Unmanned aerial vehicle based wind turbine blade inspection local dynamic path planning method and system

Abstract

The application discloses a fan blade inspection local dynamic path planning method and system based on a UAV, relates to the technical field of unmanned aerial vehicles, and comprises the following steps: S1, a template generation stage: generating a standard inspection template based on a three-dimensional model of a blade, wherein the standard inspection template defines a globally optimal inspection sequence covering the surface of the blade and a starting point thereof; S2, a path mapping stage: mapping the standard inspection template to an actual blade and generating a three-dimensional space path maintaining a safe distance from the curved surface of the blade based on a preset rule; and S3, an online execution stage: a UAV executing the three-dimensional space path.The application solves the problem that existing UAV path planning methods do not fully consider the physical limitation that the UAV must maintain approximate horizontal flight during the inspection process, thereby causing the UAV to be forced to frequently adjust the attitude and substantially reducing the inspection efficiency.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) technology, specifically to a method and system for local dynamic path planning for wind turbine blade inspection based on UAVs. Background Technology

[0002] With the rapid development of drone technology, path planning has been widely used in many fields, such as agricultural monitoring, disaster relief, logistics and transportation, and urban air traffic. The purpose of path planning is to ensure that drones can avoid obstacles, reduce flight time and optimize energy consumption, and ensure the flight safety and efficient completion of missions. However, existing drone path planning still has some shortcomings.

[0003] The invention patent with publication number CN103809597A discloses a flight path planning method for a drone and the drone itself. The method includes the following steps: acquiring depth information of the drone's flight environment and generating a two-dimensional grid map of the flight environment, including obstacles; constructing a potential function for each grid cell and its corresponding obstacle based on the position of each grid cell and obstacle in the two-dimensional grid map; obtaining a weighted graph of the grid cells in the two-dimensional grid map based on the drone's position and the potential function; and determining the flight path based on the weighted graph. Constructing the potential function using the two-dimensional grid map and determining the flight path using the weighted graph obtained from the potential function improves computational efficiency and has good scalability. Existing drone path planning methods cannot efficiently inspect the three blades of a wind turbine, and they do not fully consider the physical constraint that the drone must maintain near-horizontal flight during inspection, forcing the drone to frequently adjust its attitude and significantly reducing inspection efficiency. Traditional path planning focuses on obstacle avoidance or area coverage, failing to fully consider the no-fly path problem in wind turbine blade inspection and cannot optimize the no-fly path. Summary of the Invention

[0004] The purpose of this invention is to address the problem that existing UAV path planning methods are not suitable for wind turbine blade inspection, fail to fully consider the physical limitation that UAVs must maintain near-horizontal flight during inspection, and thus cannot obtain clear inspection images, leading to frequent attitude adjustments by the UAVs and a significant reduction in inspection efficiency. This invention provides a local dynamic path planning method and system for wind turbine blade inspection based on UAVs.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a local dynamic path planning method for wind turbine blade inspection based on unmanned aerial vehicles (UAVs), comprising the following steps:

[0006] S1. Template generation stage: Based on the three-dimensional model of the blade, a standard inspection template is generated. The standard inspection template defines the globally optimal inspection sequence covering the blade surface and its starting point.

[0007] S2, Path Mapping Stage: Map the standard inspection template to the actual blade, and generate a three-dimensional spatial path that maintains a safe distance from the blade surface based on preset rules;

[0008] S3. Online execution phase: The UAV executes the three-dimensional spatial path and performs local rolling optimization based on the real-time environmental perception and motion control model;

[0009] The path mapping stage includes:

[0010] S201. Using the adaptive surface equidistant field algorithm, calculate and generate a three-dimensional spatial path that maintains a dynamically non-uniform safe distance from the blade surface.

[0011] S202. Establish a digital twin for blade inspection. This twin continuously receives perception data from the UAV during the inspection process and performs elastic scaling and real-time correction for unexecuted path segments.

[0012] The path mapping stage includes a step of establishing a circle center coordinate system:

[0013] Control the drone to fly sequentially to the preset starting points on the three blades;

[0014] Based on the spatial coordinates of three starting points, the rotation center of the wind turbine impeller is calculated.

