Intelligent inspection system, method and equipment and storage medium

Abstract

The invention discloses an intelligent inspection system, method and device and a storage medium, and the system comprises the steps: an inspection robot dog inspects a monitoring area according to a preset inspection path, carries out the equipment abnormality detection and gas leakage detection through the inspection data collected by a sensor, and uploads a detection result and the inspection data to a large model service platform; when an abnormal condition exists in the detection area, the large model service platform generates an inspection task according to a detection result; dividing the detection area into a plurality of sub-areas, calculating the risk levels of the sub-areas according to the inspection data, determining whether to perform risk early warning, constructing a risk thermodynamic diagram according to the risk levels, and performing inspection path optimization according to the risk thermodynamic diagram and the inspection task. And the optimized inspection path and inspection task are sent to an inspection robot dog for execution. According to the inspection robot dog, multi-task processing can be realized, and the inspection quality is improved; and the inspection task and the inspection path are optimized through the large model service platform, so that the inspection efficiency is improved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent patrol inspection, and particularly relates to an intelligent patrol inspection system, method, device, and storage medium. Background Art

[0002] Traditional industrial patrol inspection methods mainly rely on manual labor or fixed sensors, and there are many significant problems. Manual patrol inspection is inefficient, limited by the physical strength and energy of personnel, with a limited coverage area, and it is difficult to achieve full - range and real - time monitoring. In high - risk scenarios, personnel face huge safety risks. For example, pipeline leaks in chemical industrial parks may lead to the leakage of toxic gases, explosions, etc., and sudden changes in gas concentration in mines are also extremely likely to trigger explosion accidents.

[0003] Although existing mobile robots can assist in patrol inspection to a certain extent, they have deficiencies in complex environment adaptability, multi - task processing ability, and real - time decision - making accuracy. For example, in the complex terrain and equipment layout of chemical industrial parks, robots may have difficulty accurately identifying obstacles and equipment abnormalities; for multiple tasks, such as simultaneously conducting gas detection and equipment status inspection, the processing ability is limited; in the face of emergencies, the accuracy and timeliness of real - time decision - making cannot meet the dynamic requirements of high - risk scenarios. Intelligent robot dogs have also been applied to a certain extent in the patrol inspection tasks of dangerous scenarios, but most of them have technical limitations and single functions. Existing robot dogs mainly rely on sensors for environmental perception and task execution, usually rely on preset paths for patrol inspection, lack in - depth analysis of patrol inspection data and intelligent decision - making support, resulting in low task execution efficiency. Summary of the Invention

[0004] This application provides an intelligent patrol inspection system, method, device, and storage medium, which are used to improve the technical problems in the prior art that robot dogs have single functions, rely on preset paths for patrol inspection, lack in - depth analysis of patrol inspection data and intelligent decision - making support, resulting in low task execution efficiency.

[0005] In view of this, the first aspect of this application provides an intelligent patrol inspection system, including: a patrol inspection robot dog and a large - model service platform;

[0006] The patrol inspection robot dog is used to patrol and monitor an area according to a preset patrol path, detect equipment abnormalities and gas leaks through patrol inspection data collected by sensors, and upload the obtained detection results and the patrol inspection data to the large - model service platform;

[0007] The large model service platform is used to generate inspection tasks according to the detection results when there are abnormal situations in the detection area; divide the detection area into multiple sub-areas, calculate the risk level of each sub-area according to the inspection data, determine whether to issue a risk warning according to the risk level, construct a risk heat map according to the risk level, optimize the inspection path according to the risk heat map and the inspection tasks, and send the optimized inspection path and the inspection tasks to the inspection robot dog for execution.

[0008] Optionally, the inspection data includes video data and infrared thermal imaging images of the equipment;

[0009] When the inspection robot dog is used for equipment anomaly detection based on the inspection data, it specifically includes:

[0010] Process the video data through an image recognition method to detect whether there are appearance anomalies in the equipment and obtain the equipment appearance detection results;

[0011] Construct a Gaussian mixture model for the temperature values of the pixel points in the infrared thermal imaging image;

[0012] Initialize the weights, means, and variances of the Gaussian distributions in the Gaussian mixture model, and calculate the responsiveness of each Gaussian distribution to the temperature values of each pixel point;

[0013] Update the weights, means, and variances of the Gaussian distributions through the responsiveness of each Gaussian distribution to the temperature values of each pixel point;

[0014] Sort the Gaussian distributions in descending order according to the ratio of the weight to the standard deviation, select the first target number of Gaussian distributions to calculate the probability that the temperature value of each pixel point in the infrared thermal imaging image belongs to the normal temperature. If the probability is less than the preset threshold, determine that the pixel point is an overheated anomaly point.

