Inspection method, system and equipment of intelligent inspection robot and medium

Abstract

The invention discloses an inspection method, system and device of an intelligent inspection robot and a medium, and belongs to the technical field of intelligent inspection, and the method comprises the steps that the intelligent inspection robot receives an inspection task instruction, and analyzes and obtains an inspection site, inspection time and inspection content; providing a navigation map; planning a plurality of polling paths to be selected, and calculating an optimal path through a dynamic planning algorithm; the intelligent inspection robot performs autonomous movement according to the optimal path, and performs data acquisition on an inspection object and an environment where an inspection point is located at a set point location; the intelligent inspection robot processes and analyzes the collected data, preliminarily recognizes potential abnormities, and transmits all the data to the cloud; generating alarm information and an inspection report; and the intelligent inspection robot executes automatic nest-returning charging operation according to the inspection period and the residual electric quantity state to guarantee continuous completion of the task. The task adaptability, the risk response efficiency and the continuous operation capability of the intelligent inspection robot in a complex environment are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent inspection technology, and in particular to an inspection method, system, equipment and medium for an intelligent inspection robot. Background Art

[0002] An intelligent inspection robot is an intelligent inspection robot that can conduct equipment inspections autonomously or semi-autonomously, and uses sensors, algorithms and other means to monitor, diagnose and maintain equipment. Its main features are: high efficiency: intelligent inspection robots can replace manual labor to perform repetitive and long-term work, thereby improving inspection efficiency; real-time: intelligent inspection robots can monitor equipment status in real time and detect abnormal situations in a timely manner; accuracy: through data analysis and learning, intelligent inspection robots can more accurately judge equipment failures and performance degradations, avoiding missed inspections and misjudgments; safety: in dangerous or remote environments, intelligent inspection robots can replace manual labor for inspections, reducing safety risks.

[0003] The present invention coordinates environmental perception with map construction, path planning, and optimization steps, so that the intelligent inspection robot can dynamically adjust its route according to preset inspection tasks, environmental obstacle information, and current position during inspection, avoid obstacles in real time, and improve the recognition ability and accuracy of the intelligent inspection robot.

[0004] Through the coordination between data processing and analysis, abnormal alarm and report generation steps, and the use of deep learning and image recognition artificial intelligence technology, the status of the inspection object can be automatically identified, alarm information can be generated immediately, and the alarm information can be quickly sent to the command center to ensure that the problem is handled in a timely manner and improve the speed at which the intelligent inspection robot obtains dangerous situations during inspections.

[0005] By equipping it with anti-collision protection, fall detection, high-temperature power-off, and autonomous return-to-nest charging functions, the intelligent inspection robot can ensure its own safety during inspections. It can adapt to complex tasks and work scenarios, while ensuring its own sustainable inspection operations and improving inspection efficiency. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is: how to enable the intelligent inspection robot to dynamically adjust its route according to the preset inspection tasks, environmental obstacle information and current position during inspection, avoid obstacles in real time, and improve the recognition ability and recognition accuracy of the intelligent inspection robot.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions: an inspection method for an intelligent inspection robot, which includes: the intelligent inspection robot receives inspection task instructions, and parses and obtains the inspection location, inspection time and inspection content; uses the onboard sensors to sense the information of the inspection environment, and constructs or updates the point cloud map of the inspection environment to provide a navigation map; the intelligent inspection robot uses the potential field and graph search fusion algorithm and genetic algorithm to perform path planning, plans multiple inspection paths to be selected, and calculates an optimal path through a dynamic programming algorithm; the intelligent inspection robot performs autonomous movement according to the optimal path , and collects data on the inspection objects and the environment where the inspection points are located at set points; the intelligent inspection robot processes and analyzes the collected data, preliminarily identifies potential anomalies, and transmits all data to the cloud; the intelligent inspection robot uses a lightweight multimodal flame detection model and a risk detection model based on the input information of sensors and cameras to identify anomalies, and analyzes the state evolution trend through the spatiotemporal graph convolutional network ST-GCN to generate alarm information and inspection reports; the intelligent inspection robot performs automatic return to the nest charging operation according to the inspection cycle and the remaining power status to ensure the continuous completion of the task.

[0009] As a preferred solution of the inspection method of the intelligent inspection robot described in the present invention, the providing of the navigation map includes providing the navigation map by rasterizing the point cloud map, which contains information on whether all spatial points are occupied.

[0010] As a preferred embodiment of the inspection method of an intelligent inspection robot described in the present invention, the autonomous movement includes the intelligent inspection robot moving to the set inspection points in sequence according to the planned path, performing inspection operations and collecting data at the inspection points.

