Wind generating set and variable pitch bearing crack detection system thereof

Abstract

The variable-pitch bearing crack detection system comprises an image acquisition module and a crack detection module, the image acquisition module is installed on the inner wall of a hub of the wind generating set, the image acquisition module can be positioned to the multi-angle shooting position of each variable-pitch bearing, and the crack detection module is installed on the inner wall of the hub of the wind generating set. Multi-angle images of the corresponding variable-pitch bearing are collected at the shooting position; and the crack detection module inputs the image into the crack identification model to obtain crack information of the corresponding variable-pitch bearing in the image.

Description

Technical Field

[0001] The present disclosure generally relates to the field of wind power generation technology, and more specifically, to a wind turbine generator set and a pitch bearing crack detection system thereof. Background Art

[0002] Wind turbines are an important part of the renewable energy sector, and their efficient and stable operation is crucial to ensuring energy supply and reducing negative impacts on the environment. As one of the key components in wind turbines, pitch bearings are responsible for adjusting the blade angles to adapt to different wind speed conditions, thereby optimizing power generation efficiency. However, pitch bearings may develop cracks due to fatigue, wear, corrosion and other factors during long-term operation. If these cracks are not discovered and handled in time, they may cause serious accidents. Therefore, the development of efficient and accurate pitch bearing crack detection technology is of great significance to ensure the safe operation of wind turbines.

[0003] With the development of robotics and computer vision technology, visual inspection systems can now be used for crack detection, that is, using industrial cameras and image processing algorithms to detect cracks on the bearing surface. However, the visual inspection system is highly dependent on lighting conditions and image quality, which may affect the accuracy of the detection. At present, camera equipment is often set at a fixed position on the inner wall of the hub to collect images to monitor the variable pitch bearing. Due to the inflexibility of the fixed position and the close distance to the running parts, blind spots are prone to occur and unclear images are captured, resulting in the inability to accurately detect cracks on the bearing surface. Summary of the invention

[0004] The embodiments of the present disclosure provide a wind turbine generator set and a pitch bearing crack detection system thereof, which can effectively solve the problem that the prior art cannot accurately detect cracks in the bearings.

[0005] In a general aspect, a pitch bearing crack detection system for a wind turbine generator set is provided, comprising an image acquisition module and a crack detection module. The image acquisition module is mounted on the inner wall of a hub of the wind turbine generator set. The image acquisition module can locate a multi-angle shooting position for each pitch bearing and capture multi-angle images of the corresponding pitch bearing at the shooting position. The crack detection module inputs the image into a crack recognition model to obtain crack information of the corresponding pitch bearing in the image.

[0006] Optionally, the image acquisition module includes a base, a mechanical arm, a camera head and a camera device. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set through the base. The mechanical arm includes a first mechanical arm and a second mechanical arm that are rotatably connected. The other side of the first mechanical arm is connected to the base through a second rotating shaft, and the other side of the second mechanical arm is connected to the camera head through a third rotating shaft. The base can rotate along the first rotating shaft, wherein the first rotating shaft is perpendicular to the inner wall of the hub; the first mechanical arm can rotate along the second rotating shaft, wherein the second rotating shaft is parallel to the inner wall of the hub; the camera head can rotate along the third rotating shaft and support the camera device to capture multi-angle images of the corresponding variable pitch bearing at the shooting position, wherein the third rotating shaft is perpendicular to the cross-section of the mechanical arm.

[0007] Optionally, the image acquisition module includes a base, a robotic arm, a camera head and a camera device. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set through the base. One side of the robotic arm is connected to the base through a fifth rotating shaft, and the other side is connected to the camera head through a sixth rotating shaft. The base can rotate along a fourth rotating shaft, wherein the fourth rotating shaft is perpendicular to the inner wall of the hub; the robotic arm can rotate along a fifth rotating shaft, wherein the fifth rotating shaft is parallel to the inner wall of the hub; the camera head can rotate along the sixth rotating shaft and support the camera device to capture multi-angle images of the corresponding variable pitch bearing at the shooting position, wherein the sixth rotating shaft is parallel to the cross-section of the robotic arm.

[0008] Optionally, the image acquisition module also includes a bracket, through which the base is mounted on the inner wall of the hub of the wind turbine generator set, and the bracket can be deformed to adjust the distance between the base and the inner wall of the hub; wherein the minimum distance between the base and the inner wall of the hub is zero, and the maximum distance between the base and the inner wall of the hub is the maximum height of the bracket.

[0009] Optionally, the robotic arm can be telescopic to adjust the arm length of the robotic arm.

[0010] Optionally, the image acquisition module is installed at a predetermined position on the inner wall of the hub of the wind turbine generator set, wherein the predetermined position is equidistant from all pitch bearings.

[0011] Optionally, the crack recognition model includes a target detector and a classifier, wherein the crack detection module inputs the image into the crack recognition model to obtain crack information of the corresponding variable pitch bearing in the image, including: inputting the image into the target detector to determine the position of the main object in the image, wherein the object includes cracks, bolts and numbers; based on the position of the main object in the image, using an anchor frame to mark the main object in the image; inputting the main object marked by the anchor frame into the classifier to obtain crack information.

[0012] Optionally, the crack recognition model is trained in the following manner: obtaining a first training sample set, wherein the first training sample set includes multiple variable pitch bearing images; performing enhancement processing on the multiple variable pitch bearing images respectively to obtain multiple enhanced images; adding the multiple enhanced images to the first training sample set to obtain a second training sample set; and training the crack recognition model based on the second training sample set.

[0013] Optionally, the above system also includes an image preprocessing module and a data processing module, the image preprocessing module acquires the image acquired by the image acquisition module and preprocesses the image; the crack detection module inputs the preprocessed image into a crack recognition model to obtain crack information of the corresponding variable pitch bearing in the image; the data processing module acquires crack information from the crack detection module, acquires the size and position of the crack based on the crack information, and determines maintenance recommendations based on the size and position of the crack.

[0014] In another general aspect, a wind turbine generator set is provided, comprising the pitch bearing crack detection system of the wind turbine generator set as described above.

[0015] According to the wind turbine generator set and its variable pitch bearing crack detection system of the embodiment of the present disclosure, the system includes an image acquisition module and a crack detection module, wherein the image acquisition device is installed on the inner wall of the hub of the wind turbine generator set, and can be positioned to the multi-angle shooting position of each variable pitch bearing, so that the image acquisition device is no longer fixed in a fixed position, and can be flexibly positioned to the multi-angle shooting position of each variable pitch bearing, thereby avoiding the problem of blind spots in the field of vision and unclear shooting. At the same time, the present disclosure also combines machine learning algorithms to identify and locate cracks, which can effectively reduce missed detection and false detection phenomena, and significantly improve the accuracy and reliability of crack detection. Therefore, through the present disclosure, the problem that the existing technology cannot accurately detect cracks in bearings can be effectively solved.