[0015] Optionally, the inspection data includes gas concentration;

[0016] When the inspection robot dog is used for gas leakage detection based on the inspection data, it specifically includes:

[0017] Calculate the gas concentration gradient according to the gas concentrations detected by gas detection sensors at various positions;

[0018] Construct the relationship between the gas concentration at each position of the gas detection sensor and the position of the leakage source through a gas Gaussian diffusion model to obtain the position likelihood function of the leakage source;

[0019] Take the logarithm of the position likelihood function of the leakage source to obtain the position log-likelihood function of the leakage source;

[0020] Maximize the logarithmic likelihood function of the position of the leakage source through an optimization method, and solve for the position and intensity of the leakage source.

[0021] Optionally, the body of the inspection robot dog is made of aluminum alloy and an insulating coating;

[0022] The inspection robot dog uses rubber or silicone rubber sealing rings to fill the connection between the equipment interface and the connecting wire plug;

[0023] The junction box of the inspection robot dog is made of aluminum alloy. The surface of the junction box is subjected to anti-corrosion treatment. The inlet of the junction box is equipped with sealing clay and a sealing stuffing box. After the wire penetrates into the inlet, tighten the nut on the sealing stuffing box, and the sealing clay tightly wraps the wire.

[0024] Optionally, the large model service platform is also used to determine the priority of the inspection task, specifically including:

[0025] Train a reinforcement learning model through historical inspection data and historical inspection tasks; predict the priority of the inspection task according to the current inspection data and the inspection task through the reinforcement learning model;

[0026] Among them, the training process of the reinforcement learning model includes:

[0027] Construct a state space and an action space. The state space includes the gas concentration, equipment temperature, list of inspected equipment, and list of uninspected equipment in the current environment; the action space includes executable inspection tasks;

[0028] Calculate the reward value for each action according to the current state; update the network parameters through the reward value.

[0029] Optionally, the large model service platform is also used to compare the environmental data collected in real time by the sensor in the preset environment with the standard value of the environmental data in the database, calculate the data acquisition deviation value of the sensor by the least squares method; adjust the calibration parameters of the sensor through the data acquisition deviation value.

[0030] Optionally, the large model service platform is also used to obtain the operation data of the inspection robot dog in real time, perform fault identification on the operation data through a fault diagnosis model, and determine whether there is a fault in the inspection robot dog.

[0031] The second aspect of this application provides an intelligent inspection method, including:

[0032] The inspection robot dog inspects the monitoring area according to the preset inspection path, performs equipment anomaly detection and gas leakage detection through the inspection data collected by the sensor, and uploads the obtained detection results and the inspection data to the large model service platform;

[0033] When there is an abnormal situation in the detection area, the large model service platform generates an inspection task according to the detection result; divides the detection area into multiple sub-areas, calculates the risk level of each sub-area according to the inspection data, determines whether to issue a risk warning according to the risk level, constructs a risk heat map according to the risk level, optimizes the inspection path according to the risk heat map and the inspection task, and sends the optimized inspection path and the inspection task to the inspection robot dog for execution.

[0034] The third aspect of the present application provides an electronic device, which includes a processor and a memory;

[0035] The memory is used to store program code and transmit the program code to the processor;

[0036] The processor is used to execute the intelligent inspection method described in the second aspect according to the instructions in the program code.

[0037] The fourth aspect of the present application provides a computer-readable storage medium, which is used to store program code, and when the program code is executed by a processor, it implements the intelligent inspection method described in the second aspect.

[0038] From the above technical solutions, it can be seen that the present application has the following advantages:

[0039] In the intelligent inspection system provided by the present application, after the inspection robot dog obtains the inspection data, it can perform equipment abnormality detection and gas leakage detection simultaneously, can achieve multi-task processing, and improves the inspection quality; through further analysis and processing of the inspection data by the large model service platform, the inspection task and the inspection path are optimized, and the inspection efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application 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 application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0041] Figure 1 It is a structural schematic diagram of an intelligent inspection system provided by an embodiment of the present application;

[0042] Figure 2 It is a flowchart of an intelligent inspection method provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] To enable those skilled in the art to better understand the solution of this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0044] For ease of understanding, please refer to Figure 1 , this application embodiment provides an intelligent inspection system, including: an inspection robot dog and a large model service platform;

[0045] The inspection robot dog is used to inspect the monitoring area according to the preset inspection path, perform equipment anomaly detection and gas leakage detection on the inspection data collected by the sensor, and upload the obtained detection results and inspection data to the large model service platform;

[0046] The large model service platform is used to generate an inspection task according to the detection result when there is an abnormal situation in the detection area; divide the detection area into multiple sub-areas, calculate the risk level of each sub-area according to the inspection data, determine whether to issue a risk warning according to the risk level, and construct a risk heat map according to the risk level, optimize the inspection path according to the risk heat map and the inspection task, and send the optimized inspection path and inspection task to the inspection robot dog for execution.

