Intelligent robot for power inspection and inspection method thereof

Abstract

The present invention provides an intelligent robot for power inspection and an inspection method thereof, comprising a quadruped robot, wherein a laser radar, a voiceprint sensor and an infrared sensor are provided at the front end of the quadruped robot; a pan-tilt base is provided at the middle of the quadruped robot, and a camera is provided on the pan-tilt base; a six-degree-of-freedom mechanical arm is also provided at the tail of the quadruped robot, and a mechanical claw is provided at the end of the mechanical arm. The robot flexibly plans tasks based on the layout of the substation, sets inspection cycles for different equipment according to their importance, and automatically generates accurate task plans, thereby avoiding the tediousness and omissions of manual planning and improving planning. The advanced path planning algorithm enables the robot to quickly plan the optimal path, integrates multiple positioning technologies to ensure accurate navigation, and effectively reduces travel time. Multiple inspection modes are carried out in parallel, non-stop inspections are dynamic screening, routine inspections are comprehensive inspections, and non-preset inspections are flexibly targeted at specific points, which greatly improves the coverage efficiency of the inspection and avoids detection blind spots.

Description

Technical Field

[0001] The present invention relates to the field of electric power inspection, and in particular to an intelligent robot for electric power inspection and an inspection method thereof. Background Art

[0002] As a key infrastructure of modern society, the safe and stable operation of the power system is crucial to economic development and social life. As the core hub of the power system, the substation has numerous and complex equipment, including various primary equipment such as transformers, circuit breakers, mutual inductors, busbars, and secondary equipment such as protection devices, measurement and control devices, and communication equipment. The normal operation of these equipment is the basis for ensuring the reliability of power supply.

[0003] Traditional power inspections mainly rely on manual methods, with operation and maintenance personnel regularly inspecting, checking, recording data, and performing maintenance operations on substation equipment. However, as the scale of the power system continues to expand and the number of substations continues to increase, manual inspections face many severe challenges.

[0004] Taking State Grid Corporation of China as an example, the number of substations has increased significantly over the past few years, while the number of substation operation and maintenance personnel has shown a downward trend. This has led to a sharp increase in the per capita operation and maintenance workload, excessive workload of operation and maintenance personnel, and long working hours. Long-term and high-intensity work not only affects the work efficiency of operation and maintenance personnel, but also easily causes fatigue and negligence, increases the risk of missed detection and false detection, makes it difficult to improve the level of refined management, and cannot meet the requirements of high-quality development of power companies.

[0005] Substation equipment is of various types and has complex and diverse operating conditions. Equipment is distributed in different areas, including indoor distribution rooms and outdoor equipment areas. Some equipment is located at high places or in narrow spaces, making it difficult for manual inspections to fully cover them. At the same time, the operating status of equipment is affected by many factors, such as temperature, humidity, and electromagnetic interference. Some potential faults are difficult to detect in time through simple manual observation. For example, overheating hazards and slight discharge phenomena inside the equipment may not have obvious appearance characteristics in the early stage and are easily overlooked by manual inspections, thus burying safety hazards and affecting the stable operation of the power system.

[0006] Manual inspections have an inherent contradiction in terms of efficiency and quality. In order to ensure the safety of equipment, regular and comprehensive inspections are required, but this consumes a lot of time and manpower. In actual operation, due to time and manpower constraints, it may not be possible to perform high-frequency and high-precision inspections on all equipment. For example, in some large substations, it may take hours or even days to complete a comprehensive inspection, and during this period the equipment status may change, resulting in the inability to grasp the real-time operating status of the equipment in a timely manner, making it difficult to meet the current requirements of strong smart grids for the safe and stable operation of substation equipment.

[0007] In order to cope with the shortcomings of manual inspections, some automated inspection technologies have gradually been applied, but there are still many limitations.

[0008] The existing inspection robots have a high failure rate in actual applications, resulting in low attendance and online rates, which affects the continuity and reliability of inspection work. For example, the robot's mechanical structure may be easily damaged in complex environments, the electrical system may fail, and the sensor accuracy may decrease over time. At the same time, the robot's inspection accuracy needs to be improved. The robot's ability to identify the status of some complex equipment is limited, and it cannot accurately judge the subtle abnormalities of the equipment.

[0009] In terms of detection methods, although some equipment such as cameras and infrared sensors have been equipped, the data processing and analysis methods are not smart and efficient enough. For image recognition, infrared spectrum analysis and voiceprint data analysis, there is a lack of deep integration and accurate judgment capabilities. For example, in image recognition, it may not be possible to accurately distinguish between normal changes in the appearance of the equipment and abnormal defects; infrared spectrum analysis may not be able to accurately determine the specific cause of abnormal temperature distribution inside the equipment; voiceprint data analysis may have difficulty identifying potential faults represented by subtle changes in the sound of equipment operation in complex environments.

[0010] In the current inspection system, the integration and coordination between the various components are not ideal. For example, the video surveillance system, robot inspection system, drone inspection system, etc. are often independent of each other and lack effective data sharing and linkage mechanisms. This results in different systems duplicating work or having detection blind spots during the inspection process, making it impossible to achieve all-round, multi-level collaborative inspections, reducing the overall inspection efficiency and effectiveness. Summary of the invention

[0011] The main purpose of the present invention is to provide an intelligent robot for power inspection and its inspection method, so as to solve the problem that traditional power inspection mainly relies on manual methods, and operation and maintenance personnel regularly inspect, check, record data and perform maintenance operations on substation equipment, and manual inspection faces many severe challenges.

[0012] In order to solve the above technical problems, the technical solution adopted by the present invention is: an intelligent robot for power inspection, including a quadruped robot, a laser radar, a voiceprint sensor and an infrared sensor are arranged at the front end of the quadruped robot; A pan-tilt base is provided in the middle of the quadruped robot, and a camera is provided on the pan-tilt base; A six-degree-of-freedom robotic arm is also provided at the tail of the quadruped robot, and a robotic claw is provided at the end of the robotic arm.

[0013] In the preferred embodiment, a communication module is also provided inside the quadruped robot, and the communication module is connected to the control center.

[0014] In a preferred embodiment, the method comprises: S1. Inspection task formulation: select inspection areas, intervals, equipment and points according to the actual situation of the substation, and set the inspection cycle; The system generates a mission template and plan that includes routes, stops, inspection items, and data collection requirements; S2, inspection execution: The robot starts and checks its own status, establishes a communication connection with the inspection host to receive task instructions; Use fusion positioning and navigation technology to navigate autonomously, avoid obstacles, and adapt to complex terrain; Accurately dock and adjust the PTZ according to the calculated parameters, collect data, and report any abnormalities promptly; Dynamically detect or merge stops while driving to reduce the number of stops; Plan the path and posture according to the selected target point and conduct inspections to collect data; S3, data processing and analysis: real-time transmission of collected data and classification and storage to establish archives; Use AI algorithms to analyze images, infrared spectra, and voiceprint data and convert them into structured data; The associated data and equipment records automatically generate inspection reports and can be queried and exported; S4. Special situation handling: automatic door opening and closing and narrow dead-end road scenarios: identify the door status or sense the environment when passing through special areas, and promptly report problems; Multi-floor indoor station inspection: determine the floor and choose the appropriate method to go to the target floor for inspection; Communication interruption and fault recovery: Start the local emergency program when communication is interrupted, stop running and report information when a fault occurs, and continue inspection after waiting for repair.

[0015] In the preferred embodiment, the method includes: the specific method of step S1 is: S11. The operation and maintenance personnel select the areas, intervals, equipment and specific points that need to be inspected in the monitoring system according to the actual layout and equipment distribution of the substation; Different inspection cycles can be set for different areas, intervals, equipment and points, including daily, weekly or monthly inspections to meet the importance and operating status monitoring needs of different equipment; S12, Task template and plan generation: The system automatically generates a task template based on the inspection content selected by the operation and maintenance personnel. The template includes the inspection route, stop points, inspection items and data collection requirements; The docking point includes automatically calculated docking pose and gimbal parameters; S13. According to the task template and the set cycle, the system automatically generates a task plan to determine the execution time, execution order and execution equipment of each inspection task, that is, the quadruped robot.

[0016] In a preferred embodiment, the method comprises: S11.1. Area, bay, equipment, and point selection and representation: Defining substation area sets , interval set ,in For Region The number of intervals in the device collection ,in For interval The number of devices in the ,in For equipment The number of points on The selection operation of the operation and maintenance personnel is carried out through a selection matrix To record, the elements Defined as: ; This selection matrix can clearly show the selection of each area, interval, equipment and point by the operation and maintenance personnel, providing a clear basis for subsequent task planning and facilitating the system to generate task templates and plans in a targeted manner; S11.2. Inspection cycle setting and matrix construction: Construction of inspection cycle matrix , whose elements Indicates the corresponding point The inspection cycle is: ; in is the inspection cycle set for this point; S12.1. Task template and plan generation: Stop point related calculation: The robot is located at two adjacent points and The cost function of moving between , which is related to the distance between points, terrain complexity, and obstacle distribution factors; For the distance factor, let the Euclidean distance between two points be ; in and are the coordinates of two points respectively, and the terrain complexity coefficient is , obtained through analysis of terrain data, including slope and flatness; The obstacle influence factor is , according to the number, size and position of obstacles between two points, the movement cost function can be expressed as: ; The mobility cost function is used to calculate the difficulty and cost of the robot moving between different points, so that when planning the inspection route, the path with the lowest total cost can be selected to improve the inspection efficiency and avoid the robot from getting stuck in difficult-to-pass areas. According to the mobile cost function, the dynamic programming algorithm is used to calculate the optimal path from the starting point to each point, and the path matrix is ​​obtained. , where the elements Indicates from the starting point to the point The previous point on the optimal path; The path matrix provides clear route guidance for the robot's navigation during the inspection process, ensuring that the robot can visit each point that needs to be inspected in sequence according to the optimal path; For each point ,According to the robot's kinematic model and sensor parameters, the parking posture and gimbal parameters are calculated; Assume the robot's body coordinate system is , the world coordinate system is , the transformation matrix from the world coordinate system to the body coordinate system is , which is a function of the robot’s position and attitude, and is updated in real time based on the robot’s positioning and orientation information; Assume the coordinates of the device in the world coordinate system are , calculate the pitch angle of the gimbal according to the visual sensor's field of view and detection accuracy requirements , yaw angle And the attitude angle and roll angle of the robot body in the world coordinate system , Pitch angle , yaw angle ; Enables the testing equipment to accurately align with the equipment for testing; The pitch angle can be calculated based on the device height and the distance between the robot and the device using trigonometric functions. Assume that the horizontal distance between the robot and the device is , the equipment height is ,but ; Accurate docking posture and gimbal parameter calculation can ensure that the robot obtains high-quality detection data at each inspection point, improving the accuracy and reliability of detection; S12.2. Determine the test items and data collection requirements: Construct a test item matrix , where the elements Defined as: ; The inspection item matrix specifies the inspection items that need to be performed at each inspection point, so that the robot can start the corresponding inspection equipment in a targeted manner; For each test project, determine the data collection requirements matrix based on its type and equipment requirements. , where the elements Contains the frequency, duration, and resolution parameters of data collection; The data collection requirement matrix ensures that the robot can operate within the specified parameters when collecting data; A task template can be represented as a structure ,in is a matrix containing the parking posture and gimbal parameters of each point; S13.1. Execution time calculation: According to the inspection cycle matrix and the current system time , calculate the execution time matrix of each inspection task ; Set inspection cycle The time interval is , then for each point , its execution time It can be calculated as: ; The above formula can accurately determine the execution time of each inspection task according to the set inspection cycle and current time, ensuring that the tasks are carried out in an orderly manner according to the predetermined plan and realizing regular monitoring of the equipment; S13.2. Execution order determination: Assign a priority value to each point based on the importance of the equipment and the risk factor of failure. ; Priority can be determined through expert experience, statistical analysis of historical failure data, etc.; For important points on key equipment, the priority value is higher; For non-critical equipment or points that are not prone to failure, the priority value is lower; Sort all points according to the priority value to get the execution order matrix , where the elements Indicates point Position in the inspection sequence; S13.3. Execution equipment allocation: Total quadruped robots are available, and their state vector is ,in Indicates The status of a robot, with a value of 0 or 1, 0 means failure or busy, and 1 means available; According to the robot's status and location information, as well as the distribution of tasks, the execution equipment is assigned to each inspection task; A greedy algorithm is used to give priority to the robot that is closest to the starting point of the task and is in an available state; suppose the robot The position coordinates are , the coordinates of the task starting point are , then the distance between the robot and the starting point of the task is ; The process of selecting an execution device can be expressed as an allocation function , which returns the robot number assigned to the task; Executing the equipment allocation function can reasonably allocate inspection tasks to available robots and improve the utilization of robot resources; The task plan can be represented as a structure ,in The execution equipment matrix, whose elements are the robot numbers assigned to each task, clarifies the execution time, sequence, and execution equipment of each inspection task, providing comprehensive guidance for the robot's inspection execution; In the preferred embodiment, the specific method of step S2 is: S21. When the inspection task start time arrives, the quadruped robot automatically starts and performs initialization operations, including checking whether its own sensors, including cameras, infrared sensors, voiceprint sensors, communication modules, and power systems are working properly; Establish a stable communication connection with the inspection host and receive task instructions and related parameters, including inspection routes and stop point information; S22. According to the pre-planned route, the quadruped robot autonomously navigates to the inspection area using fusion positioning and navigation technology; During the journey, the vehicle monitors the surrounding environment in real time and avoids obstacles; When encountering complex terrain, the quadruped robot can safely pass through with its off-road obstacle-crossing ability, adjusting its pace and posture to adapt to terrain changes to ensure stable progress; S23. When arriving at the preset stop point, the quadruped robot stops accurately according to the automatically calculated stop position to ensure that the best detection viewing angle can be obtained; at the same time, the direction of the pan-tilt is adjusted according to the calculated pan-tilt parameters so that the detection device camera, voiceprint sensor and infrared sensor are aimed at the target device or area; According to the task requirements, start the corresponding detection equipment to collect data; use the visible light camera to capture the appearance of the equipment, use the infrared sensor to collect the temperature map of the equipment, and use the voiceprint sensor to record the sound characteristics of the equipment operation; During the inspection process, if any abnormal conditions are found, including obvious deformation of the appearance, abnormal temperature rise, or abnormal sound, the abnormal information will be immediately sent back to the inspection host and marked locally for further inspection; S24. For non-stop inspections, provided that the inspection requirements are met, the quadruped robot uses advanced image recognition and data acquisition technology to dynamically capture key information, operating status, and appearance features of the equipment during its movement; By quickly photographing and analyzing the status of the logo and indicator lights on the equipment, and using infrared sensors to monitor the temperature changes of the equipment in real time; For inspections with few stops, the robot uses offline marking technology to rationally plan the stop locations and merge adjacent stops with similar inspection requirements. When arriving at the merged stops, the quadruped robot adjusts its posture and pan / tilt direction to complete the inspection of multiple targets at one time, thereby reducing the number of stops and improving inspection efficiency. S25. When non-preset inspection is required, the operation and maintenance personnel select any target point in the monitoring system (it can be a specific part of the equipment or any position within the area); The quadruped robot automatically plans the best path and parking position based on its real-time position, gimbal parameters and the spatial position of the target point using an algorithm based on a three-dimensional model or point cloud. During the planning process, the robot's motion capability, environmental factors and the viewing angle requirements of the detection equipment are considered to ensure that the target point can be reached accurately and high-quality detection data can be obtained. The patrol system goes to the target point according to the planned route for inspection and data collection. The collected data is also transmitted back to the patrol host in real time for analysis and processing.

