Intelligent detection robot system for valve hall assembly of power converter station and detection method

Abstract

The invention discloses an intelligent detection robot system for a valve hall assembly of a power converter station and a detection method. A robot body adopts an overhead six-axis series mechanical arm structure and is suspended in the center of the top of a valve hall; the intelligent path planning module performs optimization by combining a probability path diagram method with an improved A * bidirectional search algorithm, and ensures that the robot has efficient obstacle avoidance capability and an optimal moving track in a complex environment; the point cloud data detection and processing module adopts a multi-sensor fusion technology, combines an ICP algorithm to complete point cloud registration, and generates a three-dimensional model with labels. Compared with traditional manual detection, the system has the advantages that the detection efficiency and safety are improved, the labor cost is remarkably reduced, the development of intelligent maintenance means of electric power facilities is promoted, and technical guarantee is provided for stable operation of a converter station.

Description

Technical Field

[0001] The present invention relates to the technical field of power system automation, and in particular to an intelligent detection robot system and a detection method for valve hall components of a power converter station. Background Art

[0002] With the continuous growth of global electricity demand, the safe and stable operation of the power system is particularly important. In the modern power transmission network, the converter station is the key equipment for realizing AC-DC conversion and the core node of the DC transmission project. It plays a vital role in the high-voltage direct current transmission (HVDC) and flexible direct current transmission (VSC-HVDC) systems. Among them, the valve hall is the core area of ​​the converter station, which contains multiple key equipment, such as converter valves, busbars, insulators, cooling systems, etc. These equipment operate in high voltage, high current and complex electromagnetic environments, and are subject to electrical stress and environmental factors for a long time, and are prone to aging, partial discharge, corrosion and other problems. Therefore, in order to ensure the stable operation of the valve hall equipment, regular status detection and maintenance are particularly important. Traditional valve hall detection mainly relies on manual inspection, regular maintenance and semi-automatic equipment detection. It requires staff to enter the valve hall with the help of portable measuring instruments, such as infrared thermal imagers, temperature and humidity sensors, etc., to detect key parts such as converter valve components, wire connection points and cooling systems one by one, which cannot achieve comprehensive automation and intelligence. This method is not only time-consuming and labor-intensive, but also has many shortcomings. First, the operating environment in the valve hall is complex and there is a high-voltage electric field. The staff needs to cut off the power or take strict isolation measures when inspecting. They need to wear a safety rope when working at height, which increases the complexity of operation and safety risks. Second, the manual inspection cycle is long and the coverage is limited. Especially in large converter stations, manual inspections are difficult to detect all hidden dangers in a timely manner, and the detection data is heavily dependent on personnel experience and is highly subjective, making it difficult to achieve accurate quantitative detection. Third, manual inspections are usually performed regularly. Generally, converter stations are inspected once a year. It is impossible to monitor the equipment status in real time, continuously, and effectively, and it is easy to miss early signs of failure.

[0003] In recent years, with the rapid development of robotics, sensor technology and artificial intelligence, intelligent detection robots have gradually become an important tool for solving the problem of power equipment detection. Intelligent detection robots can move autonomously in high-risk environments, use a variety of high-precision sensors to achieve comprehensive perception of equipment status, and transmit detection data to the data processing platform in real time through wireless communication. This method can greatly improve detection efficiency and accuracy and reduce the safety risks of personnel participation. At present, there are some detection robots used in power systems, but most of them are concentrated on the inspection of substations and lines. Special detection robots for valve hall components of converter stations are still in their infancy. Although this type of detection robot has improved safety, its multi-degree-of-freedom movement ability and detection accuracy are still insufficient, and it is difficult to meet the flexible operation and high-precision requirements in the complex environment of the valve hall, especially in multi-layer structures or narrow areas. In addition, the existing robot detection system mostly relies on manual control of fixed positions, lacks the ability to intelligently identify targets and autonomously avoid obstacles, cannot efficiently plan paths, and is difficult to effectively deal with equipment obstacles and complex spatial structures in the valve hall. At the same time, the degree of intelligence is low, and some detection robots only have a single function, such as thermal imaging or simple video monitoring, which is difficult to meet the comprehensive detection needs of valve hall components. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a power converter station valve hall component intelligent detection robot system, the robot body adopts a top-mounted six-axis mechanical arm, which is suspended at the top center of the valve hall;

[0005] Intelligent path planning module, which uses the probabilistic road map method combined with the improved A* bidirectional search algorithm to plan the robot's motion trajectory;

[0006] A point cloud data detection and processing module, which acquires point cloud data through a camera at a collection point, and constructs a three-dimensional model after preprocessing the point cloud data;

[0007] The anti-collision and warning module uses electronic skin or airbag collision detection sensors to sense contact force or air pressure changes in real time and trigger path adjustments;

[0008] The communication and remote monitoring module consists of a client, a main server and at least seven sub-servers;

[0009] The intelligent power management module adopts modular battery pack and dynamic energy scheduling strategy to dynamically adjust energy distribution.

[0010] Furthermore, the point cloud data detection and processing module filters and removes noise from the point cloud data, uses an iterative closest point algorithm to align the data, extracts the surface, performs point cloud segmentation, and generates a three-dimensional model.

[0011] Furthermore, the airbag collision detection sensors are arranged on the surface of the robotic arm using a Delaunay triangulation algorithm, and the surface of the robotic arm is divided into triangular grids.

[0012] Furthermore, the electronic skin integrates a pressure sensor, a temperature sensor and a proximity sensing sensor, and three-modal data to predict collision risks through a Doppler effect formula.

[0013] Furthermore, it also includes an intelligent power management module, which adopts a modular battery pack and energy scheduling strategy to monitor the battery power, temperature and discharge status, dynamically adjust energy distribution, and realize power health status prediction and management.

[0014] A detection method of an intelligent detection robot system for valve hall components of a power converter station is also provided, comprising the following steps:

[0015] S1. Task initialization and path planning: Generate detection tasks according to the equipment layout of the valve hall. After the robot self-checks, it generates a coverage path map through the probabilistic road map method and the improved A* bidirectional search algorithm, and calibrates the position and posture of each joint of the robot arm;

[0016] S2. Multi-point detection and data collection: Control the robot to move along the planned path, activate multiple sensors to collect temperature, three-dimensional point cloud and surface discharge data, and automatically trigger local fine scanning when abnormal temperature is detected;

[0017] S3. Collision protection: Real-time monitoring of contact signals through electronic skin or airbag sensors, combined with path planning algorithms to dynamically adjust the trajectory of the robotic arm. If a collision occurs, the robot stops moving and replans the path.