[0047] The inspection robot dog in this application integrates a binocular camera, an infrared thermal imager, and an ultrasonic sensor. The inspection robot dog inspects the inspection area according to the preset inspection path and inspection task (which can be an inspection task pre-set by the staff). During the inspection, it collects inspection data (including video data, infrared thermal imaging images, and ultrasonic waves, etc.) through these sensors. The binocular camera can collect rich image and video data, and identify the appearance anomalies of the equipment through image recognition methods, such as cracks in the tank body and loose valves. The infrared thermal imaging sensor can detect the temperature distribution on the surface of the equipment and timely discover potential faults such as overheating of the equipment. The ultrasonic sensor is used for close-range obstacle detection to provide more accurate distance information.

[0048] The inspection robot dog obtains the temperature distribution image (i.e., the infrared thermal imaging image) of the equipment surface through the infrared thermal imager, and then performs overheating detection on the equipment based on an improved Gaussian mixture model. The traditional Gaussian mixture model has certain limitations in dealing with complex backgrounds and dynamic scenes. An adaptive learning rate and an online update mechanism are introduced to adapt to the temperature changes during the operation of the equipment.

[0049] Let the temperature value of each pixel point in the infrared thermal imaging image be T, and assume that this temperature value follows a mixture of K Gaussian distributions, that is:

[0050]

[0051] In the formula, is the weight of the th Gaussian distribution, satisfying , and ; is a Gaussian distribution with a mean of and a variance of , and its probability density function is:

[0052]

[0053] In the initial stage, the parameters , and of each Gaussian distribution are randomly initialized. is initialized to , is initialized to a random sample of the temperature values in the infrared thermal imaging image, is initialized to a value within a preset temperature range, covering a wide temperature range.

[0054] After obtaining each frame of the infrared thermal imaging image, the model parameters of the Gaussian mixture model can be updated using the expectation-maximization method. In the embodiments of the present application, first, the responsiveness of each Gaussian distribution to the temperature value of the current pixel is calculated:

[0055]

[0056] Then the weight is updated:

[0057]

[0058] In the formula, is the adaptive learning rate, which is dynamically adjusted according to the matching degree between the current pixel and the established model, is the total number of pixels in the infrared thermal imaging image. The mean and the variance are updated as follows:

[0059]

[0060] In the formula, is the adaptive learning rate, similar to , and is adjusted according to the actual situation.

[0061] For the temperature value of each pixel, calculate the probability that it belongs to the background (normal temperature range), that is, the sum of the probabilities of several Gaussian distributions with large weights and small variances among all Gaussian distributions. If is less than the preset threshold , it is determined that there is an overheating abnormality in the device part corresponding to this pixel point. Arrange all Gaussian distributions in descending order, and select the first Gaussian distributions ( determined according to the actual situation), and calculate :

[0062]

[0063] When , mark this pixel point as an overheating abnormality point. Based on the judgment of all pixel points in the entire infrared thermal imaging image, it is possible to determine whether there is an overheating area in the device, as well as the location and scope of the overheating area.

[0064] In this application, gas detection sensors such as micro spectrometers, electrochemical sensors, and photoionization sensors are deployed in each monitoring area to monitor the concentrations of various gases such as methane, CO, and H2S in real time. The micro spectrometer analyzes the absorption characteristics of gases for light of specific wavelengths to achieve high-precision detection of gas components and concentrations; the electrochemical sensor measures the gas concentration using the electrical signal generated by a chemical reaction; the photoionization sensor detects gases such as volatile organic compounds (VOCs) by ionizing gas molecules. The inspection robot dog obtains the gas concentrations detected by the gas detection sensors in real time for gas leakage detection. The gas leakage detection based on the gas concentration specifically includes:

[0065] Assume that there are N sensors deployed in the detection area, and their position coordinates are respectively , . The gas concentrations measured at these positions are respectively . The components of the concentration gradient vector in the , , directions can be calculated by differential approximation:

[0066] (when is adjacent to in the direction);

[0067] (when is adjacent to in the direction);

[0068] (when is adjacent to in the when they are adjacent in the direction);

[0069] In the formula, , , are respectively the gas concentrations detected by the sensors adjacent to at the corresponding adjacent positions in the direction. The direction of the gas concentration gradient points to the leakage source.

[0070] Analyze the relationship between the measured values of the gas detection sensors and the leakage source position and intensity based on the likelihood function. Let the leakage source position be , and the leakage source intensity be . Then the relationship between the gas concentration at the position of the gas detection sensor and the leakage source can be obtained through the relevant gas Gaussian diffusion model:

[0071]

[0072] In the formula, is the average wind speed (unit: ); and are the diffusion parameters in the horizontal and vertical directions (unit: m), respectively, which are related to the distance and meteorological conditions.

[0073] The position likelihood function of the leakage source is:

[0074]

[0075] In the formula, is the actually measured gas concentration at the position of the gas detection sensor , is the theoretical gas concentration calculated based on the leakage source position and leakage source intensity, is the measurement noise standard deviation of the th gas detection sensor (unit: ppm), reflecting the uncertainty of the sensor measurement. Different types of sensors have different measurement noise characteristics, which can be determined through experimental calibration.