[0011] As a preferred solution of the inspection method of an intelligent inspection robot described in the present invention, the automatic return to nest charging operation is performed according to the inspection cycle and the remaining power status, including starting the inspection task according to the preset inspection cycle, and triggering automatic navigation to the charging pile for charging when the power conditions are met.

[0012] As a preferred solution of the inspection method of an intelligent inspection robot described in the present invention, the path planning includes generating an initial path by adopting a fusion potential field algorithm and a graph search algorithm, the graph search algorithm searches through a node total cost function, the total cost function includes a heuristic cost and an actual cost obtained by calculating the Euclidean distance, an artificial potential field method is used to calculate the attraction and repulsion of the path points, the initial path is adjusted, and a trajectory with continuous curvature is generated by interpolation of a quintic B-spline curve; a path is constructed by GPU-accelerated calculation of a genetic algorithm, and a path cost function is constructed based on the path length, smoothness and the distance from the path point to the obstacle, and dynamic programming is introduced to calculate the energy consumption per unit distance of the path, and a path with a unit distance power consumption that does not exceed the power consumption threshold is preferentially selected as the final inspection path.

[0013] This preferred solution constructs the initial path by integrating the potential field algorithm and the graph search algorithm, and combines it with quintic B-spline interpolation to generate a curvature continuous trajectory, which can significantly reduce sharp turns and oscillations in the path, and improve the feasibility and execution smoothness of the path; at the same time, through the genetic algorithm of GPU accelerated computing combined with dynamic programming energy consumption constraints, the path with power consumption per unit distance lower than the set threshold is preferentially selected, which helps to achieve full-task coverage path planning under low-energy conditions and improve the endurance stability of the intelligent inspection robot in complex task cycles.

[0014] As a preferred embodiment of the inspection method of an intelligent inspection robot described in the present invention, the method includes: performing autonomous movement according to the optimal path and collecting data on the inspection object and the environment where the inspection point is located at the set point, including: the intelligent inspection robot autonomously moving according to the planned path and the set inspection point, and the intelligent inspection robot determines whether it has reached the inspection point by comparing its own positioning information in the global map with the position information of the set inspection point in the global map. If it has arrived, it will conduct an inspection at the set point, and use the onboard detection equipment to collect data on the inspection object and the environment where the inspection point is located, and can also record and broadcast the live image through the camera.

[0015] This preferred solution compares and judges the positioning information based on the global map with the set inspection points to achieve accurate judgment of the arrival status and improve the reliability of autonomous navigation and positioning; it combines multiple types of detection equipment to collect environmental and object status data, and records and broadcasts real-time images to provide synchronous on-site perception support for the inspection process, enhancing remote supervision and rapid response capabilities.

[0016] As a preferred solution of the inspection method of an intelligent inspection robot described in the present invention, the abnormality identification includes, on the intelligent inspection robot side, using sensors and cameras as input, utilizing deep learning and image recognition artificial intelligence technology, deploying a lightweight multimodal flame detection model for identifying open flames and smoldering fires, automatically identifying the status of the inspection object, and alarming for abnormal conditions, while generating a log report and uploading it to the cloud, using gas, temperature, and image information as input, deploying a multimodal risk detection model, and analyzing whether there are risk problems through data comparison; on the cloud, using the data uploaded by the intelligent inspection robot as input, integrating multiple information such as gas, temperature, and image, and analyzing the state evolution trend through a spatiotemporal graph convolutional network.

[0017] This preferred solution deploys a lightweight multimodal flame detection model and a risk detection model on the intelligent inspection robot side, integrating image, temperature, gas and other data to achieve dual detection of open flames and smoldering fires, with short response time and accurate alarm; introducing the spatiotemporal graph convolutional network ST-GCN in the cloud to perform trend modeling on the uploaded data, which can predict the evolution process of abnormal states in advance and enhance the system's ability to detect potential risks in advance and actively defend against them.

[0018] Another object of the present invention is to provide an inspection system for an intelligent inspection robot.

[0019] To solve the above technical problems, the present invention provides the following technical solutions: an inspection system for an intelligent inspection robot, comprising: a task parsing module, an environment perception module, a path planning module, an autonomous inspection module, a data processing module, an anomaly recognition module, and a charging control module; the task parsing module is used for the intelligent inspection robot to receive inspection task instructions and parse and obtain the inspection location, inspection time, and inspection content; the environment perception module is used to use the onboard sensors to perceive information about the inspection environment, and to construct or update a point cloud map of the inspection environment to provide a navigation map; the path planning module is used for the intelligent inspection robot to perform path planning using a potential field and graph search fusion algorithm and a genetic algorithm, plan multiple candidate inspection paths, and calculate an optimal path through a dynamic programming algorithm; The autonomous inspection module is used for the intelligent inspection robot to perform autonomous movement according to the optimal path, and to collect data on the inspection object and the environment where the inspection point is located at the set point; the data processing module is used for the intelligent inspection robot to process and analyze the collected data, preliminarily identify potential anomalies, and transmit all data to the cloud; the anomaly recognition module is used for the intelligent inspection robot to identify anomalies based on the input information of sensors and cameras, using a lightweight multimodal flame detection model and a risk detection model, and analyze the state evolution trend through the spatiotemporal graph convolutional network ST-GCN to generate alarm information and inspection reports; the charging control module is used for the intelligent inspection robot to perform automatic homing and charging operations according to the inspection cycle and the remaining power status to ensure the continuous completion of the task.