[0018] S4. Data processing and analysis: De-noise, normalize and iteratively register the collected data using the closest point algorithm, extract surface normal vectors and curvature features, segment point cloud subsets and fuse semantic labels to generate a 3D defect model;

[0019] S5. Intelligent power management: Adopt modular battery packs and dynamic energy scheduling strategies to dynamically adjust energy distribution;

[0020] S6. Data transmission and remote monitoring: The detection data is uploaded to the cloud database in real time via wireless network, and the remote platform displays the robot status, battery information and equipment health assessment report.

[0021] Furthermore, in step S1, a coverage path graph is generated by a probabilistic path graph method and an improved A* bidirectional search algorithm, specifically:

[0022] S11. Initialize the scan to determine the starting point and end point position of the six-axis robot;

[0023] S12. Determine whether the inverse solution is successful. If not, redetermine the robot's starting point and end point position and posture; if successful, proceed to the next step;

[0024] S13. Randomly generate a sampling point from the configuration space C ,judge Is it free space? If it does not belong to, then re-execute step S13; if it does belong to, then change the sampling point Add random point set U and calculate the current neighborhood radius , select several from the set U Points within the neighborhood radius;

[0025] S14. Traverse the current sampling point The neighborhood of the current sampling point With this sampling point The points in the neighborhood of are connected to form a path, and collision sampling detection is performed on the path formed by the points. If there is no collision in the path, the path formed by the points is added to the path set L and the next step is entered; if there is a collision in the path, the next step is directly entered;

[0026] S15. Determine whether the traversal is completed. If the traversal is not completed, repeat step S14; if the traversal is completed, proceed to the next step;

[0027] S16. Obtain the path network Road Map, import the position and posture of the starting point and the end point into the Road Map, use the improved A* bidirectional search algorithm to select a suitable path, determine whether the path is feasible, if so, proceed to the next step, if not, re-search the path and execute step S13;

[0028] S17. Test the system path through visual simulation to generate the final planning path.

[0029] Furthermore, in step S16, the improved A* bidirectional search algorithm adopts dynamic weights , the cost function f(n) is:

[0030]

[0031] Where g(n) is the actual cost from the starting point or target point to the current node n, h(n) is the estimated cost from the current node to the target node or starting point, and the dynamic weight for

[0032]

[0033] in is the basic weight; is the adjustment coefficient; is the current node depth; is the estimated maximum search depth.

[0034] Furthermore, in step S4, feature extraction is performed on the registered point cloud data. For the converter valve, insulator, and capacitor wire components, a local surface feature calculation method based on principal component analysis is used to extract their surface normal vectors, curvature, and point density distribution characteristics. The RANSAC geometric model is used to fit the point cloud to achieve point cloud segmentation and distinguish different component categories.

[0035] Furthermore, the specific steps of the intelligent power management module for power management are:

[0036] S51 real-time monitoring of battery status, the battery status including power, temperature;

[0037] S52. Determine whether the battery status is abnormal. If the battery status is abnormal, trigger an early warning and notify the control center; if the battery status is normal, the temperature management system turns on the heat dissipation device; the abnormal battery status includes the battery power being lower than a preset threshold or the temperature being outside a preset temperature range;

[0038] S53. Determine whether the temperature adjustment fails. If the temperature adjustment fails, trigger the emergency power-off mechanism; otherwise, allocate power to ensure that the core functions are powered first; put non-critical modules into low-power mode, and adjust the robot task priority according to the remaining power;

[0039] S54. Determine whether a serious fault occurs. If a serious fault occurs, perform emergency power-off protection; if no serious fault occurs, provide real-time feedback on the operating status and battery consumption.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] The intelligent detection robot system of the present invention optimizes the structure and mobility of the robot body in terms of hardware, and adapts to the complex environment of the valve hall of the converter station; in terms of software, it develops path planning based on the probabilistic road map method (PRM) and improved A* bidirectional search, point cloud data acquisition and detection methods, and anti-collision and early warning system design; in terms of system integration, it realizes multi-sensor data fusion and real-time remote communication, and has high reliability and efficiency. The robot can complete the detection of equipment defects and faults such as abnormal wires in the valve hall and surface discharge of structural parts, and has the ability to operate around the clock. Compared with traditional manual detection, the present invention can efficiently, safely and accurately complete the comprehensive detection of valve hall components, solving the problems of poor adaptability, single function, low intelligence level and weak anti-interference ability in the prior art, and promoting the development of power facilities towards intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The detection process of the six-axis intelligent robot of the present invention;

[0043] Figure 2 This is a structural diagram of a six-axis intelligent detection robot;

[0044] Figure 3 Provide a flowchart for the robot's intelligent path planning process;

[0045] Figure 4 This is a flowchart of the improved A* bidirectional search algorithm;

[0046] Figure 5 This is a flowchart of point cloud data detection and processing;

[0047] Figure 6 This is a flowchart of the working principle of the robot's electronic skin;

[0048] Figure 7 It is the block diagram of communication and remote monitoring structure;

[0049] Figure 8 This is a flowchart of the working principle of the intelligent power management module. DETAILED DESCRIPTION

[0050] In order to describe the purpose, technical scheme and advantages of the present invention more clearly, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0051] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] In this embodiment, as shown in the attached Figure 1 The figure shows the detection process of the six-axis intelligent robot of the present invention. The system of the present invention is mainly composed of a robot body, an intelligent path planning module, a point cloud data detection and processing module, an anti-collision and warning module, a communication and remote monitoring module, and an intelligent power management module. The robot adopts an overhead six-axis mechanical arm, which is suspended on the top of the valve hall through a high-strength track; the path planning module realizes obstacle avoidance and optimal trajectory generation in complex environments based on an improved algorithm (PRM+A*); the point cloud detection module fuses multi-sensor data and constructs a three-dimensional model with defect annotations through ICP registration; the anti-collision module is equipped with an electronic skin array or an airbag collision detection sensor; the communication module adopts a hierarchical network architecture to support multi-channel independent signal transmission; the power management module dynamically adjusts energy consumption.

