An autonomous inspection method and inspection robot for the interior of a wind turbine blade

Abstract

The application discloses a kind of wind power generator blade internal autonomous inspection method and inspection robot, method is through the synergistic effect of the steps such as point navigation, contour extraction and obstacle detection, safety corridor calculation, double-section obstacle judgment, generating obstacle avoidance path and waypoint record original road return, realizes the autonomous navigation, safety obstacle avoidance and safe return of wind power blade internal inspection trolley in unknown long and narrow space, can effectively improve the automation degree, traffic safety, obstacle avoidance stability and task completion ability of blade internal inspection;Each step forms a complete blade internal automatic inspection closed loop, and the closed loop process enables the robot to actively advance and inspect in the blade, and also timely avoid obstacles or exit according to environmental changes, improving the reliability, continuity and safety of the entire inspection system.The configuration of the inspection robot can realize high-quality image acquisition and safe autonomous inspection in the complex environment of the blade.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine blade inspection technology, and more specifically, relates to an autonomous inspection method and inspection robot for the interior of wind turbine blades. Background Technology

[0002] During long-term service, wind turbine blades are subjected to various factors such as aerodynamic loads, gravitational loads, cyclic fatigue, humid and hot environments, and lightning strikes. The internal composite material structure is prone to defects such as cracks, delamination, porosity, adhesive failure, cable abnormalities, and protruding fasteners. If these defects are not detected in time, they may further lead to blade performance degradation, downtime for maintenance, or even safety accidents.

[0003] Current blade internal inspection still primarily relies on manual entry into the blade cavity for inspection. Due to the narrow and elongated interior space of the blade, poor lighting conditions, and the curvature of the ground with gradually decreasing height in certain areas, manual inspection faces problems such as high safety risks, low inspection efficiency, poor repeatability, and limited inspection depth. Personnel typically have difficulty reaching the deep, narrow areas of the blade, resulting in blind spots in the inspection process.

[0004] As the length of large wind turbine blades continues to increase, internal inspection tasks place higher demands on equipment: First, it needs a smaller external size and higher maneuverability to adapt to the constricted space deep within the blade; second, it needs to be equipped with high-resolution imaging equipment and stable lighting to obtain clear images that can be used for defect analysis; and third, it needs to achieve safe and stable autonomous navigation and return in a closed space where GNSS (Global Navigation Satellite System; the internal environment of the blade is a GNSS-unavailable environment), textures are scarce, lighting is insufficient, and geometry is repetitive.

[0005] like Figure 1As shown, the general structure of a wind turbine blade can be divided into two bonded aerospace shells and a shear web in the middle. The two aerospace shells constitute the surface structure of the blade. These two shells are fixed by the shear web. Common shear webs include three types: closed shell, box girder, and load-bearing shell. The first two types divide the cavity into two parts: the trailing edge cavity and the leading edge cavity. The last type divides it into three parts, adding a main beam cavity compared to the first two types. For inspection tasks, each cavity needs to be inspected. Therefore, for conventional manual inspection, these shear webs not only obstruct personnel movement, but also, because they divide the space into multiple cavities, require inspectors to make multiple trips to inspect a single blade. According to data provided by Kuobo Intelligent, it takes about 1 to 2 hours to manually inspect a blade, while using a robot for internal inspection only takes 15 to 30 minutes, greatly improving efficiency. Furthermore, the deeper one goes, the smaller the space becomes. For example, on a blade 61.5 meters long, at about two-thirds of its length, the width is about 3 meters and the thickness is about 0.54 meters. Therefore, manual inspection can only cover a maximum of two-thirds of the blade's area. However, using a robot for inspection can reach a depth of about three-quarters of the blade, easily reaching areas that are inaccessible to humans.

[0006] Existing technologies include UAV-based blade internal inspection solutions. These solutions typically employ small multi-rotor UAVs equipped with cameras, lighting modules, and ranging / positioning sensors to enter the blade's internal cavity. They capture internal images through hovering, forward flight, or fixed-point acquisition, and then combine this with 3D reconstruction, defect localization, or remote control to achieve the inspection function. The advantages of this approach are that it does not rely on ground contact, can traverse some small-scale ground obstacles, and can flexibly change its viewing angle. However, due to the narrow internal space of the blade, the flight platform requires high levels of control stability, obstacle avoidance accuracy, and collision resistance.

[0007] The disadvantages of this type of solution include: limited payload capacity, constrained by the size and endurance of the drone, making it difficult to equip it with high-resolution cameras, high-power flashes, and powerful computing units, resulting in limitations in shooting resolution, lighting quality, and continuous operation capability; insufficient safety, as the narrow internal space of the blades and significant changes in wall curvature make the drone prone to instability and crashing or equipment damage in the event of a collision; and insufficient long-distance deep detection capability, as access to narrow areas usually requires a reduction in the size of the drone, but this reduction further limits the configuration of batteries and sensors, making it difficult to balance depth, efficiency, and detection quality.

[0008] Existing technologies also include manual inspection or remote-controlled ground inspection equipment solutions. Manual inspection typically involves operators carrying flashlights, cameras, or simple inspection tools to walk into the blade's interior for inspection; remote-controlled ground platforms, on the other hand, use tracked or wheeled chassis to mount cameras, thermal imaging components, and other equipment, allowing for forward and backward movement and image capture under operator control. These solutions usually offer strong load capacity, are easy to mount inspection equipment, and some systems can even transmit images back to the operator via wired / wireless means.

