A monitoring method, device, equipment and medium for a wind turbine pitch control system

Abstract

The present invention relates to the field of wind turbine monitoring technology, and specifically to a monitoring method, device, equipment and medium for a wind turbine pitch system. The method comprises: collecting image data of the wind turbine pitch system; grayscale processing the image data based on an adaptive grayscale algorithm adjusted by parameters; filtering the grayscale image data using median filtering, Gaussian filtering and wavelet filtering to obtain filtered image data; edge detection and feature and feature point recognition of the filtered image data, wherein the feature point recognition adopts SIFT and SURF fusion algorithms; based on the recognized feature points, the angle of the wind turbine pitch system is calculated according to a geometric relationship algorithm, or, based on the recognized features, the angle of the wind turbine pitch system is calculated according to a template matching algorithm. This method overcomes many drawbacks of traditional sensor monitoring methods, and can timely and accurately grasp the operating status of the pitch system, effectively reducing the probability of wind turbine shutdown due to pitch system failure.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine monitoring, and in particular to a monitoring method, device, equipment and medium for a wind turbine pitch control system. Background Art

[0002] Traditional monitoring methods for wind turbine pitch systems primarily rely on sensors. For example, various physical sensors, such as angle sensors and displacement sensors, are installed on the pitch mechanism. While this sensor-based monitoring approach can obtain relevant parameter information about the pitch system to a certain extent, it suffers from numerous significant drawbacks. First, the sensor installation process is extremely complex, requiring a certain degree of modification and adaptation to the existing wind turbine structure. This not only significantly increases installation costs but also presents numerous technical challenges and safety risks during construction. Second, sensors, exposed to the harsh operating environment of wind turbines for long periods of time, are susceptible to corrosion and impact from extreme environmental factors such as strong turbine vibration, complex electromagnetic interference, and high temperatures, high humidity, and dust. This inevitably leads to a gradual decrease in sensor measurement accuracy, significantly compromising reliability, and even frequent failures, seriously impacting the accuracy, integrity, and continuity of monitoring data. Third, the monitoring dimension of the sensor monitoring method is relatively single, and it can only obtain a limited number of specific parameter information. It is difficult to comprehensively, deeply and intuitively reflect the overall operating status of the variable pitch system and various potential fault hazards, and it is unable to provide sufficient data support and decision-making basis for the refined operation and maintenance of wind turbines. Summary of the Invention

[0003] In view of this, the present invention provides a method, device, equipment and medium for monitoring a wind turbine pitch system to solve one of the problems existing in the prior art when using sensors to monitor a wind turbine pitch system.

[0004] In a first aspect, the present invention provides a monitoring method for a wind turbine pitch system, the method comprising: collecting image data of the wind turbine pitch system; performing grayscale processing on the image data based on an adaptive grayscale algorithm with parameter adjustment to obtain grayscale image data; filtering the grayscale image data using median filtering, Gaussian filtering, and wavelet filtering to obtain filtered image data; performing edge detection and feature and feature point recognition on the filtered image data, the feature point recognition using SIFT and SURF fusion algorithms; based on the identified feature points, calculating the angle of the wind turbine pitch system according to a geometric relationship algorithm, or, based on the identified features, calculating the angle of the wind turbine pitch system according to a template matching algorithm.

[0005] The monitoring method for the wind turbine pitch system provided in the embodiment of the present invention adopts the method of image acquisition and processing of the wind turbine pitch system for monitoring and angle calculation, which overcomes many disadvantages of the traditional sensor monitoring method, and can timely and accurately grasp the operating status of the pitch system, discover potential faults in advance and take corresponding measures, effectively reduce the probability of wind turbine shutdown due to pitch system failure, ensure the safe and stable operation of the wind turbine, improve the reliability and power generation efficiency of the wind power generation system, reduce operation and maintenance costs, and have significant economic and social benefits.

[0006] In an optional embodiment, before collecting image data of the wind turbine pitch system, the method also includes: setting multiple groups of observation points in a pre-established three-dimensional model of the wind turbine pitch system; determining evaluation parameters corresponding to each group of observation points based on image data observed at each group of observation points, the image data including image data observed when the wind turbine is operating under different working conditions; screening preliminary candidate observation points from the multiple groups of observation points based on the evaluation parameters; arranging image acquisition devices in the wind turbine pitch system based on the preliminary candidate observation points; adjusting the positions of the preliminary candidate observation points based on a comparison result of the image data collected by the image acquisition device and the image data observed at the preliminary candidate observation points; determining the final observation points based on the adjusted preliminary candidate observation points to arrange the image acquisition device, the image acquisition device being used to collect image data of the wind turbine pitch system, the image acquisition device being an industrial-grade CCD or an industrial-grade CMOS.