[0015] With the rotation center as the origin, establish a center-blade polar coordinate system to uniformly describe the spatial pose of the three blades;

[0016] The path mapping stage is based on this coordinate system, and the path of the first blade is offset by the azimuth angle to simultaneously generate the inspection paths of the other two blades.

[0017] The method for generating the globally optimal inspection sequence employs a path simplification algorithm based on empty-run optimization, including:

[0018] Discretize the blade surface into a grid of detection units to identify the effective detection area and the no-load area;

[0019] Construct a graph theory-based path optimization model to transform the detection path planning into a constrained traveling salesman problem;

[0020] A multi-objective optimization algorithm is used to solve for the optimal detection sequence. The objective function simultaneously considers detection coverage, path length, and empty driving time.

[0021] As a further aspect of the present invention: in the adaptive surface isometric field algorithm, the isometric distance From the formula:

[0022]

[0023] in:

[0024] As a baseline safety distance;

[0025] To improve detection accuracy, the equidistant distance is automatically reduced in areas of high curvature to account for the local curvature of the blade.

[0026] Assign regional importance weights and automatically reduce equidistant distances at key locations;

[0027] , This is the adjustment coefficient;

[0028] This is the horizontal flight constraint correction amount, used to compensate for the viewpoint offset caused by the UAV in maintaining horizontal flight attitude;

[0029] The regional importance weight The methods for determining this include:

[0030] By integrating historical damage data, a damage evolution map of the blade is constructed;

[0031] For historically high-frequency defect areas and potential fatigue areas predicted by mechanical models, higher regional importance weights are automatically assigned. .

[0032] As a further aspect of the present invention: the online execution phase includes:

[0033] S301. Based on real-time wind field perception data and wind turbine attitude, a model predictive control (MPC) framework is used for local rolling optimization. The optimization objectives include utilizing the downwind aerodynamic shadow area of ​​the blades.

[0034] S302. When the airborne defect identification system detects a preset high-priority defect, it dynamically generates a local fine scanning path based on the online-tuned control parameters.

[0035] The online execution phase also includes a dynamic obstacle avoidance step, an online task replanning step, and an image quality optimization step based on biomimetic learning.

[0036] The dynamic obstacle avoidance steps include:

[0037] Real-time monitoring of sudden dynamic obstacles within the patrol airspace;

[0038] When a collision risk is detected, immediately plan an emergency escape corridor to deflect the drone off its original path;

[0039] After the danger is cleared, a smooth path is calculated using the optimal reconnection algorithm, allowing the drone to reconnect to the original inspection path.

[0040] The online task replanning steps include:

[0041] Continuously assess the completion status of inspection tasks and the remaining system resources;

[0042] When it is necessary to interrupt the inspection, an efficient convergence path is dynamically generated based on the remaining resources and area priority.

[0043] The image quality optimization steps based on biomimetic learning include:

[0044] An image quality prediction model is constructed, which predicts the blur and quality score of the image based on the real-time vibration state of the UAV, the relative speed with the blades, and the ambient light angle.

[0045] As a further aspect of the present invention, the method for generating the globally optimal inspection sequence includes:

[0046] Discretize the blade surface into a mesh and transform it into a graph theory problem;

[0047] Using the shortest total flight time, lowest total energy consumption, and smoothest gimbal attitude change as the multi-objective optimization function, a heuristic search algorithm is used to find an optimal Hamiltonian path, and the starting point of this path is the globally optimal starting point.

[0048] When multiple inspection drones exist within the system, a task negotiation mechanism based on distributed consensus is adopted, including:

[0049] Task Issuance: The globally optimal inspection sequence is divided into multiple sub-tasks, and a task descriptor containing location, required resources, and priority is generated for each sub-task;

[0050] Utility assessment and bidding: After receiving the task descriptor, each UAV calculates its expected utility value for performing the task based on its own local state information;

[0051] Task decision-making: Based on the expected utility values ​​submitted by each drone, subtasks are assigned to the optimal drone through consensus rules;

[0052] Commitment and Execution: The winning drone commits to the mission and updates its local mission queue, while the remaining drones are released to bid for other missions;

[0053] The distributed consensus task negotiation mechanism also supports dynamic task insertion:

[0054] When any UAV detects a high-priority defect, an emergency task descriptor containing the defect location and scanning requirements is immediately generated.