[0053] As attached Figure 2As shown in the figure, the robot consists of a top-mounted six-axis serial robot with a moving slide rail and a high-precision 3D structured light camera integrated at the end of the robot. The overall configuration of the intelligent detection robot is a top-mounted six-axis serial robot structure, which is connected to the high-strength track at the top center of the valve hall through a slider. Its installation position is at the center above the working platform. Due to the particularity of its installation position, its robot arm can evenly reach the position of each corner to detect the target. The top-mounted installation method not only avoids the interference of ground obstacles, but also improves the detection ability of high-altitude equipment. The track runs through the entire valve hall to ensure that the robot's movement path can cover all component equipment areas. The track is made of high-strength steel material, which has the advantages of light weight, high strength, and strong corrosion resistance. It can be used for a long time in a complex electromagnetic environment and a high-humidity valve hall environment without deformation or corrosion. The surface of the track is anodized to further improve the wear resistance and service life. In order to meet the stability and accuracy of the robot's operation, the track section is a closed box beam, which has high torsional strength and bearing capacity, and reduces the deflection of the track over long distances. The following is the design of the overall configuration of the robot body.

[0054] 1. Track section parameters and material selection

[0055] The cross-section of the track adopts a closed box beam (height h=150mm, width h=200mm, wall thickness t=200mm), and the track material is Q460 low-alloy high-strength steel. Its mechanical properties are: elastic modulus E=210GPa, yield strength =460Mpa, density ρ=7850kg / m 3 , linear expansion coefficient .

[0056] 1.1 Track load calculation

[0057] The track deadweight load is defined as , robot body weight (including robot arm and sensor), closed box beam (height h=150mm, width h=200mm, wall thickness t=200mm)

[0058] ,

[0059] The value to be brought in is: .

[0060] Define the robot dynamic load as , the accelerating inertia force is , maximum acceleration ,so

[0061] .

[0062] Take the impact coefficient according to ISO8686-1 standard =1.2, according to the formula

[0063]

[0064] Bring in the value .

[0065] The design method under the limit state is used to combine the two loads and define the valve hall span , For combined loads, according to the combined formula:

[0066] .

[0067] 2. Verification of strength and deflection

[0068] 2.1 Calculation of flexural strength

[0069] Define the section moment of inertia as , the maximum bending moment is , the bending stress is , the allowable stress of Q460 steel is According to the formula of section moment of inertia

[0070] ,

[0071] The value of the input

[0072] The maximum bending moment

[0073]

[0074] Bring in the value

[0075] The bending stress is defined as , is the vertical distance from the neutral axis of the track section to the outermost edge of the section. Since the neutral axis is at the center of the section, the distance to the upper and lower edges is half of the height. According to the calculation formula of bending stress

[0076]

[0077] Bring in the value , its stress is much smaller than the allowable stress of Q460 steel, and the design is safe.

[0078] 2.2 Calculation of shear strength

[0079] Define the area moment above the neutral moment as the section static moment, denoted by G, and the maximum shear force is The actual shear stress is , the allowable shear stress is , calculate the static moment of section

[0080] .

[0081] The maximum shear force , according to the shear stress calculation formula

[0082]

[0083] Substitute the numerical value for the actual shear stress , its actual shear stress is much smaller than the allowable shear stress, and the design is safe.

[0084] 2.3 Deflection Verification

[0085] The uniform load deflection is defined as , the concentrated load deflection is , E is the elastic modulus of Q460 steel, according to the formula of uniform load deflection

[0086]

[0087] The value of the input

[0088] The formula for the deflection under concentrated load is

[0089]

[0090] Bring in the value

[0091] Therefore, the total deflection , and the allowable deflection is ,The actual deflection is much smaller than the allowable deflection, and the design is safe.

[0092] In summary, it has been verified that the load capacity and reliability of the robot's overall configuration meet the operation and maintenance requirements in a high-pressure environment.

[0093] The robot intelligent path planning module described in this embodiment adopts a probabilistic road map method (as shown in the attached Figure 3 As shown) combined with the improved A* bidirectional search algorithm (as shown in the attached Figure 4 The robot's motion trajectory is planned to ensure that the robot has efficient obstacle avoidance capabilities and optimal movement trajectory in complex environments. The process includes the following steps:

[0094] 1) Obtain the preset activity planning route of the target robot, determine the position and posture of the robot's starting point and end point, and use the camera at the end of the robotic arm to collect information about obstacles (cooling system pipes, detection components) on the six-axis robot's motion route.

[0095] 2) After collecting obstacle information to the robot end effector, the kinematic model and inverse kinematics of the robot equipped with a six-degree-of-freedom robotic arm are solved to obtain the motion information of the robot's six joint angles. In the six-degree-of-freedom robotic arm, the first three joints Used to position the end effector, and the following three joints It is used to adjust the direction and posture of the end effector. The Denavit-Hartenberg (DH) parameter method is used to establish the kinematic equation. First, the DH parameters of the robot arm need to be defined. These parameters are used to construct the transformation matrix between each joint, as shown in the following table:

[0096]

[0097] in is the connecting rod length, is the connecting rod torsion angle, is the connecting rod offset, is the joint angle, the target position of the end effector is defined as P=(x,y,z), and the rotation matrix of the end effector is Joint 1 mainly affects the projection of the end effector on the XY plane, so It can be solved by the following geometric relationship:

[0098]

[0099] Defines the horizontal distance from the base to the target location , then the connecting rod length can be expressed as , solve and

[0100]

[0101]

[0102] The calculation of the last three joints depends on the decomposition of the rotation matrix, so the rotation matrix of the first three joints needs to be calculated first. , and then calculate the rotation angles of the last three joints through its inverse matrix. Suppose the target rotation matrix of the end effector is:

[0103]

[0104] The rotation matrices of the first three joints can be calculated by (DH) transformation:

[0105]

[0106] The rotation matrix of the last three joints It is calculated by the following formula:

[0107]

[0108] According to the properties of orthogonal matrices, the inverse matrix of a rotation matrix is ​​equal to its transpose, so

[0109]

[0110] Therefore, the rotation matrix of the last three joints can be expressed as:

[0111]

[0112] from From the third line:

[0113]

[0114]

[0115] so

[0116]

[0117]

[0118] .

[0119] 3) Then determine the parameters of the space where the robot is located, define the configuration space position C, and the free space is , the obstacle space is , construct an undirected graph path network RoadMap to store the random point set U and the path point set L, initialize U and L to empty sets, and then define the sampling function:

[0120] .

[0121] 4) Randomly generate a sampling point from the configuration space C , is the current sampling point. , then the current sampling point Add to the random point set U; if , then the sampling point is retrieved and the current sampling point is The neighborhood points of are taken as the connection objects. First, according to the sampling point neighborhood formula:

[0122]

[0123] If a point in the random point set U With sampling point The distance is less than or equal to , then it is called The neighboring points of The current sampling point The set of neighboring points of is a point in the random point set U, is the current neighborhood radius, calculate the current neighborhood radius ,as follows:

[0124]

[0125] in, is the neighborhood radius, n is the number of samples currently sampled, d is the spatial dimension, It is free space The volume of is the volume of the unit sphere in d-dimensional Euclidean space.