[0009] The disadvantages of this type of solution include: manual methods pose significant safety risks, and the efficiency of the operation is greatly affected by the experience, physical strength and environmental conditions of the personnel, resulting in poor repeatability; the ability of manual personnel to enter deep areas is limited, and it is usually difficult to continue to go deeper after the blades gradually narrow, resulting in insufficient detection coverage; remote control platforms rely on continuous intervention by the operator, which is a heavy burden to operate in long-distance and narrow environments, and it is difficult to achieve stable and standardized inspections. Summary of the Invention

[0010] The purpose of this invention is to address the above-mentioned shortcomings by providing an autonomous inspection method and robot for the interior of wind turbine blades, thereby solving problems such as reliance on manual remote control, lack of stable autonomous navigation basis, easy early obstacle avoidance after misjudgment of obstacles, easy jamming of the vehicle when the internal space of the blade narrows, and inability to safely exit.

[0011] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0012] In a first aspect, the present invention provides an autonomous inspection method for the interior of wind turbine blades, applied to a mobile robot equipped with lidar, the method comprising:

[0013] During the robot's inspection inside the blade, the robot's waypoints are continuously recorded;

[0014] The robot acquires the lidar point cloud inside the blade in real time, extracts several sets of blade inner wall feature lines at different heights on the left and right sides of the robot, fits the vanishing point based on the intersection of the feature lines on both sides, and calculates the final vanishing point direction, using the final vanishing point direction as the basic navigation direction.

[0015] The cross-sectional point set of the internal space of the blade in front of the robot is formed by the point cloud of the lidar, the spatial contour of the blade in front is obtained, and it is determined whether the robot's passage requirements are met based on the blade spatial contour.

[0016] If the robot's passage requirements are not met, the robot will be controlled to return along the recorded waypoints.

[0017] If the robot can pass through, obstacles are detected within the spatial contour of the blade using a clustering method.

[0018] If no obstacle is detected, a safe passage is calculated based on the position of the fitted vanishing point, and the robot is controlled to move along the safe passage.

[0019] If an obstacle is detected, a dual-section judgment is adopted: at the first section where an obstacle is first detected, the obstacle point information is cached and the robot continues to move in the current direction, and obstacle information is continuously accumulated during the movement; when the robot reaches the second section corresponding to the widest part of the obstacle, it is determined whether there is a safe passage for the robot to pass through based on the accumulated obstacle point information and the blade spatial contour.

[0020] If a safe passage exists, generate a smooth obstacle avoidance path from the robot's current position to the entrance of the safe passage and control the robot to pass through the obstacle area along the path; otherwise, control the robot to return along the recorded waypoints.

[0021] Furthermore, the waypoints include the robot's planar position and heading information; the waypoints are recorded at preset distance intervals.

[0022] Furthermore, the extraction of several sets of blade inner wall feature lines at different heights on the left and right sides of the robot, fitting vanishing points based on the intersection positions of the feature lines on both sides, and fusing and calculating the final vanishing point direction includes:

[0023] Several sets of blade inner wall point clouds at different heights on the left and right sides are selected from the lidar point cloud, and straight line fitting is performed on them to obtain the left blade feature line and the right blade feature line. The intersection position of each set of left and right blade feature lines is calculated, and this intersection position is used as the vanishing point. Then, the coordinates of several sets of vanishing points are fused to obtain the final vanishing point coordinates. This final vanishing point represents the extension direction of the blade's internal space. Due to difficulties such as the blade inner wall and the relatively inclined driving surface, the direction of the final vanishing point will be further used to construct a safety passage. The safety passage will select a relatively flat surface near the direction of the final vanishing point and maintain a certain distance from the inner wall, allowing the robot to drive inside the safety passage and preventing collisions with walls or rollovers caused by driving on planes with high inclination angles, thus achieving normal inspection navigation.

[0024] Furthermore, the step of forming a set of cross-sectional points of the internal space of the blade in front of the robot based on the lidar point cloud to obtain the spatial contour of the blade in front includes:

[0025] Based on the laser radar point cloud, a cross section is set in front of the robot, and the cross section point cloud is projected onto the lateral-height plane to form a cross section point set of the internal space of the blade;

[0026] The boundary point selection, missing point completion, and smoothing processes are performed on the cross-sectional point set inside the blade to obtain the spatial contour of the forward blade.

[0027] Furthermore, the step of determining whether the robot passage requirements are met based on the spatial contour of the blade includes:

[0028] Set up a lateral passage assessment zone near the height of the robot body or the radar, and calculate the passable width within this height zone;

[0029] Based on the passable width and top clearance, determine whether the robot's passage requirements are met.

[0030] Furthermore, the step of detecting obstacles within the blade's spatial contour using a clustering method includes:

[0031] Point cloud points that may belong to obstacles are selected as candidate points within the spatial contour of the blade, and a clustering method is used to group the candidate points to obtain several clusters of points.

[0032] If a cluster of points meets the preset conditions of number of points and spatial size, then it is determined that there is a candidate region for an obstacle ahead.