[0016] Additional aspects and / or advantages of the present general inventive concept will be set forth in part in the following description and in part will be apparent from the description or may be learned through practice of the present general inventive concept. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects and features of the embodiments of the present disclosure will become more apparent through the following description in conjunction with the accompanying drawings showing the embodiments, in which:

[0018] Figure 1 is a diagram showing the installation position of a related art camera device;

[0019] Figure 2 A pitch bearing crack detection system for a wind turbine generator set according to an embodiment of the present disclosure is shown;

[0020] Figure 3 is a schematic diagram showing the transmission of a collected image according to an embodiment of the present disclosure;

[0021] Figure 4 is a schematic diagram showing the installation position of the image acquisition module of an embodiment of the present disclosure;

[0022] Figure 5 is a schematic diagram showing a structure of an image acquisition module according to an embodiment of the present disclosure;

[0023] Figure 6 is a schematic diagram showing another structure of an image acquisition module according to an embodiment of the present disclosure;

[0024] Figure 7 is a schematic structural diagram showing a bracket according to an embodiment of the present disclosure;

[0025] Figure 8 is a schematic diagram showing model training and online deployment of an embodiment of the present disclosure;

[0026] Fig. 9 is a schematic diagram showing model training and prediction according to an embodiment of the present disclosure;

[0027] Fig.10 It is a schematic diagram of an image showing a pitch bearing with crack information marked thereon according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] The following specific embodiments are provided to help the reader obtain a comprehensive understanding of the methods, devices and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the methods, devices and / or systems described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0029] The features described herein can be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein, which will be clear after understanding the disclosure of the present application.

[0030] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more.

[0031] Although terms such as "first", "second", and "third" may be used herein to describe various members, components, regions, layers, or portions, these members, components, regions, layers, or portions should not be limited by these terms. Instead, these terms are only used to distinguish one member, component, region, layer, or portion from another member, component, region, layer, or portion. Therefore, without departing from the teachings of the examples described herein, the first member, first component, first region, first layer, or first portion referred to in the examples may also be referred to as the second member, second component, second region, second layer, or second portion.

[0032] In the specification, when an element (such as a layer, a region, or a substrate) is described as being “on”, “connected to”, or “coupled to” another element, the element may be directly “on”, “connected to”, or “coupled to” another element, or one or more other elements may be present therebetween. Conversely, when an element is described as being “directly on”, “directly connected to”, or “directly coupled to” another element, there may be no other elements present therebetween.

[0033] The terms used herein are only used to describe various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. The terms "comprise", "include" and "have" indicate the presence of the described features, quantities, operations, components, elements and / or combinations thereof, but do not exclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0034] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by a person of ordinary skill in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0035] Furthermore, in the description of examples, when it is considered that a detailed description of a well-known related structure or function would cause vague interpretation of the present disclosure, such a detailed description will be omitted.

[0036] Wind Turbine Generator System: A device that converts wind energy into electrical energy, generally including a nacelle, blades, a pitch system, a generator, a tower, a foundation and other components.

[0037] Pitch Bearing: A part of a wind turbine that is responsible for adjusting the angle of the blades to optimize wind energy capture and control the rotation speed to ensure the operating efficiency and safety of the wind turbine.

[0038] At present, the pitch bearing crack detection technologies include the following:

[0039] 1. Ultrasonic testing technology: This technology detects cracks by emitting high-frequency sound waves and receiving their reflected and scattered signals inside the material; ultrasonic testing has high resolution and high sensitivity, and can detect tiny internal defects. Moreover, ultrasonic testing technology can provide detailed information about the size and shape of the crack, which helps to assess the severity of the crack. Although this technology is very effective in detecting internal defects, it usually requires good acoustic contact with the object being tested, which may be difficult to achieve in actual operation; moreover, ultrasonic testing equipment is usually large in size, not suitable for mobile testing, and requires high skills of the operator.

[0040] 2. Magnetic particle testing technology: This technology uses the phenomenon that magnetic powder gathers around cracks under the action of a magnetic field to detect cracks; magnetic particle testing is particularly suitable for detecting surface and near-surface defects of metal materials. This technology is simple to operate, low-cost, and has no strict requirements on the size and shape of the test object. Although it performs well in detecting surface and near-surface defects, it is not applicable to non-magnetic materials. Moreover, magnetic particle testing has difficulty detecting cracks hidden in multi-layer structures or complex shapes.

[0041] 3. Eddy current detection technology: This technology is based on the principle of electromagnetic induction. It identifies cracks by generating eddy currents on the surface of conductive materials and detecting their changes. Eddy current detection is highly sensitive to surface and near-surface cracks in conductive materials and is not affected by material thickness and shape. Moreover, eddy current detection technology can realize online detection, which helps to monitor the operating status of variable pitch bearings in real time. Although its potential in automated detection systems is gradually being explored, the challenge faced by eddy current detection technology during the operation of wind turbines is how to eliminate misjudgments caused by electromagnetic noise. In addition, eddy current detection has specific requirements for material thickness and conductivity, which limits its scope of application.

[0042] 4. X-ray detection technology: This technology uses the penetrating power of X-rays to identify cracks by detecting their attenuation inside the material; X-ray detection can detect defects inside the material, including cracks, pores, and inclusions, and this technology is applicable to a variety of materials, especially for materials that are difficult to detect with other types of tests (such as thick-walled parts). However, as a high-precision detection method, X-ray detection technology has high professional requirements for equipment and operators, resulting in high application costs. In addition, the use of X-rays is also accompanied by radiation safety issues, requiring special protective measures.

[0043] 5. Acoustic emission detection technology: This technology is based on the principle that materials release energy to generate sound waves under stress, and cracks and other defects are judged by monitoring these sound waves; acoustic emission detection can monitor the crack propagation of variable pitch bearings in real time during operation, which helps to discover and deal with potential problems in a timely manner. Although this technology can be used for online monitoring, it has high requirements for the complexity of signal processing and is easily interfered by environmental noise.

[0044] It can be seen that although the pitch bearing crack detection technology is quite mature in some aspects, each pitch bearing crack detection technology has its specific advantages and limitations, and still faces some challenges in practical application, such as detection speed, cost, environmental adaptability, detection accuracy, and applicability to complex components. Moreover, currently, camera equipment is often set at a fixed position on the inner wall of the hub, such as Figure 1 As shown, when image acquisition is performed to monitor the pitch bearing, due to the inflexibility of the fixed position and the close proximity to the running parts, blind spots may occur and unclear images may be captured, which may result in the inability to accurately detect cracks on the bearing surface.

[0045] In order to overcome the blind spots in the field of vision, unclear images, and improve the detection accuracy and reliability, the present application introduces an image acquisition module, which can locate the shooting positions of each variable pitch bearing at multiple angles and collect images of the corresponding angles of the corresponding variable pitch bearing at each shooting position, thereby overcoming the problems of blind spots in the field of vision and unclear images; at the same time, combined with the crack recognition model, the crack information of the variable pitch bearing in the collected image is identified, that is, the machine learning algorithm in artificial intelligence is combined to improve the automation and intelligence level of detection, which is of great significance for further improving the performance of crack detection of variable pitch bearings, and also provides a more reliable guarantee for the safe operation of wind turbines.

