-fault data or normal data for different grid areas.

[0158] In one embodiment of the present invention, the machine dog scheduling module performs support scheduling for the machine dogs in different grid areas through the edge data management node according to the fault degree analysis results of the grid areas and the detection progress of the grids in the grid areas, including the following steps:

[0159] Obtain the fault degree analysis results of the grid areas and the detection progress of the grids in the grid areas;

[0160] If the fault degree analysis result of the grid area is normal data, no scheduling is performed for the machine dogs in different grid areas;

[0161] If the fault degree analysis result of the grid area is pre-fault data, check whether the detection progress of the grid in the grid area is greater than 50%. If so, no scheduling is performed;

[0162] If not, check the detection progress of the grid areas in the same area, select the grid area with the analysis result of normal data and the detection progress greater than 50% as the alternative grid area. If there is no alternative grid area, no scheduling is performed;

[0163] If there is an alternative grid area, select the machine dog of the grid area closest to the pre-fault grid area, and send a support scheduling command and the machine dog number to the edge data management node.

[0164] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A traffic tunnel inspection system based on a robot dog platform based on AI technology, characterized in that: include: Edge and cloud management centers; The edge end includes: Area allocation module: obtains channel data of the cross-sea channel area, divides the target area into wind tower channel area, pipe gallery channel area and external channel area, and sets robot dogs that comply with channel management rules of different areas in different areas, and divides grid areas for different detection types in different areas; Data collection module: Set different data collection strategies for different grids. The robot dogs in each grid area collect data according to the data collection strategies of different grid areas and send them to the edge data management node; Edge data management node: an edge data management node is set at each safe communication distance in the cross-sea channel area; the edge data management node performs scheduling control by communicating with the robot dog within the safe communication distance, and performs security analysis on different types of grids based on the data collected by the robot dog in the grid area, and issues security alarms for grids whose security analysis is below the threshold; The cloud management center includes: Model building module: Based on the safety analysis results of grids in different areas, safety degree analysis models for different grid areas are established, and safety degree analysis is performed on each grid area; Robot dog scheduling module: According to the fault degree analysis results of the grid area and the detection progress of the grid in the grid area, support scheduling of robot dogs in different grid areas is carried out through the edge data management node.

2. According to the AI-based robot dog platform traffic tunnel inspection system of claim 1, it is characterized in that: Area allocation module, including: Area division unit: obtain channel data of the cross-sea channel area, including: channel detection area data, channel accident records, channel type and channel map data; divide the channel connected to the wind tower into the wind tower channel area, divide the channel accommodating various pipelines into the pipeline corridor channel area, and divide the channel accommodating the pedestrian and vehicle traffic area into the external channel area according to the channel detection area data and channel map data; Robot dog allocation unit: according to the detection area area in the channel detection area data of the wind tower channel area, the pipe gallery channel area and the external channel area, robot dogs that comply with the channel management rules of different areas are set in different areas; Grid division unit: According to the channel data of each area, the wind tower channel area is divided into equipment grid area, channel grid area and wind tower monitoring grid area; the corridor channel area is divided into channel grid area, equipment grid area and pipeline detection grid area; the external channel area is divided into channel grid area and equipment grid area.

3. According to the AI-based robot dog platform traffic tunnel inspection system of claim 1, it is characterized in that: The data collection module sets different data collection strategies for different grids, including: For all grid areas in the wind tower channel area, pipe gallery channel area and external channel area, basic operation data are tested, including: wind speed, temperature, humidity, high-definition image data and infrared image data of the channel; For the equipment grid area in the wind tower channel area, pipe gallery channel area and external channel area, the data collection strategy is as follows: Additionally collect high-definition images and infrared images of the equipment in the equipment grid area, as well as infrared images and electrical parameters of the equipment wiring points; For the channel grid area of ​​the wind tower channel area and the pipeline corridor channel area, the data collection strategy is: Additional collection of smoke conditions in the channel grid area; For the pipe network detection grid area in the pipe gallery passage area, the data collection strategy is: Additional collection of high-definition images and polarization images of cables and pipelines in the pipeline network inspection grid area; For the channel grid area in the external channel region, the data collection strategy is: The salt spray concentration of the channel in the channel grid area is additionally collected.

