Power transmission line unmanned aerial vehicle autonomous inspection path planning method and quality inspection system

Abstract

The invention relates to the technical field of unmanned aerial vehicle autonomous control, and discloses a power transmission line unmanned aerial vehicle autonomous inspection path planning method and a quality inspection system, and the method comprises the following steps: obtaining or constructing and dynamically evolving a cognitive digital twin containing geometric, semantic and uncertainty information; based on the first task intention, aiming at reducing the uncertainty of the cognitive digital twinning, generating a first flight path by optimizing information gain; controlling the unmanned aerial vehicle to fly along the first flight path, and performing in-service quality inspection on the acquired data to obtain a quantized target component defect suspected degree; and when the defect suspected degree meets a preset risk triggering condition, dynamically generating a second task intention with a higher priority, and generating a second flight path for fine detailed investigation based on the intention so as to replace or modify the first flight path. According to the invention, the unmanned aerial vehicle is converted into an active diagnosis agent from a passive path executor, and the autonomy, reliability and efficiency of routing inspection are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of autonomous control technology of unmanned aerial vehicles (UAVs), and in particular to a transmission line UAV autonomous inspection path planning method and quality inspection system. Background Art

[0002] Unmanned aerial vehicles (UAVs) have been increasingly widely used in the inspection and maintenance of power transmission lines. Compared with traditional manual inspections, UAV operations have shown significant advantages in efficiency, safety, and accessibility.

[0003] Currently, mainstream drone inspection methods typically rely on pre-planned flight paths. Based on the transmission line's Geographic Information System (GIS) data or historical survey data, technicians pre-set a flight path in the ground station software that covers all targets to be inspected. During operation, the drone strictly follows this pre-set route, collecting data using onboard sensors such as cameras and lidar. After the flight mission is complete, the massive amount of data collected is downloaded to ground computers, where professionals conduct offline analysis and diagnosis to determine if any defects exist.

[0004] However, this inspection model, which relies heavily on preset trajectories, has inherent limitations. This model completely separates perception from the decision-making process, and the drone acts solely as a passive "data collector" during flight, lacking the ability to understand and respond to the environment and its own collected data in real time. Because inspection sites are complex and ever-changing, and because prior GIS data may be biased, a fixed flight path cannot guarantee the acquisition of high-quality diagnostic data at the optimal distance and angle at critical locations. For example, images may be motion blurred due to gusts of wind, or overexposed due to sudden changes in lighting. These problems cannot be identified and corrected at the time of data collection and can only be discovered during subsequent analysis. By then, the optimal opportunity for re-shooting has often been missed, and a costly second flight may be necessary.

[0005] More importantly, this model lacks an immediate response mechanism to risks. When a drone flies over an area with obvious defects or serious hidden dangers, it fails to "realize" the importance of the area and instead continues along its established route to complete the remaining routine inspections. This "one size fits all" inspection strategy is not only inefficient but can also lead to delayed responses to major and urgent defects. Ultimately, existing technology lacks a closed-loop control mechanism that enables drones to think and make autonomous decisions while in service, limiting the intelligence and reliability of inspection operations. Summary of the Invention

[0006] The technical problem to be solved by the present invention is that most existing transmission line drone inspection methods rely on preset rigid flight paths and cannot adapt to dynamic changes in the inspection site; at the same time, their data collection and quality analysis processes are separated, resulting in the inability to ensure data quality in real time, key defects may be missed due to invalid data, and the autonomy and reliability of the inspection are insufficient.

[0007] In order to solve the above technical problems, the present invention provides a transmission line UAV autonomous inspection path planning method and quality inspection system.

[0008] A first aspect of the present invention provides a method for planning a transmission line UAV autonomous inspection path.

[0009] The core of this approach lies in building a closed-loop intelligent system capable of real-time perception, cognition, and decision-making. By introducing a dynamically evolving cognitive digital twin as the system's cognitive foundation, this method transforms inspection path planning from executing a fixed sequence of waypoints to completing dynamic mission intent. This method can instantly adjust task priorities and flight strategies based on real-time findings from in-service quality inspections, transforming passive flight into active exploration and diagnosis. This fundamentally addresses the disconnect between planning and execution, and between data collection and analysis, inherent in traditional approaches.

[0010] Specifically, the method includes: First, obtain or build a cognitive digital twin that represents the physical environment of the transmission line.