[0076] Take the logarithm of the position likelihood function of the leakage source to obtain the position log-likelihood function of the leakage source:

[0077]

[0078] Maximize the position log-likelihood function of the leakage source through optimization methods (such as Newton-Raphson method, quasi-Newton method, etc.) to solve the following equations:

[0079]

[0080] ​

[0081]

[0082]

[0083] The obtained solution is the estimated value of the leakage source intensity and the leakage source location.

[0084] When the patrol robot dog detects the leakage of dangerous gases or equipment abnormalities, it can respond and give early warnings in a timely manner.

[0085] Furthermore, the body of the patrol robot dog in the embodiment of the present application is made of an aluminum alloy and insulating coating composite material. The aluminum alloy has good strength and heat dissipation performance, while the insulating coating can effectively prevent static electricity generation and avoid explosion caused by electric sparks. The circuit system of the robot dog is fully sealed and uses electrical components and packaging processes with explosion protection grades compliant with ATEX / IECEx certifications to achieve safe operation in flammable and explosive environments.

[0086] The interfaces and connection wires of the patrol robot dog equipment are also subject to special explosion protection treatments to prevent external dangerous gases from entering the equipment interior. The explosion protection treatments for the interfaces and connection wires are as follows:

[0087] The interface is designed with a seal. A rubber or silicone rubber sealing ring with high strength and good sealing performance is used and filled at the connection between the equipment interface and the connection wire plug. For example, a fluororubber sealing ring is used, which has excellent tolerance to various chemical substances, a wide working temperature range, and maintains good sealing performance between -40°C and 200°C.

[0088] The patrol robot dog uses an explosion-proof junction box compliant with ATEX / IECEx certifications. The internal space of the junction box is reasonably designed to accommodate wire connection points and has sufficient electrical clearances and creepage distances to prevent electric sparks caused by poor electrical connections. The junction box is made of high-strength aluminum alloy material, and its surface is treated with anti-corrosion. Its inlet is equipped with sealant clay and a sealing stuffing box. When the wire penetrates, the nut on the sealing stuffing box is tightened, and the sealant clay tightly wraps the wire to achieve a good sealing and explosion-proof effect.

[0089] The large model service platform in the embodiments of this application can deploy open-source large models, such as DeepSeek, LLaMA, etc. models, to jointly analyze environmental text (such as pipeline markings), images, and gas data. Based on the improved DeepSeek and LLaMA models, it understands and processes text information in the environment, such as identifying markings on pipelines, operation instructions of equipment, etc.; adopts a multimodal data fusion analysis method to generate the optimal patrol path and patrol task priority, such as determining the patrol order of the robot dog and the key inspection areas according to the importance of the equipment, historical failure records, and current abnormal situations, improving the patrol efficiency.

[0090] The large model service platform obtains environmental data of the detection area, including equipment location, pipeline layout, potential danger areas, etc., and constructs a graph model based on the environmental data. In the graph model, nodes represent equipment or detection points, and edges represent the connection paths between nodes. The weights of the edges can represent factors such as path length and passage difficulty. The large model service platform, based on the graph model, adopts a classic path planning algorithm (such as Dijkstra's algorithm), starting from the starting node, and gradually explores the shortest paths to other nodes. Let the graph model be , where is the set of nodes, is the set of edges. For each node , maintain a distance value , representing the shortest distance from the starting node to node . Initially set ( is the starting node), for other nodes , . In each iteration, select the node with the smallest distance value and not yet visited. For the node adjacent to , update to , where is the weight of the edge . Through continuous iteration, the shortest paths from the starting node to all other nodes are obtained, thus obtaining the initial patrol path (i.e., the preset patrol path). The large model service platform sends the above initial patrol path to the patrol robot dog, so that the patrol robot dog patrols the monitoring area according to this patrol path. During the patrol process, the patrol robot dog uploads the patrol data and detection results collected by the sensors to the large model service platform in real time.

[0091] When the large model service platform determines that there is an abnormal situation in the detection area based on the detection results, it generates a patrol task according to the detection results. For example, if the detection results show that a certain device has an abnormal appearance or overheating fault, or a dangerous gas leak is detected at a certain location, the large model service platform can generate a patrol task to inspect the device with abnormal appearance or faulty device or handle the dangerous gas leak.

[0092] Furthermore, the large model service platform is also used to determine the priority of the patrol task.

[0093] In one embodiment, the large model service platform can determine the priority of the patrol task based on the analytic hierarchy process. By constructing a judgment matrix , which is used to compare the relative importance of different patrol tasks (such as patrolling different devices, handling different types of abnormalities, etc.):

[0094]

[0095] In the formula, represents the patrol task relative to the patrol task importance ratio scale. Usually, the 1-9 scale method is adopted. 1 means the two are equally important, 9 means the task relative to the task absolutely important, and the intermediate values represent different degrees of relative importance. Calculate the maximum eigenvalue and the corresponding eigenvector , the eigenvector after normalization is the relative weight vector of each task, representing the priority of the task. is the number of patrol tasks.