[0055] Through the aforementioned negotiation mechanism, the most suitable drone within the system can quickly bid and execute the emergency mission.

[0056] As a further aspect of the present invention: the method for generating the globally optimal inspection sequence adopts a path simplification algorithm based on empty-run optimization, and the objective function is defined as:

[0057]

[0058] in To effectively detect path length, This is the length of the empty driving route. For the number of path turns, To detect attitude repositioning time. , , , These are the weighting coefficients;

[0059] The formula for calculating the empty driving route length is:

[0060]

[0061] in, For the first The endpoint of each detection area, For the first The starting point of each detection area This is the obstacle avoidance correction function.

[0062] As a further aspect of the present invention: the path optimization model adopts a hierarchical planning strategy.

[0063] First layer: The detection area is divided into multiple detection sub-regions based on the geometric features of the blade, and each sub-region adopts a continuous scanning mode;

[0064] The second layer: Optimize the connection order of sub-intervals and use a genetic algorithm to solve for the optimal access sequence;

[0065] The third layer: An improved Hilbert curve filling algorithm is used within the sub-region to generate continuous, non-intersecting detection paths;

[0066] The improved Hilbert curve filling algorithm includes:

[0067] The fill density is adaptively adjusted based on the Gaussian curvature of the blade surface, with high-density fill used in high-curvature areas.

[0068] Step size of the fill path Determined by the following formula

[0069]

[0070] in As the reference step size, For local Gaussian curvature, For the maximum Gaussian curvature, Curvature sensitivity coefficient (0 < <1).

[0071] A path planning system is provided for implementing a local dynamic path planning method for wind turbine blade inspection based on UAVs. The system includes a digital template library module, a path mapping module, and an online execution module. The path mapping module includes an adaptive equidistant path generator, a digital twin management module, an emergency avoidance planner, a task replanning engine, a center positioning and synchronous planning module, a damage knowledge base module, and an image quality closed-loop optimizer.

[0072] As a further aspect of the present invention, the online execution module includes a pneumatic-motion coupled planner, a dynamic point of interest responder, a collaborative management module, and a reinforcement learning parameter tuning module.

[0073] As a further aspect of the present invention, the system also includes an empty-run optimization engine for executing a path simplification algorithm based on empty-run optimization, including a detection region division unit, a path sequence optimization unit, and a path filling generation unit.

[0074] As a further aspect of the present invention: the reinforcement learning parameter tuning module dynamically optimizes the weight coefficients in the empty-running optimization algorithm using the Q-learning algorithm. , , , The reward function is defined as:

[0075] in To improve detection coverage, Total inspection time. For energy consumption, , , These are the weighting coefficients of the reward function itself.

[0076] Compared with the prior art, the beneficial effects of the present invention are:

[0077] 1. By establishing a center-blade polar coordinate system and utilizing the 120° symmetry of the three blades, a single planning process was implemented, allowing for simultaneous inspection of all three blades and resolving the issue of low efficiency in multi-blade inspections. The starting points on the three blades are used to quickly calculate the turbine's rotation center, constructing a unified coordinate system. Within this coordinate system, planning the path for only one blade is sufficient to generate complete inspection paths for the other two blades through simple azimuth offset, reducing the computational load of path planning for all three blades by approximately two-thirds. This significantly improves the efficiency and speed of inspection planning and ensures consistent compliance with inspection standards across all three blades.

[0078] 2. A path simplification algorithm based on no-fly optimization and a hierarchical planning strategy were adopted to achieve global optimization and local efficiency in wind turbine blade inspection paths, significantly improving inspection efficiency. The path planning was decomposed into three levels: detection sub-region division, sub-region sequence optimization, and sub-region filling. The length of invalid no-fly paths, number of turns, and attitude repositioning time of the UAV were all included in the optimization objectives. Through no-fly optimization, the algorithm solves for the optimal Hamiltonian path sequence and uses adaptive Hilbert curves for continuous, cross-scanning, maximizing the simplification of the flight path at both global and local levels, significantly reducing invalid flights and time losses.