4. According to the AI-based robot dog platform traffic tunnel inspection system of claim 1, it is characterized in that: The edge data management node performs scheduling control by communicating with the robot dog within the safe communication distance, including the following steps: The edge data management node establishes a communication connection with the robot dog within a safe communication distance; After receiving the robot dog scheduling control instruction between different grid areas in the same area sent by the robot dog scheduling module, the data collection strategy of the corresponding block is sent to the corresponding robot dog according to the scheduling control instruction, and an instruction to go to the corresponding block for data collection is sent.

5. The AI-based robot dog platform traffic tunnel inspection system according to claim 1 is characterized in that: The edge data management node performs security analysis on different types of grids based on the data collected by the robot dog in the grid area, and issues a security alert for grids whose security analysis is below a threshold, including the following steps: Obtain high-definition image data, infrared image data and polarization images with surface anomalies in the historical detection data of the cross-sea channel, as well as high-definition image data, infrared image data and polarization images without surface anomalies, and mark the locations of surface anomalies; The deep learning model is trained by using image data from historical detection data to identify and annotate abnormal locations in high-definition image data, infrared image data, and polarization images, and calculate the area of ​​the abnormal region; Obtain high-definition image data, infrared image data, and polarization image data of each area in normal operation in historical data, detect the area of ​​the abnormal area through the trained deep learning model, and take the average value as the maximum abnormal fault area threshold of high-definition image data, infrared image data, and polarization image; For all grid areas in the wind tower channel area, pipe gallery channel area and external channel area, basic operation data is tested during the preset test time interval, and basic grid safety analysis is performed using the following formula: Among them, S is the basic safety analysis value of the grid, Sv is the safety analysis value of the grid wind speed, Sth is the safety analysis value of the grid temperature and humidity, Sp is the safety analysis value of the grid image anomaly, w1, w2 and w3 are the corresponding weights of the grid wind speed safety analysis value, the grid temperature and humidity safety analysis value and the grid image anomaly safety analysis value respectively; v is the mean wind speed of the corresponding grid in the preset detection time interval, v0 is the abnormal wind speed threshold of the corresponding grid area, T0 is the mean temperature of the corresponding grid in the preset detection time interval, H0 is the mean humidity of the corresponding grid in the preset detection time interval, σt is the standard deviation of the temperature of the corresponding grid in the preset detection time interval, σh is the standard deviation of the humidity of the corresponding grid in the preset detection time interval, fai, fbi and fci are the abnormal areas detected by the high-definition image data, infrared image data and polarization image in the grid in the preset detection time interval. If no abnormality is detected or there is no corresponding detection data for the corresponding grid, the corresponding abnormal area is set to zero, famax, fbmax and fcmax are the maximum abnormal fault area thresholds for high-definition image data, infrared image data and polarization image detection; Based on the results of the basic safety analysis of the grid, additional safety analysis is conducted for different areas; The basic safety analysis results and the additional safety analysis results of the grid are weighted averaged to obtain the final safety analysis result of the grid, and a safety alarm is issued for grids whose safety analysis is below the threshold.