[0011] This cognitive digital twin is not a static 3D model, but a dynamic data structure that incorporates geometry, semantics, component states, and uncertainty information. Uncertainty information is key to the invention, quantifying the degree of uncertainty in the system's understanding of all information in the twin world.

[0012] Subsequently, the system generates a first task intention with the content of inspecting a specific target component.

[0013] Based on this first mission intention, and with the core goal of reducing the uncertainty information related to this intention in the cognitive digital twin, the first flight path is generated.

[0014] A specific implementation of this process is to define the reduction in uncertainty as an information gain reward function, and through decision-making algorithms such as reinforcement learning, solve the flight action that can maximize the reward, thereby forming a path. For example, an information gain reward function It can be expressed as: ; in, It is at the moment Execute an action Rewards received after and The total uncertainty of the relevant area in the cognitive digital twin before and after the action is executed, and the difference between the two That is information gain; To perform an action This reward function encourages the drone to actively explore the area with the richest information.

[0015] The UAV flies along the generated first flight path and collects data of the target component at appropriate observation points.

[0016] The key is that the system immediately conducts in-service quality control on the collected data. The quality control process is twofold: First, assess the physical quality of the data (e.g., clarity) to ensure that the data is usable; Then, an onboard defect diagnosis model is used to analyze the qualified data to obtain the quantitative defect suspicion of the target component. .

[0017] When the defect suspicion degree is obtained When the preset risk trigger conditions are met, such as exceeding a set attention threshold , the system will trigger the core dynamic re-planning mechanism. This mechanism will immediately generate a second task intention for the purpose of detailed inspection of the high-risk target component. The priority of this second task intention is It is set to be higher than all regular first-task intentions, and its priority can be calculated as follows: ; in, is a benchmark value that is higher than the priority of regular tasks. is the weight coefficient, is the defect suspicion degree, is the attention threshold.

[0018] The system then regenerates a second flight path for refined diagnosis based on this second mission intention with the highest priority to replace or modify the first flight path based on the original one.

[0019] The drone will immediately perform the diagnostic task until the uncertainty of the component status is reduced to below the confirmation threshold, thus forming a complete in-service closed loop from discovering suspicious points to confirming the diagnosis.

[0020] In addition, during the entire flight process, the drone always uses the data collected by sensors in real time to continuously update various types of information of the cognitive digital twin, enabling it to continuously and closely reflect the real state of the physical world.

[0021] A second aspect of the present invention provides a transmission line drone autonomous inspection path planning and quality inspection system, which is configured to perform all or part of the steps of the aforementioned method. Specifically, the system includes: A cognitive digital twin module, whose function is to acquire, build, and maintain a cognitive digital twin representing the physical environment of the transmission line. This module is the data and cognitive foundation of the entire system, continuously integrating sensor data and providing a unified world model containing geometric, semantic, state, and uncertainty information to other modules.

[0022] A mission planning module, which communicates with the cognitive digital twin module, is responsible for generating a first mission intent based on the inspection objectives. Based on this intent and uncertainty information obtained from the cognitive digital twin module, it generates a first flight path for the drone by optimizing information gain.

[0023] An in-service quality inspection module receives and processes data collected from target components in real time while the drone is in flight. It performs physical quality assessment and preliminary defect diagnosis, ultimately outputting a quantified defect suspicion score that is passed to the dynamic replanning module.

[0024] A dynamic replanning module serves as the system's decision-making hub. It receives defect suspicion levels from the in-service quality inspection module and determines whether they meet risk trigger conditions. If so, the module immediately generates a second mission intent with the highest priority (i.e., a diagnostic mission) and instructs the mission planning module to generate a second flight path for a more detailed inspection based on this new intent. This module then leads to adaptive adjustments in the system's mission priorities and flight behavior.

[0025] The present invention provides a transmission line UAV autonomous inspection path planning method and quality inspection system. It has the following beneficial effects: 1. This invention constructs a dynamic cognitive digital twin that incorporates uncertainty information and sets the path planning goal to reduce this uncertainty, freeing drone flight decisions from relying on fixed, pre-set routes. When unexpected obstacles or environmental changes appear at the inspection site, the system identifies them as areas of high uncertainty within the cognitive digital twin, guiding the drone to avoid or explore them rather than relying on passive obstacle avoidance. This mechanism enables the system to be highly adaptable to complex and dynamic real-world environments, significantly improving the reliability and safety of autonomous drone flight.