[0046] The wind turbine generator set and its variable pitch bearing crack detection system disclosed in the present invention are described in detail below with reference to the accompanying drawings.

[0047] The present invention provides a wind turbine pitch bearing crack detection system. Figure 2 The invention is a pitch bearing crack detection system for a wind turbine generator set according to an embodiment of the present invention. Figure 2 The variable pitch bearing crack detection system of the wind turbine generator set includes an image acquisition module and a crack detection module. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set. The image acquisition module can locate the multi-angle shooting position of each variable pitch bearing and collect multi-angle images of the corresponding variable pitch bearing at the shooting position; the crack detection module inputs the image into the crack recognition model to obtain the crack information of the corresponding variable pitch bearing in the image.

[0048] As an example, the image acquisition module can be installed on the inner wall of the hub of the wind turbine generator set, and can be moved and positioned to the shooting position of each variable pitch bearing, and collect images of corresponding angles of the corresponding variable pitch bearing at the shooting position. It should be noted that the shooting position includes a position for shooting the variable pitch bearing at multiple angles, that is, the image acquisition module can collect images of each variable pitch bearing at multiple angles, that is, shoot the variable pitch bearing from multiple angles (such as the front, side, 45-degree angle, etc.) to capture cracks in different directions.

[0049] As an example, when collecting images, this embodiment may also consider the following conditions, which are not limited by the present disclosure: images under different lighting conditions may be collected to improve the robustness of the crack recognition model to different lighting changes. Different lighting conditions include natural light, artificial lighting, dark light, etc., which are not limited by the present disclosure; background diversity may also be ensured, such as including various colors and textures, to prevent the crack recognition model from being biased against the background color; image resolution may also be kept consistent (such as 1280×1280 pixels), or adjusted uniformly in the preprocessing stage; and a suitable acquisition frequency may also be set according to the movement speed of the image acquisition module and the detection requirements to ensure that all target areas are covered.

[0050] As an example, the crack detection module can be set up in the cloud, such as being installed on Figure 3 The cloud control center shown in the figure, Figure 3 The acquisition device is the image acquisition module in the above embodiment, that is, each wind turbine generator set has its own image acquisition module. The image acquired by the acquisition device is transmitted to the cloud control center via the intelligent edge computing box, the production operation center, etc., and then the crack detection module in the cloud control center inputs the acquired image into the crack recognition model to obtain the crack information of the corresponding variable pitch bearing in the image.

[0051] As an example, the above-mentioned crack recognition model can adopt the YOLO V10 algorithm model, which has a deeper network and higher detection capability, and other deep learning target detection algorithm models can also be used, such as SSD (Single Shot MultiBox Detector) or Faster R-CNN (Faster Region-based Convolutional Neural Network), which is not limited to this disclosure.

[0052] As an example, take the crack recognition model using YOLO V10 as an example. Specifically, the convolution layer architecture of the crack recognition model selects the YOLO V10 architecture, which can have a deeper network and higher detection capability; the convolution kernel of the convolution layer can use the size of 3×3 pixels and 1×1 pixels to extract features of different scales and complexities; the activation function of the convolution layer can use Leaky ReLU to increase nonlinear characteristics and improve the expression ability of the model. The feature extraction layer of the crack recognition model can choose the feature pyramid (FPN), because through the FPN structure, feature maps from different layers can be combined to enhance the model's detection ability for cracks of different sizes; and the FPN structure uses a top-down and lateral connection method to fuse high-resolution and low-resolution feature maps to improve detection accuracy and speed.

[0053] It should be noted that YOLO (You Only Look Once) is a target detection algorithm that can detect and classify multiple targets in an image in real time in a single forward propagation, thereby improving detection speed and accuracy.

[0054] According to an embodiment of the present disclosure, the image acquisition module is installed at a predetermined position on the inner wall of the hub of the wind turbine generator set, wherein the predetermined position is equidistant from all the variable pitch bearings. Through this embodiment, the image acquisition module is set at a position equidistant from all the variable pitch bearings, which can avoid being too far away from any variable pitch bearing and wasting shooting time, and the installation position is used as the spatial symmetry center of the three variable pitch bearings, which can effectively reduce the complexity of subsequent spatial trajectory planning and attitude control.

[0055] As an example, Figure 4 As shown, the image acquisition module is installed at the hub opening position, so that the image acquisition module is at an equal distance from all pitch bearings. It should be noted that after being installed in the wind turbine generator set, the opening position will be closed, that is, the image acquisition module can be installed at the opening position on the hub.

[0056] According to an embodiment of the present disclosure, the image acquisition module may include a base, a mechanical arm, a camera head and a camera device. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set through the base. The mechanical arm includes a first mechanical arm and a second mechanical arm that are rotatably connected. The other side of the first mechanical arm is connected to the base through a second rotating shaft, and the other side of the second mechanical arm is connected to the camera head through a third rotating shaft. The base can rotate along the first rotating shaft, wherein the first rotating shaft is perpendicular to the inner wall of the hub; the first mechanical arm can rotate along the second rotating shaft, wherein the second rotating shaft is parallel to the inner wall of the hub; the camera head can rotate along the third rotating shaft and support the camera device to capture multi-angle images of the corresponding variable pitch bearing at the shooting position, wherein the third rotating shaft is perpendicular to the cross-section of the mechanical arm.

[0057] Through this embodiment, the image acquisition module can position the camera device at multiple angles through the base, the mechanical arm and the camera pan / tilt platform, thereby facilitating the acquisition of multi-angle images of the variable pitch bearing.

[0058] As an example, in order to obtain high-quality images to improve detection accuracy, the above-mentioned camera device can be a high-definition camera or an industrial camera combined with a laser scanner, which is not limited by the present disclosure. The high-definition camera and the industrial camera can transmit real-time data with the control system through a USB or GigE interface, supporting high-speed image transmission.

[0059] As an example, the above-mentioned rotational connection can be a connection with any degree of freedom or a hinge, which is not limited by the present disclosure.

[0060] As an example, the base can be a cylindrical turntable, one bottom surface of the cylinder is mounted on the inner wall of the hub of the wind turbine generator set through a first rotating shaft, and the other bottom surface of the cylinder is connected to the mechanical arm through a second rotating shaft. It should be noted that the base can also be other shapes, as long as it can achieve a functional structure similar to that of the turntable of the cylindrical pile, and this disclosure is not limited to this.

[0061] As an example, the image acquisition module can adopt a 3-rotation axis structure, allowing flexible movement in the horizontal, vertical and around vertical axes. Each joint is driven by a high-precision servo motor, which can achieve high-resolution position control and fast response.