[0020] The present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the inspection method of the intelligent inspection robot are implemented.

[0021] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of the inspection method of the intelligent inspection robot are implemented.

[0022] Beneficial effects of the present invention: The present invention cooperates with environmental perception and map construction, path planning and optimization steps, so that the patrol intelligent inspection robot can dynamically adjust its route according to the preset patrol tasks, environmental obstacle information and current position during patrol, avoid obstacles in real time, and improve the recognition ability and recognition accuracy of the intelligent patrol robot

[0023] The present invention automatically identifies the status of inspection objects through the mutual coordination between data processing and analysis, abnormal alarm and report generation steps, and utilizes deep learning and image recognition artificial intelligence technology to instantly generate alarm information and quickly send the alarm information to the command center to ensure that problems are handled in a timely manner and improve the speed at which intelligent inspection robots acquire dangerous situations during inspections.

[0024] The present invention is equipped with anti-collision protection, fall detection, high-temperature power-off and autonomous return-to-nest charging functions to ensure the self-safety of the intelligent inspection robot during inspection. It can adapt to complex tasks and work scenarios, and at the same time ensure its own sustainable inspection operations and improve inspection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 An overall flow chart of an inspection method for an intelligent inspection robot provided in one embodiment of the present invention.

[0027] Figure 2 A schematic diagram of a patrol system for an intelligent patrol robot according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0029] Example 1, reference Figure 1 , is an embodiment of the present invention, which provides an inspection method of an intelligent inspection robot, comprising:

[0030] S1. The intelligent inspection robot receives the inspection task instruction and parses it to obtain the inspection location, inspection time and inspection content.

[0031] S2. Use the onboard sensors to perceive information about the inspection environment, and build or update a point cloud map of the inspection environment to provide a navigation map.

[0032] S3. The intelligent inspection robot uses the potential field and graph search fusion algorithm and genetic algorithm for path planning, plans multiple inspection paths to be selected, and calculates the optimal path through the dynamic programming algorithm.

[0033] S4. The intelligent inspection robot moves autonomously according to the optimal path and collects data on the inspection object and the environment where the inspection point is located at the set point.

[0034] S5. The intelligent inspection robot processes and analyzes the collected data, preliminarily identifies potential anomalies, and transmits all data to the cloud.

[0035] S6, the intelligent inspection robot, uses a lightweight multimodal flame detection model and a risk detection model to identify anomalies based on input information from sensors and cameras, and analyzes state evolution trends through the spatiotemporal graph convolutional network ST-GCN to generate alarm information and inspection reports.

[0036] S7. The intelligent inspection robot automatically returns to the nest to charge according to the inspection cycle and remaining power status to ensure the continuous completion of the task.

[0037] It should be noted that the intelligent inspection robot can cover the complete process of task instruction parsing, environment mapping, path planning, data collection, anomaly analysis and returning to the nest for charging through the coordinated execution of steps S1 to S7 in the present invention; the constructed inspection method is not only suitable for standard task execution in static scenarios, but can also cope with complex working conditions including path obstacles, energy consumption limitations, risk identification, etc., and has good feasibility and adaptability, providing a technical basis for actual deployment.

[0038] Embodiment 2 is an embodiment of the present invention, and provides an inspection method of an intelligent inspection robot based on the previous embodiment, comprising:

[0039] In the embodiment of the present application, in step S1, the intelligent inspection robot receives the inspection task instruction, and parses and obtains the inspection location, inspection time and inspection content.

[0040] The intelligent inspection robot receives inspection task instructions and parses out key information in the task, including inspection location, inspection time, and inspection content. It receives instructions using a human-computer interaction interface, including a touch screen or physical buttons, allowing on-site operators to directly input instructions or parameters to adjust the working status of the intelligent inspection robot.

[0041] For the inspection of intelligent inspection robots, on-site operators can complete the distribution of inspection content through the human-computer interaction interface (touch screen, physical buttons). The intelligent inspection robot uses 16-line laser radar, single-line obstacle avoidance laser radar, camera, and IMU as input to obtain environmental information and intelligent inspection robot motion information, and builds an environmental map through SLAM technology.