[0007] In the present invention, an industrial-grade CCD or CMOS camera is used as an image acquisition device, and multiple final observation points are determined through simulation experiments and field tests, so as to achieve all-round, no-dead-angle shooting of the key components of the variable pitch system and multi-perspective capture of the dynamic changes of the transmission components of the variable pitch mechanism. Compared with traditional sensor-based monitoring, this changes the way of acquiring monitoring data and solves the problems of traditional sensors that are complex to install, easily affected by the environment, and have a single monitoring dimension.

[0008] In an optional embodiment, collecting image data of a wind turbine pitch system includes: collecting image data of the wind turbine pitch system based on a preset image acquisition device and light source; determining shadow distribution in the image data using an image processing algorithm; adjusting the luminous intensity and angle of the light source according to the ambient light intensity and shadow distribution; and collecting image data of the wind turbine pitch system based on the preset image acquisition device and the adjusted light source.

[0009] In the present invention, the light source provided can automatically adjust the luminous intensity, angle and illumination range according to the real-time lighting conditions of the wind turbine operating environment, thereby overcoming the influence of unstable lighting conditions in the wind turbine working environment on image acquisition, avoiding shadow interference, improving image quality, and providing strong guarantee for subsequent pitch angle recognition.

[0010] In an optional embodiment, the image data is grayscaled based on an adaptive grayscale algorithm with parameter adjustment to obtain grayscale image data, including: detecting edge information in the image data based on a Canny edge detection algorithm combined with multi-scale morphological gradient operations; calculating texture features of the image data; grayscaled based on an adaptive grayscale algorithm with dynamic adjustment of grayscale conversion parameters and an adaptive threshold algorithm to obtain grayscale image data, wherein the grayscale conversion parameters are dynamically adjusted based on edge information and texture features.

[0011] In the present invention, an adaptive grayscale algorithm is used to convert color images into grayscale images, and the grayscale conversion parameters are automatically adjusted according to the image content and features. The data volume is compressed while retaining the basic structural information, edge details and texture features of the image. Compared with the traditional fixed parameter grayscale method, the processing speed and real-time performance of the entire monitoring system are improved.

[0012] In an optional embodiment, edge detection and feature and feature point identification are performed on the filtered image data, including: detecting edge information in the image data based on the Canny edge detection algorithm combined with multi-scale morphological gradient operation and a local adaptive threshold algorithm, wherein the threshold in the local adaptive threshold algorithm is adjusted based on the local statistical information, edge strength and texture features of the image data; respectively detecting feature points in the image data using the SIFT algorithm and the SURF algorithm and merging them; based on the merged feature points, fusing the separately calculated SIFT feature descriptors and SURF feature descriptors to obtain a fused feature descriptor; screening the feature points according to the local geometric features and texture features of the wind turbine pitch system, and obtaining identified feature points based on the screened feature points and feature descriptors; and extracting features of the image data using a deep convolutional neural network.

[0013] In the present invention, edge detection adopts an improved Canny edge detection algorithm, introduces multi-scale morphological gradient operation to enhance edge contrast and coherence, and combines with an adaptive threshold selection strategy to automatically adjust the threshold according to the local area features of the image, avoiding the inaccurate or broken edge detection caused by noise, uneven lighting and blurred component edges in traditional algorithms, and providing an accurate geometric basis for pitch angle calculation based on edge geometric features; feature point recognition proposes a fusion feature point extraction algorithm, which organically combines the advantages of SIFT and SURF algorithms, and comprehensively utilizes the detection capabilities of the two for feature points of different scales and types in the feature point detection stage, constructs a hybrid feature descriptor in the feature descriptor generation stage, and integrates the scale invariance of the SIFT algorithm and the speed and robustness of the SURF algorithm. It also introduces a local feature constraint mechanism to screen and optimize the feature points, solving the problems of inaccurate feature point extraction and insufficient robustness of feature descriptors in traditional feature extraction algorithms, thereby improving the accuracy and reliability of the monitoring system.