[0079] 3. By introducing an adaptive curved surface isometric field algorithm and image quality closed-loop optimization, high-quality image acquisition under the constraint of UAV horizontal flight was achieved, solving the problems of blurred and incomplete coverage of detection images. The algorithm dynamically adjusts the safe distance based on the local curvature of the blades and the weight of regional importance, automatically flying closer to high-risk, high-curvature areas. Simultaneously, the image quality prediction model is used as the optimization objective within the MPC framework, actively adjusting the flight speed to ensure that the UAV acquires consistently clear, high-quality detection images across the entire domain while maintaining a safe and stable horizontal flight attitude. Attached Figure Description

[0080] Figure 1 This is a flowchart of a local dynamic path planning method for wind turbine blade inspection based on unmanned aerial vehicles (UAVs) according to the present invention.

[0081] Figure 2 This is a block diagram of a path planning system according to the present invention;

[0082] Figure 3 This is a schematic diagram of the damage knowledge base module of the present invention;

[0083] Figure 4 This is a schematic diagram of the optimized empty-running structure of the present invention;

[0084] Figure 5 This is a schematic diagram of the path generation process of the present invention. Detailed Implementation

[0085] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0086] Example 1:

[0087] like Figures 1-5 As shown, this embodiment proposes a local dynamic path planning method for wind turbine blade inspection based on unmanned aerial vehicles (UAVs), including the following steps:

[0088] S1. Template generation stage: Based on the three-dimensional model of the blade, a standard inspection template is generated. The standard inspection template defines the globally optimal inspection sequence covering the blade surface and its starting point.

[0089] S2, Path Mapping Stage: Map the standard inspection template to the actual blade, and generate a three-dimensional spatial path that maintains a safe distance from the blade surface based on preset rules;

[0090] S3. Online execution phase: The UAV executes a three-dimensional spatial path and performs local rolling optimization based on real-time environmental perception and motion control models;

[0091] The path mapping phase includes:

[0092] S201. Using the adaptive surface equidistant field algorithm, calculate and generate a three-dimensional spatial path that maintains a dynamically non-uniform safe distance from the blade surface.

[0093] S202. Establish a digital twin for blade inspection. This twin continuously receives perception data from the UAV during the inspection process and performs elastic scaling and real-time correction for unexecuted path segments.

[0094] The path mapping stage includes a step of establishing a circle center coordinate system:

[0095] Control the drone to fly sequentially to the preset starting points on the three blades;

[0096] Based on the spatial coordinates of three starting points, the rotation center of the wind turbine impeller is calculated.

[0097] With the rotation center as the origin, establish a center-blade polar coordinate system to uniformly describe the spatial pose of the three blades;

[0098] Based on this coordinate system, the path mapping stage generates the inspection paths for the other two blades simultaneously by offsetting the path of the first blade with an azimuth angle.

[0099] The method for generating the globally optimal inspection sequence employs a path simplification algorithm based on empty-run optimization, including:

[0100] Discretize the blade surface into a grid of detection units to identify the effective detection area and the no-load area;

[0101] Construct a graph theory-based path optimization model to transform the detection path planning into a constrained traveling salesman problem;

[0102] A multi-objective optimization algorithm is used to solve for the optimal detection sequence. The objective function simultaneously considers detection coverage, path length, and empty driving time.

[0103] A path planning system includes a digital template library module, a path mapping module, and an online execution module. The path mapping module includes an adaptive isometric path generator, a digital twin management module, an emergency avoidance planner, a task replanning engine, a center positioning and synchronous planning module, a damage knowledge base module, and an image quality closed-loop optimizer.

[0104] By leveraging the symmetry of the three-blade design, efficient multi-blade path planning is achieved. Standard inspection templates and empty-run optimization algorithms ensure global optimality and local efficiency of the inspection process. Furthermore, real-time correction via digital twins enhances the accuracy and adaptability of path planning.

[0105] Example 2:

[0106] The solution in Example 1 will be further described below with reference to its specific working method.

[0107] As a preferred embodiment, based on the above method, further, in the adaptive surface isometric field algorithm, the isometric distance... From the formula:

[0108]

[0109] in:

[0110] As a baseline safety distance;

[0111] To improve detection accuracy, the equidistant distance is automatically reduced in areas of high curvature to account for the local curvature of the blade.