[0017] In the preferred embodiment, the specific method of step S2 is: S21.1. Define the sensor set of the quadruped robot ,in arrive For the camera, arrive For infrared sensor, arrive For the voiceprint sensor, and the power system state variable and communication module status variables ; For each sensor, its state detection function is ; The power system status detection can be judged by checking whether the motor speed and torque output are within the normal range, whether the battery power is sufficient, etc. Assume that the conditions for the normal operation of the power system are ,but ; The communication module status detection is determined by sending a test signal and receiving feedback. Assuming that the condition for normal communication is ; These status detection functions can accurately determine the working status of each key component of the robot, ensuring that the robot is in good operating condition before performing inspection tasks, avoiding inspection failures or inaccurate data due to equipment failures; S21.2. Overall state judgment: The overall state judgment function of the robot is: ; This function can quickly determine whether the robot is qualified to perform inspection tasks. Only when all key components are working properly can the robot proceed to the next step, ensuring the safety and reliability of the inspection process. S21.3, communication connection attempt: Assume the address of the patrol host is , the communication port is $PortNumber$; the function for establishing a communication connection between the robot and the patrol host is ; This function uses TCP / IP protocol to attempt a connection, and sets the connection timeout to ; If a stable socket connection can be established with the host within the timeout period, the connection is considered successful, otherwise the connection fails; S21.3. If the communication connection is successful, the robot receives the task instruction structure sent by the patrol host. ,in Inspection route information, including a series of path point coordinates; The docking point information includes the position, posture and gimbal parameters of each docking point; Extract the inspection route information from the task instruction structure. Assume that the route consists of m path points and the path point coordinate matrix is: ,in For the The three-dimensional coordinates of the path points; Extract the stop information, the number of stops is , the stop point information matrix is ,in , is the coordinate of the stop point, is the attitude angle of the docking position, are the gimbal parameters; S22.1, the robot's current position is , the entrance position of the target inspection area is ; Define the map environment information matrix as , where the elements Indicates the location on the map terrain and obstacle information, such as represents the feasible region, Indicates an obstacle; Improved The cost function of the algorithm is ,in: From the starting point to the node The actual cost is calculated as ,in From the starting point to the node on the path of Points, From point To point The cost of movement, the distance between the two points, and the terrain difficulty coefficient and obstacle risk factor about, that is , is the Euclidean distance between two points; For slave nodes The heuristic estimated cost to the target point is calculated using the weighted Manhattan distance, that is, ,in is the weight coefficient, which is determined according to the characteristics of the map and the motion characteristics of the robot; For Node The risk assessment value of the surrounding environment is calculated as follows: ,in For Node The surrounding neighborhood area, To determine the terrain and obstacle information The calculated risk value is higher for the obstacle area and lower for the flat feasible area. is the risk coefficient weight factor, which is used to balance the safety and efficiency of the path; S22.2, during navigation, the robot integrates visual odometer, inertial measurement unit and global positioning system GPS information for positioning; Assume that the relative displacement of the robot measured by the visual odometer is , the acceleration measured by the inertial measurement unit is , the angular velocity is , the position measured by the global positioning system is ; The extended Kalman filter algorithm is used to fuse this information to obtain a more accurate estimate of the robot's position and posture; the state vector is defined as ,in is the robot position coordinate, is the attitude angle, is the velocity vector, and are the biases of the accelerometer and gyroscope, respectively; The state transfer equation is ,in is the control input, namely the acceleration and angular velocity measured by the inertial measurement unit, is the process noise, and the observation equation is ,in is the observation vector, including the relative displacement measured by the visual odometer and the position measured by the global positioning system, is the observation noise; Through the prediction and update steps of the EKF algorithm, the robot's position and attitude estimates are continuously corrected so that it can accurately navigate along the planned path; S22.3, the image obtained by the robot's visual sensor is ,in is the timestamp, and the processed image is obtained through the image preprocessing algorithm ; Assume the pixel coordinates in the image are , the processed image is analyzed using a deep neural network to identify obstacles and feasible areas in the environment. The pixel area of ​​the obstacle in the image is , the feasible region is ; At the same time, the distance information obtained by the laser radar (6) is ,in is the scanning angle of the sensor, is the distance detected at this angle; Build an environment map based on information from visual sensors and distance sensors , where the elements Indicates location The environmental status is determined based on the obstacle area and feasible area information; Using the improved obstacle avoidance algorithm based on the dynamic window method, the current speed of the robot is , the maximum speed is , the minimum speed is , the velocity resolution is ; Generate a series of candidate velocities in velocity space , where each candidate speed satisfies the speed limit condition, i.e. ; For each candidate speed , predicting that robots will Movement trajectory within ,in The first The position coordinates of the points; Calculate the cost function for each trajectory ,in: is the distance deviation between the trajectory and the target point, calculated as ,in is the end point coordinate of the trajectory); is the collision risk between the trajectory and the obstacle, and the calculation method is ,in For location The collision risk value at the location is determined based on factors such as the distance and shape of the obstacle; is the smoothness cost of speed, calculated as , used to avoid drastic changes in the robot's speed, is the weight coefficient; Select the candidate speed with the lowest cost as the control speed of the robot ; According to the control speed, the control instructions of each joint are calculated through the robot's motion control model ,in For the The control instructions of each joint enable the robot to avoid obstacles along the predetermined trajectory and continue to move towards the target inspection area; S23.1, the current position of the robot is , the attitude angle is , according to the docking point information in the task instruction, the target docking position is ; Calculate position error , , , attitude error , , ; The proportional integral derivative (PID) control algorithm is used to calculate the control quantity of the robot's leg joints and pan / tilt; For position control, the proportionality factor is , the integral coefficient is , the differential coefficient is , then the leg joints are The control quantities of the directions are: ; ; ; For attitude control, the proportional coefficient is and the integral coefficient is , the differential coefficient is , then the control quantities of the gimbal in the attitude angle direction are: ; ; ; The PID control algorithm continuously adjusts the control amount to gradually bring the robot closer to the target parking position according to the position and posture errors, and can effectively reduce overshoot and steady-state error. S23.2. Establish a kinematic model of the robot leg and convert the calculated joint control amount into actual joint motion instructions; The robot legs have joints, and the joint angle vector is , the joint velocity vector is According to the geometric structure and kinematic relationship of the robot, the mapping function from joint space to Cartesian space can be obtained: and the Jacobian matrix ; Known joint control amount , then the joint speed command is ,in is the inverse matrix of the Jacobian matrix; by solving inverse kinematics, the target angle of each joint can be obtained ,in To control the cycle; Convert the joint target angle into a motor control signal and send it to the robot's motor driver to control the movement of the leg joints so that the robot can gradually adjust to the target docking position; The kinematic model and control command conversion ensure that the calculated control quantity can be accurately converted into the actual movement of the robot leg joints; S23.4. Define the data collection time series ,in Indicates collection time; for visible light cameras, at each collection time , get the image frame , the image resolution is , the width is , the height is , the frame rate is ; then in the time period The number of images collected in ,in is the floor rounding function; For infrared sensors, at the time of collection , get the temperature map , the temperature measurement range is , the temperature resolution is ; For the voiceprint sensor, at the time of collection , collect sound signal fragments , the sampling frequency is , the sampling time is ; In order to ensure the synchronization of multi-sensor data, a timestamp synchronization mechanism is used to give an accurate timestamp to the data collected by each sensor, so that different sensor data at the same time can be accurately matched in subsequent analysis; S23.5. The collected visible light image sequence ( ) for preprocessing, including image grayscale and filtering operations, to obtain the preprocessed image sequence ; The preprocessed image is analyzed using a deep learning-based target detection algorithm. The trained target detection model is , the test result is , which contains the detected target category and location information; Define the appearance abnormality judgment function: ; If an abnormal object is detected at any acquisition moment, the device appearance is considered abnormal; The deep learning target detection algorithm can automatically learn the characteristics of the normal appearance and abnormal conditions of the equipment, accurately identify abnormal conditions such as deformation of the equipment appearance and attachment of foreign objects, and promptly discover potential equipment problems; S23.6. For infrared temperature spectrum sequence ( ), calculate the average temperature of each spectrum ,in Indicates the temperature map at the pixel point The temperature value at Calculate the rate of temperature change ; The normal operating temperature range of the device is , the temperature change rate threshold is ; Define temperature anomaly judgment function ; If the temperature abnormality condition is met at any collection moment, the device temperature is considered abnormal; By calculating the average temperature and temperature change rate and comparing them with the set threshold, it is possible to detect overheating or abnormal temperature changes of the equipment in time and prevent equipment failure caused by overheating. S23.7. Sequence of collected sound signal segments ( ) to perform preprocessing, including noise reduction and frame division, to obtain the preprocessed sound signal sequence ; The voiceprint analysis method based on time-frequency analysis and machine learning classification algorithm is used to perform STFT transformation on the pre-processed sound signal to obtain the time-frequency diagram. ; Extracting sound feature vectors from time-frequency graphs , the trained voiceprint classification model is , the classification result is , the classification result is normal sound or abnormal sound; Define the sound abnormality judgment function ; If an abnormal sound is detected at any collection moment, it is considered that the device sound is abnormal; Time-frequency analysis and machine learning classification algorithms can effectively extract features from sound signals, identify abnormal sounds during equipment operation, and assist in determining whether there are hidden faults inside the equipment. S23.8. When any abnormal situation is detected, the abnormal information is immediately packaged, and the abnormal information includes the abnormal type, the timestamp of the abnormal occurrence, the corresponding data and the device location information; The exception information structure is: ; The abnormal information structure is sent back to the patrol host through the communication module, and marked in the robot's local storage system to record the details of the abnormality so that subsequent operation and maintenance personnel can conduct further inspection and analysis; Accurate processing and reporting of abnormal information enables operation and maintenance personnel to timely understand the abnormal status of equipment and take appropriate measures to deal with it, avoid further development of equipment failures, and ensure the safe operation of substation equipment; S24.1. Non-stop inspection technology: During the robot's movement, Continuously acquire visible light image sequences ( ); Use lightweight deep learning target detection algorithms (such as YOLOv5-Tiny) to process images in real time. This algorithm has a high processing speed while ensuring a certain degree of accuracy and can meet real-time requirements; The target detection model is , the test result is , quickly identify key equipment components and operating status in images; At the same time, infrared sensors are used to Collect temperature data series ( ), and calculate the temperature change trend in real time; suppose the sliding window size of temperature data is , then at time The temperature change trend is calculated as ; The lightweight target detection algorithm can quickly process images while the robot is moving, and timely obtain the appearance and operating status information of the equipment. Combined with the temperature change trend analysis of the infrared sensor, it can realize dynamic monitoring of the equipment and improve inspection efficiency. S24.2. Determine in real time whether the equipment is abnormal based on the image recognition results and temperature change trends; The normal judgment function of appearance state is: , the temperature state normal judgment function is ; If in multiple consecutive collection moments, or , it is considered that the device may be abnormal; When a possible abnormal situation is detected, the current image, temperature data and related detection results are immediately packaged into an abnormal information structure: and sent back to the patrol host through the communication module; At the same time, all collected data are archived and stored locally on the robot at certain time intervals, so that more detailed analysis can be performed later when needed; The real-time decision-making mechanism can detect and report equipment anomalies in a timely manner, avoiding the deterioration of equipment failures due to failure to handle abnormal situations in a timely manner. Data archiving and storage provides a data basis for subsequent in-depth analysis. S24.3, less-stop inspection technology: based on the 3D point cloud or model, offline point analysis is performed on the inspection area; the equipment set in the inspection area is , each device has a set of potential inspection points ( ); Define the similarity function between two points ,It considers factors such as the spatial distance between points, the similarity of equipment types, and the similarity of inspection items; According to the similarity function, a clustering algorithm is used to cluster all points, and the points with high similarity are merged into one group; let the clustering result be , each cluster Contains several merged stops; The optimization of offline marking and docking point merging can reduce the number of robot stops and improve inspection efficiency. At the same time, reasonable similarity calculation and clustering algorithm can ensure that the merged docking points can meet the inspection requirements without affecting the accuracy of equipment status monitoring. S24.4. When the robot reaches the merged stop point, determine the set of equipment and points to be detected based on the clustering results and task requirements; let the set of equipment and points corresponding to the stop point be ,in For equipment, is a point on the device; Adjust the robot's posture and pan / tilt direction so that the detection equipment can cover all targets that need to be detected; for visible light cameras, calculate the best shooting angle and focal length to ensure that the appearance and status of all target devices can be clearly captured; for infrared sensors, adjust the measurement angle and range to obtain accurate temperature data; for voiceprint sensors, optimize the sound collection direction and sensitivity; At the same time, all relevant detection equipment is started to collect data according to the predetermined data collection requirements; for image collection, the image resolution and frame rate are adjusted according to the number and distribution of targets to obtain sufficiently clear and comprehensive image information; The collected data is marked and sorted in real time. The marking information includes the collection time, the corresponding equipment and point number, and the data type. The sorted data is packaged and sent back to the patrol host. At the same time, it is backed up and stored locally on the robot for subsequent analysis and verification. S25.1, non-preset inspection: when the operation and maintenance personnel select any target point in the monitoring system, the coordinates of the target point in the monitoring system are , which needs to be converted into the coordinates of the robot coordinate system ; Assume that the transformation matrix from the monitoring system coordinate system to the robot coordinate system is ,but ; Transformation matrix It can be obtained by a pre-calibrated method, for example, by using the coordinates of multiple reference points of known positions in two coordinate systems and calculating them by a least square method or the like; Accurate coordinate transformation enables the robot to understand the location of the target point in its own coordinate system, providing correct target location information for subsequent path planning and navigation; S25.2, based on the robot's current position is the starting point, the target point coordinates As the end point, using improved Algorithms for path planning; Improved The cost function of the algorithm is ,in: From the starting point to the node The actual cost is calculated as ,in From the starting point to the node on the path of Points, From point To point The cost of movement, the distance between the two points, and the terrain difficulty coefficient and obstacle risk factor about, that is , is the Euclidean distance between two points; For slave nodes The heuristic estimated cost to the target point is calculated using the weighted Euclidean distance, i.e. ,in is the weight coefficient, which is determined according to the environmental characteristics and robot motion characteristics; For Node The risk assessment value of the surrounding environment is calculated as follows: ,in For Node The surrounding neighborhood area, To determine the terrain and obstacle information The calculated risk value is higher for the obstacle area and lower for the flat feasible area. is the risk coefficient weight factor, which is used to balance the safety and efficiency of the path; Improved The algorithm cost function can more accurately evaluate the cost of the path based on the map environment information, so that the robot can plan the path efficiently and safely, avoiding dangerous or difficult-to-pass areas. S25.3. The planned original path may have some broken lines or discontinuous parts, which are not conducive to the smooth driving of the robot; the spline curve interpolation algorithm is used to smooth the path; Assume the original path point sequence is ,in ), after spline curve interpolation, the smoothed path point sequence is obtained ,in , ; The cubic spline interpolation function satisfies the continuity of the function value, first-order derivative and second-order derivative at each path point, ensuring the smoothness of the path; At the same time, the smoothed path is optimized, considering the kinematic constraints of the robot; the minimum turning radius of the robot is set to , the maximum speed is ; For each turning point in the path, check whether its curvature meets the minimum turning radius requirement. If not, make local adjustments to the path; calculate the reasonable driving speed of each path segment based on the path length and the maximum speed of the robot to ensure that the robot can ensure safety and improve efficiency during driving; Path smoothing and optimization can improve the stability and safety of the robot's driving, reduce the vibration and energy consumption of the robot during driving, and at the same time meet the robot's kinematic constraints to ensure the feasibility of the path; S25.4, Posture adjustment and data collection: When the robot moves along the planned path to the target point, according to the current position and attitude angle , and the spatial position of the target point , calculate the docking position in real time; Calculate position error , , , attitude error , , ,in is the desired attitude angle at the target point, which is determined according to the detection requirements of the target device and the installation position of the robot's detection equipment; The real-time feedback control algorithm is used to calculate the control quantity of the robot's leg joints and the gimbal. The MPC algorithm predicts the robot's motion state in the future and calculates the optimal control sequence based on the prediction error and control target. The prediction time domain is assumed to be , the control time domain is , then in each control cycle, the MPC algorithm solves the following optimization problem: ; Subject to: Robot kinematic model constraints, control quantity constraints in, In the prediction time domain The posture error of the step, For the The optimal control amount at the current moment is obtained by solving this optimization problem and sent to the robot's motor driver to control the leg joints and the gimbal movement, so that the robot gradually approaches the target parking position. The model predictive control algorithm can calculate the optimal control quantity in real time according to the current state and target posture of the robot, and realize accurate posture adjustment while considering future motion trends and constraints, so as to ensure that the robot can accurately stop at the target point and obtain high-quality detection data; S25.5. When the robot reaches the target point and adjusts to the position, start the corresponding detection equipment to collect data according to the task requirements; use the visible light camera to take a high-definition image of the target device or area with an image resolution of , the frame rate is ; During the image acquisition process, the camera's focal length, exposure time and other parameters are automatically adjusted according to the characteristics of the target device and the detection focus to obtain clear and accurate image information; The infrared sensor is used to collect the temperature map of the target device. The temperature measurement range is , the temperature resolution is ; Adjust the measuring angle and range of the infrared sensor according to the shape and size of the equipment to ensure that the temperature distribution of the equipment can be fully and accurately obtained; The voiceprint sensor is used to record the sound characteristics of the target device, with a sampling frequency of , the sampling time is ;Optimize the collection direction and sensitivity of the voiceprint sensor, reduce the interference of environmental noise, and improve the quality of the sound signal; The collected data is marked and sorted in real time, and the marking information includes the collection time, target point number, and data type; the sorted data is transmitted back to the patrol host in real time through the high-speed communication module to ensure the timeliness and integrity of the data; at the same time, it is backed up and stored locally on the robot for subsequent review or further analysis when needed.