[0126] 5) Traverse the current sampling point The neighborhood of the current sampling point Connect the points in its neighborhood to form a path, perform collision detection on this path, given two free space sampling points and , you need to ensure that arrive Will the paths collide? Use a straight line to approximate the paths from arrive The movement trajectory between:

[0127]

[0128] in and are the coordinates of the sampling point and the points in its neighborhood; is the interpolation coefficient, when =0, , =1 ; is a continuous path between two points;

[0129] If the path If some points on the path fall into the obstacle area, it means that the path is not feasible. Collision detection is to check:

[0130]

[0131] in represents the obstacle space, if This indicates that the path is feasible, otherwise the path has a collision. The specific steps for detecting whether the path has a collision are as follows:

[0132] 1. Path discretization

[0133] Since the real path is continuous and the computer cannot directly determine whether the entire continuous path collides, it is necessary to discretize the path into multiple points and then check whether each point is within the obstacle. Define the sampling step size of the path ,

[0134]

[0135] ,

[0136] in The Euclidean distance between two points, is the total number of discrete points on the path, is the step size of the path discretization, that is, the distance between every two adjacent sampling points, so the points on the discrete path can be expressed as

[0137] ,

[0138] in is a sampling point in the path; is the linear interpolation parameter; k is the index of the sampling point (k=0,1,2...n).

[0139] 2. Check if each point is inside the obstacle

[0140] Once the path is discretized, we need to check each point one by one Whether it is inside the obstacle, the obstacle space set is defined as , and the obstacle set is composed of multiple obstacles. Each different type of obstacle has a different representation method. The rectangular obstacle is defined as , the circular obstacle is defined as ,in is the center of the circle, r is the radius, for each point The judgment method in rectangular obstacles is:

[0141] ,

[0142] If the above formula is satisfied, then If it is inside the rectangular obstacle, a collision occurs; otherwise, no collision occurs.

[0143] For each point The judgment method in the circular obstacle is:

[0144]

[0145] If the above formula is satisfied, then If it is inside the circular obstacle, a collision occurs; otherwise, no collision occurs.

[0146] 3. Return collision detection results

[0147] If all points None of them fall into the obstacle, indicating that the path is feasible, then the path is added to the path set L. If any point If there is an obstacle, it means that the path is not feasible. It is necessary to abandon the connection, re-traverse the newly selected sampling points, and determine a new path.

[0148] 6) Repeat steps 2) to 5) until the iteration is completed to obtain the path grid RoadMap. According to the improved A* bidirectional search algorithm, the best path connecting the starting pose and the end pose is obtained from the path grid, as shown in the attached figure. Figure 5 As shown, the specific steps are:

[0149] 1. Initialize the relevant data structures for forward and reverse searches

[0150] Define the forward search to start from the starting point start, the goal is to expand to the target point goal, the reverse search starts from the target point goal, the goal is to expand to the starting point start, define the cost function The minimum node is , whose adjacent nodes are , forward node To adjacent nodes The cost function is , the forward cost function is , define the open list starting from the starting point as the forward open list represented by start Open List, the open list starting from the end point as the reverse open list represented by End start Open List, the closed list starting from the starting point as the forward closed list represented by start reverse Open List, and the closed list starting from the end point as the reverse closed list represented by End reverse Open List. In these two search processes, the cost function of the A* algorithm is Add dynamic weight , we get an improved dynamic weight function, which is used as the core cost function of the A* algorithm, as shown below

[0151] ,

[0152] Where g(n) is the actual cost from the starting point (forward search) or the target point (reverse search) to the current node n, and h(n) is the estimated cost from the current node to the target node (forward search) or from the current node to the starting point (reverse search). is the dynamic weight, and its formula is

[0153] ,

[0154] in is the basic weight; is the adjustment coefficient; is the current node depth; is the estimated maximum search depth.

[0155] 2. Start forward and reverse search

[0156] Add the starting point to the forward open list (Open List), using dynamic weights , setting the cost of the starting point , search forward from the starting point.

[0157] Add the target point to the reverse open list, using dynamic weights , set the cost of the target point , search backward from the target point.

[0158] 3. Expand the node and check if the search frontier meets

[0159] Forward expansion: Select the cost function from the forward open list start Open List The smallest node , the cost function The smallest node Move from the forward open list start Open List to the forward closed list start reverse Open List, and adjust the cost function The smallest node The adjacent nodes of Expand and determine the forward node To the adjacent node The cost function And the forward cost function , if the adjacent node If it is not in the forward closed list start reverse Open List and the reverse open list End start Open List, then add it to the forward open list start OpenList. If it is already in the reverse open list End start Open List, the search is stopped and the forward expansion is completed.

[0160] Reverse extension: Select the cost function from the reverse open list End start Open List The smallest node , the cost function The smallest node Move from the reverse open list End start Open List to the reverse closed list End reverse Open List, and adjust the cost function The smallest node The adjacent nodes of Expand and determine the backward node To the adjacent node The cost function and the backward cost function , if the adjacent node If the adjacent node is not in the forward closed list start reverse open list and the reverse open list End start open list, it is added to the reverse open list End start open list. If it is already in the forward open list start Open List, the search is stopped and the reverse expansion is completed.

[0161] 4. Merge Paths

[0162] The forward search path and the reverse search path are merged to form the final path. When merging, the intersection is only retained once to avoid path duplication.

[0163] 7) Create a path map and embed the prediction model into the path map of the visual simulation system for theoretical testing. Adjust the preset activity planning route of the six-axis robot to generate the final planned path. When the robot reaches the target point, terminate the path planning or trigger the waiting task.

[0164] The point cloud data detection and processing module is the core functional module of the intelligent detection robot system for valve hall components in power converter stations. It is mainly used to realize high-resolution three-dimensional point cloud data acquisition and accurate defect detection of key components such as wires, capacitors, and reactors in the valve hall. The detection process flowchart is shown in the attached figure. Figure 5 As shown. The core sensor mainly used in the detection process is a high-precision 3D structured light camera. In addition, the robot chassis also integrates temperature sensors, humidity sensors, discharge sensors, water leakage sensors, image sensors, vibration sensors, and deformation sensors, which are responsible for collecting real-time data of component equipment and realizing multi-dimensional and full coverage detection of component equipment in the valve hall. The detection process includes the following steps:

[0165] 1. The control center gives the robot a detection task and sends the coordinate information of the target position to the robot. The robot uses the probabilistic road map method combined with the improved A* algorithm to generate a global path covering all detection tasks based on the coordinate information of the target position.