[0112] Assign regional importance weights and automatically reduce equidistant distances at key locations;

[0113] , This is the adjustment coefficient;

[0114] This is the horizontal flight constraint correction amount, used to compensate for the viewpoint offset caused by the UAV in maintaining horizontal flight attitude;

[0115] Regional importance weight The methods for determining this include:

[0116] By integrating historical damage data, a damage evolution map of the blade is constructed;

[0117] For historically high-frequency defect areas and potential fatigue areas predicted by mechanical models, higher regional importance weights are automatically assigned. .

[0118] As a preferred embodiment, based on the above method, the online execution phase further includes:

[0119] S301. Based on real-time wind field perception data and wind turbine attitude, a model predictive control (MPC) framework is used for local rolling optimization. The optimization objectives include utilizing the downwind aerodynamic shadow area of ​​the blades.

[0120] S302. When the airborne defect identification system detects a preset high-priority defect, it dynamically generates a local fine scanning path based on the online-tuned control parameters.

[0121] The online execution phase also includes dynamic obstacle avoidance steps, online task replanning steps, and image quality optimization steps based on biomimetic learning.

[0122] Dynamic obstacle avoidance steps include:

[0123] Real-time monitoring of sudden dynamic obstacles within the patrol airspace;

[0124] When a collision risk is detected, immediately plan an emergency escape corridor to deflect the drone off its original path;

[0125] After the danger is cleared, a smooth path is calculated using the optimal reconnection algorithm, allowing the drone to reconnect to the original inspection path.

[0126] Online task replanning steps include:

[0127] Continuously assess the completion status of inspection tasks and the remaining system resources;

[0128] When it is necessary to interrupt the inspection, an efficient convergence path is dynamically generated based on the remaining resources and area priority.

[0129] The steps for image quality optimization based on biomimetic learning include:

[0130] An image quality prediction model is constructed, which predicts the blur and quality score of the image based on the real-time vibration state of the UAV, the relative speed with the blades, and the ambient light angle.

[0131] As a preferred embodiment, based on the above method, a further method for generating the globally optimal inspection sequence includes:

[0132] Discretize the blade surface into a mesh and transform it into a graph theory problem;

[0133] Using the shortest total flight time, lowest total energy consumption, and smoothest gimbal attitude change as the multi-objective optimization function, a heuristic search algorithm is used to find an optimal Hamiltonian path, and the starting point of this path is the globally optimal starting point.

[0134] When multiple inspection drones exist within the system, a task negotiation mechanism based on distributed consensus is adopted, including:

[0135] Task Issuance: The globally optimal inspection sequence is split into multiple sub-tasks, and a task descriptor containing location, required resources, and priority is generated for each sub-task;

[0136] Utility assessment and bidding: After receiving the task descriptor, each UAV calculates its expected utility value for performing the task based on its own local state information;

[0137] Task decision-making: Based on the expected utility values ​​submitted by each drone, subtasks are assigned to the optimal drone through consensus rules;

[0138] Commitment and Execution: The winning drone commits to the mission and updates its local mission queue, while the remaining drones are released to bid for other missions;

[0139] The distributed consensus task negotiation mechanism also supports dynamic task insertion:

[0140] When any UAV detects a high-priority defect, an emergency task descriptor containing the defect location and scanning requirements is immediately generated.

[0141] Through a negotiation mechanism, the most suitable drone within the system can quickly bid and execute the emergency mission.

[0142] As a preferred implementation, based on the above method, the method for generating the globally optimal inspection sequence further employs a path simplification algorithm based on empty-run optimization, and the objective function is defined as:

[0143]

[0144] in To effectively detect path length, This is the length of the empty driving route. For the number of path turns, To detect attitude repositioning time. , , , These are the weighting coefficients;

[0145] The formula for calculating the empty driving route length is:

[0146]

[0147] in, For the first The endpoint of each detection area, For the first The starting point of each detection area This is the obstacle avoidance correction function.