[0014] In an optional embodiment, based on the identified feature points, the angle of the wind turbine pitch system is calculated according to a geometric relationship algorithm, including: training a deep neural network model based on the calibration image and camera parameters to obtain a camera parameter prediction model; matching the identified feature points with a pre-built geometric model of the wind turbine pitch system to obtain a matching result; based on the matching result, converting the two-dimensional feature points into three-dimensional geometric points using a perspective transformation algorithm and camera parameters predicted by the camera parameter prediction model; and calculating the angle of the wind turbine pitch system based on the three-dimensional geometric points and trigonometric functions.

[0015] In the present invention, when calculating the pitch angle based on geometric relationships, an automatic calibration and optimization method of camera parameters based on deep learning is proposed. By collecting a large number of calibration images and corresponding real camera parameter data to train a deep neural network model, automatic prediction and real-time optimization adjustment of camera parameters are realized. Combined with the perspective transformation algorithm and camera parameters, two-dimensional features are converted into three-dimensional geometric points to calculate the pitch angle, thereby improving the parameter accuracy and environmental adaptability in the geometric relationship calculation method.

[0016] In an optional embodiment, based on the identified features, the angle of the wind turbine pitch system is calculated according to a template matching algorithm, including: matching the features with image features of a template image in a preset template library, the preset template library including multiple template images, image features and corresponding pitch angles, and the preset template library is dynamically updated; determining the angle of the wind turbine pitch system based on the matching results, the matching results including a single template image or multiple template images that are successfully matched. When there are multiple template images, the angle of the wind turbine pitch system is a fusion value of the pitch angles corresponding to the multiple template images.

[0017] In the present invention, when calculating the pitch angle based on template matching, a dynamically updated template image library is constructed, so that the template image library can adapt to different operating states and environmental conditions, thereby enhancing the robustness of the system. At the same time, a multi-template fusion matching strategy is introduced to solve the problem that traditional template matching methods are difficult to adapt to the image changes of the wind turbine pitch system, and provide a new, efficient and accurate solution for the angle calculation of the wind turbine pitch system.

[0018] In the second aspect, the present invention provides a monitoring device for a wind turbine pitch system, the device comprising: an image acquisition module for acquiring image data of the wind turbine pitch system; a grayscale processing module for performing grayscale processing on the image data based on an adaptive grayscale algorithm with parameter adjustment to obtain grayscale image data; a filtering module for filtering the grayscale image data using median filtering, Gaussian filtering and wavelet filtering to obtain filtered image data; a feature recognition module for performing edge detection and feature and feature point recognition on the filtered image data, the feature point recognition using SIFT and SURF fusion algorithms; an angle calculation module for calculating the angle of the wind turbine pitch system based on the identified feature points according to a geometric relationship algorithm, or, based on the identified features, calculating the angle of the wind turbine pitch system according to a template matching algorithm.

[0019] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute the monitoring method for a wind turbine pitch system according to the first aspect or any corresponding embodiment thereof.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for monitoring a wind turbine pitch system according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 is a flow chart of a method for monitoring a wind turbine pitch control system according to an embodiment of the present invention;

[0023] Figure 2 is a structural block diagram of a monitoring device for a wind turbine pitch control system according to an embodiment of the present invention;

[0024] Figure 3 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0026] According to an embodiment of the present invention, an embodiment of a monitoring method for a wind turbine pitch system is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0027] In this embodiment, a monitoring method for a wind turbine pitch control system is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. Figure 1 FIG. 1 is a flow chart of a method for monitoring a wind turbine pitch control system according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0028] Step S101: Collect image data from the wind turbine pitch system. Specifically, the wind turbine pitch system adjusts blade angles based on wind speed and other conditions to achieve optimal power generation efficiency. The specific operating process can be implemented with reference to related technologies and will not be detailed here. In this embodiment, monitoring of the wind turbine pitch system primarily involves monitoring and calculating the blade angle, thereby assisting in the safe and stable operation of the wind turbine pitch system.