[0166] 2. After the robot moves to the inspection location according to the planned path, it starts the high-precision 3D structured light camera at the end of the robotic arm to collect image data and 3D point cloud data of the equipment to be inspected.

[0167] 3. Preprocess the collected image data and 3D point cloud data by denoising filtering to eliminate the noise and clutter, and use the statistical characteristics of Gaussian distribution to filter and denoise. The specific process includes:

[0168] ① Define a neighborhood for each point H in the three-dimensional point cloud, specify a spherical neighborhood with a radius of r, including all points whose distance from point H is less than r in this spherical neighborhood.

[0169] ②For each point in the neighborhood Calculate the mean vector and covariance matrix of all points in the neighborhood. The mean vector and covariance matrix describe the spatial distribution of the points. The calculation formula of the mean vector is as follows:

[0170] ,

[0171] in is the mean vector; is the component vector of the mean vector; is the coordinate of point i in the neighborhood; is the covariance matrix, which is a 3×3 matrix and its calculation formula is as follows:

[0172] ,

[0173] ③For data point H, evaluate the compactness of its distribution by calculating its Mahalanobis distance with the neighborhood mean and set the threshold To determine whether H is a noise point, its Mahalanobis distance is defined as:

[0174] ,

[0175] Among them, H is the point in the neighborhood; is the mean vector; is the covariance matrix

[0176] Then determine the threshold , and its calculation formula is:

[0177] ;

[0178] Where p, It can be obtained by checking the chi-square distribution table.

[0179] like Then H is marked as a noise point and removed, otherwise it is retained.

[0180] 4. After obtaining the preprocessed point cloud data, use ICP (iterative closest point algorithm) to align the point cloud data, including the following steps:

[0181] ① Find the nearest neighbor corresponding point

[0182] Define the source point cloud , the target point cloud is , for each point exist Find the corresponding points with the closest Euclidean distance so that .

[0183] ② Filter unreliable correspondences based on distance thresholds or statistical methods.

[0184] ③Definition , are the centroids of the source point cloud and the target point cloud respectively. The centroids of the source point cloud and the target point cloud are calculated according to the centroid formula, as shown below

[0185] ;

[0186] ;

[0187] Where K is the number of valid corresponding points.

[0188] ④ Construct the covariance equation and use singular value decomposition to solve the rotation matrix, obtain the optimal rotation matrix R and translation vector t, and output the optimal rotation matrix R and translation vector t so that the source point cloud With the target point cloud Alignment, including the following steps:

[0189] a. Calculate the covariance matrix between the source point cloud and the target point cloud , define the covariance matrix as

[0190]

[0191] Where H is a 3×3 matrix; K is the number of valid corresponding points; are points in the source point cloud respectively; is the corresponding point in the target point cloud;

[0192] b. Use singular value decomposition (SVD) to decompose this covariance matrix and obtain the rotation matrix R,

[0193] ,

[0194] Where U and V are orthogonal matrices; is a diagonal matrix, and then the rotation matrix R is solved by SVD decomposition, as shown below

[0195]

[0196] c. Determine the translation vector t through the rotation matrix and the center of mass,

[0197] ;

[0198] in is the target point cloud; is the source point cloud; R is the rotation matrix

[0199] d. Apply the rotation matrix and translation vector t to the source point cloud to complete the registration.

[0200] ;

[0201] in Represents the source point cloud after registration.

[0202] 5. Perform feature extraction on the registered point cloud data. For key components such as converter valves, insulators, capacitor wire assemblies, etc., a local surface feature calculation method based on principal component analysis (PCA) is used to extract their surface normal vectors, curvature and point density distribution characteristics. The normal vector is used to distinguish between planes (such as capacitor bases) and curved structures (such as insulator sheds), and the curvature feature is used to identify the edges of highly complex areas such as wire connections. Based on the extracted geometric and semantic features (such as normal vectors and curvature), the RANSAC geometric model is used to fit the point cloud segmentation to distinguish different component categories (such as insulators, capacitors, and wires).

[0203] 6. The segmented point cloud subsets (such as equipment components, defective areas) are fused with semantic labels (categories, states) to generate annotated 3D models that support interactive visualization. In the detection of the valve hall of the power converter station in this embodiment, the contaminated areas on the surface of the insulator or the discharge and ablation of the conductor can be highlighted by rendering, providing intuitive spatial positioning and assessment of the severity of the abnormality for the operation and maintenance personnel.

[0204] The anti-collision and warning module in this embodiment provides two optional solutions, namely, arranging an electronic skin array or an airbag collision detection sensor on the surface of the robotic arm. The electronic skin array is a flexible sensing technology with an embedded high-sensitivity pressure sensor. Covering it on the surface of the robotic arm can sense tiny contact forces or impact signals in the external environment in real time. When the robotic arm contacts an obstacle, the electronic skin can generate a warning signal through pressure changes. In addition, the electronic skin has flexible characteristics and can adapt to the complex geometric shape of the robotic arm to ensure uniform protection across the entire surface. The advantage of this technology is that it has a wide sensing range and fast response speed, which can avoid equipment damage or personal injury in high-precision tasks. The disadvantage is that its sensor has a high maintenance cost and is more suitable for low-speed and high-precision detection. The electronic skin detection principle flow chart is shown in the attached figure. Figure 6 As shown. The second solution is to use an airbag collision detection sensor to achieve collision prevention. The system wraps the airbag module in the collision-prone area of ​​the robotic arm. When a collision risk occurs during the movement of the robotic arm, the gas pressure inside the airbag will rise instantly. The high-precision gas pressure sensor installed inside the airbag measures and collects the gas pressure in real time, thereby detecting the high pressure generated by the collision. In a very short time, combined with the motion trajectory of the robotic arm, the motion plan is calculated through the kinematic model, thereby indirectly obtaining the approximate spatial coordinates of the robotic arm collision point. Compared with the electronic skin array, the airbag collision detection sensor has relatively low collision measurement accuracy, and the kinematic model is used to indirectly calculate the collision point. However, the layout cost of the airbag collision detection sensor is lower. Collision detection can be achieved with only an inflatable airbag and a gas pressure sensor, and the collision response is also very sensitive.