[0018] In the preferred embodiment, the method includes: the specific method of step S3 is: S31. The images, infrared spectra, and voiceprint data collected by the quadruped robot during the inspection process are transmitted back to the inspection host in real time through a secure and stable network; The inspection host stores the received data according to the equipment, time, and inspection items, and establishes a complete inspection data file for subsequent query, analysis, and comparison; S32, the intelligent analysis host uses the AI ​​algorithm based on deep learning to analyze the collected data, and for visible light images, it identifies the appearance status of the device, the operating status of the device, and the behavior of personnel; For infrared spectra, analyze the temperature distribution of the equipment to determine whether there is a risk of overheating; For voiceprint data, analyze the characteristics of the device's operating sound to determine whether the device has abnormal sounds; Convert analysis results into structured data, including equipment status information; S33, associating the analyzed inspection data with the equipment ledger, and corresponding the inspection results to specific equipment asset information, including name, model, number, and interval; Automatically generate inspection reports based on the associated data. The report content includes basic information of the inspection task, equipment status summary, abnormal situation analysis and recommended measures; Inspection reports can be queried and exported in the monitoring system to provide a basis for decision-making for operation and maintenance personnel, so that they can timely understand the operating status of the equipment and arrange subsequent maintenance and inspection work.

[0019] In a preferred embodiment, the method comprises: S31.1. Data transmission and storage: Pack and mark each type of data, and encrypt the data packets using an asymmetric encryption algorithm; Let the public key be , the private key is ; For image data packets , The encrypted data packet is: ; Similarly, ; ; The robot transmits the encrypted data packets back to the patrol host through a secure and stable network; After receiving the encrypted data packet, the patrol host first uses the private key to decrypt it. , , ; Design a data storage structure with the device as the root directory. Under each device, store data according to time and test items. Suppose the device set is , for equipment , create a time subdirectory under its directory , create a detection project subdirectory under each time subdirectory , corresponding to image, infrared spectrum, and voiceprint data respectively; Establish a data index table to record the storage location and related information of the data. Through the data index table, you can quickly query and retrieve the required data to improve the efficiency of data management; The data index table provides a way to quickly find data, reduces the time complexity of data query, improves the response speed of the system, and facilitates the management and use of inspection data by operation and maintenance personnel; S32.1. Image analysis based on deep learning: For the received visible light image data , first preprocessing is performed; suppose the image preprocessing function is , which includes image denoising, using algorithms such as bilateral filtering, ), image enhancement and image scaling operations; Convolutional neural networks are used for feature extraction. The feature vector contains the texture, shape, and color information of the image, which can effectively characterize the appearance and operating status of the device. Construct a classification model based on deep learning for equipment status recognition; suppose the classification model is , taking the extracted feature vector as input and outputting it as the device status category; Define the device status category set as: ; Respectively represent normal equipment appearance, abnormal appearance, normal operation, abnormal operation, personnel presence, and personnel violation behavior; For the feature vector , the standby state recognition result is ; The classification model can accurately judge the appearance and operating status of the equipment and whether there are any personnel-related situations based on image features, and promptly discover potential equipment problems and safety hazards; S32.2, infrared spectrum analysis: for the received infrared spectrum data , firstly, temperature data extraction and calibration are performed; let the temperature data extraction function be , which converts the temperature information in the infrared spectrum into a matrix of actual temperature values ,in and is the size of the infrared spectrum, is the coordinate in the map The temperature value at The temperature data is calibrated, considering the influence of factors such as ambient temperature and emissivity. The calibration function is: , the calibrated temperature matrix is: ; in Collection time Ambient temperature at: and the standard deviation of the temperature distribution ; Temperature data processing and analysis can accurately obtain the actual temperature information of the equipment. Statistics such as average temperature and standard deviation can help determine whether the equipment has overheating risks, providing an important basis for evaluating the equipment's operating status. S32.3. The normal operating temperature range of the equipment is , the temperature change rate threshold is ; Define the overheating hazard judgment function: ; If the judgment result is 1, it is considered that the device has a potential overheating risk and requires further attention and inspection; S32.4. Voiceprint feature extraction and processing: For the received voiceprint data ,First, pre-processing is performed including audio noise reduction, framing and windowing operations; Perform short-time Fourier transform on each frame of audio data to obtain a time-frequency diagram representation; the STFT transform function is: , then the time-frequency diagram matrix is: ,in is the total number of frames; Extract voiceprint feature vectors from the time-frequency graph. MFCC features can effectively characterize the spectral characteristics of sound, which is of great significance for feature extraction of device operation sound. Construct an abnormal sound classification model based on machine learning. The classification model is , taking the extracted voiceprint feature vector as input and outputting it as the voice state category; S32.5. Define the sound state category set as , for the feature vector , the sound status judgment result is ; The results of image analysis, infrared spectrum analysis and voiceprint data analysis are structured and integrated; the structured data structure is assumed to be: ,in is the device number, is the collection time, is the device status obtained by image analysis, The equipment status obtained by infrared spectrum analysis, that is, the result of overheating hazard judgment, The device status obtained by analyzing the voiceprint data. Take the value in S33.1. Data association and report generation: The equipment ledger database contains an equipment information table, including equipment number, equipment name, equipment model, equipment number, and interval number to which it belongs; Read the device information from the device inventory database and create a device information index table, where DeviceInfo searches for the corresponding device information in the device information index table based on the device number; Design inspection report templates, which include basic inspection task information, equipment status summary, abnormal situation analysis and recommended measures; Fill in the inspection report template according to the associated data structure.

[0020] In the preferred embodiment, the specific method of step S4 is: S41. When encountering an automatic door opening and closing scenario, the quadruped robot uses its visual sensor or other detection device to identify the location and status of the door; when approaching the door, it interacts with the access control system, and then safely passes through the door to enter the next area to continue inspection; if the door fails to open normally, the robot will stop moving and send an alarm message to the inspection host, waiting for remote intervention or on-site processing by the operation and maintenance personnel; In complex passage scenarios such as narrow dead-end roads, the quadruped robot uses its own sensors to accurately sense the surrounding environment, adjust its body posture and pace, and carefully pass through narrow areas; if it encounters an impassable obstacle or the end of a dead-end road, the robot will handle it according to the pre-set strategy to ensure its own safety and continue to complete the inspection task; S42. When inspecting a multi-floor indoor substation, the quadruped robot first determines its own floor and the target inspection floor. If there is an elevator or lifting equipment available, the robot communicates with the elevator control system, calls the elevator and enters the elevator safely, and continues the inspection after reaching the target floor. If there is no elevator or lifting equipment, the robot uses its powerful climbing ability to go to the target floor along the stairs or other feasible climbing paths. During the climbing process, the robot always maintains a stable posture to ensure its own safety and the normal operation of the detection equipment, while avoiding damage to the stairs or surrounding facilities. After reaching the target floor, the robot inspects the equipment on that floor and collects data according to the pre-planned inspection route and task requirements. S43. If the communication between the quadruped robot and the patrol host is interrupted during the inspection process, the robot will immediately start the local emergency procedure; First, try to re-establish the communication connection, including checking the status of the communication module and adjusting the communication parameters. At the same time, the inspection continues according to the pre-set task plan and the collected data is temporarily stored in the local memory; If the communication interruption lasts for a long time or cannot be restored, the robot will stop running and enter standby mode after completing the current inspection task or reaching a certain safe position, waiting for communication to be restored or for on-site processing by the operation and maintenance personnel; After the communication is restored, the robot will upload the locally stored data to the patrol host in a timely manner, and report the inspection situation and its own status during the communication interruption to the patrol host; S44. When the robot itself fails, it will stop running immediately and send detailed fault information to the patrol host. The operation and maintenance personnel will remotely diagnose the cause of the fault based on the fault information and decide whether it can be repaired through remote operation. If remote repair is not possible, the operation and maintenance personnel will go to the site for repair. After the fault is eliminated, the robot will perform self-inspection and functional testing to ensure that it resumes normal operation before continuing the patrol task.

[0021] The present invention provides an intelligent robot for electric power inspection and an inspection method thereof. The application of the intelligent robot for electric power inspection and the inspection method thereof brings about significant beneficial effects in many aspects.

[0022] In terms of inspection efficiency, the robot flexibly plans tasks based on the substation layout, sets inspection cycles for different equipment according to importance, and automatically generates accurate task plans, avoiding the tediousness and omissions of manual planning and improving planning. Advanced path planning algorithms enable the robot to quickly plan the optimal path, integrate multiple positioning technologies to ensure accurate navigation, and effectively reduce travel time. Multiple inspection modes are carried out in parallel, non-stop inspections are dynamic screening, routine inspections are comprehensive inspections, and non-preset inspections are flexible for specific points, which greatly improves the coverage efficiency of inspections and avoids detection blind spots.

[0023] In terms of detection accuracy, the collaborative work of multiple sensors is the key. High-definition cameras, infrared sensors, and voiceprint sensors collect equipment information from multiple dimensions, and appearance, temperature, and sound data complement each other for verification. The AI ​​algorithm based on deep learning deeply analyzes the data, accurately identifies various equipment states and personnel behaviors, accurately determines the type of fault, and converts the results into structured data, providing detailed decision-making basis for operation and maintenance, reducing the risk of misjudgment and missed judgment.

[0024] The system reliability is enhanced. The quadruped robot can adapt to complex terrains, conduct comprehensive self-inspection to ensure the normal operation of key components, monitor the status in real time, and can self-repair or report abnormalities in a timely manner. In terms of communication, encryption algorithms are used to ensure data transmission security. The inspection host reasonably classifies and stores data, establishes index tables for easy query, and redundant storage and backup strategies prevent data loss, providing reliable data support for inspections.

[0025] Resource allocation is optimized. Inspection resources are allocated based on equipment importance and status assessment, key equipment is inspected in detail, and plans are adjusted dynamically in case of abnormalities to avoid waste of resources. Robots take on daily inspection work to free up operation and maintenance personnel, and inspection reports and analysis results help them make efficient decisions and improve overall operation and maintenance efficiency.

[0026] Application scenarios have also been effectively expanded. In complex environments such as multi-floor indoor stations, the robot can determine the floor in a variety of ways and choose a suitable path to the target floor for inspection. In special areas such as automatic doors and narrow dead-end roads, the robot can accurately identify and adopt strategies to pass through. It cooperates with drones to achieve air-ground integrated inspection and covers the substation in all directions. It cooperates with the main and auxiliary equipment monitoring systems to receive alarms and feedback data, jointly ensure the safe operation of equipment, and improve the intelligence level and comprehensive performance of power inspection.

[0027] To sum up, the power inspection intelligent robot and its inspection method have effectively solved many problems of traditional inspection and provided a strong guarantee for the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a structural diagram of a quadruped robot of the present invention; Figure 2 This is a front view structural diagram of the quadruped robot of the present invention; Figure 3 This is a schematic diagram of the operation and walking of the quadruped robot of the present invention; Figure 4 This is a schematic diagram of a quadruped robot patrolling according to the present invention; Figure 5 This is a diagram of the quadruped robot of the present invention patrolling without stopping.

[0029] In the figure: quadruped robot 1; robotic arm 2; robotic claw 3; camera 4; gimbal base 5; laser radar 6; voiceprint sensor 7; infrared sensor 8. DETAILED DESCRIPTION

[0030] Example 1 like Figure 1-5 As shown, an intelligent robot for power inspection includes a quadruped robot 1, and a laser radar 6, a voiceprint sensor 7 and an infrared sensor 8 are arranged at the front end of the quadruped robot 1; A pan-tilt base 5 is provided at the middle of the quadruped robot 1, and a camera 4 is provided on the pan-tilt base 5; LiDAR 6: Used to build a 3D map of the surrounding environment, helping the robot to achieve autonomous navigation and obstacle avoidance. LiDAR can accurately measure distances and is one of the important tools for robots to perceive the environment.

[0031] Voiceprint sensor 7: It can be used to monitor the sound changes when the device is running, and detect potential faults or abnormalities by analyzing the sound patterns, which helps to find and solve problems early.

[0032] Infrared sensor 8: used to monitor temperature distribution, especially suitable for detecting problems such as overheating of electrical equipment, which is of great significance for preventing safety accidents such as fire.

[0033] The pan / tilt base 5 can rotate 360 ​​degrees, so that the camera 4 installed thereon can observe the surrounding environment in all directions, obtain clear video images, and provide support for remote monitoring.

[0034] A six-degree-of-freedom mechanical arm 2 is also provided at the tail of the quadruped robot 1, and a mechanical claw 3 is provided at the end of the mechanical arm 2. The mechanical arm can complete delicate operation tasks, greatly expanding the application scope of the robot.

[0035] In the preferred embodiment, a communication module is also provided inside the quadruped robot 1, and the communication module is connected to the control center to ensure data transmission between the robot and the control center, so that the operator can monitor the status of the robot in real time at a location far away from the site and issue necessary instructions to the robot.