[0026] 2. This invention's closed-loop mechanism of in-service quality inspection and dynamic replanning shifts data analysis from post-process analysis to in-flight analysis. By performing real-time defect suspicion assessment on collected data, the system immediately prioritizes detailed investigation of any high-risk suspected defect and generates a dedicated diagnostic flight path. This ensures that any potentially serious defects are promptly and meticulously reviewed and confirmed, avoiding the permanent loss of critical information associated with traditional methods due to poor data quality or a single observation angle, thereby significantly improving the reliability of inspection results.

[0027] 3. The system of this invention can autonomously complete the perception-cognition-decision-action cycle, elevating inspection tasks from simple data collection to intelligent information acquisition and diagnosis. By dynamically generating and prioritizing mission intent, the system intelligently allocates inspection resources, focusing valuable flight time on exploring areas of high uncertainty or high risk. This goal-driven exploratory inspection reduces redundant flights in healthy areas, improves the efficiency of problem detection, and significantly enhances the autonomy of the entire inspection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a functional architecture diagram of a transmission line UAV autonomous inspection path planning and quality inspection system according to one embodiment of the present invention; Figure 2 The figure is a flow chart of a method for autonomous inspection path planning of transmission lines by drones according to an embodiment of the present invention.

[0029] Among them, 101, cognitive digital twin module; 102, mission planning module; 103, in-service quality inspection module; 104, dynamic replanning module. DETAILED DESCRIPTION

[0030] To make the purpose, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be described in detail below. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] The following is combined with Figure 1 -Attached Figure 2 The technical solution of the present invention is described in detail.

[0032] Refer to the attached Figure 1 , Figure 1 2 is a functional architecture diagram of a transmission line UAV autonomous inspection path planning and quality inspection system according to an embodiment of the present invention.

[0033] The present invention provides a transmission line UAV autonomous inspection path planning and quality inspection system, which can be integrated into a UAV platform with autonomous flight capability in terms of hardware.

[0034] An exemplary hardware platform may include: a multi-rotor UAV body, a high-performance flight controller, a set of multi-sensor components for environmental perception, and an onboard computing unit that serves as the brain of the system.

[0035] The multi-sensor assembly includes at least one laser radar (LiDAR) for accurate acquisition of 3D geometric information and a high-resolution visible light camera for collecting image data of target components. The onboard computing unit (OCU) is an embedded device with powerful parallel computing capabilities, responsible for running the core software modules of the system in real time.

[0036] On top of this hardware platform, the core functions of this system are implemented by a series of collaborative software modules deployed in the onboard computing unit.

[0037] like Figure 1 As shown in Figure 1, the combination of these modules forms a complete, adaptive closed-loop control flow from perception and cognition to decision-making and action. The system includes a cognitive digital twin module 101, a mission planning module 102, an in-service quality inspection module 103, and a dynamic replanning module 104.

[0038] The cognitive digital twin module 101 receives raw data streams from sensors such as lidar and cameras. By running online simultaneous localization and mapping (SLAM) and image semantic segmentation algorithms, it constructs and dynamically updates a multi-layered cognitive digital twin (CDT) in real time. This cognitive digital twin includes not only a geometric model describing the three-dimensional spatial characteristics of the environment and a semantic model that assigns engineering meaning to geometric objects, but also an uncertainty model. This uncertainty model quantitatively characterizes the system's degree of certainty regarding each information element in the twin model, providing the fundamental basis for all subsequent decision-making.

[0039] The mission planning module 102 receives the first mission intent, which is set by the upper layer and contains the inspection target component, and obtains the global uncertainty information distribution map from the cognitive digital twin module 101. The core goal of this module is to transform the abstract goal of reducing cognitive uncertainty into a specific sequence of flight actions. To this end, it quantifies information gain as a reward function and, through decision-making algorithms such as deep reinforcement learning, solves a flight path that maximizes information gain, namely the first flight path. This path is not a fixed set of waypoints, but a dynamically generated strategic trajectory that explores unknown or highly uncertain areas with maximum efficiency.

[0040] When the drone, guided by the mission planning module 102, collects data (such as images) about a target component during flight, the in-service quality inspection module 103 immediately processes this data. It performs a dual assessment: first, it evaluates the data's physical quality by calculating metrics such as clarity and exposure to ensure data validity. Then, for data that meets physical quality standards, it uses a lightweight onboard defect diagnosis neural network to perform preliminary defect analysis and output a quantitative defect suspicion score, which indicates the likelihood of an abnormality in the component.