[0029] When monitoring the wind turbine pitch system, this embodiment incorporates an image acquisition device within the wind turbine pitch system to collect image data for monitoring. The image acquisition device can utilize an industrial-grade CCD or CMOS sensor. Compared to traditional cameras, these cameras offer ultra-high resolution, enabling them to capture extremely subtle structural features and component details within the pitch system with exceptional clarity. Even minute wear marks on the blade surface or slight looseness in the pitch mechanism joints can be accurately imaged. Their excellent low-light performance easily adapts to the complex and changing lighting conditions found in wind turbine operation, enabling stable and clear image acquisition in the dim light of dawn and dusk, cloudy and foggy skies, and even at night. Their fast frame rate ensures continuous and accurate capture of a series of images during dynamic operation, with high-speed rotation and frequent pitch movements, ensuring that no critical moments are missed. This provides a rich and comprehensive data source for real-time status monitoring and analysis of the pitch system.

[0030] In step S102, the image data is grayscaled based on the parameter-adjusted adaptive grayscale algorithm to obtain grayscaled image data. Specifically, after the image data is acquired, the acquired image data is a color image. Directly processing this data may result in a large amount of data, high computational complexity, and low processing efficiency. Therefore, the acquired image data is first grayscaled to reduce the data volume.

[0031] Among them, when grayscale processing is performed on image data, this embodiment adopts an adaptive grayscale algorithm. When performing grayscale processing, the grayscale conversion parameters adopted by this algorithm can be adaptively adjusted according to the characteristics of the image data, thereby achieving a significant compression of the data volume while retaining the image information to the greatest extent.

[0032] In step S103, the grayscale image data is filtered using median filtering, Gaussian filtering, and wavelet filtering to obtain filtered image data. Specifically, vibrations generated during fan operation, strong electromagnetic interference, and the influence of surrounding environmental factors can introduce various types of noise into the collected image data, severely impacting image quality and the accuracy of subsequent analysis results. In this embodiment, a hybrid filtering approach is used to filter the noise data in the image data. This hybrid filtering combines the advantages of median filtering, Gaussian filtering, and wavelet filtering to accurately remove noise of different types and intensities. For impulse noise caused by fan vibration, median filtering effectively removes noise points while maintaining image edge clarity. For Gaussian noise generated by electromagnetic interference, Gaussian filtering smoothes the noise according to its distribution characteristics. Finally, for subtle texture noise and other complex noise components present in the image, wavelet filtering decomposes and reconstructs the image at different scales, accurately removing noise while preserving important image details. This hybrid filtering strategy significantly improves image clarity and signal-to-noise ratio, providing reliable image data for subsequent accurate feature extraction and angle calculation.

[0033] When filtering image data, median filtering is first used to remove impulse noise, followed by Gaussian filtering to smooth the image, and finally wavelet filtering to remove subtle texture noise and other complex noise components. During the filtering process, filtering parameters can be dynamically adjusted based on the specific content and characteristics of the image data to ensure that more detailed information is retained while removing noise. Furthermore, after filtering, real-time feedback on the filtering effect is provided, allowing for dynamic adjustment of the filtering strategy to improve image quality.

[0034] In step S104, edge detection and feature and feature point identification are performed on the filtered image data. Feature point identification adopts SIFT and SURF fusion algorithm. Specifically, for the grayscale and filtered image data, edge detection is first performed on them to make the edges of the image data clearer, thereby providing an accurate geometric basis for the subsequent calculation of the pitch angle. After edge detection is performed on the image data, the features and feature points in the image data are extracted to provide a data basis for the subsequent calculation of the angle. Among them, when performing feature point identification, this embodiment adopts a fusion feature point extraction algorithm, that is, the SIFT algorithm and the SURF algorithm are fused to identify feature points. Therefore, in the feature point detection stage, the detection capabilities of the two algorithms for feature points of different scales and types can be comprehensively utilized to improve the detection quantity and quality of feature points; in the feature descriptor generation stage, the scale invariance of the SIFT algorithm and the speed and robustness of the SURF algorithm are integrated by constructing a hybrid feature descriptor. For feature recognition, a neural network model can be used for implementation.

[0035] Step S105, based on the identified feature points, the angle of the wind turbine pitch system is calculated according to a geometric relationship algorithm, or based on the identified features, the angle of the wind turbine pitch system is calculated according to a template matching algorithm. In this embodiment, two methods can be used to calculate the angle. Among them, when using geometric relationship calculation, the identified feature points are used and mapped to the three-dimensional model of the wind turbine pitch system, and then the angle is calculated based on the geometric relationship between the feature points. When using the template matching algorithm to calculate the angle, the identified features are matched with the image features in the template library, and the angle is determined based on the matching results.