[0036] Example 2 Further illustrate with reference to Example 1, Figure 1-5 The structure shown in the figure, the method comprises: S1, setting inspection tasks: selecting inspection areas, intervals, equipment and points according to the actual situation of the substation, and setting the inspection cycle; The system generates a mission template and plan that includes routes, stops, inspection items, and data collection requirements; S2, inspection execution: The robot starts and checks its own status, establishes a communication connection with the inspection host to receive task instructions; Use fusion positioning and navigation technology to navigate autonomously, avoid obstacles, and adapt to complex terrain; Accurately dock and adjust the PTZ according to the calculated parameters, collect data, and report any abnormalities promptly; Dynamically detect or merge stops while driving to reduce the number of stops; Plan the path and posture according to the selected target point and conduct inspections to collect data; S3, data processing and analysis: real-time transmission of collected data and classification and storage to establish archives; Use AI algorithms to analyze images, infrared spectra, and voiceprint data and convert them into structured data; The associated data and equipment records automatically generate inspection reports and can be queried and exported; S4. Special situation handling: automatic door opening and closing and narrow dead-end road scenarios: identify the door status or sense the environment when passing through special areas, and promptly report problems; Multi-floor indoor station inspection: determine the floor and choose the appropriate method to go to the target floor for inspection; Communication interruption and fault recovery: Start the local emergency program when communication is interrupted, stop running and report information when a fault occurs, and continue inspection after waiting for repair.

[0037] In the preferred embodiment, the method includes: the specific method of step S1 is: S11. The operation and maintenance personnel select the areas, intervals, equipment and specific points that need to be inspected in the monitoring system according to the actual layout and equipment distribution of the substation; Different inspection cycles can be set for different areas, intervals, equipment and points, including daily, weekly or monthly inspections to meet the importance and operating status monitoring needs of different equipment; S12, Task template and plan generation: The system automatically generates a task template based on the inspection content selected by the operation and maintenance personnel. The template includes the inspection route, stop points, inspection items and data collection requirements; The docking point includes automatically calculated docking pose and gimbal parameters; S13. According to the task template and the set cycle, the system automatically generates a task plan to determine the execution time, execution order and execution equipment of each inspection task, that is, the quadruped robot.

[0038] In a preferred embodiment, the method comprises: S11.1. Area, bay, equipment, and point selection and representation: Defining substation area sets , interval set ,in For Region The number of intervals in the device collection ,in For interval The number of devices in the ,in For equipment The number of points on The selection operation of the operation and maintenance personnel is carried out through a selection matrix To record, the elements Defined as: ; This selection matrix can clearly show the selection of each area, interval, equipment and point by the operation and maintenance personnel, providing a clear basis for subsequent task planning and facilitating the system to generate task templates and plans in a targeted manner; S11.2. Inspection cycle setting and matrix construction: Construction of inspection cycle matrix , whose elements Indicates the corresponding point The inspection cycle is: ; in is the inspection cycle set for this point; The inspection cycle matrix specifies the inspection frequency of each selected point, which helps the system to arrange inspection tasks reasonably and ensure that important equipment can be monitored more frequently. At the same time, it can also optimize the allocation of inspection resources while ensuring the safe operation of equipment.

[0039] S12.1. Task template and plan generation: Stop point related calculation: The robot is located at two adjacent points and The cost function of moving between , which is related to the distance between points, terrain complexity, and obstacle distribution factors; For the distance factor, let the Euclidean distance between two points be ; in and are the coordinates of two points respectively, and the terrain complexity coefficient is , obtained through analysis of terrain data, including slope and flatness; The obstacle influence factor is , according to the number, size and position of obstacles between two points, the movement cost function can be expressed as: ; The mobility cost function is used to calculate the difficulty and cost of the robot moving between different points, so that when planning the inspection route, the path with the lowest total cost can be selected to improve the inspection efficiency and avoid the robot from getting stuck in difficult-to-pass areas. According to the mobile cost function, the dynamic programming algorithm is used to calculate the optimal path from the starting point to each point, and the path matrix is ​​obtained. , where the elements Indicates from the starting point to the point The previous point on the optimal path; The path matrix provides clear route guidance for the robot's navigation during the inspection process, ensuring that the robot can visit each point that needs to be inspected in sequence according to the optimal path; For each point ,According to the robot's kinematic model and sensor parameters, the parking posture and gimbal parameters are calculated; Assume the robot's body coordinate system is , the world coordinate system is , the transformation matrix from the world coordinate system to the body coordinate system is , which is a function of the robot’s position and attitude, and is updated in real time based on the robot’s positioning and orientation information; Assume the coordinates of the device in the world coordinate system are , calculate the pitch angle of the gimbal according to the visual sensor's field of view and detection accuracy requirements , yaw angle And the attitude angle and roll angle of the robot body in the world coordinate system , Pitch angle , yaw angle ; Enables the testing equipment to accurately align with the equipment for testing; The pitch angle can be calculated based on the device height and the distance between the robot and the device using trigonometric functions. Assume that the horizontal distance between the robot and the device is , the equipment height is ,but ; Accurate docking posture and gimbal parameter calculation can ensure that the robot obtains high-quality detection data at each inspection point, improving the accuracy and reliability of detection; S12.2. Determine the test items and data collection requirements: Construct a test item matrix , where the elements Defined as: ; The inspection item matrix specifies the inspection items that need to be performed at each inspection point, so that the robot can start the corresponding inspection equipment in a targeted manner; For each test project, determine the data collection requirements matrix based on its type and equipment requirements. , where the elements Contains the frequency, duration, and resolution parameters of data collection; The data collection requirement matrix ensures that the robot can operate within the specified parameters when collecting data; A task template can be represented as a structure ,in is a matrix containing the parking posture and gimbal parameters of each point; S13.1. Execution time calculation: According to the inspection cycle matrix and the current system time , calculate the execution time matrix of each inspection task ; Set inspection cycle The time interval is , then for each point , its execution time It can be calculated as: ; in is the ceiling function.

[0040] The above formula can accurately determine the execution time of each inspection task according to the set inspection cycle and current time, ensuring that the tasks are carried out in an orderly manner according to the predetermined plan and realizing regular monitoring of the equipment; S13.2. Execution order determination: Assign a priority value to each point based on the importance of the equipment and the risk factor of failure. ; Priority can be determined through expert experience, statistical analysis of historical failure data, etc.; For important points on key equipment, the priority value is higher; For non-critical equipment or points that are not prone to failure, the priority value is lower; Sort all points according to the priority value to get the execution order matrix , where the elements Indicates point Position in the inspection sequence; The execution order matrix ensures that when resources are limited, important equipment and high-risk points are inspected first, which improves the pertinence and effectiveness of inspections and promptly discovers and handles problems that may affect the safe operation of the power grid.

[0041] S13.3. Execution equipment allocation: Total quadruped robots are available, and their state vector is ,in Indicates The status of a robot, with a value of 0 or 1, 0 means failure or busy, and 1 means available; According to the robot's status and location information, as well as the distribution of tasks, the execution equipment is assigned to each inspection task; A greedy algorithm is used to give priority to the robot that is closest to the starting point of the task and is in an available state; suppose the robot The position coordinates are , the coordinates of the task starting point are , then the distance between the robot and the starting point of the task is ; The process of selecting an execution device can be expressed as an allocation function , which returns the robot number assigned to the task; Executing the device allocation function can reasonably allocate inspection tasks to available robots, improve the utilization of robot resources, ensure that inspection tasks can be executed in a timely manner, and at the same time reduce the robot's idle driving distance and improve energy efficiency.

[0042] The task plan can be represented as a structure ,in It is the execution equipment matrix, whose elements are the robot numbers assigned to each task. It clarifies the execution time, sequence and execution equipment of each inspection task, providing comprehensive guidance for the robot's inspection execution.

[0043] Example 3 Further illustrate with reference to Example 1, Figure 1-5 The structure shown in FIG. 1 is as follows: S21. When the inspection task start time arrives, the quadruped robot automatically starts and performs initialization operations, including checking whether its own sensors, including the camera 4, infrared sensor 8, voiceprint sensor 7, communication module, and power system are working normally; Establish a stable communication connection with the inspection host and receive task instructions and related parameters, including inspection routes and stop point information; S22. According to the pre-planned route, the quadruped robot autonomously navigates to the inspection area using fusion positioning and navigation technology; During the journey, the vehicle monitors the surrounding environment in real time and avoids obstacles; When encountering complex terrain, the quadruped robot can safely pass through with its off-road obstacle-crossing ability, adjusting its pace and posture to adapt to terrain changes to ensure stable progress; S23. When arriving at the preset docking point, the quadruped robot docks accurately according to the automatically calculated docking posture to ensure that the best detection viewing angle can be obtained; at the same time, the direction of the pan-tilt is adjusted according to the calculated pan-tilt parameters so that the detection device camera 4, the voiceprint sensor 7 and the infrared sensor 8 are aligned with the target device or area; According to the task requirements, start the corresponding detection equipment to collect data; use the visible light camera 4 to capture the appearance image of the equipment, use the infrared sensor 8 to collect the temperature spectrum of the equipment, and use the voiceprint sensor 7 to record the sound characteristics of the equipment operation; During the inspection process, if any abnormal conditions are found, including obvious deformation of the appearance, abnormal temperature rise, or abnormal sound, the abnormal information will be immediately sent back to the inspection host and marked locally for further inspection; S24. For non-stop inspections, provided that the inspection requirements are met, the quadruped robot uses advanced image recognition and data acquisition technology to dynamically capture key information, operating status, and appearance features of the equipment during its movement; By quickly photographing and analyzing the status of the logo and indicator lights on the equipment, and using infrared sensors to monitor the temperature changes of the equipment in real time; For inspections with few stops, the robot uses offline marking technology to rationally plan the stop locations and merge adjacent stops with similar inspection requirements. When arriving at the merged stops, the quadruped robot adjusts its posture and pan / tilt direction to complete the inspection of multiple targets at one time, thereby reducing the number of stops and improving inspection efficiency. S25. When non-preset inspection is required, the operation and maintenance personnel select any target point in the monitoring system (it can be a specific part of the equipment or any position within the area); The quadruped robot automatically plans the best path and parking position based on its real-time position, gimbal parameters and the spatial position of the target point using an algorithm based on a three-dimensional model or point cloud. During the planning process, the robot's motion capability, environmental factors and the viewing angle requirements of the detection equipment are considered to ensure that the target point can be reached accurately and high-quality detection data can be obtained. The patrol system goes to the target point according to the planned route for inspection and data collection. The collected data is also transmitted back to the patrol host in real time for analysis and processing.

[0044] In the preferred embodiment, the specific method of step S2 is: S21.1. Define the sensor set of the quadruped robot ,in arrive For the camera, arrive For infrared sensor, arrive For the voiceprint sensor, and the power system state variable and communication module status variables ; For each sensor, its state detection function is ; The power system status detection can be judged by checking whether the motor speed and torque output are within the normal range, whether the battery power is sufficient, etc. Assume that the conditions for the normal operation of the power system are ,but ; The communication module status detection is determined by sending a test signal and receiving feedback. Assuming that the condition for normal communication is ; These status detection functions can accurately determine the working status of each key component of the robot, ensuring that the robot is in good operating condition before performing inspection tasks, avoiding inspection failures or inaccurate data due to equipment failures; S21.2. Overall state judgment: The overall state judgment function of the robot is: ; This function can quickly determine whether the robot is qualified to perform inspection tasks. Only when all key components are working properly can the robot proceed to the next step, ensuring the safety and reliability of the inspection process. S21.3, communication connection attempt: Assume the address of the patrol host is , the communication port is $PortNumber$; the function for establishing a communication connection between the robot and the patrol host is ; This function uses TCP / IP protocol to attempt a connection, and sets the connection timeout to ; If a stable socket connection can be established with the host within the timeout period, the connection is considered successful, otherwise the connection fails; Ensure that a reliable communication link can be established between the robot and the inspection host to provide guarantee for the subsequent receipt of task instructions and uploading of inspection data.

[0045] S21.3. If the communication connection is successful, the robot receives the task instruction structure sent by the patrol host. ,in Inspection route information, including a series of path point coordinates; The docking point information includes the position, posture and gimbal parameters of each docking point; Extract the inspection route information from the task instruction structure. Assume that the route consists of m path points and the path point coordinate matrix is: ,in For the The three-dimensional coordinates of the path points; Extract the stop information, the number of stops is , the stop point information matrix is ,in , is the coordinate of the stop point, is the attitude angle of the docking position, are the gimbal parameters; This information will directly guide the robot's navigation, docking and detection operations during the inspection process, ensuring that the robot can accurately perform inspection tasks according to the predetermined plan.

[0046] S22.1, the robot's current position is , the entrance position of the target inspection area is ; Define the map environment information matrix as , where the elements Indicates the location on the map terrain and obstacle information, such as represents the feasible region, Indicates an obstacle; Improved The cost function of the algorithm is ,in: From the starting point to the node The actual cost is calculated as ,in From the starting point to the node on the path of Points, From point To point The cost of movement, the distance between the two points, and the terrain difficulty coefficient and obstacle risk factor about, that is , is the Euclidean distance between two points; For slave nodes The heuristic estimated cost to the target point is calculated using the weighted Manhattan distance, that is, ,in is the weight coefficient, which is determined according to the characteristics of the map and the motion characteristics of the robot; For Node The risk assessment value of the surrounding environment is calculated as follows: ,in For Node The surrounding neighborhood area, To determine the terrain and obstacle information The calculated risk value is higher for the obstacle area and lower for the flat feasible area. is the risk coefficient weight factor, which is used to balance the safety and efficiency of the path; Improved The algorithm cost function can more accurately evaluate the cost of the path based on map environment information (such as terrain and obstacles), so that the path planned by the robot is both efficient and safe, avoiding dangerous or difficult-to-navigate areas.

[0047] S22.2, during navigation, the robot integrates visual odometer, inertial measurement unit and global positioning system GPS information for positioning; Assume that the relative displacement of the robot measured by the visual odometer is , the acceleration measured by the inertial measurement unit is , the angular velocity is , the position measured by the global positioning system is ; The extended Kalman filter algorithm is used to fuse this information to obtain a more accurate estimate of the robot's position and posture; the state vector is defined as ,in is the robot position coordinate, is the attitude angle, is the velocity vector, and are the biases of the accelerometer and gyroscope, respectively; The state transfer equation is ,in is the control input, namely the acceleration and angular velocity measured by the inertial measurement unit, is the process noise, and the observation equation is ,in is the observation vector, including the relative displacement measured by the visual odometer and the position measured by the global positioning system, is the observation noise; Through the prediction and update steps of the EKF algorithm, the robot's position and attitude estimates are continuously corrected so that it can accurately navigate along the planned path; The fusion positioning and navigation technology utilizes the complementarity of information from multiple sensors and effectively integrates this information through the EKF algorithm to improve the positioning accuracy and navigation stability of the robot in complex environments, ensuring that the robot can accurately reach the target inspection area.

[0048] S22.3, the image obtained by the robot's visual sensor is ,in is the timestamp, and the processed image is obtained through the image preprocessing algorithm ; Assume the pixel coordinates in the image are , the processed image is analyzed using a deep neural network to identify obstacles and feasible areas in the environment. The pixel area of ​​the obstacle in the image is , the feasible region is ; At the same time, the distance information obtained by the laser radar (6) is ,in is the scanning angle of the sensor, is the distance detected at this angle; Accurate environmental perception data processing can provide the robot with detailed information about the surrounding environment, help the robot identify obstacles and feasible paths, and provide a basis for obstacle avoidance decisions.

[0049] Build an environment map based on information from visual sensors and distance sensors , where the elements Indicates location The environmental status is determined based on the obstacle area and feasible area information; Using the improved obstacle avoidance algorithm based on the dynamic window method, the current speed of the robot is , the maximum speed is , the minimum speed is , the velocity resolution is ; Generate a series of candidate velocities in velocity space , where each candidate speed satisfies the speed limit condition, i.e. ; For each candidate speed , predicting that robots will Movement trajectory within ,in The first The position coordinates of the points; Calculate the cost function for each trajectory ,in: is the distance deviation between the trajectory and the target point, calculated as ,in is the end point coordinate of the trajectory); is the collision risk between the trajectory and the obstacle, and the calculation method is ,in For location The collision risk value at the location is determined based on factors such as the distance and shape of the obstacle; is the smoothness cost of speed, calculated as , used to avoid drastic changes in the robot's speed, is the weight coefficient; Select the candidate speed with the lowest cost as the control speed of the robot ; According to the control speed, the control instructions of each joint are calculated through the robot's motion control model ,in For the The control instructions of each joint enable the robot to avoid obstacles along the predetermined trajectory and continue to move towards the target inspection area; The improved obstacle avoidance algorithm based on DWA can calculate the optimal control speed and motion trajectory in real time by considering the robot's kinematic constraints, target point position and surrounding obstacle distribution, so that the robot can navigate safely and efficiently in complex environments and avoid collisions with obstacles.