[0033] Furthermore, the smooth obstacle avoidance path is fitted using a cubic smooth curve, and the starting and ending points are constrained. Its trajectory formula is as follows:

[0034] ;

[0035] Where the starting robot's head position coordinates are (0,0), and (x, y(x)) represent the robot's real-time position coordinates, (x... a y a The coordinates () represent the coordinates of the center of the destination, i.e., the entrance to the safety passage. A path curve is constructed using this formula. The robot travels along this curve to the center of the safety passage and then proceeds along the direction of the safety passage to complete obstacle avoidance.

[0036] Secondly, the present invention also provides an autonomous inspection robot for the interior of wind turbine blades, comprising:

[0037] The chassis is used to move through the curved space inside the blades and to carry other modules;

[0038] The image acquisition module includes a fixed camera mounted on the front of the chassis and a rotatable camera mounted on the rear of the chassis, both of which are equipped with lighting devices;

[0039] The lidar module, fixed to the chassis, is used to collect lidar point clouds inside the blades.

[0040] The control unit, connected to the chassis, image acquisition module, and lidar module, is configured to perform the autonomous inspection method as described in the first aspect.

[0041] Furthermore, the chassis can be a four-wheel or six-wheel independently driven chassis, or a six-wheel chassis, a wheeled chassis with independent suspension, a tracked chassis, or a wheel-tracked composite chassis.

[0042] Furthermore, the rotatable camera is mounted on the chassis via a rotating base, which includes a fixed base and a servo motor. The servo motor drives the rotatable camera to rotate 180 degrees to capture images of the top, left side, and right side of the blade.

[0043] Furthermore, the front and rear ends of the four-wheel drive chassis are equipped with auxiliary lighting modules, which are configured to turn on synchronously when the image acquisition module takes pictures and turn off synchronously when the pictures stop.

[0044] Furthermore, the lidar module is fixed to the chassis by four corner brackets to provide a forward detection field of view;

[0045] The chassis is also equipped with an external communication antenna to enhance the wireless communication link between the robot and the operator.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] The autonomous inspection method described in this invention achieves autonomous navigation, safe obstacle avoidance, and safe return for a wind turbine blade internal inspection vehicle in an unknown, narrow space through the coordinated action of steps such as vanishing point navigation, contour extraction and obstacle detection, safe corridor calculation, dual-section obstacle judgment, obstacle avoidance path generation, safe passage construction, and waypoint recording and return. This effectively improves the automation level, passage safety, obstacle avoidance stability, and task completion capability of blade internal inspection. By unifying normal navigation, space deficiency judgment, obstacle detection, obstacle avoidance, and return along the original path into the same process, a complete closed loop of automatic blade internal inspection can be formed. This closed loop process allows the vehicle to actively move forward for inspection inside the blade and to avoid obstacles or exit in a timely manner according to environmental changes, thereby improving the reliability, continuity, and safety of the entire inspection system.

[0048] The method described in this invention can extract feature lines of the blades on both sides of the robot and fit vanishing points. It can obtain the robot's forward navigation direction in the absence of GPS, pre-map, and artificial markers inside the blade, enabling the robot to move automatically along the internal space of the blade. Compared with the method of simply relying on manual remote control or preset path, this invention can adapt to different internal shapes of blades and different initial placement postures, improving the autonomy and environmental adaptability of the inspection robot in the operation inside the blade.

[0049] The method described in this invention can extract the spatial contour of the front blade and determine whether the robot has sufficient driving space based on the contour. This can reduce the risk of the robot getting stuck or colliding in narrow areas inside the blade. It can also take safety measures in advance when the blade space gradually narrows, the height is insufficient, or the local shape is not suitable for the robot to pass through, thereby improving the safety of the inspection process.

[0050] The method described in this invention detects obstacles through clustering, which can distinguish the point cloud of the blade wall from the point cloud of the real obstacle, reduce misjudgments caused by the inner wall of the blade, point cloud noise or local reflection anomalies, and make the obstacle detection results more stable and reliable.

[0051] The method described in this invention employs a dual-section judgment mechanism, which can avoid erroneous actions caused by the robot immediately avoiding obstacles when it detects them at a distance. In this way, the system can use the multi-frame point cloud information continuously acquired during the robot's movement to make a more complete judgment on obstacles, reducing misjudgments caused by sparse point clouds at a distance, occlusion of the viewpoint, instantaneous noise, or incomplete local shape of obstacles, thereby improving the accuracy of obstacle avoidance decisions.

[0052] The method described in this invention does not simply stop or detour after detecting an obstacle, but further determines whether there is a safe passage around the obstacle. This method enables the robot to continue to complete the inspection when it can pass through the obstacle and to exit in time when it cannot pass through the obstacle, thus balancing inspection efficiency and operational safety.

[0053] The method described in this invention constructs a vanishing point to determine the approximate direction of the robot's movement. Based on the information from contour fitting, it determines the inclination of the wall and the road surface and constructs a safe passage around the vanishing point, allowing the robot to travel within the safe passage. This improves navigation stability in obstacle-free conditions and reduces computational overhead. The method ensures that the robot's forward direction considers both the overall extension trend of the blades and the actual passable space within the current cross-section, avoiding the robot from approaching the wall due to local spatial offset or changes in the blade cross-section when simply moving towards the vanishing point. This improves the robot's centering and safety during normal navigation.