[0062] As an example, the specific shooting position of the pitch bearing can be determined by using the spatial coordinates and the focal length of the camera, such as Figure 5 As shown, the base can rotate along the first rotation axis ( Figure 5 The first rotating shaft is not shown), the first mechanical arm can rotate along the second rotating shaft, and the camera platform can rotate along the third rotating shaft, so that the image acquisition module can plan the trajectory through the rotational freedom of the base, the first mechanical arm, the second mechanical arm, and the camera platform, so as to be able to locate each determined shooting position, realize multi-angle shooting of each variable pitch bearing, and then realize that one device can collect images of all variable pitch bearings. It should be noted that Figure 5 The base, first mechanical arm, second mechanical arm and camera head shown are for illustration only and do not limit their shapes and structures.

[0063] As an example, in the system disclosed in the present invention, a control module can also be set, and the control module can be equipped with a high-performance motion controller (such as a programmable logic controller PLC or an embedded control unit), which is responsible for coordinating the motion and image acquisition of the image acquisition module (such as a mechanical arm), and using an integrated development environment for motion planning, trajectory control and image acquisition processing, which can ensure the efficiency and reliability of the system. The motion controller can accurately determine the coordinates of the measuring points of the variable pitch bearing according to the design drawings of the wind turbine generator set, and can use the inverse kinematics algorithm to convert the target Cartesian coordinates into the joint angle of the image acquisition module (such as a mechanical arm) to ensure the feasibility of the motion path. Then, according to the joint angle calculation, a motion instruction can be generated to control the servo motor to move smoothly to the target position, and the joint position is monitored in real time through a high-precision encoder to ensure that the actual position is highly consistent with the target position. It should be noted that the PID (Proportional-Integral-Differential, i.e., proportional-integral-differential) control algorithm can be applied to achieve a smooth transition of the joint, avoid vibration and errors caused by violent movement, and perform position accuracy detection after reaching the target position to ensure that the error is within a preset tolerance range.

[0064] As an example, taking an industrial camera as an example, after the image acquisition module is stable, the control module can trigger the autofocus function of the industrial camera to ensure that the image clarity is optimal, and automatically adjust the pitch angle and rotation angle of the industrial camera according to the height and angle of the target position to ensure that the best viewing angle covers the variable pitch bearing to obtain the best view and image quality. The control module can also trigger the industrial camera to collect images, obtain detailed image data of the variable pitch bearing, and transmit the collected image data to the control module in real time for subsequent analysis.

[0065] It should be noted that industrial cameras can support two modes: single-frame acquisition and continuous acquisition, which can be selected according to specific task requirements. The image data collected by industrial cameras can be transmitted through the USB or GigE interface at a maximum rate of 5Gbps to ensure that data is not lost, and lossless compression algorithms (such as JPEG2000) can be used to reduce data transmission volume and improve efficiency. Key image data can also be backed up in real time to local storage and the cloud to prevent data loss. During the transmission process, shielded cables and isolation interfaces can be used to prevent electromagnetic interference inside the wind turbine from affecting data transmission.

[0066] According to an embodiment of the present disclosure, the image acquisition module may include a base, a robotic arm, a camera head and a camera device. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set through the base. One side of the robotic arm is connected to the base through a fifth rotating shaft, and the other side is connected to the camera head through a sixth rotating shaft. The base can rotate along the fourth rotating shaft, wherein the fourth rotating shaft is perpendicular to the inner wall of the hub; the robotic arm can rotate along the fifth rotating shaft, wherein the fifth rotating shaft is parallel to the inner wall of the hub; the camera head can rotate along the sixth rotating shaft and support the camera device to capture multi-angle images of the corresponding variable pitch bearing at the shooting position, wherein the sixth rotating shaft is parallel to the cross-section of the robotic arm.

[0067] Through this embodiment, the image acquisition module can position the camera device at multiple angles through the base, the mechanical arm and the camera pan / tilt platform, thereby facilitating the acquisition of multi-angle images of the variable pitch bearing.

[0068] As an example, in order to obtain high-quality images to improve detection accuracy, the above-mentioned camera device can be a high-definition camera or an industrial camera combined with a laser scanner, which is not limited by the present disclosure. The high-definition camera and the industrial camera can transmit real-time data with the control system through a USB or GigE interface, supporting high-speed image transmission.

[0069] As an example, the base can be a cylindrical turntable, one bottom surface of the cylinder is mounted on the inner wall of the hub of the wind turbine generator set through a first rotating shaft, and the other bottom surface of the cylinder is connected to the mechanical arm through a second rotating shaft. It should be noted that the base can also be other shapes, as long as it can achieve a functional structure similar to that of the turntable of the cylindrical pile, and this disclosure is not limited to this.

[0070] As an example, the image acquisition module can adopt a 4-rotation axis structure with a three-degree-of-freedom rotation joint (R), allowing flexible movement in the horizontal, vertical and around the vertical axis. Each joint is driven by a high-precision servo motor, which can achieve high-resolution position control and fast response.

[0071] As an example, the specific shooting position of the pitch bearing can be determined by using the spatial coordinates and the focal length of the camera, such as Figure 6 As shown, the base can rotate along the fourth axis ( Figure 6 The fourth axis is not shown), the mechanical arm can rotate along the fifth axis, and the camera head can rotate along the sixth axis, so that the image acquisition module can plan the trajectory through the rotational freedom of the base, the mechanical arm, and the camera head, so as to be able to locate each determined shooting position, realize multi-angle shooting of each variable pitch bearing, and then realize that one device can collect images of all variable pitch bearings. It should be noted that Figure 6 The base, mechanical arm and camera head shown are for illustration only and do not limit their shapes and structures.

[0072] As an example, in the system disclosed in the present invention, a control module can also be set, and the control module can be equipped with a high-performance motion controller (such as a programmable logic controller PLC or an embedded control unit), which is responsible for coordinating the motion and image acquisition of the image acquisition module (such as a mechanical arm), and using an integrated development environment for motion planning, trajectory control and image acquisition processing, which can ensure the efficiency and reliability of the system. The motion controller can accurately determine the coordinates of the measuring points of the variable pitch bearing according to the design drawings of the wind turbine generator set, and can use the inverse kinematics algorithm to convert the target Cartesian coordinates into the joint angle of the image acquisition module (such as a mechanical arm) to ensure the feasibility of the motion path. Then, according to the joint angle calculation, a motion instruction can be generated to control the servo motor to move smoothly to the target position, and the joint position is monitored in real time through a high-precision encoder to ensure that the actual position is highly consistent with the target position. It should be noted that the PID (Proportional-Integral-Differential, i.e., proportional-integral-differential) control algorithm can be applied to achieve a smooth transition of the joint, avoid vibration and errors caused by violent movement, and perform position accuracy detection after reaching the target position to ensure that the error is within a preset tolerance range.