[0050] S23.1, the current position of the robot is , the attitude angle is , according to the docking point information in the task instruction, the target docking position is ; Calculate position error , , , attitude error , , ; The proportional integral derivative (PID) control algorithm is used to calculate the control quantity of the robot's leg joints and pan / tilt; For position control, the proportionality factor is , the integral coefficient is , the differential coefficient is , then the leg joints are The control quantities of the directions are: ; ; ; For attitude control, the proportional coefficient is and the integral coefficient is , the differential coefficient is , then the control quantities of the gimbal in the attitude angle direction are: ; ; ; The PID control algorithm continuously adjusts the control amount to make the robot gradually approach the target docking position according to the position and posture errors, and can effectively reduce overshoot and steady-state error to achieve precise docking.

[0051] S23.2. Establish a kinematic model of the robot leg and convert the calculated joint control amount into actual joint motion instructions; The robot legs have joints, and the joint angle vector is , the joint velocity vector is According to the geometric structure and kinematic relationship of the robot, the mapping function from joint space to Cartesian space can be obtained: and the Jacobian matrix ; Known joint control amount , then the joint speed command is ,in is the inverse matrix of the Jacobian matrix; by solving inverse kinematics, the target angle of each joint can be obtained ,in To control the cycle; Convert the joint target angle into a motor control signal and send it to the robot's motor driver to control the movement of the leg joints so that the robot can gradually adjust to the target docking position; The kinematic model and control command conversion ensure that the calculated control quantity can be accurately converted into the actual movement of the robot's leg joints, achieving precise posture adjustment and ensuring that the detection equipment can be aligned with the target device or area.

[0052] S23.4. Define the data collection time series ,in Indicates collection time; for visible light cameras, at each collection time , get the image frame , the image resolution is , the width is , the height is , the frame rate is ; then in the time period The number of images collected in ,in is the floor rounding function; For infrared sensors, at the time of collection , get the temperature map , the temperature measurement range is , the temperature resolution is ; For the voiceprint sensor, at the time of collection , collect sound signal fragments , the sampling frequency is , the sampling time is ; In order to ensure the synchronization of multi-sensor data, a timestamp synchronization mechanism is used to give an accurate timestamp to the data collected by each sensor, so that different sensor data at the same time can be accurately matched in subsequent analysis; Accurate data collection synchronization can ensure that the data from different sensors are consistent in time when analyzing the status of the device, avoiding misjudgment due to data asynchrony and improving detection accuracy.

[0053] S23.5. The collected visible light image sequence ( ) for preprocessing, including image grayscale and filtering operations, to obtain the preprocessed image sequence ; The preprocessed image is analyzed using a deep learning-based target detection algorithm. The trained target detection model is , the test result is , which contains the detected target category and location information; Define the appearance abnormality judgment function: ; If an abnormal object is detected at any acquisition moment, the device appearance is considered abnormal; The deep learning target detection algorithm can automatically learn the characteristics of the normal appearance and abnormal conditions of the equipment, accurately identify abnormal conditions such as deformation of the equipment appearance and attachment of foreign objects, and promptly discover potential equipment problems; S23.6. For infrared temperature spectrum sequence ( ), calculate the average temperature of each spectrum ,in Indicates the temperature map at the pixel point The temperature value at Calculate the rate of temperature change ; The normal operating temperature range of the device is , the temperature change rate threshold is ; Define temperature anomaly judgment function ; If the temperature abnormality condition is met at any collection moment, the device temperature is considered abnormal; By calculating the average temperature and temperature change rate and comparing them with the set threshold, it is possible to detect overheating or abnormal temperature changes of the equipment in time and prevent equipment failure caused by overheating. S23.7. Sequence of collected sound signal segments ( ) to perform preprocessing, including noise reduction and frame division, to obtain the preprocessed sound signal sequence ; The voiceprint analysis method based on time-frequency analysis and machine learning classification algorithm is used to perform STFT transformation on the pre-processed sound signal to obtain the time-frequency diagram. ; Extracting sound feature vectors from time-frequency graphs , the trained voiceprint classification model is , the classification result is , the classification result is normal sound or abnormal sound; Define the sound abnormality judgment function ; If an abnormal sound is detected at any collection moment, it is considered that the device sound is abnormal; Time-frequency analysis and machine learning classification algorithms can effectively extract features from sound signals, identify abnormal sounds during equipment operation, and assist in determining whether there are hidden faults inside the equipment. S23.8. When any abnormal situation is detected, the abnormal information is immediately packaged, and the abnormal information includes the abnormal type, the timestamp of the abnormal occurrence, the corresponding data and the device location information; The exception information structure is: ; The abnormal information structure is sent back to the patrol host through the communication module, and marked in the robot's local storage system to record the details of the abnormality so that subsequent operation and maintenance personnel can conduct further inspection and analysis; Accurate processing and reporting of abnormal information enables operation and maintenance personnel to timely understand the abnormal status of equipment and take appropriate measures to deal with it, avoid further development of equipment failures, and ensure the safe operation of substation equipment; S24.1. Non-stop inspection technology: During the robot's movement, Continuously acquire visible light image sequences ( ); Use lightweight deep learning target detection algorithms (such as YOLOv5-Tiny) to process images in real time. This algorithm has a high processing speed while ensuring a certain degree of accuracy and can meet real-time requirements; The target detection model is , the test result is , quickly identify key equipment components and operating status in images; At the same time, infrared sensors are used to Collect temperature data series ( ), and calculate the temperature change trend in real time; suppose the sliding window size of temperature data is , then at time The temperature change trend is calculated as ; The lightweight target detection algorithm can quickly process images while the robot is moving, and timely obtain the appearance and operating status information of the equipment. Combined with the temperature change trend analysis of the infrared sensor, it can realize dynamic monitoring of the equipment and improve inspection efficiency. S24.2. Determine in real time whether the equipment is abnormal based on the image recognition results and temperature change trends; The normal judgment function of appearance state is: , the temperature state normal judgment function is ; If in multiple consecutive collection moments, or , it is considered that the device may be abnormal; When a possible abnormal situation is detected, the current image, temperature data and related detection results are immediately packaged into an abnormal information structure: and sent back to the patrol host through the communication module; At the same time, all collected data (including images, temperature data, etc.) are archived and stored locally on the robot at certain time intervals (set to) so that more detailed analysis can be performed later when needed.

[0054] The real-time decision-making mechanism can promptly detect and report equipment anomalies, avoiding the deterioration of equipment failures due to failure to promptly handle abnormal situations. Data archiving and storage provides a data basis for subsequent in-depth analysis.

[0055] S24.3, less-stop inspection technology: based on the 3D point cloud or model, offline point analysis is performed on the inspection area; the equipment set in the inspection area is , each device has a set of potential inspection points ( ); Define the similarity function between two points ,It considers factors such as the spatial distance between points, the similarity of equipment types, and the similarity of inspection items; For example, spatial distance similarity can be calculated as ,in is the Euclidean distance between two points, is the distance influencing factor; the device type similarity and the detection item similarity can be calculated according to the pre-defined device type and detection item classification, set as and The total similarity is ,in is the weight coefficient.

[0056] According to the similarity function, a clustering algorithm is used to cluster all points, and the points with high similarity are merged into one group; let the clustering result be , each cluster Contains several merged stops; The optimization of offline marking and docking point merging can reduce the number of robot stops and improve inspection efficiency. At the same time, reasonable similarity calculation and clustering algorithm can ensure that the merged docking points can meet the inspection requirements without affecting the accuracy of equipment status monitoring. S24.4. When the robot reaches the merged stop point, determine the set of equipment and points to be detected based on the clustering results and task requirements; let the set of equipment and points corresponding to the stop point be ,in For equipment, is a point on the device; Adjust the robot's posture and pan / tilt direction so that the detection equipment can cover all targets that need to be detected; for visible light cameras, calculate the best shooting angle and focal length to ensure that the appearance and status of all target devices can be clearly captured; for infrared sensors, adjust the measurement angle and range to obtain accurate temperature data; for voiceprint sensors, optimize the sound collection direction and sensitivity; At the same time, all relevant detection equipment is started to collect data according to the predetermined data collection requirements; for image collection, the image resolution and frame rate are adjusted according to the number and distribution of targets to obtain sufficiently clear and comprehensive image information; The collected data is marked and sorted in real time. The marking information includes the collection time, the corresponding equipment and point number, and the data type. The sorted data is packaged and sent back to the patrol host. At the same time, it is backed up and stored locally on the robot for subsequent analysis and verification. The multi-target detection and data collection strategy can reduce the number of stops while ensuring the detection quality of multiple target devices and points, improve inspection efficiency, and accurately obtain the operating status information of the equipment.

[0057] S25.1, non-preset inspection: when the operation and maintenance personnel select any target point in the monitoring system, the coordinates of the target point in the monitoring system are , which needs to be converted into the coordinates of the robot coordinate system ; Assume that the transformation matrix from the monitoring system coordinate system to the robot coordinate system is ,but ; Transformation matrix It can be obtained by a pre-calibrated method, for example, by using the coordinates of multiple reference points of known positions in two coordinate systems and calculating them by a least square method or the like; Accurate coordinate transformation enables the robot to understand the location of the target point in its own coordinate system, providing correct target location information for subsequent path planning and navigation; S25.2, based on the robot's current position is the starting point, the target point coordinates As the end point, using improved Algorithms for path planning; Improved The cost function of the algorithm is ,in: From the starting point to the node The actual cost is calculated as ,in From the starting point to the node on the path of Points, From point To point The cost of movement, the distance between the two points, and the terrain difficulty coefficient and obstacle risk factor about, that is , is the Euclidean distance between two points; For slave nodes The heuristic estimated cost to the target point is calculated using the weighted Euclidean distance, i.e. ,in is the weight coefficient, which is determined according to the environmental characteristics and robot motion characteristics; For Node The risk assessment value of the surrounding environment is calculated as follows: ,in For Node The surrounding neighborhood area, To determine the terrain and obstacle information The calculated risk value is higher for the obstacle area and lower for the flat feasible area. is the risk coefficient weight factor, which is used to balance the safety and efficiency of the path; Improved The algorithm cost function can more accurately evaluate the cost of the path based on map environment information (such as terrain and obstacles), so that the path planned by the robot is both efficient and safe, avoiding dangerous or difficult-to-navigate areas.

[0058] S25.3. The planned original path may have some broken lines or discontinuous parts, which are not conducive to the smooth driving of the robot; the spline curve interpolation algorithm is used to smooth the path; Assume the original path point sequence is ,in ), after spline curve interpolation, the smoothed path point sequence is obtained ,in , ; The cubic spline interpolation function satisfies the continuity of the function value, first-order derivative and second-order derivative at each path point, ensuring the smoothness of the path; At the same time, the smoothed path is optimized, considering the kinematic constraints of the robot; the minimum turning radius of the robot is set to , the maximum speed is ; For each turning point in the path, check whether its curvature meets the minimum turning radius requirement. If not, make local adjustments to the path; calculate the reasonable driving speed of each path segment based on the path length and the maximum speed of the robot to ensure that the robot can ensure safety and improve efficiency during driving; Path smoothing and optimization can improve the stability and safety of the robot's driving, reduce the vibration and energy consumption of the robot during driving, and at the same time meet the robot's kinematic constraints to ensure the feasibility of the path; S25.4, Posture adjustment and data collection: When the robot moves along the planned path to the target point, according to the current position and attitude angle , and the spatial position of the target point , calculate the docking position in real time; Calculate position error , , , attitude error , , ,in is the desired attitude angle at the target point, which is determined according to the detection requirements of the target device and the installation position of the robot's detection equipment; The real-time feedback control algorithm is used to calculate the control quantity of the robot's leg joints and the gimbal. The MPC algorithm predicts the robot's motion state in the future and calculates the optimal control sequence based on the prediction error and control target. The prediction time domain is assumed to be , the control time domain is , then in each control cycle, the MPC algorithm solves the following optimization problem: ; Subject to: Robot kinematic model constraints, control quantity constraints in, In the prediction time domain The posture error of the step, For the The optimal control amount at the current moment is obtained by solving this optimization problem and sent to the robot's motor driver to control the leg joints and the gimbal movement, so that the robot gradually approaches the target parking position. The model predictive control algorithm can calculate the optimal control quantity in real time according to the current state and target posture of the robot, and realize accurate posture adjustment while considering future motion trends and constraints, so as to ensure that the robot can accurately stop at the target point and obtain high-quality detection data; S25.5. When the robot reaches the target point and adjusts to the position, start the corresponding detection equipment to collect data according to the task requirements; use the visible light camera to take a high-definition image of the target device or area with an image resolution of , the frame rate is ; During the image acquisition process, the camera's focal length, exposure time and other parameters are automatically adjusted according to the characteristics of the target device and the detection focus to obtain clear and accurate image information; The infrared sensor is used to collect the temperature map of the target device. The temperature measurement range is , the temperature resolution is ; Adjust the measuring angle and range of the infrared sensor according to the shape and size of the equipment to ensure that the temperature distribution of the equipment can be fully and accurately obtained; The voiceprint sensor is used to record the sound characteristics of the target device, with a sampling frequency of , the sampling time is ;Optimize the collection direction and sensitivity of the voiceprint sensor, reduce the interference of environmental noise, and improve the quality of the sound signal; The collected data is marked and sorted in real time, and the marking information includes the collection time, target point number, and data type; the sorted data is transmitted back to the patrol host in real time through the high-speed communication module to ensure the timeliness and integrity of the data; at the same time, it is backed up and stored locally on the robot for subsequent review or further analysis when needed.

[0059] Accurate data collection and timely data transmission can provide operation and maintenance personnel with real-time equipment status information, making it easier to detect equipment failures or abnormal conditions in a timely manner, take appropriate measures to deal with them, and ensure the safe operation of substation equipment.

[0060] Example 4 Further illustrate with reference to Example 1, Figure 1-5 The structure shown in FIG. 1 is as follows: S31. The images, infrared spectra, and voiceprint data collected by the quadruped robot during the inspection process are transmitted back to the inspection host in real time through a secure and stable network; The inspection host stores the received data according to the equipment, time, and inspection items, and establishes a complete inspection data file for subsequent query, analysis, and comparison; S32, the intelligent analysis host uses the AI ​​algorithm based on deep learning to analyze the collected data, and for visible light images, it identifies the appearance status of the device, the operating status of the device, and the behavior of personnel; For infrared spectra, analyze the temperature distribution of the equipment to determine whether there is a risk of overheating; For voiceprint data, analyze the characteristics of the device's operating sound to determine whether the device has abnormal sounds; Convert analysis results into structured data, including equipment status information; S33, associating the analyzed inspection data with the equipment ledger, and corresponding the inspection results to specific equipment asset information, including name, model, number, and interval; Automatically generate inspection reports based on the associated data. The report content includes basic information of the inspection task, equipment status summary, abnormal situation analysis and recommended measures; Inspection reports can be queried and exported in the monitoring system to provide a basis for decision-making for operation and maintenance personnel, so that they can timely understand the operating status of the equipment and arrange subsequent maintenance and inspection work.