[0042] In an optional embodiment, the intelligent inspection robot receives task information through a physical button input interface, and the operator can select a preset task number and adjust related parameters on the start panel. The parameters are loaded after the system's internal task management module calls the corresponding template.

[0043] In another optional embodiment, the intelligent patrol robot receives task instructions sent from a remote control center, and the instructions are transmitted to the intelligent patrol robot body through the Wi-Fi communication module. The received instructions include the task ID, target area number and timing trigger conditions, which are parsed and prepared by the intelligent patrol robot scheduling unit.

[0044] By combining the touch screen interface with the Chinese task template, the present invention enables operators to quickly input and confirm the inspection task content on site, avoiding communication dependence or mechanical operation errors, improving the human-computer interaction efficiency of the task receiving link and the accuracy of task data input, and providing a direct and reliable input path for task analysis and subsequent execution.

[0045] In the embodiment of the present application, in step S2, the onboard sensors are used to sense information about the inspection environment, and a point cloud map of the inspection environment is constructed or updated to provide a navigation map.

[0046] The point cloud map is rasterized to provide a navigation map, which contains information on whether all spatial points are occupied.

[0047] The intelligent inspection robot uses the onboard sensors, including lidar and cameras, to perceive the inspection environment and obtain environmental point cloud and image information. Based on the perceived environmental information, the intelligent inspection robot builds or updates the point cloud map of the inspection environment. The point cloud map is rasterized to provide a navigation map, which contains information on whether all spatial points are occupied, and is used for navigation of the intelligent inspection robot.

[0048] Using SLAM technology or preset map navigation, the intelligent inspection robot can independently plan the optimal inspection path, dynamically adjust the route according to the preset inspection task, environmental obstacle information and current position, and avoid obstacles in real time through the environmental perception system, including lidar and ultrasonic sensors, to ensure the safety and smoothness of the inspection process. Based on S2, S3 predicts the possible location to be reached in the future based on the current speed of the intelligent inspection robot, and selects the best obstacle avoidance path. The target point is regarded as the source of gravity and the obstacle as the source of repulsion. The intelligent inspection robot moves according to the force direction of the potential field. The path optimization in S3 includes smoothing the path through B-spline interpolation to reduce sharp turns and vibrations. At the same time, based on the prediction model, the control input in the future is optimized and real-time trajectory adjustment is performed.

[0049] In the embodiment of the present application, in step S3, the intelligent inspection robot uses the potential field and graph search fusion algorithm and the genetic algorithm to perform path planning, plans multiple inspection paths to be selected, and calculates an optimal path through the dynamic programming algorithm.

[0050] Path planning includes generating an initial path by fusion of potential field algorithm and graph search algorithm. The graph search algorithm searches through the node total cost function, which includes the heuristic cost and actual cost calculated by Euclidean distance. The artificial potential field method is used to calculate the attraction and repulsion of the path points, adjust the initial path, and generate a curvature-continuous trajectory through quintic B-spline curve interpolation.

[0051] The path is constructed by using GPU-accelerated genetic algorithms, and a path cost function is constructed based on the path length, smoothness, and the distance from the path point to the obstacle. At the same time, dynamic programming is introduced to calculate the energy consumption per unit distance of the path, and the path whose unit distance power consumption does not exceed the power consumption threshold is preferentially selected as the final inspection path.

[0052] According to the inspection task requirements set in step S1, the rasterized map in step S2 is used as the global map input for path planning. The intelligent inspection robot uses potential field and graph search fusion algorithms, genetic algorithms, etc. to plan paths and plans multiple candidate inspection paths. During the path planning process, obstacle avoidance, efficiency, and safety factors are taken into consideration, and an optimal path is calculated through a dynamic planning algorithm. It can also be dynamically adjusted and optimized according to actual conditions.

[0053] Specifically, the intelligent inspection robot has multiple path planning algorithms. One is a fusion of potential field algorithm and graph search algorithm. It uses 5th-order B-spline curve interpolation to plan a curvature-continuous path in real time. The graph search algorithm provides a global initial solution. The cost function when searching for nodes is:

[0054] f(n)=h(n)+g(n)

[0055] Where f(n) is the total node cost, h(n) is the heuristic cost, and g(n) is the actual cost, which is calculated by the Euclidean distance. The potential field algorithm optimizes the initial global path and obtains the force on each point through the distance to the obstacle and the distance to the target point:

[0056]

[0057] Among them, U(p) is the total force, U att (p) is the attraction, U rep (p) is the repulsive force, k att is the gravitational adjustment coefficient, p is the current position, p g is the target point position, k repis the repulsion adjustment coefficient, ρ(p) is the distance from the current position to the obstacle, and ρ0 is the influence range. The positions of the path points are adjusted using the artificial potential field method, and then a smooth and second-order continuous trajectory is generated based on the B-spline curve.