[0073] As an example, taking an industrial camera as an example, after the image acquisition module is stable, the control module can trigger the autofocus function of the industrial camera to ensure that the image clarity is optimal, and automatically adjust the pitch angle and rotation angle of the industrial camera according to the height and angle of the target position to ensure that the best viewing angle covers the variable pitch bearing to obtain the best view and image quality. The control module can also trigger the industrial camera to collect images, obtain detailed image data of the variable pitch bearing, and transmit the collected image data to the control module in real time for subsequent analysis.

[0074] It should be noted that industrial cameras can support two modes: single-frame acquisition and continuous acquisition, which can be selected according to specific task requirements. The image data collected by industrial cameras can be transmitted through USB 3.0 or GigE interface at a maximum rate of 5Gbps to ensure that data is not lost, and lossless compression algorithms (such as JPEG2000) can be used to reduce data transmission volume and improve efficiency. Key image data can also be backed up to local storage and the cloud in real time to prevent data loss. During the transmission process, shielded cables and isolation interfaces can be used to prevent electromagnetic interference inside the wind turbine from affecting data transmission.

[0075] According to an embodiment of the present disclosure, the image acquisition module may further include a bracket, through which the base is mounted on the inner wall of the hub of the wind turbine generator set, and the bracket can be deformed to adjust the distance between the base and the inner wall of the hub; wherein the minimum distance between the base and the inner wall of the hub is zero, and the maximum distance between the base and the inner wall of the hub is the maximum height of the bracket.

[0076] Through this embodiment, by adding a deformable bracket, the image acquisition module can increase a degree of freedom, so as to capture images of the variable pitch bearing at more angles.

[0077] As an example, the bracket may be X-shaped, such as Figure 7 As shown, it may also be a shape of multiple X connections, or other shapes, as long as the distance between the base and the inner wall of the hub can be adjusted, and the present disclosure does not limit this.

[0078] According to an embodiment of the present disclosure, the mechanical arm can be extended and retracted to adjust the arm length of the mechanical arm. Through this embodiment, the image acquisition module can acquire images of more angles of the variable pitch bearing.

[0079] As an example, the robotic arm may adopt a structure similar to a hydraulic support, or may adopt other structures, as long as the arm length can be telescopically adjusted, and the present disclosure does not limit this.

[0080] It should be noted that, it can be seen from the above embodiments that, by developing a multi-degree-of-freedom robotic arm with rotation and extension, equipped with a high-resolution industrial camera, and based on an image acquisition solution, it is possible to achieve multi-angle periodic monitoring of variable pitch bearings and automatic identification of cracks in the target area. This image recognition solution provides comprehensive monitoring and crack identification guarantees for the operation of wind turbines, demonstrating its potential in improving detection accuracy and reliability.

[0081] According to an embodiment of the present disclosure, a crack recognition model may include a target detector and a classifier, wherein the crack detection module inputs an image into the crack recognition model to obtain crack information of a corresponding variable pitch bearing in the image, which may include: inputting the image into the target detector to determine the position of a main object in the image, wherein the object includes cracks, bolts, and numbers; based on the position of the main object in the image, annotating the main object in the image with an anchor frame; and inputting the main object annotated by the anchor frame into the classifier to obtain size information of the crack.

[0082] Through this embodiment, a crack recognition model including a target detector and a classifier is adopted, which can accurately locate and identify cracks in the variable pitch bearing. The crack recognition model can also be used to take photos and identify at a high frequency, such as daily / weekly / monthly, to replace the original regular inspections, which can effectively reduce missed inspections and false inspections and significantly improve the accuracy and reliability of crack detection.

[0083] As an example, the above crack recognition model can use the YOLO V10 model. By using the YOLO V10 model for target detection, the crack position can be quickly and accurately identified and marked. Therefore, the YOLO V10 model is applied to the crack detection of the variable pitch bearing of the wind turbine generator set, which improves the detection accuracy and speed. Specifically, a real-time image of the variable pitch bearing is obtained; the real-time image is input into the FCOS target detector to obtain the position information of the main object, so that the main object is marked out from the image using an anchor frame; the marked main object is then input into the SE-DenseNet201 image classifier to obtain crack information, that is, the main object marked out by the anchor frame is classified into categories such as cracks, bolts, numbers, etc.

[0084] As an example, the above anchor boxes are pre-defined rectangular boxes with fixed size and aspect ratio in the target detection algorithm. In the image, these anchor boxes are placed at different positions for preliminary guessing of the possible position and size range of the main object. Specifically, the anchor boxes can define anchor boxes of different scales and aspect ratios, such as small: 10×13 pixels, medium: 16×30 pixels, large: 33×23 pixels, which are not limited by the present disclosure; and multiple anchor boxes can be generated on each predicted grid unit to cover cracks of different sizes and shapes; furthermore, the anchor box can predict the relative coordinate adjustment (offsets) of each anchor box through the regression branch, including the center point offset (tx, ty) and the width and height scaling (tw, th), and the CIoU (Complete IoU) loss function can be used to combine the distance, overlap and aspect ratio losses to improve the accuracy of the bounding box regression of the main object.

[0085] It should be noted that the anchor box that marks the main object is also called the bounding box of the main object.

[0086] As an example, the coordinate format of the above bounding box may be as follows:<class_id><x_center><y_center> <width> <height>, and can be normalized to the range of [0, 1]. Moreover, the present disclosure converts the coordinates in this format into absolute pixel coordinates (xmin, ymin, xmax, ymax) for visualization and subsequent processing.

[0087] As an example, through the classifier, each bounding box can be assigned a corresponding category label, such as a crack label, a bolt label, a digital label, etc. It should be noted that it is necessary to ensure that all labelers use a unified category definition and label assignment specification to reduce labeling errors to ensure the accuracy of labeling; moreover, in the case of multiple categories, the cross entropy loss function can be combined with Focal Loss to deal with the category imbalance problem and improve the detection performance of rare categories. Furthermore, based on the type label, subsequent maintenance personnel can only focus on the information corresponding to the crack label to avoid misjudging other label content as cracks. It should be noted that Focal Loss is a loss function widely used in computer vision tasks such as target detection.

[0088] As an example, each bounding box corresponds to a category label, and, in the case of multiple categories, one-hot encoding can be used to represent the category label for subsequent classification and statistics.

[0089] According to an embodiment of the present disclosure, a crack recognition model can be trained in the following manner: obtaining a first training sample set, wherein the first training sample set includes multiple pitch bearing images; performing enhancement processing on the multiple pitch bearing images respectively to obtain multiple enhanced images; adding the multiple enhanced images to the first training sample set to obtain a second training sample set; and training the crack recognition model based on the second training sample set. Through this embodiment, by using data enhancement technology, the diversity of the training sample set can be expanded, and the generalization ability of the crack recognition model in practical applications can be improved, so that high-performance detection effects can be maintained under different lighting, angles, and other conditions.