[0061] In a preferred embodiment, the method comprises: S31.1. Data transmission and storage: Pack and mark each type of data, and encrypt the data packets using an asymmetric encryption algorithm; Assume that the image data collected by the quadruped robot is (in is the collection time, For in time The collected images), infrared spectrum data is ( For in time The collected infrared spectrum), the voiceprint data is ( For in time The collected segment of voiceprint data).

[0062] Each type of data is packaged and marked, and the image data is packaged as ,in (DeviceID is the robot device number, TimeStamp is the acquisition timestamp, DataType indicates that the data type is image, $ImageCount$ is the number of images in the data packet), .

[0063] Infrared spectrum data is packaged as ,in (ThermalCount is the number of infrared spectra in this data package), .

[0064] Voiceprint data is packaged as ,in (AudioCount is the number of voiceprint data segments in the data packet), .

[0065] Data packaging and labeling can ensure the integrity and identifiability of data during transmission and storage, facilitating subsequent classification processing and query.

[0066] Adopt asymmetric encryption algorithm, elliptic curve encryption algorithm, ECC; Encrypt the data packet.

[0067] Let the public key be , the private key is ; For image data packets , The encrypted data packet is: ; Similarly, ; ; The robot transmits the encrypted data packets back to the patrol host through a secure and stable network; Data encryption ensures the security of data during transmission and prevents data from being stolen or tampered with. The public key and private key mechanism of the asymmetric encryption algorithm improves the security and efficiency of encryption.

[0068] After receiving the encrypted data packet, the patrol host first uses the private key to decrypt it. , , ; Design a data storage structure with the device as the root directory. Under each device, store data according to time and test items. Suppose the device set is , for equipment , create a time subdirectory under its directory , create a detection project subdirectory under each time subdirectory , corresponding to image, infrared spectrum, and voiceprint data respectively; Design a data storage structure with the device as the root directory, and store data in each device according to time and detection items. , for equipment , create a time subdirectory under its directory Each subdirectory corresponds to a collection time, and a detection project subdirectory is created under each time subdirectory, corresponding to the image, infrared spectrum, and voiceprint data respectively.

[0069] For example, for image data, Stored in Directory, the file name is ( Indicates the time images, Indicates the acquisition time); for infrared spectrum data, Stored in Directory, the file name is ; For voiceprint data, Stored in Directory, the file name is .

[0070] This data classification storage structure facilitates subsequent data query, analysis and comparison, and can quickly locate data for specific equipment, time and test items.

[0071] Establish a data index table to record the storage location and related information of the data. Through the data index table, you can quickly query and retrieve the required data to improve the efficiency of data management; The data index table provides a way to quickly find data, reduces the time complexity of data query, improves the response speed of the system, and facilitates the management and use of inspection data by operation and maintenance personnel; S32.1. Image analysis based on deep learning: For the received visible light image data , first preprocessing is performed; suppose the image preprocessing function is , which includes image denoising, using algorithms such as bilateral filtering, ), image enhancement and image scaling operations; Convolutional neural networks are used for feature extraction. The feature vector contains the texture, shape, and color information of the image, which can effectively characterize the appearance and operating status of the device. Image preprocessing can improve image quality, reduce noise interference, enhance image contrast and clarity, and facilitate subsequent feature extraction and analysis; feature extraction of convolutional neural networks can automatically learn key features in images and provide strong support for equipment status recognition.

[0072] Construct a classification model based on deep learning for equipment status recognition; suppose the classification model is , taking the extracted feature vector as input and outputting it as the device status category; Define the device status category set as: ; Respectively represent normal equipment appearance, abnormal appearance, normal operation, abnormal operation, personnel presence, and personnel violation behavior; For the feature vector , the standby state recognition result is ; The classification model can accurately judge the appearance and operating status of the equipment and whether there are any personnel-related situations based on image features, and promptly discover potential equipment problems and safety hazards; S32.2, infrared spectrum analysis: for the received infrared spectrum data , firstly, temperature data extraction and calibration are performed; let the temperature data extraction function be , which converts the temperature information in the infrared spectrum into a matrix of actual temperature values ,in and is the size of the infrared spectrum, is the coordinate in the map The temperature value at The temperature data is calibrated, considering the influence of factors such as ambient temperature and emissivity. The calibration function is: , the calibrated temperature matrix is: ; in Collection time Ambient temperature at: and the standard deviation of the temperature distribution ; Temperature data processing and analysis can accurately obtain the actual temperature information of the equipment. Statistics such as average temperature and standard deviation can help determine whether the equipment has overheating risks, providing an important basis for evaluating the equipment's operating status.

[0073] S32.3. The normal operating temperature range of the equipment is , the temperature change rate threshold is ; Define the overheating hazard judgment function: ; If the judgment result is 1, it is considered that the device has a potential overheating risk and requires further attention and inspection; The overheating hazard judgment function can timely detect possible overheating problems of the equipment based on the analysis results of the temperature data, and prevent the occurrence of equipment failures.

[0074] S32.4. Voiceprint feature extraction and processing: For the received voiceprint data ,First, pre-processing is performed including audio noise reduction, framing and windowing operations; Audio noise reduction (such as using the Wiener filter algorithm, ), framing (divide the audio signal into frames of fixed length, assuming the frame length is , the frame shift is , ) and windowing (using a Hamming window, )operate.

[0075] Perform short-time Fourier transform on each frame of audio data to obtain a time-frequency diagram representation; the STFT transform function is: , then the time-frequency diagram matrix is: ,in is the total number of frames; Extract voiceprint feature vectors from the time-frequency graph. MFCC features can effectively characterize the spectral characteristics of sound, which is of great significance for feature extraction of device operation sound. Extract the voiceprint feature vector from the time-frequency graph, for example, using Mel-frequency cepstral coefficient (MFCC) features. The MFCC extraction function is , then the voiceprint feature vector MFCC features can effectively characterize the spectral characteristics of sound and are of great significance for the feature extraction of device operation sound.

[0076] Voiceprint data preprocessing and voiceprint feature extraction can convert the original audio signal into a feature vector suitable for analysis, highlight the key information in the sound signal, and facilitate subsequent judgment of abnormal device sounds.

[0077] Construct an abnormal sound classification model based on machine learning. The classification model is , taking the extracted voiceprint feature vector as input and outputting it as the voice state category; S32.5. Define the sound state category set as , for the feature vector , the sound status judgment result is ; The abnormal sound classification model can accurately determine whether the operating sound of the equipment is normal based on the voiceprint characteristics, promptly discover potential fault hazards inside the equipment, and assist in equipment status assessment.

[0078] The results of image analysis, infrared spectrum analysis and voiceprint data analysis are structured and integrated; the structured data structure is assumed to be: ,in is the device number, is the collection time, is the device status obtained by image analysis, The equipment status obtained by infrared spectrum analysis, that is, the result of overheating hazard judgment, The device status obtained by analyzing the voiceprint data. Take the value in For a certain inspection data, the structured result is: , indicating the device In time When the device is abnormal in appearance, there is a risk of overheating, and the sound is abnormal. These structured data are stored in the database so that they can be associated with the equipment ledger and generate inspection reports.

[0079] Formula usage: Structured processing can unify the formats of different types of analysis results, facilitate data management, association and report generation, and improve data availability and readability.

[0080] S33.1. Data association and report generation: The equipment ledger database contains an equipment information table, including equipment number, equipment name, equipment model, equipment number, and interval number to which it belongs; Read the device information from the device inventory database and create a device information index table, where DeviceInfo searches for the corresponding device information in the device information index table based on the device number; Design inspection report templates, which include basic inspection task information, equipment status summary, abnormal situation analysis and recommended measures; Fill in the inspection report template according to the associated data structure.

[0081] For the analysis results in the above example, the associated data is .

[0082] Data association operations closely combine inspection data with equipment asset information, allowing operation and maintenance personnel to clearly understand the specific operating status of each device and provide an accurate basis for subsequent decision-making.

[0083] Example 5 Further illustrate with reference to Example 1, Figure 1-5 The structure shown, S41, when encountering an automatic door opening and closing scenario, the quadruped robot uses its visual sensor or other detection device to identify the location and status of the door; when approaching the door, it interacts with the access control system, and then safely passes through the door to enter the next area to continue inspection; if the door fails to open normally, the robot will stop moving forward and send an alarm message to the inspection host, waiting for remote intervention or on-site processing by the operation and maintenance personnel; In complex passage scenarios such as narrow dead-end roads, the quadruped robot uses its own sensors to accurately sense the surrounding environment, adjust its body posture and pace, and carefully pass through narrow areas; if it encounters an impassable obstacle or the end of a dead-end road, the robot will handle it according to the pre-set strategy to ensure its own safety and continue to complete the inspection task; S42. When inspecting a multi-floor indoor substation, the quadruped robot first determines its own floor and the target inspection floor. If there is an elevator or lifting equipment available, the robot communicates with the elevator control system, calls the elevator and enters the elevator safely, and continues the inspection after reaching the target floor. If there is no elevator or lifting equipment, the robot uses its powerful climbing ability to go to the target floor along the stairs or other feasible climbing paths. During the climbing process, the robot always maintains a stable posture to ensure its own safety and the normal operation of the detection equipment, while avoiding damage to the stairs or surrounding facilities. After reaching the target floor, the robot inspects the equipment on that floor and collects data according to the pre-planned inspection route and task requirements. S43. If the communication between the quadruped robot and the patrol host is interrupted during the inspection process, the robot will immediately start the local emergency procedure; First, try to re-establish the communication connection, including checking the status of the communication module and adjusting the communication parameters. At the same time, the inspection continues according to the pre-set task plan and the collected data is temporarily stored in the local memory; If the communication interruption lasts for a long time or cannot be restored, the robot will stop running and enter standby mode after completing the current inspection task or reaching a certain safe position, waiting for communication to be restored or for on-site processing by the operation and maintenance personnel; After the communication is restored, the robot will upload the locally stored data to the patrol host in a timely manner, and report the inspection situation and its own status during the communication interruption to the patrol host; S44. When the robot itself fails, it will stop running immediately and send detailed fault information to the patrol host. The operation and maintenance personnel will remotely diagnose the cause of the fault based on the fault information and decide whether it can be repaired through remote operation. If remote repair is not possible, the operation and maintenance personnel will go to the site for repair. After the fault is eliminated, the robot will perform self-inspection and functional testing to ensure that it resumes normal operation before continuing the patrol task.

[0084] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. An intelligent robot for power inspection, characterized by: The invention comprises a quadruped robot (1), wherein a laser radar (6), a voiceprint sensor (7) and an infrared sensor (8) are arranged at the front end of the quadruped robot (1); A pan-tilt base (5) is provided in the middle of the quadruped robot (1), and a camera (4) is provided on the pan-tilt base (5); A six-degree-of-freedom mechanical arm (2) is also provided at the tail of the quadruped robot (1), and a mechanical claw (3) is provided at the end of the mechanical arm (2).

2. The intelligent robot for power inspection according to claim 1 is characterized in that: A communication module is also provided inside the quadruped robot (1), and the communication module is connected to the control center.

3. According to claim 2, a method for inspecting and cleaning a visibility meter automatic cleaning device, characterized in that: The method includes: S1. Inspection task formulation: select inspection areas, intervals, equipment and points according to the actual situation of the substation, and set the inspection cycle; The system generates a mission template and plan that includes routes, stops, inspection items, and data collection requirements; S2, inspection execution: The robot starts and checks its own status, establishes a communication connection with the inspection host to receive task instructions; Use fusion positioning and navigation technology to navigate autonomously, avoid obstacles, and adapt to complex terrain; Accurately dock and adjust the PTZ according to the calculated parameters, collect data, and report any abnormalities promptly; Dynamically detect or merge stops while driving to reduce the number of stops; Plan the path and posture according to the selected target point and conduct inspections to collect data; S3, data processing and analysis: real-time transmission of collected data and classification and storage to establish archives; Use AI algorithms to analyze images, infrared spectra, and voiceprint data and convert them into structured data; The associated data and equipment records automatically generate inspection reports and can be queried and exported; S4. Special situation handling: automatic door opening and closing and narrow dead-end road scenarios: identify the door status or sense the environment when passing through special areas, and promptly report problems; Multi-floor indoor station inspection: determine the floor and choose the appropriate method to go to the target floor for inspection; Communication interruption and fault recovery: Start the local emergency program when communication is interrupted, stop running and report information when a fault occurs, and continue inspection after waiting for repair.

4. The inspection method of the automatic cleaning device of the visibility meter according to claim 3 is characterized by: The method comprises: the specific method of step S1 is: S11. The operation and maintenance personnel select the areas, intervals, equipment and specific points that need to be inspected in the monitoring system according to the actual layout and equipment distribution of the substation; Different inspection cycles can be set for different areas, intervals, equipment and points, including daily, weekly or monthly inspections to meet the importance and operating status monitoring needs of different equipment; S12, Task template and plan generation: The system automatically generates a task template based on the inspection content selected by the operation and maintenance personnel. The template includes the inspection route, stop points, inspection items and data collection requirements; The docking point includes automatically calculated docking pose and gimbal parameters; S13. According to the task template and the set cycle, the system automatically generates a task plan to determine the execution time, execution order and execution equipment of each inspection task, that is, the quadruped robot.

5. The inspection method of the automatic cleaning device of the visibility meter according to claim 4 is characterized by: The method includes: S11.

1. Area, bay, equipment, and point selection and representation: Defining substation area sets , interval set ,in For Region The number of intervals in the device collection ,in For interval The number of devices in the ,in For equipment The number of points on The selection operation of the operation and maintenance personnel is carried out through a selection matrix To record, the elements Defined as: ； This selection matrix can clearly show the selection of each area, interval, equipment and point by the operation and maintenance personnel, providing a clear basis for subsequent task planning and facilitating the system to generate task templates and plans in a targeted manner; S11.

2. Inspection cycle setting and matrix construction: Construction of inspection cycle matrix , whose elements Indicates the corresponding point The inspection cycle is: ; in is the inspection cycle set for this point; S12.

1. Task template and plan generation: Stop point related calculation: The robot is located at two adjacent points and The cost function of moving between , which is related to the distance between points, terrain complexity, and obstacle distribution factors; For the distance factor, let the Euclidean distance between two points be ; in and are the coordinates of two points respectively, and the terrain complexity coefficient is , obtained through analysis of terrain data, including slope and flatness; The obstacle influence factor is , according to the number, size and position of obstacles between two points, the movement cost function can be expressed as: ； The mobility cost function is used to calculate the difficulty and cost of the robot moving between different points, so that when planning the inspection route, the path with the lowest total cost can be selected to improve the inspection efficiency and avoid the robot from getting stuck in difficult-to-pass areas. According to the mobile cost function, the dynamic programming algorithm is used to calculate the optimal path from the starting point to each point, and the path matrix is ​​obtained. , where the elements Indicates from the starting point to the point The previous point on the optimal path; The path matrix provides clear route guidance for the robot's navigation during the inspection process, ensuring that the robot can visit each point that needs to be inspected in sequence according to the optimal path; For each point ,According to the robot's kinematic model and sensor parameters, the parking posture and gimbal parameters are calculated; Assume the robot's body coordinate system is , the world coordinate system is , the transformation matrix from the world coordinate system to the body coordinate system is , which is a function of the robot’s position and attitude, and is updated in real time based on the robot’s positioning and orientation information; Assume the coordinates of the device in the world coordinate system are , calculate the pitch angle of the gimbal according to the visual sensor's field of view and detection accuracy requirements , yaw angle And the attitude angle and roll angle of the robot body in the world coordinate system , Pitch angle , yaw angle ; Enables the testing equipment to accurately align with the equipment for testing; The pitch angle can be calculated based on the device height and the distance between the robot and the device using trigonometric functions. Assume that the horizontal distance between the robot and the device is , the equipment height is ,but ; Accurate docking posture and gimbal parameter calculation can ensure that the robot obtains high-quality detection data at each inspection point, improving the accuracy and reliability of detection; S12.