[0058] The second is the planning path obtained by using GPU accelerated calculation of genetic algorithm, the cost term is:

[0059] F(P)=ω1·L(P)+ω2·C(P)+ω3·O(P)

[0060] Among them, ω is the weight, L(P) is the path length cost, which is obtained by adding up the sum of the Euclidean distances between each point, that is:

[0061]

[0062] C(P) is the smoothness cost, and its calculation formula is:

[0063]

[0064] O(P) is the obstacle cost, which is obtained from the distance between the path point and the obstacle, that is:

[0065]

[0066] Dynamic programming is introduced to calculate the energy consumption weights of different paths, and paths with power consumption per unit distance ≤ 5W / m are prioritized. By fusion of lidar and ultrasonic sensors, intelligent inspection robots and dynamic obstacles are modeled as game participants. The movement trends of obstacles are predicted based on Nash equilibrium theory to achieve preventive collision avoidance.

[0067] Optimization in S3 also includes algorithm optimization, using multi-core processors or GPUs to accelerate algorithm calculations, pruning invalid or inefficient branches during the search process to reduce the amount of calculation, using simulation tools on simulation platforms such as ROS and Gazebo to simulate complex environments, test algorithm performance, and deploy intelligent inspection robots in real environments to collect data and evaluate the actual effects of the algorithm. At the same time, it searches for the Pareto optimal solution set through genetic algorithms, provides a variety of path planning schemes for selection, and can also convert multiple optimization objectives into a single objective function, achieving multi-objective balance by adjusting weight coefficients.

[0068] In the embodiment of the present application, in step S4, the intelligent inspection robot performs autonomous movement according to the optimal path, and collects data on the inspection object and the environment where the inspection point is located at the set point.

[0069] Autonomous movement involves the intelligent inspection robot following a planned path to the designated inspection points, performing inspection operations and collecting data at those points. The intelligent inspection robot autonomously moves along the planned path and to the designated inspection points. The intelligent inspection robot compares its own positioning information within the global map with the location information of the designated inspection points within the global map to determine whether it has reached the inspection point. If it has, it conducts an inspection at the designated point and uses its onboard detection equipment to collect data on the inspection object and the environment surrounding the inspection point. It can also record and livestream live images through cameras.

[0070] Specifically, the intelligent inspection robot moves autonomously according to the planned path and the set inspection points. The intelligent inspection robot determines whether it has reached the inspection point by comparing its own positioning information in the global map with the position information of the set inspection point in the global map. If it has arrived, it will conduct an inspection at the set point and use the onboard detection equipment, including temperature sensors, gas detectors, etc., to collect data on the inspection object and the environment where the inspection point is located. At the same time, it can record and broadcast live images through cameras.

[0071] S4 is equipped with a variety of high-precision sensors, including temperature sensors, high-definition cameras, infrared thermal imagers, and gas detection sensors, which can comprehensively collect various data such as sound, images, temperature, humidity, gas concentration, etc. in the inspection area. The high-definition camera in S4 supports real-time transmission of high-definition video. Operators can view images and videos of the inspection site in real time through the remote monitoring interface. It also has a flame target detection system based on deep learning, which can receive signals from other flame sensors. Operators can view the fire situation through remote monitoring and send flame identification signals, so that they can autonomously move to the position sent by the flame sensor, perform precise positioning and extinguish the fire, realizing remote monitoring and fire rescue work. In S5, the data will be transmitted to the cloud or local server for more in-depth analysis, and machine learning algorithms will be used to explore the laws behind the data to provide a scientific basis for decision-making.

[0072] In this embodiment, the gas detection sensor adopts an electrochemical-photoacoustic combined gas sensor, which utilizes the oxidation or reduction reaction of electrochemically active gases. The electrochemical module detects the composition and concentration of the gas through mechanisms such as primary cell type, stable potential electrolytic cell type, concentration cell type or limiting current type. It is very effective in detecting toxic and harmful gases. The photoacoustic module simultaneously detects gases such as methane; the flame sensor adopts an ultraviolet-infrared dual-band flame detection sensor, which uses an ultraviolet lamp to illuminate the target. When the target generates a flame, the flame absorbs the ultraviolet radiation and generates a spectral signal in a characteristic wavelength region. The sensor can detect this spectral signal and respond. It has a fast response speed and a long detection distance. It is suitable for detecting open flames and uses infrared thermal radiation to detect smoldering fires, thereby making the functions of the patrol intelligent inspection robot more comprehensive and avoiding the spread of fire when a fire occurs.

[0073] In the embodiment of the present application, in step S5, the intelligent inspection robot processes and analyzes the collected data, preliminarily identifies potential anomalies, and transmits all data to the cloud.