[0036] The monitoring method for the wind turbine pitch system provided in the embodiment of the present invention adopts the method of image acquisition and processing of the wind turbine pitch system for monitoring and angle calculation, which overcomes many disadvantages of the traditional sensor monitoring method, and can timely and accurately grasp the operating status of the pitch system, discover potential faults in advance and take corresponding measures, effectively reduce the probability of wind turbine shutdown due to pitch system failure, ensure the safe and stable operation of the wind turbine, improve the reliability and power generation efficiency of the wind power generation system, reduce operation and maintenance costs, and have significant economic and social benefits.

[0037] Among them, in the image processing process, compared with the traditional fixed parameter grayscale method, the adaptive grayscale algorithm is adopted, which can automatically adjust the grayscale conversion parameters according to the image content, while retaining key information. It greatly compresses the data volume, reduces the computational complexity, improves the processing speed and real-time performance, and solves the problems of large data volume and low processing efficiency of traditional image analysis methods, so that the entire monitoring system can respond more quickly to changes in the state of the variable pitch system. When filtering the image, the advantages of median filtering, Gaussian filtering and wavelet filtering are combined to accurately remove various types of noise generated by the operation of the wind turbine, improve image clarity and signal-to-noise ratio, and provide reliable image data for subsequent feature extraction and angle calculation. It avoids the problem that traditional single filtering methods cannot effectively remove complex noise and affect the accuracy of subsequent analysis, and ensures the stability and reliability of the monitoring system.

[0038] In this embodiment, a method for monitoring a wind turbine pitch control system is provided, the method comprising the following steps:

[0039] Step S201, setting multiple groups of observation points in a pre-established three-dimensional model of a wind turbine pitch system; determining evaluation parameters corresponding to each group of observation points based on image data observed at each group of observation points, the image data including image data observed when the wind turbine is operating under different working conditions; and screening preliminary candidate observation points from the multiple groups of observation points based on the evaluation parameters.

[0040] Specifically, in order to enable the collected image data to capture all components of the wind turbine pitch control system in an all-round manner, it is first necessary to determine the placement of the image acquisition device. In this embodiment, the placement is determined by simulation tests and field tests.

[0041] During the simulation test, a high-precision 3D model of the wind turbine's pitch control system is first created. This model can be created using professional computer-aided design software, such as CAD. Based on the actual design parameters of the wind turbine, a high-precision 3D model of the blades, hub, and pitch control mechanism is constructed. The model reflects the shape, size, material properties, and interconnectedness of each component, providing accurate basic data for subsequent simulation analysis.

[0042] Furthermore, to evaluate the observation points during the simulation test, it is necessary to first determine the observation targets and observation parameters. The observation targets need to include key components and operating status information of the wind turbine pitch system, such as the looseness of the blade-hub connection, the motion trajectory and angle changes of the pitch mechanism transmission components, etc. The observation parameters can be used to evaluate the observation effect, such as the coverage of the observation area, the clarity of key component features, and the recognizability under different operating conditions.

[0043] Multiple groups of observation points can be determined in advance, for example, they can be determined by relevant staff based on the analysis of the wind turbine pitch system. Each group of observation points includes multiple observation points distributed at multiple positions of the wind turbine pitch model, and the wind turbine pitch system can be fully monitored through the multiple observation points in each group of observation points. Therefore, during the simulation test, each group of observation points can be arranged in the three-dimensional model in turn. In this way, the evaluation of each group of observation points can be achieved. After arranging a group of observation points in the three-dimensional model, the operation of the wind turbine under various typical working conditions can be simulated, including different wind speeds, wind directions, pitch angles, and lighting conditions. The ray tracing algorithm is used to simulate the reflection and refraction of light on the surface of the wind turbine components, as well as the image data captured by the image acquisition device at different observation point positions.

[0044] The image data obtained after arranging each set of observation points can be evaluated using pre-determined observation targets and observation parameters. For example, parameters such as coverage of the observation area, clarity of key component features, and recognizability under different operating conditions can be calculated. Data analysis algorithms and visualization techniques can then be used to identify a set of observation points that perform well on each evaluation parameter as preliminary candidate observation points. For example, cluster analysis methods can be used to group multiple groups of observation points based on similarity, and then representative solutions with high comprehensive scores on evaluation parameters can be selected from each group as preliminary candidate observation points.