[0205] Solution 1: Electronic skin array system design

[0206] a. Topological optimization of sensor selection and layout

[0207] The core sensing unit of the electronic skin needs to take into account sensitivity, dynamic response and electromagnetic compatibility. In view of this, a piezoresistive flexible sensor array is selected as the basic sensing unit, and its resistance change rate is The relationship with the normal vector pressure F follows a piecewise linear model:

[0208] ;

[0209] Set the parameters , , F=20N, ensuring that the nonlinear error within 0.1N to 20N is ≤3%. The electromagnetic frequency band of the converter station valve hall in this embodiment is defined as 2.4GHZ at most. Since there is strong electromagnetic interference in the valve hall, in order to suppress the interference of the electromagnetic phenomenon in the valve hall to the sensor, the sensor structure adopts a three-layer design, the upper and lower layers are copper-nickel alloy (thickness t=50mm), and the middle layer is a polyimide flexible circuit. The shielding effectiveness SE of the circuit is calculated by Schelkunoff theory:

[0210] ;

[0211] Where f represents the frequency of the electromagnetic wave, which is 2.4 GHZ; Indicates the relative magnetic permeability of the material; represents the relative conductivity of the material; t is the thickness of the shielding layer; λ is the wavelength of the electromagnetic wave. Substituting the relevant values ​​into the calculation, SE>83dB is obtained, which meets the anti-interference requirements.

[0212] The sensor layout adopts a hexagonal cellular topology to minimize the blind area coverage. The single sensor spacing is d = 5cm, the effective radius is r = 2.5cm, and the definition is is the blind area coverage rate:

[0213] ;

[0214] Substituting the numerical value into the calculation , to meet the testing needs,

[0215] Assume the surface area of ​​the robot arm 0.5m 2 , according to the formula

[0216]

[0217] Substituting the numerical values ​​into the calculation, we get the number of sensors N≈200.

[0218] b. Multimodal signal acquisition and preprocessing

[0219] The electronic skin integrates pressure, temperature and proximity sensing trimodal data. When the robotic arm is working, the vibration and high temperature of the motor will cause noise in the signal. Wavelet analysis is used to remove high-frequency noise, and then a bandpass filter is used to filter out low-frequency thermal drift to filter out the real signal. The Doppler effect formula is used to predict the risk of collision, including the following steps;

[0220] ①Wavelet analysis eliminates high-frequency vibration noise

[0221] First, the Daubechies4 wavelet is used. Its waveform is similar to the mechanical impact signal and can better separate the noise. Then the pressure signal is decomposed into 5 layers. Each layer decomposes the signal into low-frequency approximate components (carrying real contact information) and high-frequency detail components (including motor harmonics and random impact noise). For the gear noise layer of 500-1000Hz, adaptive hard threshold processing is used to retain the noise exceeding the threshold ( )’s significant impact features are removed, and small fluctuations are eliminated. During reconstruction, only the denoised high-frequency components and the original low-frequency signals are retained, thus meeting the noise reduction requirements of the robot arm under real-time working conditions.

[0222] ②Bandpass filter to filter out low-frequency thermal drift

[0223] A fourth-order Butterworth filter design is used, with -40dB stopband attenuation set at 5Hz and 250Hz, maintaining signal integrity with 0.1dB passband ripple. For thermal drift of less than 1Hz, the filter uses a transfer function , achieving 24dB / oct roll-off, attenuating 0.1Hz interference to its original value , while retaining 95% of the energy of the 15Hz contact signal. Combined with temperature sensor real-time compensation ( ), so that the signal baseline drift is controlled within ±0.5N under 80℃ conditions, ensuring that the 50N range pressure detection accuracy reaches ±1%, meeting the stable sensing requirements in the high temperature environment inside the valve hall of the power converter station.

[0224] ③Predict collision risk

[0225] a. Define the target state

[0226] Assume that the state vector of the i-th obstacle detected in the k-th frame is:

[0227] ;

[0228] in is the three-dimensional position of the obstacle in the robot base coordinate system; is the speed of the obstacle in the robot base coordinate system (when the obstacle is a static obstacle, is zero)

[0229] b. Time to Collision Calculation (TTC)

[0230] Since the surface of the robot's manipulator is a polyhedron, the plane equation of the mth face is defined as:

[0231]

[0232] The normal vector direction is: , and then calculate the obstacle point The shortest distance to the robot surface is:

[0233] ;

[0234] Then use KD-Tree to spatially index the sampling points on the robot surface and quickly search for the nearest point:

[0235] ;

[0236] Where q is the coordinate position of the collision point on the robot surface.

[0237] Calculate the component of the dynamic obstacle velocity on the robot surface normal vector ,

[0238] ;

[0239] c. Determination of collision conditions

[0240] when When the robot surface collides with the obstacle, no collision occurs in other cases. Specifically:

[0241] ;

[0242] Finally, the collision risk level is evaluated through the Mahalanobis distance, and the collision risk information is sent to the monitoring center and the operator is notified to handle it. The advantage of this technology is that it has a wide perception range and fast response speed, which can avoid equipment damage or personal injury in high-precision tasks. The disadvantage is that its sensor maintenance cost is high, and it is more suitable for high-precision detection in low-speed environments.

[0243] Solution 2: Airbag collision detection sensor system design

[0244] a. Airbag layout and sensor optimization

[0245] The Delaunay triangulation algorithm is used to divide the surface of the robot arm into triangular meshes to ensure that the airbag spacing meets the following requirements:

[0246] ;

[0247] Where A represents the segment surface area; is the number of airbags; L is the distance between airbags;

[0248] Calculate coverage:

[0249] ;

[0250] Calculate the airbag coverage rate based on the airbag spacing L and the airbag effective radius r ,make sure , eliminating protection blind spots.

[0251] b. Airbag structure and material design

[0252] The airbag material and structural design uses high-strength aramid fiber-silicone composite film as the base material, with a thickness of 0.8mm, a tensile strength of ≥500MPa, an elongation at break of ≥200%, and a polytetrafluoroethylene (PTFE) coating (friction coefficient ≤0.1) on the surface to reduce friction damage. The temperature range covers -40℃ to 150℃. The airbag adopts a folded honeycomb structure design, and an annular aramid reinforcement rib with a spacing of 2cm is set on the inner wall. Combined with the Delaunay triangulation algorithm layout, the surface coverage rate of the robot arm's easy-to-collision area is ensured to be ≥95%, meeting the long-term protection needs in high-intensity impact and extreme temperature scenarios.

[0253] c. Collision Detection Algorithm

[0254] Step 1: Use the differential detection method to detect the location of the pressure mutation.