[0074] After reading the data information from all sensors, the intelligent inspection robot performs preliminary analysis through the built-in data processing unit. It preliminarily identifies potential anomalies or risk points by simple methods such as checking whether the data exceeds or falls below the set threshold, and transmits all data to the cloud for staff observation and cloud processing.

[0075] In an optional embodiment, after completing data collection at multiple inspection points, the intelligent inspection robot compares the change trend of the same inspection point based on historical record data. If the current value deviates from the historical average by more than a set ratio, it is marked as an abnormal trend point.

[0076] In another optional embodiment, after completing data collection, the intelligent inspection robot performs abnormal pattern recognition through an embedded rule base. The rule base has preset characteristics of several equipment failures (such as a combination of sudden temperature rise and increased gas concentration). If a match is successful, it is preliminarily identified as an abnormal event.

[0077] By directly comparing the sensor collected values ​​with preset thresholds, the present invention constructs a local anomaly recognition mechanism with low computational complexity and high real-time responsiveness, enabling intelligent inspection robots to quickly discover potential risk data on site and perform local marking processing, providing timely and accurate initial screening support for subsequent in-depth cloud analysis.

[0078] In the embodiment of the present application, in step S6, the intelligent inspection robot uses a lightweight multimodal flame detection model and a risk detection model to identify anomalies based on the input information of sensors and cameras, and analyzes the state evolution trend through the spatiotemporal graph convolutional network ST-GCN to generate alarm information and inspection reports.

[0079] On the intelligent inspection robot side, with sensors and cameras as input, deep learning and image recognition artificial intelligence technology are used to deploy a lightweight multimodal flame detection model to identify open flames and smoldering fires, automatically identify the status of the inspection object, and alarm for abnormal conditions. At the same time, a log report is generated and uploaded to the cloud. With gas, temperature, and image information as input, a multimodal risk detection model is deployed, and data comparison is used to analyze whether there are risk issues. On the cloud side, the data uploaded by the intelligent inspection robot is used as input, and multiple information such as gas, temperature, and images are integrated to analyze the state evolution trend through the spatiotemporal graph convolution network.

[0080] Specifically, on the intelligent inspection robot side, with flame sensors and cameras as input, deep learning and image recognition artificial intelligence technologies are used to deploy a lightweight multimodal flame detection model to identify open flames and smoldering fires, automatically identify the status of the inspection object, and alarm for abnormal conditions. At the same time, a log report is generated and uploaded to the cloud. With gas, temperature, image and other information as input, a multimodal risk detection model is deployed to analyze whether there are risk issues through data comparison; on the cloud side, the data uploaded by the intelligent inspection robot in S5 is used as input, and various information such as gas, temperature, and images are integrated to analyze the state evolution trend through the spatiotemporal graph convolution network (ST-GCN). The spatial graph convolution and temporal convolution are as follows:

[0081]

[0082] Among them, X t is the sensor data matrix, A k is the data adjacency matrix, Λ k is the weight matrix, W k is the data transformation matrix, Z t is the spatial convolution result, Δ is the window range of temporal convolution, W t is the temporal convolution kernel, Y is the result of spatiotemporal convolution, and finally connected to the Softmax layer to express the dynamic trend of risk evolution:

[0083]

[0084] Among them, FC represents the fully connected layer, which performs dimension mapping and finally predicts the risk probability.

[0085] The status of objects in S6 includes: the on / off status of equipment, instrument readings, and line damage. By comparing with preset standard images or data models, the intelligent inspection robot can accurately detect abnormal conditions, including abnormal temperature rise, gas leakage, and equipment failure, and immediately generate alarm information. At the same time, the intelligent inspection robot has a built-in wireless communication module that supports 4G / 5G and Wi-Fi communication methods, and has a stable connection with the command center or other equipment. In case of emergency, the intelligent inspection robot can quickly send an alarm information to the command center to ensure that the problem is handled in a timely manner.

[0086] In the embodiment of the present application, in step S7, the intelligent inspection robot performs an automatic return-to-nest charging operation according to the inspection cycle and the remaining power status to ensure the continuous completion of the task.

[0087] The inspection task is started according to the preset inspection cycle, and when the power conditions are met, it is triggered to automatically navigate to the charging pile for charging.

[0088] Specifically, in low power mode (<15%), the automatic return function is triggered and the vehicle returns to the charging station for charging.