[0090] As an example, the enhancement processing of the pitch bearing image includes but is not limited to the following methods:

[0091] The variable pitch bearing image can be geometrically changed, such as randomly rotating the image by ±15 degrees to simulate cracks at different shooting angles; the image can also be flipped horizontally and vertically to improve the model's insensitivity to direction; the image can also be randomly cropped and scaled to increase samples of cracks in different positions and sizes, which is not limited in the present disclosure.

[0092] The color of the variable pitch bearing image can be transformed, for example, the image brightness can be randomly adjusted by ±30% to simulate different lighting conditions; the image contrast can be randomly adjusted by ±20% to enhance the contrast diversity of the image; the image saturation can also be randomly adjusted by ±20% to improve the adaptability of the model to color changes, which is not limited in the present disclosure.

[0093] Noise may be added to the pitch bearing image, such as adding random Gaussian noise to the image to simulate image noise in a real environment; random noise points may also be added to the image to enhance the robustness of the model to image noise, which is not limited in the present disclosure.

[0094] Other detail parameters of the variable pitch bearing image can also be randomly enhanced to expand the diversity of training samples. The above other detail parameters may include but are not limited to: exposure, white balance; the variable pitch bearing image can also be rotated, translated and other transformations to generate more samples to expand the diversity of training samples.

[0095] As an example, the above-mentioned image enhancement can be implemented by an image enhancement model, which is not limited in the present disclosure.

[0096] As an example, after obtaining the second training sample set, a suitable annotation tool, such as LabelImg, VGGImage Annotator, etc., can be selected. Using the selected annotation tool, a bounding box is manually drawn for each crack and the coordinates of the bounding box are recorded. For example, the coordinates of each bounding box may include but are not limited to: category ID, center coordinates relative to the width and height of the image, width and height. The coordinate format may be as follows:<class_id><x_center><y_center> <width> <height>, and the present disclosure does not limit this; you can also manually assign corresponding category labels to each bounding box, such as crack labels, bolt labels, digital labels, etc. It should be noted that it is necessary to ensure that all labelers use unified category definitions and label assignment specifications to reduce labeling errors and ensure the accuracy of labeling.

[0097] As an example, after obtaining the second training sample set, the second training sample set can be divided, 70%-80% of the samples are selected for training the crack recognition model, that is, the training set, 10%-15% of the samples are selected for model tuning and hyperparameter selection, that is, the validation set; 10%-15% of the samples are selected for the final evaluation of the generalization ability of the model, that is, the test set. Moreover, when dividing, it is necessary to ensure that each category is evenly distributed in the training set, validation set, and test set to prevent the model from being biased towards common categories due to category imbalance; in addition, the validation set and test set should contain scenes that have not been seen in the training set to evaluate the generalization ability of the model.

[0098] As an example, the crack recognition model can be developed offline, that is, trained offline. After the training meets the required requirements, the trained model can be deployed online. Figure 8 As shown in the figure, the offline training process is similar to the online service process. The offline training process requires image acquisition, that is, obtaining training samples, extracting the features of the training samples, and using the extracted features to train the crack recognition model. After the training is completed, the trained model is deployed online. The online model receives real-time images, extracts the features of the images, and inputs them into the trained model to identify the crack information in the images. Subsequently, the training sample set can be enriched according to the real-time images to iteratively update the model.

[0099] According to an embodiment of the present disclosure, the above-mentioned system may further include an image preprocessing module and a data processing module. The image preprocessing module acquires the image acquired by the image acquisition module and preprocesses the image. The crack detection module inputs the preprocessed image into a crack recognition model to obtain crack information of the corresponding variable pitch bearing in the image. The data processing module acquires crack information from the crack detection module, acquires the size and position of the crack based on the crack information, and determines maintenance recommendations based on the size and position of the crack.

[0100] Through this embodiment, the collected images are pre-processed to improve the image quality and facilitate subsequent processing. In addition, the crack information is processed to generate maintenance suggestions, which facilitates maintenance personnel to adopt appropriate maintenance plans.

[0101] As an example, the image preprocessing module can scale the image, such as adjusting the collected image to a uniform size, such as 640×640 pixels, to meet the input requirements of the crack recognition model (such as the YOLO V10 model) to ensure the consistency of the input data; padding or cropping technology can also be used to maintain the original proportion of the image to prevent deformation from affecting the model performance, which is not limited by the present disclosure; the image preprocessing module can also use Gaussian filtering, median filtering and other methods to perform denoising, graying and other preprocessing on the collected image, because image denoising, graying and other preprocessing can improve the quality of the image and facilitate subsequent processing. It should be noted that the preprocessing method is not limited to denoising, graying, image scaling, etc., and can also be other methods, which is not limited by the present disclosure.

[0102] As an example, other image processing algorithms such as Median Filter and Gaussian Blur may be used to replace the existing denoising method, which is not limited in the present disclosure.

[0103] As an example, the data processing module can perform data analysis and record the detection results (i.e., the above-mentioned crack information), specifically, analyze the detection results, record information such as the size and location of the cracks, and perform statistical analysis. The statistical analysis can introduce a machine learning model to automatically generate maintenance recommendations, which is not limited by the present disclosure.

[0104] As an example, the severity of a crack can be evaluated from the following directions: 1) the length, width and area of ​​the crack can be evaluated, and different levels of severity (such as low, medium and high) can be defined; 2) the impact of the location of the crack on the structure can be considered, and different levels of severity (such as low, medium and high) can be defined. For example, a crack in the stress-bearing part of the bolt may affect the stability of the overall structure, which can be defined as a high level, and the present disclosure does not limit this; 3) the development direction and potential impact of the crack can be predicted through historical data and crack propagation trends, and then different levels of severity (such as low, medium and high) can be defined.

[0105] As an example, the severity of a crack can be assessed in the following ways: 1) classifying the severity of the crack based on preset rules and thresholds; 2) training a classification model (such as random forest, support vector machine (SVM)) to automatically assess the severity based on crack characteristics.

[0106] As an example, after determining the severity level of the cracks, they can be sorted according to the severity, with high-risk cracks given priority; maintenance resources can be reasonably allocated to ensure timely maintenance of key parts; and detailed maintenance plans can be formulated, including maintenance time, required resources, and expected goals. Specifically, based on conditions such as defect severity and size threshold, defect detection and assessment results can be integrated with the maintenance management system to automatically generate early warning information, and maintenance needs can be fed back to the operation and maintenance team via email, text message or system notification to ensure a timely response.