[0096] In another embodiment, in order to make the task priority strategy adapt to the dynamically changing environment, the large model service platform introduces a reinforcement learning model to determine the priority of the patrol task. The reinforcement learning model can be trained through historical patrol data and historical patrol tasks; the reinforcement learning model predicts the priority of the patrol task according to the current patrol data and patrol tasks; among them, the training process of the reinforcement learning model includes:

[0097] Construct a state space and an action space. The state space includes the gas concentration, device temperature, list of devices that have been patrolled, and list of devices that have not been patrolled in the current environment; the action space includes executable patrol tasks;

[0098] Calculate the reward value for selecting each action according to the current state; update the network parameters through the reward value.

[0099] Specifically, the patrol process can be regarded as a sequential decision-making problem. The patrol robot dog performing the patrol task selects a patrol task to execute in each state (i.e., the current environmental information and task list), and then obtains a reward feedback according to the execution result and transfers to a new state.

[0100] Define the state space , which contains all relevant information of the current environment, such as the current detected gas concentration distribution ( is the number of sensors), equipment temperature, list of equipment that has been patrolled , list of remaining equipment that has not been patrolled , etc. The action space is the set of all executable patrol tasks. Design the reward function , and give the agent a reward according to the current state and the executed action (i.e., the selected patrol task). For example, for the task of handling a high-concentration gas leak and successfully reducing the gas concentration, a higher reward is given; for handling a relatively unimportant task while there is an emergency gas leak that has not been handled, a lower reward is given. The reward function can be , where and are weight coefficients, is the change in the target state after handling the task (such as the change in gas concentration, the change in equipment temperature), is an indicator function, when handling an emergency task , otherwise .

[0101] Through the continuous interaction between the patrol robot dog performing the patrol task and the environment, learn the optimal policy according to the Bellman equation , that is, the mapping from state to action, to maximize the long-term cumulative reward. Among them, is the value function of state , is the discount factor ( ), indicating the degree of emphasis on future rewards; is the probability of transferring from state by executing action to state . By iteratively updating the value function and the policy, the patrol robot dog makes the optimal task selection decision in different environmental states, realizing the dynamic adjustment of task priorities.

[0102] Furthermore, the large model service platform is also used to dynamically optimize the inspection path according to the inspection data and inspection tasks of the inspection robot dog, and then send the optimized inspection path and inspection tasks to the inspection robot dog for execution. The large model service platform uses the large model to analyze the severity of abnormal situations, combines the current position of the inspection robot dog and the remaining uninspected nodes, and uses a local replanning method (such as the A* algorithm) to dynamically adjust the inspection path.

[0103] Specifically, the large model service platform is used to divide the detection area into grids to obtain multiple sub-areas, and each grid (i.e., each sub-area) is used as a basic unit for risk assessment. The grid size can be determined according to the complexity and accuracy requirements of the monitoring area. For example, in areas with dense equipment and rapid risk changes, smaller grids are used; in relatively open areas with relatively slow risk changes, larger grids are used. Calculate the risk level of each sub-area. Specifically, a risk factor set (risk factors can include gas concentration, equipment temperature, environmental humidity, etc.), and a comment set ( is low risk, is medium risk, is high risk, etc.). Based on historical data and real-time data, construct the membership function of each risk factor and the comment set to obtain the fuzzy relation matrix :

[0104]

[0105] (i = 1, 2,..., n, j = 1, 2,..., k) represents the membership degree of the risk factor to the comment set , which is determined by the large model's assisted statistical analysis of historical data and expert experience. For example, for gas concentration , when the concentration is below a certain threshold, the membership degree to the low-risk level is close to 1; as the concentration increases, the membership degrees to the medium-risk and high-risk levels gradually increase.

[0106] Determine the weight vector of each risk factor based on the analytic hierarchy process and . The fuzzy comprehensive evaluation result , where is the number of risk factors, determined according to the actual monitoring parameters; is the number of risk levels in the comment set, generally 3 - 5 levels. is the weight of the i-th (i = 1, 2,..., n) risk factor, and the large model adjusts the weight according to the current environment and equipment status. For example, in a flammable and explosive environment, the weight of gas concentration is relatively large. According to Determine the grid risk level based on the maximum value and map it to the color range of the heat map. For example, the range of 0 - 2 is green (low risk), the range of 2 - 5 is yellow (medium risk), and the range of 5 - 10 is red (high risk).