[0045] Step S202: Arrange an image acquisition device in the wind turbine pitch control system according to the preliminary candidate observation points; adjust the position of the preliminary candidate observation points based on the comparison result of the image data acquired by the image acquisition device and the image data observed at the preliminary candidate observation points; determine the final observation point based on the adjusted preliminary candidate observation points to arrange the image acquisition device, and the image acquisition device is used to acquire image data of the wind turbine pitch control system.

[0046] Specifically, after preliminary candidate observation points are determined through simulation experiments, further adjustments are required through field testing. During field testing, observation equipment, such as image acquisition devices and lighting equipment, are initially installed on actual wind turbines based on the preliminary candidate observation points. During installation, the stability and safety of the equipment must be ensured, and preliminary debugging is performed to ensure that the equipment is functioning properly and capturing clear image data.

[0047] During actual wind turbine operation, the installed image acquisition device is used to collect image data of the wind turbine's variable pitch system under different operating conditions. This data is combined with other traditional monitoring methods (such as sensor monitoring data) and actual on-site records, including wind turbine operating parameters, environmental conditions (such as light intensity, temperature, humidity, wind speed), and the actual space limitations of the equipment installation location, to provide comprehensive data support for subsequent data analysis.

[0048] Compare and analyze the image data collected on-site with the simulation test results in step S101 to assess the difference between the actual observation results and the expected goals. Focus on observation points that performed well in the simulation but encountered problems in the actual field test. Analysis of possible causes may include interference factors in the actual environment that were not fully considered in the simulation, the impact of equipment installation accuracy, and the impact of small vibrations during fan operation on observation. Based on the analysis results, further optimization and adjustment of the observation point location and equipment parameters can be performed, such as fine-tuning the angle, height, or distance of the observation point, and adjusting the luminous intensity and angle of the lighting equipment.

[0049] After multiple rounds of field testing, data analysis, and optimization adjustments, a final set of observation points was selected that provided stable, clear, and comprehensive visibility of the key components and operating status of the pitch system during actual operation, with all evaluation parameters achieving optimal or near-optimal results. These final observation points will serve as the official installation locations for the subsequent long-term monitoring system, providing a reliable hardware foundation for accurate image monitoring of the wind turbine pitch system.

[0050] Therefore, this embodiment, through extensive simulation experiments and field testing, innovatively identified multiple optimal observation points. These installation locations not only enable clear, all-round, and comprehensive capture of key components of the variable pitch system, such as the connection between the blades and the hub, but also capture the dynamic changes of the variable pitch mechanism's transmission components during operation from different angles. Multi-view image acquisition enables more comprehensive component spatial position information to be obtained, laying a solid foundation for subsequent accurate 3D reconstruction and angle calculation.

[0051] Step S203: collecting image data of the wind turbine pitch control system.

[0052] Specifically, the above step S203 includes:

[0053] Step S2031 : collecting image data of the wind turbine pitch control system based on a preset image acquisition device and light source.

[0054] Step S2032: Determine the shadow distribution in the image data using an image processing algorithm.

[0055] Step S2033: Adjust the luminous intensity and angle of the light source according to the ambient light intensity and shadow distribution.

[0056] Step S2034: collecting image data of the wind turbine pitch control system based on the preset image acquisition device and the adjusted light source.

[0057] Specifically, considering the high complexity of wind turbine operating environments and the extreme instability of lighting conditions, in addition to placing an image acquisition device at the determined final observation point, this embodiment also deploys a light source, such as an LED, as illumination for image acquisition. During image data acquisition, the control device can adaptively adjust the light intensity and angle of the light source. For example, the light intensity, angle, and illumination range are automatically adjusted based on the real-time light intensity, angle, and shadow distribution of the wind turbine operating environment, thereby providing highly stable, uniform, and shadow-free lighting. During wind turbine operation, whether in direct sunlight, resulting from complex shadows formed by blades, in low-light environments caused by severe weather such as storms, heavy rain, and dust storms, or even in pitch-black nighttime operating scenarios, intelligent adjustment of the light source ensures that the image acquisition area is always optimally illuminated, effectively preventing any adverse effects of shadows on image quality. High-quality, shadow-free images significantly improve the accuracy and reliability of subsequent pitch angle recognition, providing a strong guarantee for the high-precision operation of the entire monitoring system.