6. The AI-based robot dog platform traffic tunnel inspection system according to claim 5 is characterized in that: For different regions, additional safety analysis is performed in different regions, including the following steps: For the equipment grid area in each region, obtain high-definition images and infrared images of the equipment in the additional equipment grid area, as well as infrared images and electrical parameters of the equipment wiring points, and perform additional safety analysis of the equipment using the following formula: Wherein, Sep is the additional safety analysis value of the equipment, fej is the area of ​​the abnormal area in the j-th image data additionally collected in the equipment grid area of ​​each region, femaxj is the maximum abnormal fault area threshold of the image type corresponding to the j-th image data additionally collected, σek is the standard deviation of the k-th electrical parameter additionally collected in the equipment grid area of ​​each region in the preset detection time interval, and Tek is the mean value of the k-th electrical parameter additionally collected in the equipment grid area of ​​each region in the preset detection time interval; For the channel grid areas of the wind tower channel area and the pipe gallery channel area, obtain the smoke status of the channel in the additional collection channel grid area. If smoke is detected, immediately send a safety alarm signal to the corresponding grid, otherwise, do not send a safety alarm signal; For the pipe network inspection grid area in the pipe gallery passage area, obtain high-definition images and polarization images of cables and pipelines in the additional collection pipe network inspection grid area, and calculate the additional safety analysis values ​​of cables and pipelines in the same way as the calculation method of image anomaly safety analysis values ​​of the basic operation data grid; For the channel grid area in the external channel area, the salt spray concentration of the channel in the additional acquisition channel grid area is obtained, and the ratio of the average value of the detected salt spray concentration within the preset detection time interval to the abnormal salt spray concentration threshold is calculated as the additional safety analysis value of the salt spray; All the additional safety analysis values ​​calculated in each grid are added together. If the sum is zero, the additional safety analysis result of the corresponding grid is zero, and the additional safety analysis value of the grid does not participate in the subsequent calculation of the final safety analysis result; If the added result is greater than zero, the result of the addition is added by 1, and the reciprocal is taken as the additional safety analysis result of the corresponding grid, which is used in the subsequent calculation of the final safety analysis result.

7. The AI-based robot dog platform traffic tunnel inspection system according to claim 5 is characterized in that: The model building module establishes safety degree analysis models for different grid areas according to the safety analysis results of grids in different areas, and performs safety degree analysis on each grid area, including the following steps: Obtain historical detection data of different grid areas in different regions before a fault occurs, as well as historical detection data of different grid areas that have been operating normally for a long time, and annotate the historical detection data with pre-fault data and normal data; Through the deep learning model trained by the edge data management node, the area of ​​the abnormal area of ​​the image data is detected, added to the corresponding historical data to replace the original image data, and the security analysis results of the grid of the historical data are calculated; According to the historical data of different grid areas, the LSTM-fully connected hybrid model is trained separately to predict whether the safety analysis results of the grid are pre-fault data or normal data.

8. The AI-based robot dog platform traffic tunnel inspection system according to claim 7 is characterized in that: According to the historical data of different grid areas, the LSTM-fully connected hybrid model is trained separately to predict the safety analysis results of the grid as pre-fault data or normal data, including the following steps: Through the historical data of different grid areas, the data of any time period is used as input, and the data of the corresponding next time period is used as output. The LSTM multimodal time series model is trained separately to predict the future detection data and grid security analysis results based on the historical data; The output end of the LSTM multimodal time series model sets a feature processing layer and a connection layer. The feature processing layer calls the edge data management node to calculate the future grid safety analysis results as feature values ​​based on the prediction results of future detection data, and retains the annotations of pre-fault data and normal data on the estimated value; The characteristic values ​​of the future grid safety analysis results and the predicted values ​​of the safety analysis results output by the LSTM multimodal time series model are input into the fully connected classifier through the connection layer, and the safety analysis results of the grid are predicted as pre-fault data or normal data of different grid areas.

9. The AI-based robot dog platform traffic tunnel inspection system according to claim 1 is characterized in that: The robot dog scheduling module performs support scheduling for robot dogs in different grid areas through edge data management nodes according to the fault degree analysis results of the grid area and the detection progress of the grid in the grid area, including the following steps: Obtain the results of the fault degree analysis of the grid area and the detection progress of the grids in the grid area; If the fault degree analysis result of the grid area is normal data, the robot dogs in different grid areas will not be scheduled; If the result of the fault degree analysis in the grid area is the data before the fault, then check whether the detection progress of the grid in the grid area is greater than 50%. If so, no scheduling is performed; If not, the detection progress of the grid areas in the same area is tested, and the grid areas with normal analysis results and detection progress greater than 50% are selected as candidate grid areas. If there are no candidate grid areas, no scheduling is performed; If there are alternative grid areas, the robot dog in the grid area closest to the grid area before the failure is selected, and the support scheduling command and the robot dog number are sent to the edge data management node.

Images