[0107] Optimize the inspection path based on the generated risk heat map and the priority of the inspection tasks. Path planning can be used for inspection path optimization. For example, the A* algorithm can be used to focus on the high - risk areas in the risk heat map, considering factors such as the importance of equipment and the difficulty of passage, to plan an inspection path that covers high - risk areas and takes into account other high - priority inspection tasks. During the inspection, if a new high - risk area is found or the risk level increases, the large model quickly re - plans the path to guide the inspectors to conduct inspections and promptly handle potential safety hazards. According to the risk assessment results, the large model intelligently adjusts the inspection strategy, such as automatically increasing the inspection frequency for high - risk areas and giving a reasonable inspection time interval.

[0108] It should be noted that the A* algorithm uses an evaluation function to select the next node, where is the actual cost from the start node to node , and is the estimated cost from node to the target node. Based on continuously selecting the node with the minimum value for expansion, quickly find the optimal path from the current position to the target position (such as the location of abnormal equipment, high - risk areas, etc.) to achieve dynamic adjustment of the inspection path.

[0109] Combined with the above - mentioned inspection data, construct a hierarchical early - warning and large - model intelligent response mechanism. According to the risk assessment results, when the risk level reaches a certain threshold, the system automatically issues alarms of different levels, such as audible and visual alarms, text message notifications, etc. The large model participates in determining the early - warning level and accurately adjusts the early - warning level according to factors such as the speed of risk change and the scope of influence. For example, if the risk level rises rapidly, the early - warning level is increased. Give corresponding emergency measure suggestions according to the risk level and specific circumstances. For example, in the event of a gas leak in a high - risk area, give suggestions such as evacuation routes and emergency treatment procedures to assist relevant personnel in taking effective emergency measures.

[0110] Furthermore, the number of inspection robot dogs in the embodiments of the present application can be multiple. The large model service platform can adopt a distributed algorithm to achieve multi-robot dog formation inspection. Each robot dog has independent computing and decision-making capabilities and conducts real-time data sharing and communication with other robot dogs through the 5G / 6G network. The large model service platform can reasonably allocate inspection tasks based on information such as the positions of each robot dog and the task completion status, enabling multiple robot dogs to work collaboratively and cover a wider area. During the inspection process, the robot dogs can cooperate with each other to jointly complete complex tasks. For example, when one robot dog discovers an abnormal situation, it can promptly notify other robot dogs to go for support and jointly conduct further detection and processing, improving the response speed to abnormal events.

[0111] The intelligent inspection system in the embodiments of the present application provides a remote control interface for the desktop or mobile terminal. Users can view information such as the environmental pictures, gas concentration, and equipment status collected by the robot dog in real time through these devices, providing a more intuitive operation experience for users; the mobile terminal facilitates users to monitor and operate anytime and anywhere at different locations. The intelligent inspection system supports remotely controlling the robot dog to perform emergency operations, such as closing valves, starting ventilation equipment, etc. When an emergency occurs, users can quickly issue instructions through the remote control interface to avoid the further expansion of the accident.

[0112] The intelligent inspection system is equipped with an automatic charging dock. When the battery of the robot dog is insufficient, it can automatically return to the charging dock for charging to ensure the continuous operation of the system.

[0113] The system also has an online sensor calibration function, regularly calibrating the sensors to ensure the accuracy of the detected data. The system has built-in scheduled tasks, and different calibration periods are set according to the sensor type and usage environment. For example, for gas detection sensors, since they are greatly affected by environmental factors, the calibration period is set to once a week; for relatively stable sensors such as temperature and humidity, the calibration period is set to once a month. When the calibration time arrives, the system automatically starts the calibration program. The calibration process is as follows:

[0114] The large model service platform retrieves the initial calibration parameters and historical calibration data of the sensor from the database and prepares a standard gas sample (for gas sensors) or a standard environmental parameter simulator (for sensors such as temperature and humidity).

[0115] Based on the data collected by the sensor in the preset environment in real time (continuously collecting data for a period of time such as 5 minutes), compare the collected environmental data with the standard value of the environmental data in the database, and calculate the data acquisition deviation value of the sensor using algorithms such as the least squares method. According to the calculated data acquisition deviation value, adjust the calibration parameters of the sensor, such as the sensitivity coefficient, zero offset, etc. Write the updated calibration parameters into the internal storage chip of the sensor and synchronize them to the system database to complete the calibration process.

[0116] The system can also adopt a fault diagnosis algorithm to monitor the running state of the robotic dog in real time. When a fault is detected, it can automatically diagnose the cause of the fault and take corresponding measures. For example, when a certain sensor fails, the system can automatically switch to a backup sensor, notify the maintenance personnel in time for maintenance, reduce the frequency of manual intervention, and ensure continuous operation for 7×24 hours. The fault detection of the robotic dog is as follows:

[0117] Fault diagnosis algorithms based on machine learning, such as support vector machines and artificial neural networks. Collect data on the normal operation and various fault states of the robotic dog, label the fault types and characteristics, and achieve different fault mode recognition. During the real-time monitoring process, input the collected running data of the robotic dog into the trained fault diagnosis model, and judge whether there is a fault and the fault type according to the data characteristics.