[0107] In order to understand the above embodiments, Fig. 9 Describe the system from the software level. Fig. 9 Show the software level process, such as Fig. 9 As shown, when an image of a variable pitch bearing is collected, it can be first determined whether the current mode is the actual prediction stage. If the judgment result is that it is not the actual prediction stage, the training stage of the crack recognition model is entered. First, the acquired multiple variable pitch bearing images are enhanced, and the multiple enhanced images and the multiple variable pitch bearing images are combined into a training sample set (i.e., the second training sample set mentioned above). The crack recognition model is trained with the training sample set. After the training, it is determined whether to optimize the online model. If the judgment result is that the online model is not to be optimized, the training stage of the crack recognition model is returned. If the judgment result is that the online model is not to be optimized, the training stage of the crack recognition model is returned. If the online model is optimized, the trained model will be pushed to the model library to replace the original model. The subsequent crack detection module uses the new crack recognition model to identify cracks. If the judgment result is the actual prediction stage, the crack recognition prediction stage is entered. First, the collected image is preprocessed and the preprocessed image is sent to the crack detection module. The crack detection module uses the latest crack recognition model in the model library to identify cracks and obtain the detection results. Then, the detection results are input into the data processing module. The data processing module analyzes the detection results to obtain the size and position of the crack, and can generate dimension suggestions based on this information.

[0108] In order to further understand the above embodiments, the following is a systematic description starting from the combination of software and hardware. Specifically, the variable pitch bearing crack detection system locates the shooting positions of the variable pitch bearings at multiple angles of the image to be shot through the image acquisition module, and uses the high-definition camera in the image acquisition module to perform real-time image acquisition of the variable pitch bearing. The acquired images will be transmitted to the image preprocessing module, and the image preprocessing module will preprocess the acquired images, such as removing noise and converting them into grayscale images, and then input the preprocessed images into the crack detection module for crack detection. For example, the crack detection module will identify and locate the cracks in the image according to the pre-trained crack recognition model, and generate a detection result containing crack location distribution (location and scale, etc.), quantity information and other information. The data processing module further analyzes and processes the detection data, obtains the crack location, size and other information contained in the detection results, and generates a detection report (i.e., maintenance recommendations) for reference by maintenance personnel.

[0109] As an example, the image acquisition process can be as follows: pre-define the measurement points of all variable pitch bearings, store them in the task queue, and execute them in sequence; complete the following processing for each measurement point: confirm that the image acquisition and data processing of the current measurement point are completed, start position switching, read the three-dimensional coordinates of the next variable pitch bearing from the task queue, that is, the next measurement point, and after obtaining the next measurement point, execute the position planning and motion control process, move the robotic arm to the next measurement point, repeat the image acquisition and data processing steps at the new measurement point, until all measurement points are completed, generate a task completion report and enter the standby state.

[0110] It should be noted that during the image acquisition process, the measurement order can be dynamically adjusted according to the real-time acquired images and on-site conditions to improve sampling efficiency; moreover, if a measurement point fails or cannot be acquired, it will be automatically skipped and logged to continue with subsequent tasks; furthermore, based on the current measurement point and the next measurement point, the robot arm's motion path can be optimized to reduce motion time and energy consumption.

[0111] Fig.10 The image of the variable pitch bearing annotated with crack information obtained by the method disclosed in the present invention is shown, that is, during regular maintenance, the variable pitch bearing crack detection system collects a large number of images and videos of the variable pitch bearing of a wind turbine generator set. After being processed by the image preprocessing module, the crack detection module identifies the presence of a crack in one of the preprocessed images and videos, and locates the crack near the sealing ring of the variable pitch bearing, and marks it with a rectangular box diagram and a label (crack 0.91), where 0.91 is the confidence level, that is, the probability that the model determines that the object in this box is the identified object, and the higher the probability, the more accurate the model. The data processing module records this information, generates a test report, and provides it to maintenance personnel for maintenance. It should be noted that Fig.10 Numeric information (number 0.77) is also marked, where 0.77 is the confidence level. The purpose of marking the numerical information is to avoid identifying the numbers as cracks and improve the recognition accuracy.

[0112] During the entire operation process, a monitoring system can be set up for real-time monitoring to ensure the safety of the image acquisition module's movement and prevent potential collisions or interference. Safety thresholds can also be set and an alarm system can be equipped to promptly notify operators when abnormal conditions are detected.

[0113] Specifically, the monitoring system may include but is not limited to the following sensors: 1) Position sensors, which monitor the position and speed of each joint of the image acquisition module to ensure that the movement is within a safe range; 2) Collision sensors, which are installed at the joints and ends of the robotic arm to detect object contact and trigger an emergency stop; 3) Temperature sensors, which monitor the temperature of the servo motor and control system to prevent overheating; 4) Humidity sensors, which monitor the humidity inside the wind turbine to prevent electronic components from getting damp; 5) Vibration sensors, which detect the vibration frequency and amplitude of the image acquisition module and the system to ensure that the system stability is not affected.

[0114] Safety thresholds may include but are not limited to the following thresholds: 1) Position deviation, if the error exceeds ±0.1mm or ±0.05°, an alarm is triggered; 2) Speed ​​limit, if the joint speed exceeds the set upper limit, it will immediately slow down or stop; 3) Temperature limit, if the device temperature exceeds the safe working range (such as 50°C), the power supply will be automatically cut off for protection; 4) Humidity limit, if the humidity exceeds the set threshold (such as 80% RH), anti-moisture measures will be triggered; 5) Vibration limit, if the vibration amplitude and frequency exceed the set threshold, an alarm will be triggered and the robot arm motion strategy will be adjusted.

[0115] The alarm system may include but is not limited to the following forms: 1) visual alarm, where a red indicator light flashes to display the alarm status; 2) audible alarm, where a high-decibel buzzer sounds an alarm to alert on-site operators; 3) remote notification, where the alarm information is sent to the monitoring center and the mobile devices of relevant personnel through the network; 4) automatic recording, where the alarm event is recorded in the system log and the fault time and type are marked for subsequent analysis.

[0116] In addition, the image acquisition module and camera can be regularly maintained and inspected to ensure long-term stable operation of the equipment and avoid failures that affect sampling efficiency.

[0117] It can be seen that the present invention designs a multi-rotational degree of freedom image acquisition module, which can adapt to complex working environments through trajectory planning in the hub of the wind turbine generator set to ensure the acquisition of high-quality images, and realize the identification of pitch bearing cracks through visual detection technology, which is conducive to the efficient detection flow of the YOLO V10 target detection framework V10 version without NMS (Non-Maximum Suppression) detection, and trains the crack recognition model based on GPU parallel computing. Then, the trained crack recognition model can be connected to the on-site unit pitch bearing inspection photo library and video stream to realize the identification and detection of cracks, and then the identification and early warning of crack defects are applied through automated statistical analysis.

[0118] In order to verify the accuracy of the detection of the disclosed method, the present invention also uses multiple pitch bearing images to verify the method of the present invention, and finds that the method of the present invention can accurately identify the location and size of the cracks in the image, and can be marked with rectangular boxes and labels. In addition, in the experiment, it is also compared with the traditional detection method. For example, compared with the traditional detection method, the present invention reduces the image processing time by 80% and improves the real-time detection accuracy by 15%, which can effectively solve the problems of inaccurate detection and low efficiency in the traditional method.