[0041] Dynamic replanning module 104 continuously monitors the defect suspicion level output by in-service quality inspection module 103. Once this level meets a pre-set risk trigger condition, such as exceeding a certain concern threshold, the module immediately determines that a potential major defect has occurred. In response, it immediately generates a second, highest-priority mission intent, focusing on a detailed inspection of the high-risk target component. It then instructs mission planning module 102 to generate a second, more detailed inspection flight path based on this new diagnostic intent, replacing or modifying the first flight path.

[0042] During system operation, these modules form a tightly coupled whole. The cognitive digital twin module 101 provides a world model for decision-making, based on which the mission planning module 102 generates preliminary action plans. The in-service quality inspection module 103 monitors the system in real time and, through the dynamic replanning module 104, instantly converts these into new, more targeted action instructions. This makes the entire inspection process highly intelligent and adaptable, ensuring in-depth exploration and confirmation of key risk points.

[0043] Refer to the attached Figure 2 , Figure 2 The present invention provides a method for autonomous inspection of transmission lines by drones, which may include the following steps: S201. Acquire or build and dynamically evolve cognitive digital twins.

[0044] This step provides the entire method with a unified world model that contains uncertainty information, which serves as the basis for all subsequent perception, planning, and decision-making.

[0045] S202: Generate a first mission intention, and based on the intention, generate a first flight path with the goal of reducing the uncertainty of the cognitive digital twin.

[0046] This step is the core planning link in the routine inspection state, and its purpose is to improve the system's complete understanding of the environment in the most efficient way through exploratory flight.

[0047] S203 : Control the UAV to fly along the first flight path, and perform in-service quality inspection on the target component data collected during the flight to obtain a defect suspicion degree of the target component.

[0048] This step introduces a real-time quality control and risk assessment mechanism while the path is being executed.

[0049] S204. Determine whether the defect suspicion level meets a preset risk trigger condition. If the condition is met, dynamically generate a second task intention with a higher priority than the first task intention, and generate a second flight path based on the second task intention.

[0050] This step is a key embodiment of the intelligent adaptability of the present invention. Through the dynamic reordering of tasks, seamless switching from routine inspections to focused detailed inspections is achieved.

[0051] The above steps are described in detail below.

[0052] Specifically, in step S201, the process of acquiring or constructing and dynamically evolving the cognitive digital twin is as follows: This step aims to provide a unified, uncertainty-informed world model for the entire autonomous inspection approach—the cognitive digital twin (CDT). This model serves as the basis for all subsequent perception, planning, and decision-making. The CDT is not a static three-dimensional map but a continuously evolving, multidimensional data structure that integrates geometric, semantic, and uncertainty information.

[0053] During the mission initialization phase, the system first constructs the CDT. If historical survey data exists, such as geographic information system (GIS) data of transmission lines or existing lidar point clouds, the system loads this data to build an initial twin model. In this initial model, the cognitive state of all regions and components is assigned a high uncertainty value, such as a value close to 1, indicating that the accuracy of the prior information has not yet been verified by real-time perception. Without any prior data, the CDT is gradually constructed from a blank slate.

[0054] After the mission begins, the drone takes off and enters the inspection area, and the online dynamic evolution of CDT is activated. The system uses the onboard laser radar (LiDAR) and inertial measurement unit (IMU) through a tightly coupled laser inertial odometry SLAM algorithm to estimate the drone's six-degree-of-freedom position in real time and simultaneously build a geometric model of the environment. In a specific embodiment, the algorithm may use LIO-SAM, FAST-LIO2 and other schemes known in the art, or perform adaptive modifications based on these schemes.

[0055] The geometric model It is represented as a dense 3D point cloud map, which describes the spatial position and outline of objects such as transmission towers, wires, and landforms. As the drone continues to fly, the newly collected point cloud data is continuously aligned and integrated into the global map. The coverage is expanded and the details are refined.

[0056] While building the geometric model, the system collects image data using a high-resolution camera, triggered synchronously with the lidar. This image data is then used by an onboard, lightweight convolutional neural network (CNN) for real-time semantic segmentation. For example, models from the MobileNet series, ShuffleNet series, or optimized YOLO series with pruned and quantized models can be used. These models are chosen for their ability to maintain high recognition accuracy while minimizing the computing power consumption of the onboard computing platform and meeting real-time requirements.