[0118] Furthermore, during the execution of the inspection task, the inspection robotic dog feeds back the actual execution results and newly discovered abnormal situations to the large model service platform. For example, when the robotic dog discovers a previously unrecognized equipment fault type, it uploads the relevant images and data to the large model service platform. The large model service platform uses the fed-back data to continuously train and update the large model. Through the incremental learning algorithm, without affecting the original performance of the large model, the large model can continuously learn new knowledge and patterns to adapt to the changing inspection environment and task requirements.

[0119] A reliable communication protocol (such as TCP / IP) is adopted between the inspection robotic dog and the large model service platform to encapsulate and verify the transmitted data to ensure the integrity and correctness of the transmitted data. At the same time, different transmission priorities and strategies are adopted according to the importance and real-time requirements of the transmitted data to ensure that key data can be transmitted in time.

[0120] The large model service platform generates corresponding instructions according to the inspection task and real-time data, such as inspection path adjustment, equipment detection instructions, etc., and sends these instructions to the inspection robotic dog. After receiving the instructions, the inspection robotic dog parses and executes them, and feeds back the execution results to the platform. During the instruction interaction process, encryption technology is used to encrypt the instructions to prevent the instructions from being stolen or tampered with, and ensure the security of communication.

[0121] Specifically, the patrol robot dog encrypts the patrol data and detection results, and transmits the encrypted patrol data and encrypted detection results to the large model service platform. The large model service platform decrypts them to obtain the original data, then conducts further analysis and processing, and encrypts the obtained patrol tasks and optimized patrol paths and transmits them to the patrol robot dog. When there are multiple patrol robot dogs, the large model service platform can determine the priority of the transmitted data (patrol tasks and optimized patrol paths) according to the risk levels of the monitoring areas where each patrol robot dog is located, and transmit them to each patrol robot dog based on the priority of the transmitted data.

[0122] When communication between the patrol robot dog and the large model service platform is abnormal, such as network interruption, data loss, etc., an exception handling mechanism is designed. The patrol robot dog can perform emergency processing according to preset rules, such as pausing the patrol task, returning to a safe area, etc.; after the communication is restored, the large model service platform reprocesses and synchronizes the unprocessed data. At the same time, the system will record the detailed information of the abnormal event for subsequent analysis and improvement.

[0123] For the intelligent patrol system provided in this application, after the patrol robot dog obtains the patrol data, it can simultaneously perform equipment anomaly detection and gas leakage detection, enabling multitasking and improving the patrol quality; through further analysis and processing of the patrol data by the large model service platform, the patrol tasks and patrol paths are optimized, improving the patrol efficiency.

[0124] Please refer to Figure 2 , this application embodiment also provides an intelligent patrol method, which is applied to the foregoing intelligent patrol system. The method includes:

[0125] Step 210, the patrol robot dog patrols the monitoring area according to the preset patrol path, and conducts equipment anomaly detection and gas leakage detection on the patrol data collected by the sensor, and uploads the obtained detection results and patrol data to the large model service platform;

[0126] Step 220, when there is an abnormal situation in the detection area, the large model service platform generates patrol tasks according to the detection results; divides the detection area into multiple sub-areas, calculates the risk level of each sub-area according to the patrol data, determines whether to issue a risk warning according to the risk level, constructs a risk heat map according to the risk level, optimizes the patrol path according to the risk heat map and patrol tasks, and sends the optimized patrol path and patrol tasks to the patrol robot dog for execution.

[0127] This application embodiment also provides an electronic device, which includes a processor and a memory;

[0128] The memory is used to store program codes and transmit the program codes to the processor;

[0129] The processor is used to execute the intelligent patrol inspection method in the foregoing method embodiments according to the instructions in the program code.

[0130] An embodiment of the present application also provides a computer-readable storage medium, which is used to store program code, and when the program code is executed by a processor, the intelligent patrol inspection method in the foregoing method embodiments is implemented.

[0131] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific processes of the methods described above can refer to the corresponding processes in the foregoing system embodiments, and will not be elaborated herein.

[0132] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0133] It should be understood that in the present application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" may mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B may be singular or plural. The character " / " generally means that the associated objects before and after are in an "or" relationship. "At least one (item) of the following" or its similar expression means any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) of a, b or c may mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c may be single or multiple.

[0134] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0135] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0136] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0137] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of this application. And the aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (English full name: Read-Only Memory, English abbreviation: ROM), random access memories (English full name: Random Access Memory, English abbreviation: RAM), magnetic disks or optical discs and other media that can store program codes.