[0148] As a preferred implementation, based on the above method, the path optimization model further adopts a hierarchical planning strategy:

[0149] First layer: The detection area is divided into multiple detection sub-regions based on the geometric features of the blade, and each sub-region adopts a continuous scanning mode;

[0150] The second layer: Optimize the connection order of sub-intervals and use a genetic algorithm to solve for the optimal access sequence;

[0151] The third layer: An improved Hilbert curve filling algorithm is used within the sub-region to generate continuous, non-intersecting detection paths;

[0152] The improved Hilbert curve filling algorithm includes:

[0153] The fill density is adaptively adjusted based on the Gaussian curvature of the blade surface, with high-density fill used in high-curvature areas.

[0154] Step size of the fill path Determined by the following formula

[0155]

[0156] in As the reference step size, For local Gaussian curvature, For the maximum Gaussian curvature, Curvature sensitivity coefficient (0 < <1).

[0157] As a preferred embodiment, based on the above method, the online execution module further includes a pneumatic-motion coupling planner, a dynamic point of interest responder, a collaborative management module, and a reinforcement learning parameter tuning module.

[0158] As a preferred embodiment, based on the above method, the system further includes an empty-run optimization engine for executing an empty-run optimization-based path simplification algorithm, including:

[0159] The detection area division unit is used to identify the valid detection area and the empty driving area;

[0160] The path sequence optimization unit is used to solve for the optimal detection sequence;

[0161] The fill path generation unit is used to generate a continuous scan path within the detection area.

[0162] As a preferred implementation, based on the above method, the reinforcement learning parameter tuning module further optimizes the weight coefficients in the empty-running optimization algorithm using the Q-learning algorithm. , , , The reward function is defined as:

[0163] in To improve detection coverage, Total inspection time. For energy consumption, , , These are the weighting coefficients of the reward function itself.

[0164] By further refining the path planning algorithm, including an adaptive curved surface isometric field algorithm, dynamic optimization during online execution, and a multi-UAV collaborative mechanism, detection accuracy and image quality are improved while ensuring safe distances. Hierarchical planning strategies and reinforcement learning parameter tuning optimize the global and local performance of the path, achieving efficient and adaptive wind turbine blade inspection.

[0165] Example 3:

[0166] The solutions in Embodiments 1 and 2 will be further described below with reference to their specific working methods.

[0167] Specifically, when using this UAV-based local dynamic path planning method and system for wind turbine blade inspection:

[0168] like Figure 5 As shown, considering the detection band width during horizontal flight of the UAV, the selection of the three starting points should ensure coverage of the critical area of ​​the blade. Let the spatial coordinates of the three starting points be:

[0169] Blade A starting point: A( , , )

[0170] Blade B starting point: B( , , )

[0171] Blade C starting point: C( , , )

[0172] The coordinates of the center O are obtained by solving the following improved system of equations. , , )O( , , ):

[0173]

[0174] get:

[0175]

[0176] in:

[0177]

[0178]

[0179] To consider the detection band coverage when the UAV maintains horizontal flight, the detection band width is introduced. and overlap rate :

[0180] Actual coverage width of the detection strip:

[0181]

[0182] in It is the angle between the blade surface normal vector and the UAV's vertical axis.

[0183] Interval between adjacent detection bands:

[0184]

[0185] Improved path generation formula based on 120° symmetry:

[0186] In the center-blade polar coordinate system, considering the horizontal flight constraints of the UAV, the path generation formula is improved as follows:

[0187] For the path point on the first blade ,in It is the gimbal pitch angle, which is used to generate symmetrical path points on the other two blades through rotation transformation:

[0188] The second leaf corresponds to the following point:

[0189]

[0190] The second leaf corresponds to the following point:

[0191]

[0192] Among them, height correction amount Calculated by the following formula:

[0193]

[0194] in, Install the tilt angle on the blades.

[0195] Considering the horizontal flight constraints of the UAV, the contour offset formula in the polar coordinate system yields the reference contour curve. The offset profile is:

[0196]

[0197] The adaptive offset is:

[0198]

[0199] The horizontal flight correction is:

[0200]

[0201] The angle between the blade chord and the horizontal plane.