2. Determine the test items and data collection requirements: Construct a test item matrix , where the elements Defined as: ； The inspection item matrix specifies the inspection items that need to be performed at each inspection point, so that the robot can start the corresponding inspection equipment in a targeted manner; For each test project, determine the data collection requirements matrix based on its type and equipment requirements. , where the elements Contains the frequency, duration, and resolution parameters of data collection; The data collection requirement matrix ensures that the robot can operate within the specified parameters when collecting data; A task template can be represented as a structure ,in is a matrix containing the parking posture and gimbal parameters of each point; S13.

1. Execution time calculation: According to the inspection cycle matrix and the current system time , calculate the execution time matrix of each inspection task ; Set inspection cycle The time interval is , then for each point , its execution time It can be calculated as: ； The above formula can accurately determine the execution time of each inspection task according to the set inspection cycle and current time, ensuring that the tasks are carried out in an orderly manner according to the predetermined plan and realizing regular monitoring of the equipment; S13.

2. Execution order determination: Assign a priority value to each point based on the importance of the equipment and the risk factor of failure. ; Priority can be determined through expert experience, statistical analysis of historical failure data, etc.; For important points on key equipment, the priority value is higher; For non-critical equipment or points that are not prone to failure, the priority value is lower; Sort all points according to the priority value to get the execution order matrix , where the elements Indicates point Position in the inspection sequence; S13.

3. Execution equipment allocation: Total quadruped robots are available, and their state vector is ,in Indicates The status of a robot, with a value of 0 or 1, 0 means failure or busy, and 1 means available; According to the robot's status and location information, as well as the distribution of tasks, the execution equipment is assigned to each inspection task; A greedy algorithm is used to give priority to the robot that is closest to the starting point of the task and is in an available state; suppose the robot The position coordinates are , the coordinates of the task starting point are , then the distance between the robot and the starting point of the task is ; The process of selecting an execution device can be expressed as an allocation function , which returns the robot number assigned to the task; Executing the equipment allocation function can reasonably allocate inspection tasks to available robots and improve the utilization of robot resources; The task plan can be represented as a structure ,in It is the execution equipment matrix, whose elements are the robot numbers assigned to each task. It clarifies the execution time, sequence and execution equipment of each inspection task, providing comprehensive guidance for the robot's inspection execution.

6. The inspection method of the automatic cleaning device of the visibility meter according to claim 3 is characterized by: The method includes: the specific method of step S2 is: S21. When the inspection task start time arrives, the quadruped robot automatically starts and performs initialization operations, including checking whether its own sensors, including the camera (4), infrared sensor (8), voiceprint sensor (7), communication module, and power system are working normally; Establish a stable communication connection with the inspection host and receive task instructions and related parameters, including inspection routes and stop point information; S22. According to the pre-planned route, the quadruped robot autonomously navigates to the inspection area using fusion positioning and navigation technology; During the journey, the vehicle monitors the surrounding environment in real time and avoids obstacles; When encountering complex terrain, the quadruped robot can safely pass through with its off-road obstacle-crossing ability, adjusting its pace and posture to adapt to terrain changes to ensure stable progress; S23, when arriving at the preset stop point, the quadruped robot stops accurately according to the automatically calculated stop position to ensure that the best detection viewing angle can be obtained; at the same time, the direction of the pan-tilt is adjusted according to the calculated pan-tilt parameters so that the detection device camera (4) and the voiceprint sensor (7) and the infrared sensor (8) are aimed at the target device or area; According to the task requirements, start the corresponding detection equipment to collect data; use the visible light camera (4) to capture the appearance image of the equipment, use the infrared sensor (8) to collect the temperature spectrum of the equipment, and use the voiceprint sensor (7) to record the sound characteristics of the equipment operation; During the inspection process, if any abnormal conditions are found, including obvious deformation of the appearance, abnormal temperature rise, or abnormal sound, the abnormal information will be immediately sent back to the inspection host and marked locally for further inspection; S24. For non-stop inspections, provided that the inspection requirements are met, the quadruped robot uses advanced image recognition and data acquisition technology to dynamically capture key information, operating status, and appearance features of the equipment during its movement; By quickly photographing and analyzing the status of the logo and indicator lights on the equipment, and using infrared sensors to monitor the temperature changes of the equipment in real time; For inspections with few stops, the robot uses offline marking technology to rationally plan the stop locations and merge adjacent stops with similar inspection requirements. When arriving at the merged stops, the quadruped robot adjusts its posture and pan / tilt direction to complete the inspection of multiple targets at one time, thereby reducing the number of stops and improving inspection efficiency. S25. When non-preset inspection is required, the operation and maintenance personnel select any target point in the monitoring system (it can be a specific part of the equipment or any position within the area); The quadruped robot automatically plans the best path and parking position based on its real-time position, gimbal parameters and the spatial position of the target point using an algorithm based on a three-dimensional model or point cloud. During the planning process, the robot's motion capability, environmental factors and the viewing angle requirements of the detection equipment are considered to ensure that the target point can be reached accurately and high-quality detection data can be obtained. The patrol system goes to the target point according to the planned route for inspection and data collection. The collected data is also transmitted back to the patrol host in real time for analysis and processing.

7. The inspection method of the automatic cleaning device of the visibility meter according to claim 6 is characterized by: The method includes: the specific method of step S2 is: S21.

1. Define the sensor set of the quadruped robot ,in arrive For the camera, arrive For infrared sensor, arrive For the voiceprint sensor, and the power system state variable and communication module status variables ; For each sensor, its state detection function is ; The power system status detection can be judged by checking whether the motor speed and torque output are within the normal range, whether the battery power is sufficient, etc. Assume that the conditions for the normal operation of the power system are ,but ； The communication module status detection is determined by sending a test signal and receiving feedback. Assuming that the condition for normal communication is ； These status detection functions can accurately determine the working status of each key component of the robot, ensuring that the robot is in good operating condition before performing inspection tasks, avoiding inspection failures or inaccurate data due to equipment failures; S21.

2. Overall state judgment: The overall state judgment function of the robot is: ； This function can quickly determine whether the robot is qualified to perform inspection tasks. Only when all key components are working properly can the robot proceed to the next step to ensure the safety and reliability of the inspection process. S21.3, communication connection attempt: Assume the address of the patrol host is , the communication port is $PortNumber$; the function for establishing a communication connection between the robot and the patrol host is ; This function uses TCP / IP protocol to attempt a connection, and sets the connection timeout to ; If a stable socket connection can be established with the host within the timeout period, the connection is considered successful, otherwise the connection fails; S21.

3. If the communication connection is successful, the robot receives the task instruction structure sent by the patrol host. ,in Inspection route information, including a series of path point coordinates; The docking point information includes the position, posture and gimbal parameters of each docking point; Extract the inspection route information from the task instruction structure. Assume that the route consists of m path points and the path point coordinate matrix is: ,in For the The three-dimensional coordinates of the path points; Extract the stop information, the number of stops is , the stop point information matrix is ,in , is the coordinate of the stop point, is the attitude angle of the docking position, are the gimbal parameters; S22.1, the current position of the robot is , the entrance position of the target inspection area is ; Define the map environment information matrix as , where the elements Indicates the location on the map terrain and obstacle information, such as represents the feasible region, Indicates an obstacle; Improved The cost function of the algorithm is ,in: From the starting point to the node The actual cost is calculated as ,in From the starting point to the node on the path of Points, From point To point The cost of movement, the distance between the two points, and the terrain difficulty coefficient and obstacle risk factor about, that is , is the Euclidean distance between two points; For slave nodes The heuristic estimated cost to the target point is calculated using the weighted Manhattan distance, that is, ,in is the weight coefficient, which is determined according to the characteristics of the map and the motion characteristics of the robot; For Node The risk assessment value of the surrounding environment is calculated as follows: ,in For Node The surrounding neighborhood area, To determine the terrain and obstacle information The calculated risk value is higher for the obstacle area and lower for the flat feasible area. is the risk coefficient weight factor, which is used to balance the safety and efficiency of the path; S22.2, during navigation, the robot integrates visual odometer, inertial measurement unit and global positioning system GPS information for positioning; Assume that the relative displacement of the robot measured by the visual odometer is , the acceleration measured by the inertial measurement unit is , the angular velocity is , the position measured by the global positioning system is ; The extended Kalman filter algorithm is used to fuse this information to obtain a more accurate estimate of the robot's position and posture; the state vector is defined as ,in is the robot position coordinate, is the attitude angle, is the velocity vector, and are the biases of the accelerometer and gyroscope, respectively; The state transfer equation is ,in is the control input, namely the acceleration and angular velocity measured by the inertial measurement unit, is the process noise, and the observation equation is ,in is the observation vector, including the relative displacement measured by the visual odometer and the position measured by the global positioning system, is the observation noise; Through the prediction and update steps of the EKF algorithm, the robot's position and attitude estimates are continuously corrected so that it can accurately navigate along the planned path; S22.3, the image obtained by the robot's visual sensor is ,in is the timestamp, and the processed image is obtained through the image preprocessing algorithm ; Assume the pixel coordinates in the image are , the processed image is analyzed using a deep neural network to identify obstacles and feasible areas in the environment. The pixel area of ​​the obstacle in the image is , the feasible region is ; At the same time, the distance information obtained by the laser radar (6) is ,in is the scanning angle of the sensor, is the distance detected at this angle; Build an environment map based on information from visual sensors and distance sensors , where the elements Indicates location The environmental status is determined based on the obstacle area and feasible area information; Using the improved obstacle avoidance algorithm based on the dynamic window method, the current speed of the robot is , the maximum speed is , the minimum speed is , the velocity resolution is ; Generate a series of candidate velocities in velocity space , where each candidate speed satisfies the speed limit condition, i.e. ; For each candidate speed , predicting that robots will Movement trajectory within ,in The first The position coordinates of the points; Calculate the cost function for each trajectory ,in: is the distance deviation between the trajectory and the target point, calculated as ,in is the end point coordinate of the trajectory); is the collision risk between the trajectory and the obstacle, and the calculation method is ,in For location The collision risk value at the location is determined based on factors such as the distance and shape of the obstacle; is the smoothness cost of speed, calculated as , used to avoid drastic changes in the robot's speed, is the weight coefficient; Select the candidate speed with the lowest cost as the control speed of the robot ; According to the control speed, the control instructions of each joint are calculated through the robot's motion control model ,in For the The control instructions of each joint enable the robot to avoid obstacles along the predetermined trajectory and continue to move towards the target inspection area; S23.1, the current position of the robot is , the attitude angle is , according to the docking point information in the task instruction, the target docking position is ; Calculate position error , , , attitude error , , ; The proportional integral derivative (PID) control algorithm is used to calculate the control quantity of the robot's leg joints and pan / tilt; For position control, the proportionality factor is , the integral coefficient is , the differential coefficient is , then the leg joints are The control quantities of the directions are: ； ； ； For attitude control, the proportional coefficient is and the integral coefficient is , the differential coefficient is , then the control quantities of the gimbal in the attitude angle direction are: ； ； ； The PID control algorithm continuously adjusts the control amount to gradually bring the robot closer to the target parking position according to the position and posture errors, and can effectively reduce overshoot and steady-state error. S23.

2. Establish a kinematic model of the robot leg and convert the calculated joint control amount into actual joint motion instructions; The robot legs have joints, and the joint angle vector is , the joint velocity vector is According to the geometric structure and kinematic relationship of the robot, the mapping function from joint space to Cartesian space can be obtained: and the Jacobian matrix ; Known joint control amount , then the joint speed command is ,in is the inverse matrix of the Jacobian matrix; by solving inverse kinematics, the target angle of each joint can be obtained ,in To control the cycle; Convert the joint target angle into a motor control signal and send it to the robot's motor driver to control the movement of the leg joints so that the robot can gradually adjust to the target docking position; The kinematic model and control command conversion ensure that the calculated control quantity can be accurately converted into the actual movement of the robot leg joints; S23.

4. Define the data collection time series ,in Indicates collection time; for visible light cameras, at each collection time , get the image frame , the image resolution is , the width is , the height is , the frame rate is ; then in the time period The number of images collected in ,in is the floor rounding function; For infrared sensors, at the time of acquisition , get the temperature map , the temperature measurement range is , the temperature resolution is ; For the voiceprint sensor, at the time of collection , collect sound signal fragments , the sampling frequency is , the sampling time is ; In order to ensure the synchronization of multi-sensor data, a timestamp synchronization mechanism is used to give an accurate timestamp to the data collected by each sensor, so that the data of different sensors at the same time can be accurately matched in subsequent analysis; S23.

5. The collected visible light image sequence ( ) for preprocessing, including image grayscale and filtering operations, to obtain the preprocessed image sequence ; The preprocessed image is analyzed using a deep learning-based target detection algorithm. The trained target detection model is , the test result is , which contains the detected target category and location information; Define the appearance abnormality judgment function: ； If an abnormal object is detected at any acquisition moment, the device appearance is considered abnormal; The deep learning target detection algorithm can automatically learn the characteristics of the normal appearance and abnormal conditions of the equipment, accurately identify abnormal conditions such as deformation of the equipment appearance and attachment of foreign objects, and promptly discover potential equipment problems; S23.

6. For infrared temperature spectrum sequence ( ), calculate the average temperature of each spectrum ,in Indicates the temperature map at the pixel point The temperature value at Calculate the rate of temperature change ; The normal operating temperature range of the device is , the temperature change rate threshold is ; Define temperature anomaly judgment function ; If the temperature abnormality condition is met at any collection moment, the device temperature is considered abnormal; By calculating the average temperature and temperature change rate and comparing them with the set threshold, it is possible to detect overheating or abnormal temperature changes of the equipment in time and prevent equipment failure caused by overheating. S23.

7. Sequence of collected sound signal segments ( ) to perform preprocessing, including noise reduction and frame division, to obtain the preprocessed sound signal sequence ; The voiceprint analysis method based on time-frequency analysis and machine learning classification algorithm is used to perform STFT transformation on the preprocessed sound signal to obtain the time-frequency diagram. ; Extracting sound feature vectors from time-frequency graphs , the trained voiceprint classification model is , the classification result is , the classification result is normal sound or abnormal sound; Define the sound abnormality judgment function ; If an abnormal sound is detected at any collection moment, it is considered that the device sound is abnormal; Time-frequency analysis and machine learning classification algorithms can effectively extract features from sound signals, identify abnormal sounds during equipment operation, and assist in determining whether there are hidden faults inside the equipment. S23.

8. When any abnormal situation is detected, the abnormal information is immediately packaged, and the abnormal information includes the abnormal type, the timestamp of the abnormal occurrence, the corresponding data and the device location information; The exception information structure is: ； The abnormal information structure is sent back to the patrol host through the communication module, and marked in the robot's local storage system to record the details of the abnormality so that subsequent operation and maintenance personnel can conduct further inspection and analysis; Accurate processing and reporting of abnormal information enables operation and maintenance personnel to timely understand the abnormal status of equipment and take appropriate measures to deal with it, avoid further development of equipment failures, and ensure the safe operation of substation equipment; S24.

1. Non-stop inspection technology: During the robot's movement, Continuously acquire visible light image sequences ( ); Use lightweight deep learning target detection algorithms (such as YOLOv5-Tiny) to process images in real time. This algorithm has a high processing speed while ensuring a certain degree of accuracy and can meet real-time requirements; The target detection model is , the test result is , quickly identify key equipment components and operating status in images; At the same time, infrared sensors are used to Collect temperature data series ( ), and calculate the temperature change trend in real time; suppose the sliding window size of temperature data is , then at time The temperature change trend is calculated as ; The lightweight target detection algorithm can quickly process images while the robot is moving, and timely obtain the appearance and operating status information of the equipment. Combined with the temperature change trend analysis of the infrared sensor, it can realize dynamic monitoring of the equipment and improve inspection efficiency. S24.