[0054] The inspection robot described in this invention effectively solves the problems of small load capacity, poor safety, low efficiency, and limited reach of existing technologies by integrating a chassis, image acquisition module, lidar module, and control unit. The front and rear fixed and rotatable cameras, combined with lighting equipment, enable multi-angle, high-definition imaging of the blade interior in a dark environment, meeting the detection requirements for small defects such as cracks and delamination. The lidar module provides reliable environmental perception capabilities for enclosed spaces with no GNSS, weak light, and sparse texture. The control unit executes an autonomous inspection method, enabling the robot to complete the entire process autonomously without continuous human control, significantly improving the automation and safety of blade interior inspection. Attached Figure Description

[0055] Figure 1 This diagram shows the structure of a wind turbine blade and the type of its web.

[0056] Figure 2 A flowchart of an autonomous inspection method for the interior of a wind turbine blade provided in Embodiment 1 of the present invention;

[0057] Figure 3 This is a schematic diagram of the vanishing point fitting for an autonomous inspection method for the interior of a wind turbine blade provided in Embodiment 1 of the present invention;

[0058] Figure 4 This is a schematic diagram of the safety channel fitting based on the vanishing point direction, which is part of the autonomous inspection method for the interior of a wind turbine blade provided in Embodiment 1 of the present invention.

[0059] Figure 5 This is a schematic diagram of the obstacle detection and avoidance method provided in Embodiment 1 of the present invention, which is a method for autonomous inspection of the interior of a wind turbine blade.

[0060] Figure 6 This is a schematic diagram of the structure of an autonomous inspection robot for the inside of a wind turbine blade provided in Embodiment 2 of the present invention;

[0061] Figure 7 This is a front view of an autonomous inspection robot for the interior of a wind turbine blade, provided in Embodiment 2 of the present invention.

[0062] In the diagram: 1. Chassis; 2. Fixed camera; 3. Rotatable camera; 4. Front fill light; 5. Rear fill light; 6. LiDAR; 7. Mini fill light; 8. External communication antenna; 9. Rotating base. Detailed Implementation

[0063] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0064] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0065] Example 1

[0066] This embodiment provides an autonomous inspection method for the interior of wind turbine blades. This method is applied to the Kiwi (Kart Inside Windblade Inspection) automatic inspection robot and implemented through the BIN (Blade Internal Navigation) automatic inspection algorithm. The inspection robot comprises four parts: a four-wheel drive chassis, an image acquisition module, a LiDAR module, and a control unit. The BIN algorithm is executed by the control unit. The BIN algorithm includes two core modules: LVPN (LiDAR Vanishing Point Navigation) and SOD (Space and Obstacle Detection).

[0067] The BIN algorithm begins when the robot is placed at the blade inlet, activating the two core modules, LVPN and SOD, and continuously recording waypoints throughout the inspection process. The LVPN module outputs vanishing point navigation information, while the SOD module outputs the spatial contour of the blade ahead and obstacle detection information. The system first determines whether the contour space is sufficient for passage; if the space is insufficient, it returns along the recorded waypoints; if the space is sufficient, it continues to determine whether an obstacle has been detected. If no obstacle is detected, the system uses the safety corridor module to calculate the orientation based on the contour and vanishing points, enabling the robot to navigate normally; if an obstacle is detected, the system enters a dual-section judgment mechanism and decides whether to initiate an obstacle avoidance procedure or return along the original path based on whether a safe passage can be found.

[0068] like Figure 2 As shown, the specific steps of the autonomous inspection method for the interior of wind turbine blades described in this embodiment are as follows:

[0069] Step S1: Robot Placement and System Startup

[0070] First, place the robot at the inlet of the wind turbine blade, ensuring its forward direction is towards the inside of the blade. After verifying that everything is correct, start the robot. It first uses the lidar module to collect lidar point cloud data inside the blade, while the local computing unit simultaneously starts the LVPN module, SOD module, navigation control module, and trajectory recording module.

[0071] Step S2: Continuously record waypoints

[0072] From the moment the robot begins operation, the trajectory recording module continuously records waypoints in the odometry coordinate system. Each waypoint includes the robot's planar position and heading information. Recording can be performed at preset intervals to avoid generating too many redundant waypoints when the robot is stationary or vibrating at low speed. The recorded waypoints are used for subsequent return along the same route.

[0073] Step S3: LVPN extracts feature lines of the blades on both sides.

[0074] The LVPN module selects several sets of blade inner wall point clouds at different heights on the left and right sides of the robot from the LiDAR point cloud, and performs linear fitting on these sets of point clouds on both sides to obtain the feature lines of the left and right blades. These feature lines are used to describe the structural trend of the blade's internal space along its length.

[0075] Step S4: LVPN fitting vanishing point

[0076] RVPN calculates the intersection of the left and right blade feature lines in each point cloud and uses this intersection as the vanishing point. Then, it performs weighted fusion on these n vanishing point coordinates to obtain the final vanishing point coordinates. This final vanishing point represents the extension direction of the blade's internal space. The navigation control module can calculate the basic navigation orientation based on the position of the final vanishing point relative to the robot's current coordinate system.

