Autonomous inspection method and system based on path planning

Abstract

The invention relates to an autonomous inspection method and system based on path planning, and belongs to the field of path planning. According to the method, firstly, a plurality of candidate paths from a starting point to an end point are generated by using an A * algorithm based on a three-dimensional model of a to-be-inspected area, then the plurality of candidate paths are evaluated by using a Q learning algorithm, the path with the highest Q value in the plurality of paths is taken as the optimal path, and finally inspection is performed according to the optimal path. The priority of the to-be-inspected equipment is considered when the A * algorithm heuristic function is designed, and the operation state of the non-inspected equipment is also considered when the state space of the Q learning algorithm is designed, so that the path planned by the method comprehensively considers the importance and the operation risk of the equipment, and the inspection efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to an autonomous inspection method and system based on path planning, belonging to the field of path planning. Background Technology

[0002] In modern industrial production, infrastructure management, and public safety, the stable operation of equipment and facilities is crucial. Traditional manual inspection methods face numerous challenges, such as high labor costs, low inspection efficiency, and significant susceptibility to subjective human factors, making it difficult to comprehensively and promptly identify potential equipment problems. While automated inspection technologies have developed, such as drone inspection and fixed-track robot inspection, they have significant drawbacks. Drones have limited positioning accuracy in complex terrain or environments with strong electromagnetic interference, making detailed equipment inspection impossible; fixed-track robots lack flexibility and cannot adapt flexibly to environmental changes or equipment layout adjustments. Existing inspection technologies rely heavily on preset fixed paths for path planning, failing to dynamically optimize paths based on the importance of the inspected equipment, real-time environment, and equipment status. Furthermore, human-machine interaction methods are limited, often using flat, two-dimensional operations, lacking intuitiveness and immersion, making it difficult for operators to fully and deeply understand the inspection situation, severely restricting inspection efficiency and the accuracy and timeliness of fault diagnosis. Summary of the Invention

[0003] The purpose of this invention is to provide an autonomous inspection method and system based on path planning to solve the problem of low inspection efficiency in automated inspection.

[0004] To achieve the above objectives, the present invention includes:

[0005] The present invention provides an autonomous inspection method based on path planning, comprising the following steps: 1) generating multiple candidate paths from the starting point to the ending point using the A* algorithm based on a three-dimensional model of the area to be inspected; 2) evaluating the multiple candidate paths using the Q-learning algorithm, and selecting the path with the highest Q value as the optimal path; 3) performing inspection based on the optimal path.

[0006] The open list in the A* algorithm contains all the devices to be inspected in the area to be inspected. The heuristic function of the A* algorithm is the sum of the distance estimation function and the device priority function. The device priority function is the priority coefficient set for the device to be inspected.

[0007] The evaluation process involves using the location of the currently inspected device, the operating status of the inspected devices, and the uninspected devices in the candidate path as the state space for Q-learning; using the movement direction and path length from the location of the currently inspected device to the next device to be inspected in the candidate path as the action space for Q-learning; and iteratively updating the Q-value based on a pre-defined reward function.

[0008] Furthermore, the reward function is:

[0009] R=ω1·R 效率 +ω2·R 覆盖 +ω3·R 能耗 +ω4·R 风险

[0010] Where R is the reward function; R 效率 The value of the function is negatively correlated with the path length; R 覆盖 The function's value is the sum of the inspection importance function and the important equipment coverage function. The inspection importance function is positively correlated with the importance of the next piece of equipment to be inspected, while the important equipment coverage function is negatively correlated with the importance of any equipment missed during the inspection process. R 能耗 The value of the function is negatively correlated with the energy consumption during inspection; R 风险 The value of the function is positively correlated with the operational risk level of the next piece of equipment to be inspected; ω1, ω2, ω3 and ω4 are the corresponding reward weight coefficients.

[0011] Furthermore, the heuristic function is:

[0012] h(n) = α·d manhattan (n,goal)+β·P(goal)

[0013] Where h(n) is the heuristic function; n is the currently inspected device; goal is the next device to be inspected; d manhattan (n,goal) is the distance estimation function, representing the Manhattan distance from the currently inspected device to the next device to be inspected; P(goal) is the device priority function; α and β are the corresponding set heuristic weight coefficients.

[0014] Furthermore, during the inspection process, if any abnormal situation occurs, the optimal path is replanned according to the pre-set dynamic path planning algorithm, and then the inspection is carried out according to the replanned optimal path; abnormal situations include abnormal operating status of uninspected equipment and / or obstacles in the inspection path.