[0057] The network recognizes predefined key components or environmental elements in the image, such as "insulator string", "shock absorber", "conductor", "tower", etc. The system projects the semantic labels obtained by segmentation onto the 3D geometric model based on the current precise drone pose and camera internal and external parameter matrix. This process establishes semantic associations for each point or each point cluster in the 3D point cloud, thus forming a semantic model. .

[0058] In addition, this step also includes uncertainty information The system is a model for each basic unit in the twin space (for example, a 3D voxel). ) maintains an uncertainty value. This value combines multiple sources of uncertainty. A specific calculation example is: ; in: is a voxel The final comprehensive uncertainty value is in the range of [0,1], where 0 means complete certainty and 1 means complete uncertainty.

[0059] is the observation uncertainty, which characterizes the degree to which the voxel is effectively observed geometrically. A voxel that is never covered by any sensor’s field of view has a . Whenever the voxel is effectively observed by the lidar or camera, its The value will decay once according to the quality factors such as the observed distance and angle. For example, its decay process can follow the following update rule: ,in is the learning rate, and is a variable with observation distance Increase and observation angle The quality factor that decreases as it increases (away from the normal direction).

[0060] is the semantic uncertainty, which comes from the confidence of the semantic segmentation network output. For example, if the network identifies a region as category The softmax output probability is , then the semantic uncertainty of the region can be defined as Low-confidence recognition results will lead to high semantic uncertainty.

[0061] and is a preset weight coefficient used to balance the importance of geometric observation and semantic recognition in comprehensive uncertainty assessment, and satisfies .

[0062] Through the above process, during the entire flight, every movement and observation of the drone will cause the corresponding voxels within its perception range to and changes, and continuously update the global Uncertainty Map. In this way, CDT can dynamically reflect the system's current cognitive state, not only including the geometric and semantic information of the environment, but also providing a quantitative, real-time updated confidence assessment for this information, thus providing data support for subsequent advanced intelligent decision-making steps.

[0063] Specifically, in step S202, a first mission intention is generated, and based on the intention, with the goal of reducing the uncertainty of the cognitive digital twin, a process of generating a first flight path is as follows: This step is performed in regular inspection mode. Its purpose is to transform a high-level "first mission intent" (for example, inspecting the route between towers 15 and 17) into a dynamic planning process focused on proactive information acquisition, which is used to generate a first flight path. This first flight path is specifically represented by a sequence of flight maneuvers determined in real time to improve the system's complete understanding of the environment.

[0064] A specific implementation method is to construct the path planning problem as a partially observable Markov decision process (POMDP). In this decision framework, the state space of the UAV is Defined by the current state of the cognitive digital twin (CDT) built in step S201, which includes geometric, semantic, and global uncertainty maps Action Space It is a set of discrete basic flight actions that can be accurately executed by the flight control system, such as {fly forward one meter, move horizontally one meter to the left, move vertically one meter upward, and yaw 15 degrees to the left in place}, etc.

[0065] This decision process introduces a reward function The design is used to quantify the abstract goal of "reducing uncertainty" into the immediate reward that the agent can obtain after each action. A specific reward function At the moment Can be defined as: ; in: It's the drone at the moment Execute an action The immediate reward value obtained after.

[0066] The total information entropy or uncertainty of the area in the CDT related to the current task intention before and after the action is performed. The difference between the two constitutes the information gain term. The information entropy can be calculated as the total information entropy of all voxels in the task area. The uncertainty value of The sum of A positive information gain value corresponds to an effective observation behavior, which reduces the system's unknown degree of the environment.

[0067] Is to perform an action A cost function is created that quantifies and imposes negative incentives on behaviors such as flight time, energy consumption, or entering high-risk areas. The purpose of introducing the cost term is to impose constraints on the economy and safety of flight while optimizing information gain. In one embodiment, the cost function can be specifically expressed as: ,in and Execute actions The estimated energy and time required, and is the corresponding weight coefficient.

[0068] The goal of the system is to find an optimal flight strategy , this strategy can guide the drone to choose an action in any state to maximize the future expected cumulative discounted reward ,in is the discount factor. The optimal strategy It can be solved using deep reinforcement learning algorithms (e.g., training a Deep Q-Network (DQN) offline or using Monte Carlo Tree Search (MCTS) online).

[0069] Specifically, when using DQN, the uncertainty map of the local area around the drone can be The network takes the state as input and outputs the Q-value of each action in the discrete action space. When using MCTS, at each decision point, starting from the current state, a Monte Carlo simulation is performed to proactively evaluate the long-term information gain of different action sequences.