[0077] Step S5: SOD extracts the spatial contour of the front blade.

[0078] The SOD module sets a cross-section in front of the robot based on the LiDAR point cloud and projects the cross-section point cloud onto the lateral-height plane to form a cross-sectional point set of the blade's internal space. Subsequently, the SOD module performs boundary point selection, missing point completion, and smoothing on the cross-sectional point set to obtain the blade's spatial contour. This contour represents the boundary of the passable space inside the blade at a certain distance in front of the robot.

[0079] Step S6: Determine if the spatial outline is sufficient for passage.

[0080] The system determines whether the robot has sufficient driving space based on the blade spatial profile output by SOD. During this determination, a lateral passage assessment zone can be set up near the robot's body height or radar height, and the passable width within this height zone can be calculated. Simultaneously, it can also determine whether the overhead clearance meets the robot's passage requirements. If the profile space is insufficient for the robot to drive safely, the navigation control module immediately triggers a return along the original route.

[0081] Step S7: SOD clustering to detect obstacles

[0082] The SOD module filters point cloud points within the blade's spatial contour as candidate points, and uses a clustering method to group these candidate points into several clusters. If a cluster meets the preset point count and spatial size conditions, it is determined that there is a candidate obstacle region ahead.

[0083] This embodiment employs a clustering method that combines KD-tree (a multidimensional spatial index structure used to accelerate point cloud nearest neighbor search) and DBSCAN (a density-based clustering algorithm used for clustering obstacle candidate points).

[0084] Step S8: Normal navigation when no obstacle is detected

[0085] If the contour space is sufficient and SOD does not detect any obstacles, the system enters the normal navigation process. The safety passage module first constructs a rectangular safety passage based on the vanishing point direction fitted by LVPN. Then, combining the contour information obtained by SOD, it determines whether the current safety passage maintains a certain safe distance from the wall within the contour, and also determines the inclination of the safety passage. If it is too close to the wall or the inclination is too high, a new safety passage will be searched. Specifically, new safety passage center points are generated at equal intervals near the vanishing point. It is then determined whether the safety passages generated by these center points meet the requirements. If multiple safety passages meet the requirements, the safety passage closest to the original vanishing point is selected as the final safety passage. Subsequently, the navigation control module controls the robot to move along the passage, achieving automatic navigation in obstacle-free conditions.

[0086] Step S9: When an obstacle is detected, enter the dual-section judgment mechanism.

[0087] If the SOD detects an obstacle, the system does not immediately perform obstacle avoidance. Instead, it enters a dual-section judgment mechanism: when the obstacle is first detected in the first section, the system only caches the obstacle point information and keeps the robot moving in the normal navigation direction. This avoids premature obstacle avoidance caused by sparse point clouds at long distances, occlusion, or momentary false detections.

[0088] Step S10: Continuously detect and accumulate obstacle point information

[0089] As the robot continues to move forward, the SOD module continuously detects obstacles and accumulates the obstacle point information detected at different times. The accumulated obstacle point information is used to form a more complete and stable obstacle spatial distribution.

[0090] Step S11: Determining the passage after the second section reaches the widest obstacle position.

[0091] Once the second section reaches the widest obstacle position, the system re-evaluates the obstacle-occupied area based on the previously accumulated obstacle point information and calculates whether a safe passage for the robot exists, taking into account the blade's spatial profile. This safe passage must at least meet the robot's width, height, and safety margin requirements.

[0092] Step S12: If a safe passage cannot be found, return via the original route.

[0093] If a safe passage cannot be found based on the accumulated obstacle point information, it means the robot cannot safely pass through the obstacle area. At this time, the navigation control module stops moving forward and calls the waypoint sequence in the trajectory recording module, causing the robot to return along the recorded waypoints until it returns to the vicinity of the entrance or completes the return journey.

[0094] Step S13: Activate obstacle avoidance procedure when a safe passage can be found.

[0095] If a safe passage is found, the system initiates the obstacle avoidance procedure. The navigation control module generates a smooth obstacle avoidance path in the current vehicle coordinate system, and the obstacle avoidance trajectory is fitted using a cubic smooth curve. The trajectory formula is as follows:

[0096] The starting point, the center of the robot's front, is (0,0), and the ending point, the center of the safe passage that allows passage over obstacles, is (x,0). a y a (x, y(x)) represents the robot's real-time position coordinates, and a path curve can be obtained through this formula.

[0097] Step S14: Reach the entrance to the safety passage and pass through the obstacle area.

[0098] The navigation control module guides the robot to follow the aforementioned cubic curve path. Upon reaching the entrance to the safety passage, the robot continues forward along the passage, directly traversing the obstacle area. After passing through the obstacle area, the system clears the current obstacle avoidance status and resumes normal navigation.

[0099] Step S15: Execute the inspection process repeatedly.

[0100] After normal navigation is restored, the system continues to perform steps such as LVPN vanishing point navigation, SOD contour extraction, obstacle detection, safe corridor calculation, and waypoint recording. If insufficient space or impassable obstacles occur again, the system will trigger a return along the original route.