[0015] Furthermore, during the inspection process, the system also collects the operating data of the equipment to be inspected. Through preset fault diagnosis rules and machine learning models, it determines whether the equipment to be inspected is abnormal. If an abnormality is found, an alarm signal is sent.

[0016] Furthermore, the optimal path can be displayed using AR technology.

[0017] Furthermore, the optimal path can be adjusted using AR technology.

[0018] Furthermore, during the inspection process, AR technology is used to display the status information of the equipment to be inspected, including real-time operating parameters, historical fault records, and maintenance guidance information.

[0019] An autonomous inspection system based on path planning includes a processor that executes a computer program to implement the steps of the method described above.

[0020] The beneficial effects of this invention are as follows: This invention is a pioneering invention. Firstly, based on a 3D model of the area to be inspected, it uses the A* algorithm to generate multiple candidate paths from the starting point to the ending point. Then, it uses the Q-learning algorithm to evaluate these candidate paths, selecting the path with the highest Q-value as the optimal path. Finally, inspection is performed based on the optimal path. Because the priority of the equipment to be inspected is considered when designing the heuristic function of the A* algorithm, and the operating state of uninspected equipment is also considered when designing the state space of the Q-learning algorithm, the path planned by this invention comprehensively considers the importance of the equipment and operational risks, significantly improving inspection efficiency. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of a system architecture according to the present invention;

[0022] Figure 2 This is a schematic diagram of a path planning process according to the present invention;

[0023] Figure 3 This is a schematic diagram of an autonomous inspection process according to the present invention;

[0024] Figure 4 This is a three-dimensional interactive schematic diagram of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0026] The concept of this invention is to use heuristic search combined with the importance of inspection equipment to quickly search for multiple candidate paths; then use reinforcement learning algorithm to select the optimal path that meets the corresponding conditions from the multiple candidate paths, and then perform inspection based on the optimal path.

[0027] System Implementation Method:

[0028] This embodiment provides an autonomous inspection system based on path planning, such as... Figure 1 As shown, the system includes a data acquisition module, a path planning and decision-making module, an inspection execution module, and a three-dimensional interactive display module.

[0029] Furthermore, the data acquisition module is used to construct a 3D model of the area to be inspected, collect real-time operating data of the equipment to be inspected, and obtain geographical layout and environmental data of the area. High-resolution LiDAR is used to quickly and accurately acquire 3D spatial point cloud data of the area to be inspected by emitting and receiving laser beams. A multispectral imaging camera is configured to not only acquire conventional visible light images but also capture abnormal characteristics of the equipment under specific spectra, such as early aging and corrosion marks of materials. A high-precision infrared thermal imaging sensor is used to accurately measure the surface temperature distribution of the equipment to be inspected and obtain its thermal status information. In addition, vibration sensors and sound sensors are equipped to monitor the mechanical vibration and sound characteristics of the equipment to be inspected, comprehensively perceiving its operating status. Data fusion software integrates and processes this data to construct an accurate 3D model of the area to be inspected, closely linking detailed parameters of each piece of equipment in the area, such as equipment volume, pressure rating, pipe material, and pipe diameter, to the corresponding equipment in the 3D model. The sensors used are distributed at key locations in the area to be inspected and on the corresponding equipment, forming a distributed data acquisition system. The collected data is transmitted to the data processing center in real time via wired or wireless communication. For sensors with short range, large data volume, and high stability requirements, such as some LiDAR sensors, wired Ethernet transmission is used. For widely distributed and flexibly installed multispectral cameras and infrared sensors, wireless communication technologies such as Wi-Fi, Bluetooth Low Energy (BLE), or 4G / 5G are employed. During transmission, the TCP / IP protocol is used in conjunction with data verification algorithms to ensure the accuracy and integrity of the data.

[0030] Furthermore, the path planning and decision-making module generates an initial optimal inspection path based on the 3D model using an algorithm that integrates reinforcement learning and heuristic search. The algorithm, which integrates reinforcement learning and heuristic search, uses "heuristic search to quickly narrow down the solution space + reinforcement learning to dynamically optimize path weights" as its core logic. The specific process is as follows: Figure 2 As shown.

[0031] First, the A* algorithm, a heuristic search algorithm, is used to generate candidate paths. This algorithm combines the completeness of Dijkstra's algorithm with the efficiency of greedy best-first search. By introducing an evaluation function to guide the search direction, it can quickly find the optimal path from the starting point to the target point in complex grids or graphs.