[0255] ;

[0256] in is the instantaneous rate of change of the gas pressure inside the airbag; is the internal pressure of the airbag at time t; Indicates the time difference between two samples. Here the fixed sampling period is 1ms. The trigger conditions are:

[0257] ;

[0258] Step 2: Use the weighted centroid method to locate the collision point

[0259] Assume the positions of n adjacent trigger sensors are ( ), the pressure increment is , then the coordinates of the collision point can be expressed as:

[0260] ;

[0261] Where k represents the weight power, k=2; is the pressure increment of the i-th sensor;

[0262] Step 3: Calculate the real-time position and posture of the robot arm, and determine the corresponding collision strategy based on the three-dimensional coordinates of the collision point. When a collision is detected, the robot arm stops moving immediately and adjusts the path according to the location of the collision point.

[0263] The airbag collision detection sensor system can detect the collision risk of the robot arm in real time, and calculate the spatial coordinates of the collision point in combination with the kinematic model, thereby achieving safe motion control of the robot arm. The system has the advantages of high sensitivity, fast response and low cost, and also has the advantages of stability in high temperature and high electromagnetic interference environments, providing a reliable active protection solution for harsh scenarios such as power converter stations.

[0264] In this embodiment, the communication and remote monitoring module is composed of a client, at least seven servers, and a six-axis robot. The communication and remote monitoring structure diagram is shown in the attached figure. Figure 7 As shown. The client and the server where the six-axis robot is located are connected through the Internet network using the TCP / IP protocol. The six-axis robot server is connected to the sub-server where each detection module is located through the MODBUS protocol. The sub-server first establishes a data connection with the six-axis robot, and then establishes a data connection with each sub-module installed on the robot detection end. The six-axis robot collects the various status data of the robotic arm in real time and sends it to the server. At the same time, it also collects the data read by each sub-module and sends it to the server. The server waits for the client's viewing request. After the client connects to the server through the network, it sends a viewing request to the server. After the server verifies its legitimacy, it sends the monitoring data to the client. After the client connects to the server through the network, it sends a control command to the six-axis robot. The control command is forwarded to the six-axis robot by the server. After receiving the control command, the six-axis robot verifies it and executes the corresponding instruction after the verification is successful. After the client connects to the server through the network, it sends a collection command to the detection module of the server. After receiving the collection command, the detection module performs verification and executes the corresponding instructions after successful verification. At the same time, the client can also monitor multiple signals from each server, such as the motion status information of the robotic arm, the image information collected by the camera, and the temperature, humidity, deformation, discharge, water leakage and other signals in the detection module. These multiple signals are independent of each other, and when one signal is interrupted, it will not affect the transmission and monitoring of other signals.

[0265] The robot is connected to the remote monitoring platform. The connection is established through a wireless communication network (such as Wi-Fi, 5G, LTE, etc.). After the remote monitoring system is initialized, it will automatically obtain the robot's location information, task progress, and current detection status. The remote monitoring platform displays the robot's current working status in real time, including battery power, task progress, current detection location, etc. The monitoring personnel can check whether the robot is in normal operation at any time and obtain key parameter data. All data collected by the robot will be uploaded to the cloud or local database in real time for storage to ensure long-term data preservation and historical record management. These data include but are not limited to data collected by sensors, task execution records, equipment status information, etc., and the data storage system supports efficient data management and retrieval. Operators can query historical records at any time and use data analysis to predict equipment health trends, diagnose faults, and make optimization decisions.

[0266] This embodiment is equipped with an intelligent power management module. The working principle flowchart of the intelligent power management module is shown in the attached figure. Figure 8 As shown in the figure, its main task is to monitor the battery status in real time to ensure stable power supply, dynamically adjust energy distribution to extend battery life, and protect equipment and data security under extreme conditions. The specific operation process is as follows:

[0267] 1) System startup

[0268] Start the temperature management system to ensure that the heat dissipation mechanism of key components is normal, initialize the energy distribution controller, load the power demand of the current task, and enter the inspection task preparation phase after the system reports that the status is normal.

[0269] 2) Battery monitoring module works in real time

[0270] When the robot enters the working mode, the intelligent power management system first initializes the initial state of the detection battery, including power (80%), voltage (12.6V), current (0A, static state) and temperature (27℃). After the robot starts the inspection, the battery monitoring module collects power, voltage, current and temperature data once per second. Assume that the following changes occur during the inspection process: Phase 1 (initial operation) The battery consumption is slow, the power drops to 75%, the temperature remains below 30℃, the current increases to 1.5A, and the power supply is stable. Phase 2 (load increase) When high-complexity components are detected, the industrial camera and laser sensor work at the same time, and the current increases to 3.2A. The temperature rises to 36℃, and the system starts the fan for heat dissipation. The energy distribution controller prioritizes the power supply of the detection module and switches the non-core system to low power mode. Phase 3 (abnormal warning) The battery temperature is detected to reach 45℃ (higher than the set threshold of 40℃), and the system triggers an alarm, instructing the robot to suspend high-load operation and enter temperature management mode.

[0271] 3) Operation of temperature management system

[0272] a. Active cooling: The fan runs at full speed and cooperates with the cooling fins to quickly reduce the battery temperature. The system records temperature changes in real time and adjusts the fan speed.

[0273] b. Load regulation: Suspend the functions of non-critical modules (such as communication and status indicator lights), reduce heat generation, and focus energy on critical sensors (such as the imaging function of high-precision industrial cameras).

[0274] c. Temperature recovery: Within 5 minutes, when the temperature drops below 40°C, the system automatically resumes normal working mode.

[0275] 4) Optimal scheduling of energy distribution controller

[0276] During the entire inspection task, the energy distribution controller optimizes the power distribution according to the following rules: detection module > motion module > communication module > auxiliary module. When the power is low (below 30%), the auxiliary module is turned off to extend the working time of key tasks. When the power drops to 10%, the system is forced to enter low-power mode, notify the remote monitoring platform and plan the robot to return to the charging station.

[0277] 5) Dynamic task allocation and execution

[0278] The system dynamically adjusts the inspection tasks based on the current battery status and task progress.

[0279] a. Sufficient power (>50%): Complete all inspection tasks, including high-resolution image acquisition and complex data analysis.

[0280] b. Medium power (30%-50%): Prioritize the inspection of important components and reduce the inspection accuracy of non-critical areas.

[0281] c. Low battery level (<30%): Suspend detection of non-critical areas and notify the operator to intervene.