[0070] Therefore, the first flight path generated is not a static trajectory calculated once, but is calculated at each decision moment. Next best flight maneuver generated in real time This mechanism enables the path planning results to respond to changes in the cognitive digital twin state in real time, thereby guiding the UAV to move preferentially towards areas with higher uncertainty and perform observation tasks.

[0071] Specifically, in step S203, the process of controlling the UAV to fly along the first flight path and performing in-service quality inspection on the target component data collected during the flight to obtain the defect suspicion degree of the target component is as follows: This step embeds real-time data quality control and risk assessment into the path execution process. In this step, the system controls the drone to fly along the first flight path generated in step S202. During flight, it uses onboard sensors such as cameras to collect image data of target transmission line components.

[0072] The key to this step is to conduct in-service quality inspection on the collected image data before the task is completed. In a specific embodiment, the in-service quality inspection includes two evaluation aspects: First, a physical quality assessment is performed to ensure the validity of the image data.

[0073] Before using an image for defect analysis, the system quickly calculates and evaluates its basic physical properties. A specific implementation involves calculating the image's clarity (e.g., Laplace operator-based variance) and exposure metrics and determining whether these metrics fall within a preset acceptable range. The clarity threshold can be determined by statistically analyzing a set of sample images and selecting the inflection point that effectively distinguishes blurry from sharp images.

[0074] If the image data fails physical quality due to motion blur, loss of focus, or over / underexposure, the data will be marked as invalid and the system can choose to re-capture it at a nearby location.

[0075] Secondly, for image data that meets physical quality standards, a lightweight defect diagnosis model deployed on an onboard computing unit is called for analysis.

[0076] The model, such as a convolutional neural network (CNN) with optimized structure and model compression, can perform real-time, preliminary defect recognition on the image content without transmitting the data back to the ground. After analysis, the model outputs a quantitative defect suspicion degree. Specifically, the It can be the softmax output probability value of the defect diagnosis model for the defect category, or a comprehensive anomaly score output by the model.

[0077] Defect suspicion It represents the probability of abnormality or potential failure of the target component in numerical form (e.g., a continuous value between 0 and 1). A value of indicates that a significant defect feature is found in the current observation data. The value will be used as a direct input for decision-making in the subsequent step S204 and is the core basis for triggering dynamic replanning.

[0078] Specifically, in step S204, the process of determining whether the defect suspicion level meets the preset risk triggering condition and performing dynamic replanning accordingly is as follows: This step defines a dynamic re-planning mechanism, which is based on the defect suspicion output in step S203. Conduct an assessment and, when the preset conditions are met, switch the task from routine inspection to focused inspection.

[0079] One or more risk trigger conditions are preset in the system. In one embodiment, the condition is the degree of defect suspicion. Exceeding a preset attention threshold The system will continue to obtain and When comparing, judging The risk trigger condition is met when . The satisfaction of the condition indicates that a high-risk feature that requires further investigation has been identified in the current observation.

[0080] When the risk trigger condition is met, the system automatically generates a second task intention with the content of detailed inspection of the high-risk target component. The second task intention is given a higher priority than all the first task intentions currently being executed or to be executed.

[0081] In order to quantify and dynamically assign the priority of the diagnostic task, its priority It can be calculated according to the following formula: ; in: is the priority of the newly generated second task intention.

[0082] It is a preset priority reference value, which is higher than the highest possible priority of all conventional first task intentions, thereby ensuring the priority execution of the diagnostic task.

[0083] It is a positive weight coefficient used to adjust the contribution of defect suspicion to priority improvement.

[0084] The role of the item is to give tasks whose defect suspicion exceeds the threshold more priority, thereby achieving differentiated treatment of different risk levels.

[0085] After generating the second mission intention with the highest priority, the system instructs the mission planning module to immediately generate a second flight path based on the new diagnostic intention.

[0086] This second flight path can be generated using a different strategy than the first. For example, a path generator based on a preset geometric template could be used, such as one that generates a circular trajectory centered on the target component or a sequence of hovering observation points from multiple perspectives. Alternatively, the POMDP framework could be retained but the reward function could be modified to reward only the reduction of epistemic uncertainty about the specific target component and to increase the reward for viewpoint diversity.