[0101] The autonomous inspection method provided in this embodiment can solve the problems of existing blade internal inspection processes, such as reliance on manual remote control, lack of stable autonomous navigation data, easy premature obstacle avoidance after misjudging obstacles, robot jamming when the internal space of the blade narrows, and inability to safely exit. Its specific beneficial effects are as follows:

[0102] (1) such as Figure 3 As shown, by using the LVPN module to extract feature lines from the blades on both sides of the robot and fitting vanishing points, the robot's forward navigation direction can be obtained even without GPS, pre-maps, or human markers inside the blades. Specifically, through the LVPN steps of extracting feature lines from the blades on both sides and fitting vanishing points, the system can calculate the extension direction of the internal space of the blade in real time based on the geometric structure of the point cloud on the inner wall of the blade, enabling the robot to move automatically along the internal space of the blade. Compared with methods that rely solely on manual remote control or preset paths, this invention can adapt to different internal shapes of blades and different initial placement postures, improving the autonomy and environmental adaptability of the inspection robot operating inside the blades.

[0103] (2) By using the SOD module to extract the spatial contour of the preceding blade and determining whether the robot has sufficient driving space based on the contour, the risk of the robot getting stuck or colliding in narrow areas inside the blade can be reduced. Specifically, through the steps of extracting the spatial contour of the preceding blade using SOD and determining whether the contour space is sufficient for passage, the system does not judge the passage status based solely on a single point or local distance, but rather assesses the passable width and height based on the overall cross-sectional contour of the preceding blade. When the contour space is insufficient, the robot will not continue to penetrate into the blade but will directly trigger a return along the original path. Therefore, this invention can take safety measures in advance when the blade space gradually narrows, the height is insufficient, or the local shape is unsuitable for the robot to pass through, thereby improving the safety of the inspection process.

[0104] (3) such as Figure 5 As shown, the SOD module is used to detect obstacles within the forward spatial contour using a clustering method, which can distinguish the point cloud of the blade wall from the point cloud of the actual obstacle. Specifically, through the obstacle detection clustering step, the system first filters candidate obstacle points within the blade spatial contour, and then performs clustering processing on the candidate points to obtain the spatial distribution range of the obstacles. This method can reduce false detections caused by blade inner wall noise, point cloud noise, or local reflection anomalies, making the obstacle detection results more stable and reliable.

[0105] (4) The dual-section judgment mechanism can avoid erroneous actions caused by the robot immediately avoiding obstacles when it detects them at a distance. Specifically, when an obstacle is first detected at the first section, the system does not start the obstacle avoidance procedure, but continues to maintain normal navigation and continuously accumulate obstacle point information; when the second section reaches the position of the widest obstacle, it then judges whether there is a safe passage based on the accumulated obstacle point information. In this way, the system can make a more complete judgment on obstacles by using the multi-frame point cloud information continuously acquired during the robot's movement, reducing misjudgments caused by sparse point clouds at a distance, occlusion of the view, instantaneous noise, or incomplete local shape of the obstacle, thereby improving the accuracy of obstacle avoidance decisions.

[0106] (5) Upon detecting an obstacle, instead of simply stopping or detouring, the system further determines whether a safe passage exists around the obstacle. Specifically, by determining whether a safe passage can be found, the system combines the blade spatial contour and accumulated obstacle point information to calculate whether there is a passable space within the obstacle area that meets the robot's width, height, and safety margin requirements. If a safe passage exists, obstacle avoidance is initiated; if no safe passage exists, the robot returns along the original path. This method enables the robot to continue inspection when it can pass through an obstacle and to exit promptly when it cannot pass through an obstacle, balancing inspection efficiency and operational safety.

[0107] (6) Utilizing the safety corridor module in conjunction with contour and vanishing point calculations to determine the robot's orientation can improve navigation stability in obstacle-free conditions. Specifically, for example... Figure 4 As shown, when no obstacle is detected, the safety corridor module uses the forward contour information obtained from SOD to find a suitable spatial passage for the robot to pass through within the contour, and calculates the target orientation together with the vanishing point position fitted by LVPN. Therefore, the robot's forward direction considers both the overall extension trend of the blades and the actual passable space within the current cross-section, avoiding the robot approaching the wall due to local spatial offset or changes in the blade cross-section when simply moving towards the vanishing point, thus improving the robot's centering and safety during normal navigation.

[0108] In some other embodiments, the straight line fitting of the point clouds on both sides in LVPN can be done by total least squares, RANSAC straight line fitting, spline curve fitting or piecewise straight line fitting, etc. As long as it can characterize the convergence trend of the two sides of the blade and obtain the direction guidance point, it is an alternative implementation that can be covered by the present invention.

[0109] In some other embodiments, the vanishing points of multiple height planes can be used to generate the final target direction by empirical weighting, confidence weighting, adaptive weighting, Kalman filter fusion, or robust statistical fusion.

[0110] In some other embodiments, candidate point clustering can be replaced by DBSCAN with Euclidean clustering, mean shift, voxel / raster-based clustering, or graph clustering; nearest neighbor search can also be replaced by KD-tree (a data structure used to accelerate nearest neighbor search) with ball tree, voxel hash, or other efficient spatial indexing structures.

[0111] In some other embodiments, the contour may be represented as a discrete point set, B-spline curve, Bezier curve, spline envelope, polygonal boundary, or deep learning segmentation contour.