[0032] Search objective: Based on the spatial constraints of the received 3D model, generate a set of candidate paths that satisfy "no collisions and feasible passages" from the starting point (such as the docking point of the inspection equipment) to the positions of all equipment to be inspected. That is, in the 3D model, the starting and ending points of the corresponding pipes are input into the A* algorithm. The open list used by the A* algorithm contains all equipment to be inspected, thus ensuring that the generated multiple candidate paths can pass through all the equipment to be inspected.

[0033] The heuristic function design of the A* algorithm: A hybrid heuristic function h(n) based on weighted Manhattan distance and device priority correction is adopted, and the corresponding formula is:

[0034] h(n) = α·d manhattan (n,goal)+β·P(goal) (1)

[0035] Where n is the current node (the device currently being inspected); goal is the target device (the next device to be inspected); d manhattan (n,goal) is the distance estimation function, specifically the Manhattan distance from the currently inspected device to the next device to be inspected (efficient estimation adapted to rectangular channel scenarios); P(goal) is the device priority function, which sets a corresponding priority coefficient for the device to be inspected. For example, it is set according to the importance of the device to be inspected, with a value of 1.2 for high-importance devices and 1.0 for ordinary devices; α and β are the corresponding set heuristic weight coefficients, for example, α is 0.7 and β is 0.3, thus ensuring that the search is directed towards high-priority devices. The importance of the devices to be inspected is classified according to their impact on the system; for example, the importance of core reaction devices is much higher than that of auxiliary cooling devices. As a specific embodiment of the present invention, importance is specifically divided into 5 levels, with level 1 being the lowest and level 5 the highest, and level 3 and above being high-importance devices.

[0036] Furthermore, Q-learning from reinforcement learning is used to filter the generated candidate paths and select the optimal path. The core function of the Q-learning algorithm is to "rank" the candidate paths based on multiple dimensions, that is, to comprehensively consider factors such as the importance of the equipment to be inspected, the level of operational risk, and the real-time operating status. The selected optimal path fully considers inspection efficiency, comprehensive equipment coverage, and energy consumption optimization, ensuring that the effective inspection of key equipment is completed within a limited time.

[0037] Specifically, the parameters of the Q-learning algorithm are defined based on the obtained candidate paths. The state space (S) of the Q-learning algorithm is defined as "current inspection position + set of covered devices (set of inspected devices) + state of remaining inspected devices (running state of uninspected devices)". For example, S = ({x represents the position of the currently inspected device}, {y represents the ID of the inspected device}, {z represents the real-time temperature / vibration value of the remaining devices}).

[0038] Motion space (A): defined as "the direction of movement and path length from the current location of the inspected equipment to the next equipment to be inspected".

[0039] The reward function (R), designed by integrating multiple dimensions of factors, is the core driver of path optimization. Its formula is:

[0040] R=ω1·R 效率 +ω2·R 覆盖 +ω3·R 能耗 +ω4·R 风险 (2)

[0041] in, This is an efficiency function, and its value is negatively correlated with path length (the shorter the path, the higher the reward). For example, , For path length, The efficiency coefficient is set. The coverage function is the sum of the inspection importance function and the important equipment coverage function. The value of the inspection importance function is positively correlated with the importance of the next equipment to be inspected, and the value of the important equipment coverage function is negatively correlated with the importance of the equipment to be inspected that was missed during the inspection process. For example, if the next equipment to be inspected is "highly important equipment that was not inspected", the reward is +50; if a highly important equipment is missed, the reward is -100. This is an energy consumption function, and its value is negatively correlated with the energy consumption during inspection. For example, , The set energy consumption coefficient, Energy consumption; This is a risk function whose value is positively correlated with the operational risk level of the next piece of equipment to be inspected. For example, risk level = (current temperature - rated temperature) / rated temperature + failure rate. If the risk level of the next piece of equipment to be inspected is greater than or equal to 0.8, the inspection reward is +30. , , and The corresponding reward weighting coefficients are 25%, 30%, 20%, and 25%, respectively.

[0043] Iterative optimization: Q-value update through reinforcement learning:

[0044] Q(S,A)←Q(S,A)+α[R+γmaxQ(S′,A′)-Q(S,A)])(3)

[0045] Each of the multiple candidate paths is given a "value score", and the path with the highest Q value is selected as the initial optimal path.

[0046] To ensure that route planning aligns with actual needs, it is necessary to quantitatively model factors such as "route length, equipment importance, equipment energy consumption, and operational risk level" and assign appropriate weights to them.