[0089] For disaster detection, a dual-band ultraviolet-infrared flame sensor is used. It integrates ultraviolet spectra, infrared thermal radiation, and camera image data to collect and annotate flame data. A lightweight multimodal flame detection model is trained and deployed on the intelligent patrol robot to achieve dual detection of open and smoldering fires. The detection response time is <0.5 seconds, and the false alarm rate is <1%. Upon detecting a flame, the intelligent patrol robot triggers an alarm, saves the current image, generates a log report, and uploads it to the cloud. An electrochemical-photoacoustic gas sensor is used. The electrochemical module detects toxic gases such as CO and H2S (with an accuracy of ±1ppm), while the photoacoustic module simultaneously monitors methane (with a sensitivity of 0.1%LEL), eliminating cross-interference. The gas sensor triggers an alarm and uploads it to the cloud when the detected gas concentration exceeds or falls below a set threshold. Using data from multiple sensors as input, the intelligent patrol robot deploys a risk detection model that compares the data to determine whether there are any risks. After abnormal data is uploaded to the cloud, the cloud platform integrates various information such as gas, temperature, and images, analyzes state evolution trends through a spatiotemporal graph convolutional network (ST-GCN), and predicts risk probability (confidence ≥ 90%). A sensor self-calibration mechanism is implemented, fusing multi-sensor data through a Kalman filter to correct measurement deviations caused by temperature drift and aging in real time. In the event of a sensor failure, a redundant data reconstruction algorithm is activated, using LSTM to predict missing data.

[0090] S7 supports flexible inspection task configuration. It can conduct regular inspections according to preset cycles to ensure daily maintenance and upkeep of equipment. It can also conduct special inspections based on specific needs, including troubleshooting and status comparison before and after equipment replacement. To ensure the safety of the inspection process, the intelligent inspection robot is equipped with multiple safety protection measures, including anti-collision protection, fall detection, and high-temperature power-off. When the power level is lower than the preset threshold, it can automatically return to the charging station for charging, thereby ensuring the continuity of the inspection task.

[0091] After the inspection is completed, the intelligent inspection robot can automatically generate an inspection report, including detailed information such as inspection time, location, inspection results, abnormal records, etc. These reports can be sent to relevant personnel via email, text message, etc. to facilitate subsequent analysis and processing. At the same time, all data collected by the intelligent inspection robot will be properly stored in the cloud or local server, supporting data query, export, analysis and other functions.

[0092] Example 3, reference Figure 2 , is an embodiment of the present invention, which provides an inspection system for an intelligent inspection robot, including a task analysis module, an environment perception module, a path planning module, an autonomous inspection module, a data processing module, an abnormality recognition module and a charging control module.

[0093] The task parsing module is used by the intelligent inspection robot to receive inspection task instructions and parse and obtain the inspection location, inspection time and inspection content.

[0094] The environmental perception module is used to use the onboard sensors to perceive the information of the inspection environment, build or update the point cloud map of the inspection environment, and provide a navigation map.

[0095] The path planning module is used for intelligent inspection robots to perform path planning using potential field and graph search fusion algorithm and genetic algorithm, plan multiple inspection paths to be selected, and calculate the optimal path through dynamic programming algorithm.

[0096] The autonomous inspection module is used for the intelligent inspection robot to perform autonomous movement according to the optimal path, and to collect data on the inspection object and the environment where the inspection point is located at the set point.

[0097] The data processing module is used by the intelligent inspection robot to process and analyze the collected data, preliminarily identify potential anomalies, and transmit all data to the cloud.

[0098] The anomaly recognition module is used by the intelligent inspection robot to identify anomalies based on the input information of sensors and cameras, using a lightweight multimodal flame detection model and a risk detection model, and analyzing the state evolution trend through the spatiotemporal graph convolutional network ST-GCN to generate alarm information and inspection reports.

[0099] The charging control module is used for the intelligent inspection robot to automatically return to the nest for charging according to the inspection cycle and remaining power status to ensure the continuous completion of the task.

[0100] This embodiment also provides an electronic device suitable for a patrol method of an intelligent patrol robot, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement a patrol method of an intelligent patrol robot as proposed in the above embodiment.

[0101] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, an inspection method of an intelligent inspection robot as proposed in the above embodiment is implemented.

[0102] The storage medium proposed in this embodiment and the inspection method for implementing an intelligent inspection robot proposed in the above embodiment belong to the same inventive concept. Technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0103] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general-purpose hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0104] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An inspection method for an intelligent inspection robot, characterized in that: include, The intelligent inspection robot receives the inspection task instructions and parses them to obtain the inspection location, inspection time and inspection content; Utilize the onboard sensors to perceive the information of the inspection environment, and build or update the point cloud map of the inspection environment to provide a navigation map; The intelligent inspection robot uses a potential field and graph search fusion algorithm and a genetic algorithm for path planning. It plans multiple candidate inspection paths and calculates the optimal path through a dynamic programming algorithm. The intelligent inspection robot moves autonomously along the optimal path and collects data on the inspection object and the environment at the set points; The intelligent inspection robot processes and analyzes the collected data, preliminarily identifies potential anomalies, and transmits all data to the cloud; Based on input from sensors and cameras, the intelligent inspection robot uses a lightweight multimodal flame detection model and a risk detection model to identify anomalies. It also uses a spatiotemporal graph convolutional network (ST-GCN) to analyze state evolution trends and generate alarm information and inspection reports. The intelligent inspection robot automatically returns to its nest to recharge based on the inspection cycle and remaining power status to ensure the continuous completion of tasks.