[0112] Example 2

[0113] This embodiment provides an autonomous inspection robot for the interior of wind turbine blades, such as... Figure 6 and Figure 7 As shown, it includes:

[0114] A four-wheel independent drive chassis 1, wherein the chassis 1 is equipped with four wheels, each of which is equipped with an independent drive motor;

[0115] The image acquisition module includes a fixed camera 2 installed at the front of the chassis 1 and a rotatable camera 3 installed at the rear of the chassis 1. Both the fixed camera 2 and the rotatable camera 3 are equipped with lighting devices.

[0116] The lidar module 6 is fixed on the chassis 1 and is used to collect lidar point clouds inside the blades.

[0117] The control unit, connected to the four-wheel independent drive chassis, image acquisition module, and lidar module 6, is configured to execute the autonomous inspection method as described in Example 1.

[0118] In this embodiment, considering the comprehensive requirements of robot accessibility, stability, and load capacity for blade interior inspection tasks, a four-wheel independent drive chassis 1 structure is adopted. The overall dimensions of the robot are approximately 45cm in length, 23cm in width, and 25cm in height. The four-point support structure provides good posture stability and anti-rollover capability on the curved surface of the blade cavity, while providing ample space for the image acquisition module, LiDAR 6 module, and control unit. The four wheels are driven by four servo motors, and the control core uses an STM32F407VET6 microcontroller as the main control chip to control the drive of the four motors. Wheel speed information is acquired in real time through motor encoders, and the robot's motion state is estimated accordingly. Based on this, a PID control algorithm is used to adjust the motor output online to reduce speed fluctuations that may occur under load changes, ground friction differences, and low-speed conditions, thereby improving the robot's speed tracking accuracy and motion repeatability during the slow inspection process inside the blade.

[0119] In some other embodiments, the four-wheel chassis can be replaced by a six-wheel chassis, a wheeled chassis with independent suspension, a tracked chassis, or a wheel-tracked composite chassis. As long as it can stably pass through the curved space inside the blade and carry the testing equipment, the same or similar inventive objectives can be achieved.

[0120] In this embodiment, to achieve autonomous navigation and forward environmental perception within the narrow space inside the blade, the inspection robot integrates a miniaturized radar module to acquire geometric information of the space in front of the robot. Since the interior of the blade lacks stable lighting and the visual texture may be sparse, the radar can provide reliable point cloud information without relying on ambient light. The data collected by the lidar 6 is connected via cable to the control board (control unit) located inside the chassis 1. In this embodiment, a Raspberry Pi 4B is used as the control board to process the point cloud data and execute the navigation algorithm BIN described in Embodiment 1. The entire lidar 6 is fixed to the chassis 1 by a four-corner bracket, ensuring a good detection field of view.

[0121] In this embodiment, the image acquisition module adopts a combination of fixed front-end imaging and steerable rear-end imaging. The fixed camera 2 is fixedly mounted at the front of the chassis 1, capturing an overall image of the robot's forward direction, including the space in front and the ground area. The rotatable camera 3 is mounted at the rear of the chassis 1 via a rotating base 9, facing the top and side surfaces of the blade. The rotating base 9 consists of a fixed base and a servo mechanism. The servo mechanism allows the rotatable camera 3 to rotate freely 180 degrees, enabling it to align with the target area from different angles, thus achieving multi-directional imaging of the blade's top, left side, and right side walls. This design expands the imaging coverage without significantly increasing the platform's volume, achieving full-angle imaging of the wind turbine blade's interior.

[0122] To address the typical working condition of no light inside the blades, two sets of mini flashlights 7 are arranged at each of the front and rear ends of the chassis 1 as auxiliary lighting devices. During inspection tasks, these lights are activated simultaneously when the image acquisition module takes pictures, providing auxiliary light; they are also deactivated simultaneously when the imaging work stops, saving power. Meanwhile, to address potential wireless signal attenuation or obstruction issues in enclosed spaces, an external communication antenna 8 is installed at the rear of the chassis 1 to enhance the wireless communication link between the robot and the operator, improving connection reliability and control availability in complex, enclosed environments.

[0123] One of the core objectives of blade internal inspection is to acquire high-quality images that can support defect analysis. Compared to the scale of the robot, the blade tip is relatively large, while defects such as cracks and delamination often have small-scale characteristics. If ordinary industrial cameras or conventional resolution cameras are used, it is difficult to capture them clearly even under ideal lighting conditions, thus affecting the reliability of defect interpretation and subsequent quantitative analysis. Based on the above imaging requirements, this embodiment uses two high-resolution cameras with an effective pixel count of approximately 26 MP, which is far higher than that of ordinary industrial cameras and can capture defect images with sufficient resolution. Secondly, since the interior of the blade is a dark or low-light environment, in addition to the auxiliary mini supplementary lights 7 at the front and rear ends of the chassis 1, this embodiment also equips each camera with a large flash to provide instantaneous high-intensity illumination, including a front supplementary light 4 and a rear supplementary light 5.

[0124] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and principles of the described embodiments, and these modifications and variations should also be considered within the scope of protection of the present invention.