[0047] (1) Path length (weight 25%)

[0048] Calculations are based on distance and are negatively correlated with path length; R0 效率 = -0.1·L (L is the path length, in meters).

[0049] (2) Equipment importance (weight 30%)

[0050] Equipment is categorized into 5 levels based on its impact on the system (Level 1 is the lowest, Level 5 is the highest). The quantified value is: I = 0.2 * Level (Level 5 corresponds to I = 1.0, Level 1 to I = 0.2, and highly critical equipment I > 0.8). The time interval between the last inspection also affects the coverage function, quantified as "interval time / standard inspection cycle". For example, if the standard cycle is 24 hours, and a piece of equipment has not been inspected for 36 hours, then T = 36 / 24 = 1.5 (T ≥ 0.8 indicates a serious timeout; equipment exceeding the timeout setting R...). 覆盖 *1.5).

[0051] (3) Equipment energy consumption (weight 20%)

[0052] Negatively correlated with energy consumption, R 能耗 = -0.2·E (E is energy consumption, unit: W·h).

[0053] (4) Operational risk level (weight 25%)

[0054] Based on historical fault data and real-time status calculations: R i = 0.6 * Failure Rate + 0.4 * Real-time Deviation Rate (Real-time Deviation Rate = (Current Parameter - Rated Parameter) / Rated Parameter). The real-time operating status also affects the risk function. Real-time data transmitted from sensors (such as infrared temperature and vibration amplitude) is binary-coded (1: exceeding the warning threshold, e.g., temperature > 80℃; 0: normal). Devices marked as 1 trigger R. 风险 Emergency Bonus (Additional +20).

[0055] The obtained initial optimal path is sent to the inspection execution module, which (inspection equipment, such as an inspection robot) performs inspections according to the corresponding path. Figure 2 and Figure 3 As shown, during the inspection process, environmental and equipment status data are monitored in real time. The intelligent diagnostic module built into the inspection equipment (with preset fault diagnosis rules and machine learning models) analyzes the collected data in real time. Once an equipment malfunction is detected or an obstacle appears on the inspection path, the system immediately replans the path in a very short time according to a pre-set dynamic path planning algorithm, such as Rapid Exploration Random Tree (RRT*), thus bypassing the abnormal area and ensuring the continuity and efficiency of the inspection task. Simultaneously, if an abnormal condition is detected in the equipment to be inspected, an alarm is immediately issued, and detailed anomaly information is transmitted to the three-dimensional interactive module and the subsequent maintenance system.

[0056] Specifically, the inspection equipment integrates a deep learning-based intelligent obstacle avoidance system, capable of identifying various obstacles in real time. For example, when encountering obstacles such as temporary food supplies on the road, the intelligent obstacle avoidance system immediately activates. Through data fusion analysis of cameras and LiDAR, it identifies the type and location of the obstacle, automatically adjusting its direction and speed to bypass it and continue the inspection. Simultaneously, during the inspection process, the equipment continuously collects operational data from the equipment under inspection using its various onboard sensors, including electrical parameters, mechanical vibrations, and sound characteristics. This data is then analyzed in real time through a built-in intelligent diagnostic module.

[0057] Furthermore, such as Figure 4 As shown, the 3D interactive display module utilizes AR technology to build a powerful visualization platform. The system backend pushes planned paths and data and employs advanced graphics rendering and virtual reality (VR) and augmented reality (AR) technologies to present 3D models, real-time inspection paths, and equipment status data to operators in a high-definition and highly realistic manner. Operators can use immersive VR devices, such as VR headsets, to virtually "enter" the inspection scene, freely move between the equipment to be inspected, view the inspection path from any angle, and flexibly adjust the planned path through natural interaction methods such as gesture recognition and voice commands.

[0058] Specifically, operators wear AR devices and enter a virtual 3D scene of the area to be inspected. Since the data collected by sensors establishes a correspondence with the corresponding equipment in the 3D model, the inspection equipment allows operators to see detailed information overlaid on the equipment in real time by scanning its QR code. This includes real-time operating parameters, historical fault records, and maintenance guidance information. When a minor leak is detected in a pipeline, operators can see the leak location, leak severity assessment, and emergency handling guidelines overlaid on the pipeline in the real-world scene using AR glasses, facilitating rapid organization of repair work. Simultaneously, operators can conduct real-time video calls with remote experts through the AR interface to obtain professional guidance. The platform also provides data analysis and decision support functions, offering equipment maintenance suggestions and decision-making references based on equipment operating data and historical records, facilitating quick and accurate equipment diagnosis and decision-making. Furthermore, in the inspection scene, operators can clearly see the planned inspection path and flexibly modify the planned path using gestures, such as clenching a fist to confirm the path and waving to adjust the path direction.