[0138] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent inspection system, characterized in that, Including: An inspection robot dog and a large model service platform; The inspection robot dog is used to inspect a monitoring area according to a preset inspection path, detect equipment abnormalities and gas leaks through inspection data collected by sensors, and upload the obtained detection results and the inspection data to the large model service platform; The large model service platform is used to generate an inspection task according to the detection results when there are abnormalities in the detection area; Divide the detection area into multiple sub-areas, calculate the risk level of each sub-area according to the inspection data, determine whether to issue a risk warning according to the risk level, construct a risk heat map according to the risk level, optimize the inspection path according to the risk heat map and the inspection task, and send the optimized inspection path and the inspection task to the inspection robot dog for execution.

2. The intelligent inspection system according to claim 1, wherein The inspection data includes video data and infrared thermal imaging images of equipment; When the inspection robot dog is used to detect equipment abnormalities based on the inspection data, it specifically includes: Processing the video data through an image recognition method to detect whether there are appearance abnormalities in the equipment and obtaining an equipment appearance detection result; Constructing a Gaussian mixture model for the temperature values of pixel points in the infrared thermal imaging image; Initializing the weights, means, and variances of each Gaussian distribution in the Gaussian mixture model, and calculating the responsiveness of each Gaussian distribution to the temperature values of each pixel point; Updating the weights, means, and variances of each Gaussian distribution through the responsiveness of each Gaussian distribution to the temperature values of each pixel point; Sorting each Gaussian distribution in descending order according to the ratio of the weight to the standard deviation, selecting the first target number of Gaussian distributions to calculate the probability that the temperature value of each pixel point in the infrared thermal imaging image belongs to the normal temperature. If the probability is less than a preset threshold, it is determined that the pixel point is an overheated abnormal point.

3. The intelligent inspection system according to claim 1, wherein, The inspection data includes gas concentration; When the inspection robot dog is used to detect gas leaks based on the inspection data, it specifically includes: Calculating the gas concentration gradient according to the gas concentrations detected by gas detection sensors at various positions; Constructing the relationship between the gas concentration at each position of the gas detection sensor and the position of the leakage source through a gas Gaussian diffusion model to obtain the position likelihood function of the leakage source; Taking the logarithm of the position likelihood function of the leakage source to obtain the position log-likelihood function of the leakage source; Maximizing the position log-likelihood function of the leakage source through an optimization method to solve for the position and intensity of the leakage source.

4. The intelligent inspection system according to claim 1, wherein The body of the inspection robot dog is made of aluminum alloy and an insulating coating; The inspection robot dog uses rubber or silicone rubber seals to fill the connection between the equipment interface and the connection plug of the connecting wire; The junction box of the inspection robot dog is made of aluminum alloy. The surface of the junction box is subjected to anti-corrosion treatment. The inlet of the junction box is equipped with sealant clay and a sealed stuffing box. After the wire penetrates into the inlet, the nut on the sealed stuffing box is tightened, and the sealant clay tightly wraps the wire.

5. The intelligent inspection system according to claim 1, wherein The large model service platform is also used to determine the priority of the inspection task, specifically including: Train a reinforcement learning model using historical patrol inspection data and historical patrol inspection tasks; predict the priority of the patrol inspection task according to the current patrol inspection data and the patrol inspection task by means of the reinforcement learning model; Among them, the training process of the reinforcement learning model includes: Construct a state space and an action space, where the state space includes the gas concentration, equipment temperature, list of inspected equipment, and list of uninspected equipment in the current environment; the action space includes executable patrol inspection tasks; Calculate the reward value for each action according to the current state; update the network parameters through the reward value.

6. The intelligent inspection system according to claim 1, wherein, The large model service platform is also used to compare the environmental data collected in real time by the sensor with the environmental data standard value in the database, and calculate the data acquisition deviation value of the sensor by the least squares method; adjust the calibration parameters of the sensor through the data acquisition deviation value.

7. The intelligent inspection system according to claim 1, wherein The large model service platform is also used to obtain the operation data of the patrol inspection robot dog in real time, and perform fault identification on the operation data through a fault diagnosis model to determine whether there is a fault in the patrol inspection robot dog.

8. An intelligent inspection method, characterized in that, Including: The patrol inspection robot dog patrols the monitoring area according to the preset patrol inspection path, and performs equipment anomaly detection and gas leakage detection through the patrol inspection data collected by the sensor, and uploads the obtained detection results and the patrol inspection data to the large model service platform; When there is an abnormal situation in the detection area, the large model service platform generates a patrol inspection task according to the detection result; Divide the detection area into multiple sub-areas, calculate the risk level of each sub-area according to the patrol inspection data, determine whether to issue a risk warning according to the risk level, construct a risk heat map according to the risk level, optimize the patrol inspection path according to the risk heat map and the patrol inspection task, and send the optimized patrol inspection path and the patrol inspection task to the patrol inspection robot dog for execution.

9. An electronic device, characterized in that, The device includes a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the intelligent patrol inspection method according to the instructions in the program code as claimed in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code, and when the program code is executed by a processor, the intelligent patrol inspection method as claimed in claim 8 is implemented.

Images