[0087] The planning goal of the second flight path is to optimize the collection of local information for specific high-risk components, such as planning a flight trajectory around the component or a sequence of waypoints for hovering observation from multiple specific angles. The purpose is to obtain multi-perspective, high-resolution data for accurate diagnosis.

[0088] This second flight path immediately replaces or modifies the first flight path, causing the drone to interrupt its regular inspection mission and switch to executing this detailed diagnostic inspection. This process establishes a response mechanism from the discovery of suspicious points in service to the execution of a detailed inspection. The detailed inspection mission will continue until a preset termination condition is met. A specific termination condition is that after executing the second flight path and continuously updating the cognitive digital twin, the uncertainty about the health status of the target component decreases below a preset confirmation threshold. Once this condition is met, the system ends the current diagnostic mission and can resume the previous regular inspection mission.

[0089] Attention threshold The specific value of the confirmation threshold can be adjusted according to the security level of the actual application scenario and the tolerance for missed reports and false positives. For example, in a high-demand scenario, a lower attention threshold can be set. To increase sensitivity.

[0090] In order to more clearly illustrate the technical solution of the present invention, the overall execution process of the method is described in detail below in conjunction with a specific application scenario.

[0091] Scene setting: Mission objective: The drone needs to conduct autonomous inspections of the first-stage transmission line and its components between towers 15 and 16.

[0092] Initial state: The system loaded the low-precision GIS data of the line and built the initial framework of the cognitive digital twin (CDT), but the comprehensive uncertainty of all areas They are all initialized to high values ​​close to 1.0. The drone is located at the takeoff point near tower 15.

[0093] The specific steps include: S1. Mission initiation and preliminary exploration: The drone received “Inspect Tower 15-16 Line” as its first mission intent.

[0094] The system starts and the drone takes off. According to step S201, the drone starts to use LiDAR and camera to collect data in real time and build a geometric map through SLAM algorithm. And perform semantic segmentation while continuously updating the global uncertainty map .

[0095] At this moment, according to step S202, the task planning module starts working. are very high, the reward function The calculation results show that flying towards the nearest large unknown area (i.e., Tower 15) can obtain the greatest information gain. Therefore, the first flight path generated by the system points to Tower 15, and the drone begins to fly towards the tower structure.

[0096] In the process of flying close to and around Tower 15, the geometric and semantic information of the tower, crossarms, insulators and other components were quickly built into the CDT. The value drops rapidly.

[0097] S2. Routine inspection and in-service quality inspection: After completing the preliminary modeling of Tower 15, the area with the highest uncertainty in the CDT shifted to the conductor and its associated equipment connecting Towers 15 and 16. The mission planning module ( S202 ) therefore generated a flight path along the conductor, and the drone entered the routine inspection phase.

[0098] The drone flies at a constant speed along the wire, and the onboard camera continuously collects images. According to step S203, the in-service quality inspection module processes each frame of images containing key components (such as the anti-vibration hammer and suspension clamp) in real time.

[0099] When the drone flies over the first anti-vibration hammer, the image obtained is clear and the exposure is normal, and the physical quality assessment is qualified. The defect diagnosis model outputs the defect suspicion degree after analysis Since 0.08 is much smaller than the preset attention threshold (For example, ), the system determines that the component is in normal condition. The drone does not change its flight path and continues to move forward.

[0100] S3. Anomaly detection and dynamic replanning: The drone continued to fly, and when observing the second shock absorber, the in-service quality inspection module was activated again. This time, the defect diagnosis model detected the suspected loose bolt feature from the image and output a defect suspicion degree of .

[0101] According to step S204, the system determines , the risk triggering conditions are met.

[0102] The system immediately performs dynamic replanning: Generate the second task intention: a task of "detailed inspection of shock-absorbing hammer No. 2" is generated. Its priority According to the formula is calculated and placed at the top of the task queue.

[0103] Interruption of the original path: The original first flight path, which continued along the wire to Tower 16, was suspended.

[0104] Generate a Second Flight Path: The mission planning module generates a new second flight path based on the precise three-dimensional coordinates of the shock absorber No. 2. This path is designed as a small-radius circular flight trajectory and hovers at three key diagnostic viewpoints (front, side, and top) to obtain the highest quality diagnostic data.

[0105] S4. Detailed inspection of execution and task recovery: The drone immediately executed the second flight path, flew near the No. 2 shock absorber, and began to perform detailed circling and hovering photography.

[0106] Multiple high-resolution, multi-angle images were collected and sent back to the system. This high-quality data not only made the geometric and semantic models of the shock absorber in CDT extremely refined, but more importantly, significantly reduced its state uncertainty.