[0202] To ensure the inspection strip completely covers the blade surface, the drone's flight altitude needs to be dynamically adjusted:

[0203]

[0204] The blade cross-section height, The blade twist angle, For safety margin;

[0205] This method achieves precise control over the calculation of the center coordinates and path generation. By considering the detection band width, overlap rate, and horizontal flight constraints, it ensures the coverage integrity and flight stability of the inspection path. The introduction of altitude correction and adaptive offset optimizes the inspection effect of UAVs on complex curved surfaces, improving the practicality and reliability of the system.

[0206] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A local dynamic path planning method for wind turbine blade inspection based on unmanned aerial vehicles (UAVs), comprising the following steps: S1. Template generation stage: Based on the three-dimensional model of the blade, a standard inspection template is generated. The standard inspection template defines the globally optimal inspection sequence covering the blade surface and its starting point. S2, Path Mapping Stage: Map the standard inspection template to the actual blade, and generate a three-dimensional spatial path that maintains a safe distance from the blade surface based on preset rules; S3. Online execution phase: The UAV executes the three-dimensional spatial path and performs local rolling optimization based on the real-time environmental perception and motion control model; The path mapping stage includes: S201. Using the adaptive surface equidistant field algorithm, calculate and generate a three-dimensional spatial path that maintains a dynamically non-uniform safe distance from the blade surface. S202. Establish a digital twin for blade inspection. This twin continuously receives perception data from the UAV during the inspection process and performs elastic scaling and real-time correction for unexecuted path segments. The path mapping stage also includes the establishment of a circle center coordinate system: In the adaptive surface isometric field algorithm, the isometric distance From the formula: in: As a baseline safety distance; To improve detection accuracy, the equidistant distance is automatically reduced in areas of high curvature to account for the local curvature of the blade. Assign regional importance weights and automatically reduce equidistant distances at key locations; , This is the adjustment coefficient; This is the horizontal flight constraint correction amount, used to compensate for the viewpoint offset caused by the UAV in maintaining horizontal flight attitude; The regional importance weight The methods for determining this include: By integrating historical damage data, a damage evolution map of the blade is constructed; For historically high-frequency defect areas and potential fatigue areas predicted by mechanical models, higher regional importance weights are automatically assigned. .

2. The method for local dynamic path planning of wind turbine blade inspection based on UAV according to claim 1, characterized in that, The steps for establishing the circular center coordinate system are as follows: Control the drone to fly sequentially to the preset starting points on the three blades; Based on the spatial coordinates of three starting points, the rotation center of the wind turbine impeller is calculated. With the rotation center as the origin, establish a center-blade polar coordinate system to uniformly describe the spatial pose of the three blades; The path mapping stage is based on this coordinate system, and the path of the first blade is offset by the azimuth angle to simultaneously generate the inspection paths of the other two blades. The method for generating the globally optimal inspection sequence employs a path simplification algorithm based on empty-run optimization, including: Discretize the blade surface into a grid of detection units to identify the effective detection area and the no-load area; Construct a graph theory-based path optimization model to transform the detection path planning into a constrained traveling salesman problem; A multi-objective optimization algorithm is used to solve for the optimal detection sequence. The objective function simultaneously considers detection coverage, path length, and empty driving time.

3. The method for local dynamic path planning of wind turbine blade inspection based on UAV according to claim 1, characterized in that, The online execution phase includes: S301. Based on real-time wind field perception data and wind turbine attitude, a model predictive control (MPC) framework is used for local rolling optimization. The optimization objectives include utilizing the downwind aerodynamic shadow area of ​​the blades. S302. When the airborne defect identification system detects a preset high-priority defect, it dynamically generates a local fine scanning path based on the online-tuned control parameters. The online execution phase also includes a dynamic obstacle avoidance step, an online task replanning step, and an image quality optimization step based on biomimetic learning. The dynamic obstacle avoidance steps include: Real-time monitoring of sudden dynamic obstacles within the patrol airspace; When a collision risk is detected, immediately plan an emergency escape corridor to deflect the drone off its original path; After the danger is cleared, a smooth path is calculated using the optimal reconnection algorithm, allowing the drone to reconnect to the original inspection path. The online task replanning steps include: Continuously assess the completion status of inspection tasks and the remaining system resources; When it is necessary to interrupt the inspection, an efficient convergence path is dynamically generated based on the remaining resources and area priority. The image quality optimization steps based on biomimetic learning include: An image quality prediction model is constructed, which predicts the blur and quality score of the image based on the real-time vibration state of the UAV, the relative speed with the blades, and the ambient light angle.