[0282] 6) Triggering and restoring of emergency power-off protection

[0283] Assume that a sudden failure occurs in the system in the third stage, and the industrial camera circuit short-circuits, causing the current to surge to 10A instantly. The system immediately cuts off the camera power supply and activates emergency power-off protection to prevent other modules from being damaged. The faulty module name and abnormal current data are stored and uploaded to the remote platform. The system notifies maintenance personnel to check the equipment. After confirming that the fault has been eliminated, the system reallocates power and resumes inspection tasks.

[0284] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent detection robot system for valve hall components of a power converter station, characterized in that: include: The robot body adopts a top-mounted six-axis robotic arm, which is suspended at the top center of the valve hall; Intelligent path planning module, which uses the probabilistic road map method combined with the improved A* bidirectional search algorithm to plan the robot's motion trajectory; A point cloud data detection and processing module, which acquires point cloud data through a camera at a collection point, and constructs a three-dimensional model after preprocessing the point cloud data; The anti-collision and warning module uses electronic skin or airbag collision detection sensors to sense contact force or air pressure changes in real time and trigger path adjustments; The communication and remote monitoring module consists of a client, a main server and at least seven sub-servers; The intelligent power management module adopts modular battery pack and dynamic energy scheduling strategy to dynamically adjust energy distribution.

2. The intelligent detection robot system for valve hall components of a power converter station according to claim 1 is characterized in that: The point cloud data detection and processing module filters and removes noise from the point cloud data, uses an iterative closest point algorithm to align the data, extracts the surface, performs point cloud segmentation, and generates a three-dimensional model.

3. The intelligent detection robot system for valve hall components of a power converter station according to claim 2 is characterized in that: The airbag collision detection sensors are arranged on the surface of the robotic arm using a Delaunay triangulation algorithm, and the surface of the robotic arm is divided into triangular grids.

4. The intelligent detection robot system for valve hall components of a power converter station according to claim 2 is characterized in that: The electronic skin integrates pressure sensor, temperature sensor and proximity sensing sensor, tri-modal data, and predicts collision risk through the Doppler effect formula.

5. The intelligent detection robot system for valve hall components of a power converter station according to claim 3 or 4, characterized in that: It also includes an intelligent power management module that uses a modular battery pack and energy scheduling strategy to monitor battery charge, temperature and discharge status, dynamically adjust energy distribution, and realize power health status prediction and management.

6. A detection method for an intelligent detection robot system for valve hall components of a power converter station according to any one of claims 1 to 5, characterized in that: The following steps are involved: S1. Task initialization and path planning: Generate detection tasks according to the equipment layout of the valve hall. After the robot self-checks, it generates a coverage path map through the probabilistic road map method and the improved A* bidirectional search algorithm, and calibrates the position and posture of each joint of the robot arm; S2. Multi-point detection and data collection: Control the robot to move along the planned path, activate multiple sensors to collect temperature, three-dimensional point cloud and surface discharge data, and automatically trigger local fine scanning when abnormal temperature is detected; S3. Collision protection: Real-time monitoring of contact signals through electronic skin or airbag sensors, combined with path planning algorithms to dynamically adjust the trajectory of the robotic arm. If a collision occurs, the robot stops moving and replans the path. S4. Data processing and analysis: De-noise, normalize and iteratively register the collected data using the closest point algorithm, extract surface normal vectors and curvature features, segment point cloud subsets and fuse semantic labels to generate a 3D defect model; S5. Intelligent power management: Adopt modular battery packs and dynamic energy scheduling strategies to dynamically adjust energy distribution; S6. Data transmission and remote monitoring: The detection data is uploaded to the cloud database in real time via wireless network, and the remote platform displays the robot status, battery information and equipment health assessment report.

7. The detection method according to claim 6, characterized in that: In step S1, a coverage path graph is generated by a probabilistic path graph method and an improved A* bidirectional search algorithm, specifically: S11. Initialize the scan to determine the starting point and end point position of the six-axis robot; S12. Determine whether the inverse solution is successful. If not, redetermine the robot's starting point and end point position and posture; If successful, proceed to the next step; S13. Randomly generate a sampling point from the configuration space C ,judge Is it free space? If it does not belong to, then re-execute step S13; if it does belong to, then change the sampling point Add random point set U and calculate the current neighborhood radius , select several from the set U Points within the neighborhood radius; S14. Traverse the current sampling point The neighborhood of the current sampling point With this sampling point The points in the neighborhood of are connected to form a path, and collision sampling detection is performed on the path formed by the points. If there is no collision on the path, the path formed by the points is added to the path set L and the next step is entered; If there is a collision on the path, go directly to the next step; S15. Determine whether the traversal is completed. If the traversal is not completed, repeat step S14; if the traversal is completed, proceed to the next step; S16. Obtain the path network Road Map, import the position and posture of the starting point and the end point into the Road Map, use the improved A* bidirectional search algorithm to select a suitable path, determine whether the path is feasible, if so, proceed to the next step, if not, re-search the path and execute step S13; S17. Test the system path through visual simulation to generate the final planning path.

8. The detection method according to claim 7, characterized in that: In step S16, the improved A* bidirectional search algorithm uses dynamic weights , the cost function f(n) is: ； Where g(n) is the actual cost from the starting point or target point to the current node n, h(n) is the estimated cost from the current node to the target node or starting point, and the dynamic weight for ; in is the basic weight; is the adjustment coefficient; is the current node depth; is the estimated maximum search depth.

9. The detection method according to claim 8, characterized in that: In step S4, feature extraction is performed on the registered point cloud data. For the converter valve, insulator, and capacitor wire components, a local surface feature calculation method based on principal component analysis is used to extract their surface normal vectors, curvature, and point density distribution characteristics. The RANSAC geometric model is used to fit the point cloud to achieve point cloud segmentation and distinguish different component categories.

10. The detection method according to claim 9, characterized in that: The specific steps of the intelligent power management module for power management are: S51 real-time monitoring of battery status, the battery status including power, temperature; S52. Determine whether the battery status is abnormal. If the battery status is abnormal, trigger an early warning and notify the control center; if the battery status is normal, the temperature management system turns on the heat dissipation device; the abnormal battery status includes the battery power being lower than a preset threshold or the temperature being outside a preset temperature range; S53. Determine whether the temperature adjustment fails. If the temperature adjustment fails, trigger the emergency power-off mechanism; otherwise, allocate power to ensure that the core functions are powered first; Put non-critical modules into low-power mode and adjust robot task priorities based on remaining power; S54. Determine whether a serious fault occurs. If a serious fault occurs, perform emergency power-off protection; if no serious fault occurs, provide real-time feedback on the operating status and battery consumption.

Images