Claims

1. A method for autonomous inspection of the interior of a wind turbine blade, characterized in that, The method, applied to a mobile robot equipped with a lidar, includes: During the robot's inspection inside the blade, the robot's waypoints are continuously recorded; The robot acquires the lidar point cloud inside the blade in real time, extracts several sets of blade inner wall feature lines at different heights on the left and right sides of the robot, fits the vanishing point based on the intersection of the feature lines on both sides, and calculates the final vanishing point direction, using the final vanishing point direction as the basic navigation direction. The cross-sectional point set of the internal space of the blade in front of the robot is formed by the point cloud of the lidar, the spatial contour of the blade in front is obtained, and it is determined whether the robot's passage requirements are met based on the blade spatial contour. If the robot's passage requirements are not met, the robot will be controlled to return along the recorded waypoints. If the robot can pass through, obstacles are detected within the spatial contour of the blade using a clustering method. If no obstacle is detected, a safe passage is calculated based on the position of the fitted vanishing point, and the robot is controlled to move along the safe passage. If an obstacle is detected, a dual-section judgment is adopted: at the first section where an obstacle is first detected, the obstacle point information is cached and the robot continues to move in the current direction, and obstacle information is continuously accumulated during the movement; when the robot reaches the second section corresponding to the widest part of the obstacle, it is determined whether there is a safe passage for the robot to pass through based on the accumulated obstacle point information and the blade spatial contour. If a safe passage exists, generate a smooth obstacle avoidance path from the robot's current position to the entrance of the safe passage and control the robot to pass through the obstacle area along the path; otherwise, control the robot to return along the recorded waypoints.

2. The method for autonomous inspection of the interior of wind turbine blades according to claim 1, characterized in that, The waypoints include the robot's planar position and heading information; the waypoints are recorded at preset distance intervals.

3. The method for autonomous inspection of the interior of wind turbine blades according to claim 1, characterized in that, The extraction robot extracts several sets of blade inner wall feature lines at different heights on both sides, fits vanishing points based on the intersection of the feature lines on both sides, and calculates the final vanishing point direction, including: Several sets of blade inner wall point clouds at different heights on the left and right sides are selected from the lidar point cloud, and straight lines are fitted to them respectively to obtain the left blade feature line and the right blade feature line; the intersection position of each set of left blade feature lines and right blade feature lines is calculated, and the intersection position is taken as the vanishing point. Then, the coordinates of several sets of vanishing points are fused to obtain the final vanishing point coordinates, which represent the extension direction of the blade's internal space.

4. The method for autonomous inspection of the interior of wind turbine blades according to claim 1, characterized in that, The step of forming a set of cross-sectional points of the internal space of the blade in front of the robot based on the lidar point cloud to obtain the spatial contour of the blade in front includes: Based on the laser radar point cloud, a cross section is set in front of the robot, and the cross section point cloud is projected onto the lateral-height plane to form a cross section point set of the internal space of the blade; The boundary point selection, missing point completion, and smoothing processes are performed on the cross-sectional point set inside the blade to obtain the spatial contour of the forward blade.

5. The method for autonomous inspection of the interior of wind turbine blades according to claim 1, characterized in that, The step of detecting obstacles within the spatial contour of the blade using a clustering method includes: Point cloud points that may belong to obstacles are selected as candidate points within the spatial contour of the blade, and a clustering method is used to group the candidate points to obtain several clusters of points. If a cluster of points meets the preset conditions of number of points and spatial size, then it is determined that there is a candidate region for an obstacle ahead.

6. The method for autonomous inspection of the interior of wind turbine blades according to claim 1, characterized in that, The smooth obstacle avoidance path is fitted using a cubic smooth curve, with constraints on the start and end points. Its trajectory formula is as follows: ； Where the starting robot's head position coordinates are (0,0), and (x, y(x)) represent the robot's real-time position coordinates, (x... a y a () indicates the coordinates of the endpoint, i.e., the center of the safety passage entrance.

7. An autonomous inspection robot for the interior of wind turbine blades, characterized in that, include: The chassis is used to move through the curved space inside the blades and to carry other modules; The image acquisition module includes a fixed camera mounted on the front of the chassis and a rotatable camera mounted on the rear of the chassis, both of which are equipped with lighting devices; The lidar module, fixed to the chassis, is used to collect lidar point clouds inside the blades. The control unit, connected to the chassis, image acquisition module, and lidar module, is configured to execute the autonomous inspection method as described in any one of claims 1 to 6.

8. The autonomous inspection robot for the interior of wind turbine blades according to claim 7, characterized in that, The rotatable camera is mounted on the chassis via a rotating base, which includes a fixed base and a servo motor. The servo motor drives the rotatable camera to rotate 180 degrees to capture images of the top, left side, and right side of the blade.

9. The autonomous inspection robot for the interior of wind turbine blades according to claim 7, characterized in that, The front and rear ends of the four-wheel drive chassis are also equipped with auxiliary lighting modules, which are configured to turn on synchronously when the image acquisition module takes pictures and turn off synchronously when the pictures stop.

10. The autonomous inspection robot for the interior of wind turbine blades according to claim 7, characterized in that, The lidar module is fixed to the chassis by four corner brackets to provide a forward detection field of view; The chassis is also equipped with an external communication antenna to enhance the wireless communication link between the robot and the operator.

Images