4. The method for local dynamic path planning of wind turbine blade inspection based on UAV as described in claim 1, characterized in that, The method for generating the globally optimal inspection sequence includes: Discretize the blade surface into a mesh and transform it into a graph theory problem; Using the shortest total flight time, lowest total energy consumption, and smoothest gimbal attitude change as the multi-objective optimization function, a heuristic search algorithm is used to find an optimal Hamiltonian path, and the starting point of this path is the globally optimal starting point. When multiple inspection drones exist within the system, a task negotiation mechanism based on distributed consensus is adopted, including: Task Issuance: The globally optimal inspection sequence is divided into multiple sub-tasks, and a task descriptor containing location, required resources, and priority is generated for each sub-task; Utility assessment and bidding: After receiving the task descriptor, each UAV calculates its expected utility value for performing the task based on its own local state information; Task decision-making: Based on the expected utility values ​​submitted by each drone, subtasks are assigned to the optimal drone through consensus rules; Commitment and Execution: The winning drone commits to the mission and updates its local mission queue, while the remaining drones are released to bid for other missions; The distributed consensus task negotiation mechanism also supports dynamic task insertion: When any UAV detects a high-priority defect, an emergency task descriptor containing the defect location and scanning requirements is immediately generated. Through the aforementioned negotiation mechanism, the most suitable drone within the system can quickly bid and execute the emergency mission.

5. The method for local dynamic path planning for wind turbine blade inspection based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The method for generating the globally optimal inspection sequence adopts a path simplification algorithm based on empty-run optimization, and the objective function is defined as: in To effectively detect path length, This is the length of the empty driving route. For the number of path turns, To detect attitude repositioning time. , , , These are the weighting coefficients; The formula for calculating the empty driving route length is: in, For the first The endpoint of each detection area, For the first The starting point of each detection area This is the obstacle avoidance correction function.

6. The method for local dynamic path planning for wind turbine blade inspection based on unmanned aerial vehicles (UAVs) according to claim 5, characterized in that, The path optimization model employs a hierarchical planning strategy: First layer: The detection area is divided into multiple detection sub-regions based on the geometric features of the blade, and each sub-region adopts a continuous scanning mode; The second layer: Optimize the connection order of sub-intervals and use a genetic algorithm to solve for the optimal access sequence; The third layer: An improved Hilbert curve filling algorithm is used within the sub-region to generate continuous, non-intersecting detection paths; The improved Hilbert curve filling algorithm includes: The fill density is adaptively adjusted based on the Gaussian curvature of the blade surface, with high-density fill used in high-curvature areas. Step size of the fill path Determined by the following formula in As the reference step size, For local Gaussian curvature, For the maximum Gaussian curvature, Curvature sensitivity coefficient (0 < <1).

7. A path planning system for implementing the UAV-based local dynamic path planning method for wind turbine blade inspection as described in any one of claims 1-6, characterized in that, It includes a digital template library module, a path mapping module, and an online execution module. The path mapping module includes an adaptive equidistant path generator, a digital twin management module, an emergency avoidance planner, a task replanning engine, a center positioning and synchronous planning module, a damage knowledge base module, and an image quality closed-loop optimizer.

8. A path planning system according to claim 7, characterized in that, The online execution module includes a pneumatic-motion coupled planner, a dynamic point of interest responder, a collaborative management module, and a reinforcement learning parameter tuning module.

9. A path planning system according to claim 8, characterized in that, The system also includes an empty-run optimization engine for executing a path simplification algorithm based on empty-run optimization, including a detection region segmentation unit, a path sequence optimization unit, and a path filling generation unit.

10. A path planning system according to claim 9, characterized in that, The reinforcement learning parameter tuning module dynamically optimizes the weight coefficients in the empty-running optimization algorithm using the Q-learning algorithm. , , , The reward function is defined as: in To improve detection coverage, Total inspection time. For energy consumption, , , These are the weighting coefficients of the reward function itself.

Images