[0119] In summary, the present invention constructs an automated crack detection system, including modules such as image acquisition, image preprocessing, crack detection and data processing, which realizes the automation of the entire process from image input to detection results, thereby reducing the frequency and intensity of manual inspections, such as reducing the risks of manual participation and high-altitude operations, reducing the use of human resources, and further reducing the maintenance cost of wind turbines and improving the operating efficiency of wind turbines.

[0120] Moreover, the present invention adds a multi-degree-of-freedom mechanism with image acquisition or other sensor equipment inside the wind turbine, and realizes the acquisition of the variable pitch bearing image of the wind turbine through self-space planning, as well as the identification of cracks, to provide a solution combining robot technology and computer image technology for the stable operation of the wind turbine. The modular design can realize the adaptation of the entire mechanism or components to the space of different machine models, and eliminate the interference effect with the spatial planning algorithm. Through one mechanism, different monitoring data of multiple components and positions can be collected, providing the real status of the front line for subsequent analysis, perception, decision-making, alarm and other management; moreover, through modular design, the telescopic arm and rotary joint parameters of the mechanism can be optimized as needed to realize the trajectory planning of a larger space in the wheel hub, and the terminal image acquisition device can also replace and expand other sensor equipment to realize internal monitoring with more monitoring means.

[0121] Furthermore, the present disclosure utilizes the non-NMS (Non-Maximum Suppression) detection feature of the YOLO V10 algorithm, which can complete the processing and analysis of large-scale images in a short time, and is suitable for online monitoring and abnormality identification. Therefore, it can realize real-time crack detection of the variable pitch bearing of the wind turbine generator set, which helps to find problems in time and take countermeasures, thereby improving the safety and operation efficiency of the wind turbine generator set. Therefore, the present disclosure has significant technical advantages in the detection of cracks in the variable pitch bearing of the wind turbine generator set, can effectively solve the problems existing in the prior art, and provide a more efficient, accurate and reliable solution.

[0122] It should be noted that the present invention is not only applicable to the crack detection of variable pitch bearings of wind turbines, but can also be extended to the crack and defect detection of other mechanical components, such as the blade roots of wind turbines, generators and other core components, etc. It can also be promoted and applied in other industrial fields and has good market prospects.

[0123] In another general aspect of the present disclosure, a wind turbine generator set is provided, comprising the pitch bearing crack detection system of the wind turbine generator set as described above.

[0124] Although some embodiments of the present disclosure have been shown and described, it will be appreciated by those skilled in the art that modifications may be made to the embodiments without departing from the principles and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents.< / height> < / width> < / height> < / width>

Claims

1. A wind turbine pitch bearing crack detection system, characterized in that: It includes an image acquisition module and a crack detection module. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set. The image acquisition module can locate a multi-angle shooting position of each variable pitch bearing, and acquire multi-angle images of the corresponding variable pitch bearing at the shooting position; The crack detection module inputs the image into a crack recognition model to obtain crack information of the corresponding pitch bearing in the image.

2. The pitch bearing crack detection system according to claim 1, characterized in that: The image acquisition module includes a base, a mechanical arm, a camera platform and a camera device. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set through the base. The mechanical arm includes a first mechanical arm and a second mechanical arm that are rotatably connected. The other side of the first mechanical arm is connected to the base through a second rotating shaft, and the other side of the second mechanical arm is connected to the camera platform through a third rotating shaft. The base is capable of rotating along a first rotation axis, wherein the first rotation axis is perpendicular to the inner wall of the hub; The first mechanical arm is capable of rotating along the second rotation axis, wherein the second rotation axis is parallel to the inner wall of the hub; The camera platform can rotate along the third rotation axis and support the camera device to collect multi-angle images of the corresponding variable pitch bearing at the shooting position, wherein the third rotation axis is perpendicular to the cross-section of the mechanical arm.

3. The pitch bearing crack detection system according to claim 1, characterized in that: The image acquisition module includes a base, a mechanical arm, a camera platform and a camera device. The image acquisition module is installed on the inner wall of the hub of the wind turbine generator set through the base. One side of the mechanical arm is connected to the base through a fifth rotating shaft, and the other side is connected to the camera platform through a sixth rotating shaft. The base can rotate along a fourth rotation axis, wherein the fourth rotation axis is perpendicular to the inner wall of the hub; The mechanical arm can rotate along the fifth rotation axis, wherein the fifth rotation axis is parallel to the inner wall of the hub; The camera platform can rotate along the sixth rotation axis and support the camera device to collect multi-angle images of the corresponding variable pitch bearing at the shooting position, wherein the sixth rotation axis is parallel to the cross-section of the mechanical arm.

4. The pitch bearing crack detection system according to claim 2 or 3, characterized in that: The image acquisition module also includes a bracket, and the base is mounted on the inner wall of the hub of the wind turbine generator set through the bracket. The bracket is deformable to adjust the distance between the base and the inner wall of the hub; The minimum distance between the base and the inner wall of the hub is zero, and the maximum distance between the base and the inner wall of the hub is the maximum height of the bracket.

5. The pitch bearing crack detection system according to claim 2 or 3, characterized in that: The mechanical arm can be extended and retracted to adjust the arm length of the mechanical arm.

6. The pitch bearing crack detection system according to claim 1, characterized in that: The image acquisition module is installed at a predetermined position on the inner wall of the hub of the wind turbine generator set, wherein the predetermined position is equidistant from all pitch bearings.

7. The pitch bearing crack detection system according to claim 1, characterized in that: The crack recognition model includes a target detector and a classifier, The crack detection module inputs the image into a crack recognition model to obtain crack information of the corresponding pitch bearing in the image, including: Inputting the image into the object detector to determine the location of a main object in the image, wherein the object includes a crack, a bolt, and a number; Based on the position of the main object in the image, annotating the main object in the image using an anchor frame; The main object annotated by the anchor frame is input into the classifier to obtain the crack information.

8. The pitch bearing crack detection system according to claim 7, characterized in that: The crack recognition model is trained in the following way: Acquire a first training sample set, wherein the first training sample set includes a plurality of pitch bearing images; Performing enhancement processing on the multiple pitch bearing images respectively to obtain multiple enhanced images; Adding the plurality of enhanced images to the first training sample set to obtain a second training sample set; The crack identification model is trained based on the second training sample set.

9. The pitch bearing crack detection system according to claim 1, characterized in that: It also includes an image preprocessing module and a data processing module. The image preprocessing module obtains the image acquired by the image acquisition module and preprocesses the image; The crack detection module inputs the preprocessed image into the crack recognition model to obtain crack information of the corresponding pitch bearing in the image; The data processing module obtains crack information from the crack detection module, obtains the size and position of the crack based on the crack information, and determines maintenance suggestions based on the size and position of the crack.

10. A wind turbine generator set, characterized in that: A pitch bearing crack detection system for a wind turbine generator set comprising the system as claimed in any one of claims 1 to 9.

Images