2. The inspection method of an intelligent inspection robot according to claim 1, characterized in that: Providing the navigation map includes providing the navigation map by rasterizing the point cloud map, which includes information on whether all spatial points are occupied.

3. The inspection method of an intelligent inspection robot according to claim 2, characterized in that: The autonomous movement includes the intelligent inspection robot moving to the set inspection points in sequence according to the planned path, performing inspection operations and collecting data at the inspection points.

4. The inspection method of an intelligent inspection robot according to claim 3, characterized in that: The automatic return to home charging operation is performed according to the inspection cycle and the remaining power status, including: The inspection task is started according to the preset inspection cycle, and when the power conditions are met, it is triggered to automatically navigate to the charging pile for charging.

5. The inspection method of an intelligent inspection robot according to claim 4, characterized in that: The path planning includes generating an initial path using a fusion potential field algorithm and a graph search algorithm. The graph search algorithm searches using a node total cost function, wherein the total cost function includes a heuristic cost and an actual cost calculated by Euclidean distance. An artificial potential field method is used to calculate the attractive and repulsive forces acting on the path points, the initial path is adjusted, and a curvature-continuous trajectory is generated through quintic B-spline curve interpolation. The path is constructed by using GPU-accelerated genetic algorithms, and a path cost function is constructed based on the path length, smoothness, and the distance from the path point to the obstacle. At the same time, dynamic programming is introduced to calculate the energy consumption per unit distance of the path, and the path whose unit distance power consumption does not exceed the power consumption threshold is preferentially selected as the final inspection path.

6. The inspection method of an intelligent inspection robot according to claim 4, characterized in that: The autonomous movement is performed according to the optimal path, and data collection is performed on the inspection object and the environment where the inspection point is located at the set point, including: The intelligent inspection robot moves autonomously according to the planned path and the set inspection points. The intelligent inspection robot determines whether it has reached the inspection point by comparing its own positioning information in the global map with the position information of the set inspection point in the global map. If it has arrived, it will conduct an inspection at the set point and use the onboard detection equipment to collect data on the inspection object and the environment where the inspection point is located. At the same time, it can record and broadcast live images through cameras.

7. The inspection method of an intelligent inspection robot according to claim 4, characterized in that: The abnormality identification includes deploying a lightweight multimodal flame detection model on the intelligent inspection robot side, using sensors and cameras as input, and utilizing deep learning and image recognition artificial intelligence technologies to identify open flames and smoldering fires, automatically identifying the status of the inspection object, and alarming for abnormal conditions. At the same time, a log report is generated and uploaded to the cloud. Using gas, temperature, and image information as input, a multimodal risk detection model is deployed to analyze whether there are risk issues through data comparison; In the cloud, data uploaded by intelligent inspection robots is used as input, and multiple information such as gas, temperature, and images are integrated to analyze the state evolution trend through the spatiotemporal graph convolutional network.

8. An inspection system for an intelligent inspection robot, using the inspection method for an intelligent inspection robot according to any one of claims 1 to 7, characterized in that: include: Task analysis module, environmental perception module, path planning module, autonomous inspection module, data processing module, anomaly identification module and charging control module; The task parsing module is used for the intelligent inspection robot to receive inspection task instructions and parse and obtain the inspection location, inspection time and inspection content; The environmental perception module is used to use the onboard sensors to perceive the information of the inspection environment, and to construct or update the point cloud map of the inspection environment and provide a navigation map; The path planning module is used for the intelligent inspection robot to perform path planning using the potential field and graph search fusion algorithm and the genetic algorithm to plan multiple inspection paths to be selected, and calculate an optimal path through the dynamic programming algorithm; The autonomous inspection module is used for the intelligent inspection robot to perform autonomous movement according to the optimal path and collect data on the inspection object and the environment where the inspection point is located at the set point; The data processing module is used by the intelligent inspection robot to process and analyze the collected data, preliminarily identify potential anomalies, and transmit all data to the cloud; The anomaly recognition module is used by the intelligent inspection robot to identify anomalies based on the input information from sensors and cameras, using a lightweight multimodal flame detection model and a risk detection model, and analyze the state evolution trend through the spatiotemporal graph convolutional network ST-GCN to generate alarm information and inspection reports; The charging control module is used for the intelligent inspection robot to perform automatic return-to-nest charging operations according to the inspection cycle and remaining power status to ensure continuous completion of the task.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the inspection method of the intelligent inspection robot according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the inspection method of an intelligent inspection robot according to any one of claims 1 to 7 are implemented.

Images