[0059] The inspection equipment used in this invention employs various methods, including intelligent inspection robots with advanced navigation and motion control capabilities, high-precision positioning drones, and track-mounted inspection vehicles. Intelligent inspection robots are suitable for complex ground environments, flexibly moving between equipment for close-range, detailed inspections; drones can quickly cover large areas, inspecting high-altitude equipment or hard-to-reach areas; and track-mounted inspection vehicles are used for periodic, repetitive inspections of specific areas. These devices autonomously move and inspect in complex environments based on paths generated by the path planning and decision-making module.

[0060] Method implementation:

[0061] This embodiment provides an autonomous inspection method based on path planning. The system described in the system implementation is designed using this method. Since the description of the system is already clear enough, it will not be repeated here.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific embodiments of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A path planning-based autonomous inspection method, characterized in that, The process includes the following steps: 1) Based on the 3D model of the area to be inspected, use the A* algorithm to generate multiple candidate paths from the starting point to the ending point; 2) Use the Q-learning algorithm to evaluate the multiple candidate paths and select the path with the highest Q value as the optimal path; 3) Perform inspection according to the optimal path. The open list in the A* algorithm contains all the devices to be inspected in the area to be inspected. The heuristic function of the A* algorithm is the sum of the distance estimation function and the device priority function, where the device priority function is the priority coefficient set for the device to be inspected. The evaluation process involves using the location of the currently inspected device, the operating status of the inspected devices, and the uninspected devices in the candidate path as the state space for Q-learning. The movement direction and path length from the current inspected device's location to the next inspected device in the candidate path are used as the action space for Q-learning. The Q-value is iteratively updated based on a pre-defined reward function.

2. The autonomous inspection method based on path planning according to claim 1, characterized in that, The reward function is: R = ω1 · R 效率 + ω2 · R 覆盖 + ω3 · R 能耗 + ω4 · R 风险 Where R is the reward function; R 效率 The value of the function is negatively correlated with the path length; R 覆盖 The function value is the sum of the inspection importance function and the important equipment coverage function. The value of the inspection importance function is positively correlated with the importance of the next piece of equipment to be inspected, and the value of the important equipment coverage function is negatively correlated with the importance of any equipment missed during the inspection process. R 能耗 The value of the function is negatively correlated with the energy consumption during inspection; R 风险 The value of the function is positively correlated with the operational risk level of the next piece of equipment to be inspected; ω1, ω2, ω3 and ω4 are the corresponding reward weight coefficients.

3. The autonomous inspection method based on path planning according to claim 1, characterized in that, The heuristic function is: h(n)=α·d manhattan (n,goal)+β·P(goal) Where h(n) is the heuristic function; n is the currently inspected device; and goal is the next device to be inspected. d manhattan (n,goal) is the distance estimation function, representing the Manhattan distance from the currently inspected device to the next device to be inspected; P(goal) is the device priority function; α and β are the corresponding set heuristic weight coefficients.

4. The autonomous inspection method based on path planning according to claim 1, characterized in that, During the inspection process, if any abnormal situation occurs, the optimal path is replanned according to the pre-set dynamic path planning algorithm, and then the inspection is carried out according to the replanned optimal path; the abnormal situation includes the abnormal operating status of the uninspected equipment and / or the presence of obstacles on the inspection path.

5. The autonomous inspection method based on path planning according to claim 1, characterized in that, During the inspection process, the system also collects the operating data of the equipment to be inspected. Through preset fault diagnosis rules and machine learning models, it determines whether the equipment to be inspected is abnormal. If an abnormality is found, an alarm signal is sent.

6. The autonomous inspection method based on path planning according to claim 1, characterized in that, The optimal path is displayed using AR technology.

7. The autonomous inspection method based on path planning according to claim 6, characterized in that, Adjust the optimal path using AR technology.

8. The autonomous inspection method based on path planning according to claim 1, characterized in that, During the inspection, AR technology is also used to display the status information of the equipment to be inspected, including real-time operating parameters, historical fault records, and maintenance guidance information.

9. An autonomous inspection system based on path planning, comprising a processor, characterized in that, The processor executes a computer program to implement the steps of the method according to any one of claims 1 to 8.

Images