[0107] The in-service quality inspection module used this new data to conduct another diagnosis, ultimately confirming the loose bolt defect and updating the component's status. When the overall uncertainty regarding the component's health status fell below a preset "confirmation threshold," the detailed inspection was completed.

[0108] The system then checked the task queue and found that the original first task intention, "Inspect Towers 15-16," had not yet been completed. So, starting from its current location, the system replanned a route to Tower 16, resuming its regular inspection mission.

[0109] It can be seen from this embodiment that the method of the present invention makes the drone no longer a tool that passively executes a preset trajectory, but an intelligent inspection subject that can actively perceive, continuously think, and dynamically make decisions, thereby greatly improving the autonomy, reliability, and efficiency of transmission line inspections.

[0110] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for autonomous inspection path planning of transmission lines by drones, characterized in that: The following steps are involved: Obtaining or constructing a cognitive digital twin representing the physical environment of the transmission line, wherein the cognitive digital twin includes uncertainty information of at least one component; Generate a first task intention with the inspection target component as the content; Based on the first mission intent and with the goal of reducing uncertainty information of the cognitive digital twin, generating a first flight path; controlling the UAV to fly along the first flight path and collect data of the target component; Performing in-service quality inspection on the collected data of the target component to obtain a suspected defect degree of the target component; When the defect suspicion level meets the preset risk trigger conditions, a second task intention is dynamically generated with the goal of detailed inspection of the target component and with a higher priority than the first task intention, and a second flight path is generated based on the second task intention to replace or modify the first flight path.

2. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: The cognitive digital twin includes: A geometric model that describes the spatial geometric characteristics of the transmission line environment; a semantic model that assigns semantic information to objects in the geometric model; Component status model that records key component status information; The uncertainty information quantifies the degree of certainty of information in the geometric model, semantic model, and component state model.

3. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: The step of generating a first flight path based on the first mission intention and aiming to reduce uncertainty information of the cognitive digital twin includes: Converting the uncertainty reduction information of the cognitive digital twin into an information gain reward function; A flight action sequence that maximizes the information gain reward function is solved through a reinforcement learning decision-making process to form the first flight path.

4. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: The step of performing in-service quality inspection on the collected data of the target component includes: Before performing defect suspicion analysis, the physical quality of the data is first evaluated; If and only if the physical quality of the data meets the preset quality standard, the subsequent defect suspicion analysis is performed.

5. The transmission line UAV autonomous inspection path planning method according to claim 4 is characterized in that: The step of obtaining the defect suspicion degree of the target component is specifically as follows: The data of the target component is input into a pre-trained defect diagnosis model, and the model outputs the defect suspicion degree representing the possibility of abnormality of the target component.

6. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: The preset risk trigger conditions are: The defect suspicion level exceeds a preset attention threshold.

7. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: After the UAV completes the detailed inspection of the target component along the second flight path, the method further includes the following steps: Continuously collect and analyze data until the uncertainty information about the state of the target component in the cognitive digital twin is reduced to below a preset confirmation threshold, and then end the flight based on the second mission intention.

8. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: The method further comprises: During the flight of the drone, the geometric model, semantic model and uncertainty information of the cognitive digital twin are continuously updated using real-time collected data.

9. The transmission line UAV autonomous inspection path planning method according to claim 1 is characterized in that: The method further comprises: Adaptively adjust the flight control parameters of the UAV based on local uncertainty information related to the current position of the UAV in the cognitive digital twin.

10. A transmission line drone autonomous inspection path planning and quality inspection system, used to execute the method according to any one of claims 1 to 9, characterized in that: include: A cognitive digital twin module, configured to acquire or construct a cognitive digital twin representing the physical environment of the transmission line, wherein the cognitive digital twin includes uncertainty information of at least one component; A mission planning module, configured to generate a first mission intent with the inspection target component as the content, and generate a first flight path based on the first mission intent with the goal of reducing the uncertainty information of the cognitive digital twin; An in-service quality inspection module is used to perform quality inspection on the data of the target component collected during the flight of the UAV to obtain the defect suspicion degree of the target component; A dynamic replanning module is used to dynamically generate a second task intention with the goal of detailed inspection of the target component and a priority higher than the first task intention when the defect suspicion level meets the preset risk trigger condition, and instruct the task planning module to generate a second flight path to replace or modify the first flight path.

Images