2. Determine in real time whether the equipment is abnormal based on the image recognition results and temperature change trends; The normal judgment function of appearance state is: , the temperature state normal judgment function is ; If in multiple consecutive collection moments, or , it is considered that the device may be abnormal; When a possible abnormal situation is detected, the current image, temperature data and related detection results are immediately packaged into an abnormal information structure: and sent back to the patrol host through the communication module; At the same time, all collected data are archived and stored locally on the robot at certain time intervals, so that more detailed analysis can be performed later when needed; The real-time decision-making mechanism can detect and report equipment anomalies in a timely manner, avoiding the deterioration of equipment failures due to failure to handle abnormal situations in a timely manner. Data archiving and storage provides a data basis for subsequent in-depth analysis. S24.3, less-stop inspection technology: based on the 3D point cloud or model, offline point analysis is performed on the inspection area; the equipment set in the inspection area is , each device has a set of potential inspection points ( ); Define the similarity function between two points ,It considers factors such as the spatial distance between points, the similarity of equipment types, and the similarity of inspection items; According to the similarity function, a clustering algorithm is used to cluster all points, and the points with high similarity are merged into one group; let the clustering result be , each cluster Contains several merged stops; The optimization of offline marking and docking point merging can reduce the number of robot stops and improve inspection efficiency. At the same time, reasonable similarity calculation and clustering algorithm can ensure that the merged docking points can meet the inspection requirements without affecting the accuracy of equipment status monitoring. S24.

4. When the robot reaches the merged stop point, determine the set of equipment and points to be detected based on the clustering results and task requirements; let the set of equipment and points corresponding to the stop point be ,in For equipment, is a point on the device; Adjust the robot's posture and pan / tilt direction so that the detection equipment can cover all targets that need to be detected; for visible light cameras, calculate the best shooting angle and focal length to ensure that the appearance and status of all target devices can be clearly captured; for infrared sensors, adjust the measurement angle and range to obtain accurate temperature data; for voiceprint sensors, optimize the sound collection direction and sensitivity; At the same time, all relevant detection equipment is started to collect data according to the predetermined data collection requirements; for image collection, the image resolution and frame rate are adjusted according to the number and distribution of targets to obtain sufficiently clear and comprehensive image information; The collected data is marked and sorted in real time. The marking information includes the collection time, the corresponding equipment and point number, and the data type. The sorted data is packaged and sent back to the patrol host. At the same time, it is backed up and stored locally on the robot for subsequent analysis and verification. S25.1, non-preset inspection: when the operation and maintenance personnel select any target point in the monitoring system, the coordinates of the target point in the monitoring system are , which needs to be converted into the coordinates of the robot coordinate system ; Assume that the transformation matrix from the monitoring system coordinate system to the robot coordinate system is ,but ; Transformation matrix It can be obtained by a pre-calibrated method, for example, by using the coordinates of multiple reference points of known positions in two coordinate systems and calculating them by a least square method or the like; Accurate coordinate transformation enables the robot to understand the location of the target point in its own coordinate system, providing correct target location information for subsequent path planning and navigation; S25.2, based on the robot's current position is the starting point, the target point coordinates As the end point, using improved Algorithms for path planning; Improved The cost function of the algorithm is ,in: From the starting point to the node The actual cost is calculated as ,in From the starting point to the node on the path of Points, From point To point The cost of movement, the distance between the two points, and the terrain difficulty coefficient and obstacle risk factor about, that is , is the Euclidean distance between two points; For slave nodes The heuristic estimated cost to the target point is calculated using the weighted Euclidean distance, i.e. ,in is the weight coefficient, which is determined according to the environmental characteristics and robot motion characteristics; For Node The risk assessment value of the surrounding environment is calculated as follows: ,in For Node The surrounding neighborhood area, To determine the terrain and obstacle information The calculated risk value is higher for the obstacle area and lower for the flat feasible area. is the risk coefficient weight factor, which is used to balance the safety and efficiency of the path; Improved The algorithm cost function can more accurately evaluate the cost of the path based on the map environment information, so that the robot can plan the path efficiently and safely, avoiding dangerous or difficult-to-pass areas. S25.

3. The planned original path may have some broken lines or discontinuous parts, which are not conducive to the smooth driving of the robot; the spline curve interpolation algorithm is used to smooth the path; Suppose the original path point sequence is ,in ), after spline curve interpolation, the smoothed path point sequence is obtained ,in , ; The cubic spline interpolation function satisfies the continuity of the function value, first-order derivative and second-order derivative at each path point, ensuring the smoothness of the path; At the same time, the smoothed path is optimized, considering the kinematic constraints of the robot; the minimum turning radius of the robot is set to , the maximum speed is ; For each turning point in the path, check whether its curvature meets the minimum turning radius requirement. If not, make local adjustments to the path; calculate the reasonable driving speed of each path segment based on the path length and the maximum speed of the robot to ensure that the robot can ensure safety and improve efficiency during driving; Path smoothing and optimization can improve the stability and safety of the robot's driving, reduce the vibration and energy consumption of the robot during driving, and at the same time meet the robot's kinematic constraints to ensure the feasibility of the path; S25.4, Posture adjustment and data collection: When the robot moves along the planned path to the target point, according to the current position and attitude angle , and the spatial position of the target point , calculate the docking position in real time; Calculate position error , , , attitude error , , ,in is the desired attitude angle at the target point, which is determined according to the detection requirements of the target device and the installation position of the robot's detection equipment; The real-time feedback control algorithm is used to calculate the control quantity of the robot's leg joints and the gimbal. The MPC algorithm predicts the robot's motion state in the future and calculates the optimal control sequence based on the prediction error and control target. The prediction time domain is assumed to be , the control time domain is , then in each control cycle, the MPC algorithm solves the following optimization problem: ； Subject to: Robot kinematic model constraints, control quantity constraints in, In the prediction time domain The posture error of the step, For the The optimal control amount of the current moment is obtained by solving this optimization problem and sent to the robot's motor driver to control the leg joints and the gimbal movement, so that the robot gradually approaches the target parking position. The model predictive control algorithm can calculate the optimal control quantity in real time according to the current state and target posture of the robot, and realize accurate posture adjustment while considering future motion trends and constraints, so as to ensure that the robot can accurately stop at the target point and obtain high-quality detection data; S25.

5. When the robot reaches the target point and adjusts to the position, start the corresponding detection equipment to collect data according to the task requirements; use the visible light camera to take a high-definition image of the target device or area with an image resolution of , the frame rate is ; During the image acquisition process, the camera's focal length, exposure time and other parameters are automatically adjusted according to the characteristics of the target device and the detection focus to obtain clear and accurate image information; The infrared sensor is used to collect the temperature map of the target device. The temperature measurement range is , the temperature resolution is ; Adjust the measuring angle and range of the infrared sensor according to the shape and size of the equipment to ensure that the temperature distribution of the equipment can be fully and accurately obtained; The sound characteristics of the target device are recorded using the voiceprint sensor, with a sampling frequency of , the sampling time is ;Optimize the collection direction and sensitivity of the voiceprint sensor, reduce the interference of environmental noise, and improve the quality of the sound signal; The collected data is marked and sorted in real time, and the marking information includes the collection time, target point number, and data type; the sorted data is transmitted back to the patrol host in real time through the high-speed communication module to ensure the timeliness and integrity of the data; at the same time, it is backed up and stored locally on the robot for subsequent review or further analysis when needed.

8. The inspection method of the automatic cleaning device of a visibility meter according to claim 1 is characterized by: The method includes: the specific method of step S3 is: S31. The images, infrared spectra, and voiceprint data collected by the quadruped robot during the inspection process are transmitted back to the inspection host in real time through a secure and stable network; The inspection host stores the received data according to the equipment, time, and inspection items, and establishes a complete inspection data file for subsequent query, analysis, and comparison; S32, the intelligent analysis host uses AI algorithms based on deep learning to analyze the collected data. For visible light images, it identifies the appearance status of the device, the operating status of the device, and the behavior of personnel; For infrared spectra, analyze the temperature distribution of the equipment to determine whether there is a risk of overheating; For voiceprint data, analyze the characteristics of the device's operating sound to determine whether the device has abnormal sounds; Convert analysis results into structured data, including equipment status information; S33, associating the analyzed inspection data with the equipment ledger, and corresponding the inspection results to specific equipment asset information, including name, model, number, and interval; Automatically generate inspection reports based on the associated data. The report content includes basic information of the inspection task, equipment status summary, abnormal situation analysis and recommended measures; Inspection reports can be queried and exported in the monitoring system to provide a basis for decision-making for operation and maintenance personnel, so that they can timely understand the operating status of the equipment and arrange subsequent maintenance and inspection work.

9. The inspection method of the automatic cleaning device of the visibility meter according to claim 8 is characterized by: The method includes: S31.

1. Data transmission and storage: Pack and mark each type of data, and encrypt the data packets using an asymmetric encryption algorithm; Let the public key be , the private key is ; For image data packets , The encrypted data packet is: ； Similarly, ; ; The robot transmits the encrypted data packets back to the patrol host through a secure and stable network; After receiving the encrypted data packet, the patrol host first uses the private key to decrypt it. , , ; Design a data storage structure with the device as the root directory. Under each device, store data according to time and test items. Suppose the device set is , for equipment , create a time subdirectory under its directory , create a detection project subdirectory under each time subdirectory , corresponding to image, infrared spectrum, and voiceprint data respectively; Establish a data index table to record the storage location and related information of the data. Through the data index table, you can quickly query and retrieve the required data to improve the efficiency of data management; The data index table provides a way to quickly find data, reduces the time complexity of data query, improves the response speed of the system, and facilitates the management and use of inspection data by operation and maintenance personnel; S32.

1. Image analysis based on deep learning: For the received visible light image data , first preprocessing is performed; suppose the image preprocessing function is , which includes image denoising, using algorithms such as bilateral filtering, ), image enhancement and image scaling operations; Convolutional neural networks are used for feature extraction. The feature vector contains the texture, shape, and color information of the image, which can effectively characterize the appearance and operating status of the device. Construct a classification model based on deep learning for equipment status recognition; suppose the classification model is , taking the extracted feature vector as input and outputting it as the device status category; Define the device status category set as: ; Respectively indicate normal equipment appearance, abnormal appearance, normal operation, abnormal operation, personnel presence, and personnel violation behavior; For the feature vector , the standby state recognition result is ; The classification model can accurately judge the appearance and operating status of the equipment and whether there are any personnel-related situations based on image features, and promptly discover potential equipment problems and safety hazards; S32.2, infrared spectrum analysis: for the received infrared spectrum data , firstly, temperature data extraction and calibration are performed; let the temperature data extraction function be , which converts the temperature information in the infrared spectrum into a matrix of actual temperature values ,in and is the size of the infrared spectrum, is the coordinate in the map The temperature value at The temperature data is calibrated, considering the influence of factors such as ambient temperature and emissivity. The calibration function is: , the calibrated temperature matrix is: ； in Collection time Ambient temperature at: and the standard deviation of the temperature distribution ; Temperature data processing and analysis can accurately obtain the actual temperature information of the equipment. Statistics such as average temperature and standard deviation can help determine whether the equipment has overheating risks, providing an important basis for evaluating the equipment's operating status. S32.

3. The normal operating temperature range of the equipment is , the temperature change rate threshold is ; Define the overheating hazard judgment function: ； If the judgment result is 1, it is considered that the device has a potential overheating risk and needs further attention and inspection; S32.

4. Voiceprint feature extraction and processing: For the received voiceprint data ,First, pre-processing is performed including audio noise reduction, framing and windowing operations; Perform short-time Fourier transform on each frame of audio data to obtain a time-frequency diagram representation; the STFT transform function is: , then the time-frequency diagram matrix is: ,in is the total number of frames; Extract voiceprint feature vectors from the time-frequency graph. MFCC features can effectively characterize the spectral characteristics of the sound, which is of great significance for the feature extraction of the device operation sound. Construct an abnormal sound classification model based on machine learning. The classification model is , taking the extracted voiceprint feature vector as input and outputting it as the voice state category; S32.

5. Define the sound state category set as , for the feature vector , the sound status judgment result is ; The results of image analysis, infrared spectrum analysis and voiceprint data analysis are structured and integrated; the structured data structure is assumed to be: ,in is the device number, is the collection time, is the device status obtained by image analysis, The equipment status obtained by infrared spectrum analysis, that is, the result of overheating hazard judgment, The device status obtained by analyzing the voiceprint data. Take the value in S33.

1. Data association and report generation: The equipment ledger database contains an equipment information table, including equipment number, equipment name, equipment model, equipment number, and interval number to which it belongs; Read the device information from the device inventory database and create a device information index table, where DeviceInfo searches for the corresponding device information in the device information index table based on the device number; Design inspection report templates, which include basic inspection task information, equipment status summary, abnormal situation analysis and recommended measures; Fill in the inspection report template according to the associated data structure.

10. The inspection method of the automatic cleaning device of a visibility meter according to claim 3 is characterized by: The method includes: the specific method of step S4 is: S41. When encountering an automatic door opening and closing scenario, the quadruped robot uses its visual sensor or other detection device to identify the location and status of the door; when approaching the door, it interacts with the access control system, and then safely passes through the door to enter the next area to continue inspection; if the door fails to open normally, the robot will stop moving and send an alarm message to the inspection host, waiting for remote intervention or on-site processing by the operation and maintenance personnel; In complex passage scenarios such as narrow dead-end roads, the quadruped robot uses its own sensors to accurately sense the surrounding environment, adjust its body posture and pace, and carefully pass through narrow areas; if it encounters an impassable obstacle or the end of a dead-end road, the robot will handle it according to the pre-set strategy to ensure its own safety and continue to complete the inspection task; S42. When inspecting a multi-floor indoor substation, the quadruped robot first determines its own floor and the target inspection floor. If there is an elevator or lifting equipment available, the robot communicates with the elevator control system, calls the elevator and enters the elevator safely, and continues the inspection after reaching the target floor. If there is no elevator or lifting equipment, the robot uses its powerful climbing ability to go to the target floor along the stairs or other feasible climbing paths. During the climbing process, the robot always maintains a stable posture to ensure its own safety and the normal operation of the detection equipment, while avoiding damage to the stairs or surrounding facilities. After reaching the target floor, the robot inspects the equipment on that floor and collects data according to the pre-planned inspection route and task requirements. S43. If the communication between the quadruped robot and the patrol host is interrupted during the inspection process, the robot will immediately start the local emergency procedure; First, try to re-establish the communication connection, including checking the status of the communication module and adjusting the communication parameters. At the same time, the inspection continues according to the pre-set task plan and the collected data is temporarily stored in the local memory; If the communication interruption lasts for a long time or cannot be restored, the robot will stop running and enter standby mode after completing the current inspection task or reaching a certain safe position, waiting for communication to be restored or for on-site processing by the operation and maintenance personnel; After the communication is restored, the robot will upload the locally stored data to the patrol host in a timely manner, and report the inspection situation and its own status during the communication interruption to the patrol host; S44. When the robot itself fails, it will stop running immediately and send detailed fault information to the patrol host. The operation and maintenance personnel will remotely diagnose the cause of the fault based on the fault information and decide whether it can be repaired through remote operation. If remote repair is not possible, the operation and maintenance personnel will go to the site for repair. After the fault is eliminated, the robot will perform self-inspection and functional testing to ensure that it resumes normal operation before continuing the patrol task.

Images