[0058] An adaptive adjustment algorithm can be used to intelligently adjust the light source. First, a photoresistor or photodiode can be used as a light intensity sensor to measure ambient light intensity in real time. These sensors convert optical signals into electrical signals, which are then converted to digital signals via an analog-to-digital converter (ADC) and fed into a control device for processing. Next, an image processing algorithm is used to detect shadow areas within the image data captured by the image acquisition device. Specifically, the Canny edge detection algorithm can be used, combined with multi-scale morphological gradient operations, to enhance edge contrast and continuity and detect shadow edges within the image. The image is then segmented into multiple regions, and the shadow area of each region is calculated to determine the shadow distribution.

[0059] During light source adjustment, the control device can use pulse-width modulation technology to adjust the light source's luminous intensity. Specifically, the duty cycle of the PWM signal is dynamically adjusted based on the output of the photosensor, thereby controlling the brightness of the LED. Furthermore, the control device can use a small stepper motor or servo motor to control the angle and illumination range of the light source. Based on the shadow distribution, a control signal is sent to the motor to adjust the motor's rotation angle, thereby adjusting the angle and illumination range of the light source. During the adjustment process, the control device continuously monitors the ambient light intensity and shadow distribution from image data, adjusting the light source in real time to ensure that the light source is always in optimal condition. Furthermore, a specific adjustment strategy for adjusting the motor angle based on shadow distribution can be pre-determined, allowing adjustments to be made directly during adjustment. This adjustment strategy can also be adjusted based on the real-time shadow distribution to optimize the lighting effect.

[0060] Step S204 , performing grayscale processing on the image data based on the adaptive grayscale algorithm with adjusted parameters to obtain grayscale image data.

[0061] Specifically, the above step S204 includes:

[0062] Step S2041: Detect edge information in the image data based on the Canny edge detection algorithm combined with multi-scale morphological gradient operations. Specifically, the Canny edge detection algorithm is combined with multi-scale morphological gradient operations to enhance edge contrast and coherence and detect edge information in the image.

[0063] Step S2042: Calculate texture features of the image data. Specifically, Local Binary Patterns (LBP) or Gray-level co-occurrence matrix (GLCM) can be used to analyze the texture information of the image.

[0064] In step S2043, the image data is grayscaled using an adaptive grayscale conversion algorithm and an adaptive threshold algorithm with dynamically adjusted grayscale conversion parameters, generating grayscale image data. The grayscale conversion parameters are dynamically adjusted based on edge information and texture features. Specifically, when adjusting the grayscale conversion parameters based on edge information and texture features, the following adjustments can be made: for example, the weight of the green channel can be increased for edge regions to retain more detail information; for texture regions, the weight of the blue channel can be appropriately adjusted.

[0065] When using grayscale conversion parameters for adaptive grayscale processing, the grayscale value can be calculated using weighted average, maximum value method, minimum value method, or average value method in combination with the grayscale conversion parameters. In addition, during the grayscale processing, an adaptive threshold algorithm such as the OSTU (Otsu algorithm) can be used to dynamically calculate the threshold of each area of the image data, thereby ensuring that the grayscaled image retains details while reducing noise and blur.

[0066] The adaptive grayscale conversion algorithm employed in this paper automatically adjusts grayscale conversion parameters based on image content and features, significantly reducing data size while preserving the image's essential structural information, edge details, and texture features. Compared to traditional fixed-parameter grayscale conversion methods, this algorithm better adapts to the differences in color, material, and light reflection characteristics of various components in wind turbine pitch system images. This provides a high-quality, low-data-volume image foundation for subsequent edge detection and feature extraction, significantly improving the processing speed and real-time performance of the entire monitoring system.

[0067] Step S205: Use median filtering, Gaussian filtering and wavelet filtering to filter the grayscale image data to obtain filtered image data. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0068] Step S206: edge detection and feature and feature point recognition are performed on the filtered image data. Feature point recognition uses SIFT and SURF fusion algorithms. Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0069] Step S207: Based on the identified feature points, the angle of the wind turbine pitch system is calculated using a geometric relationship algorithm, or based on the identified features, the angle of the wind turbine pitch system is calculated using a template matching algorithm. Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0070] In this embodiment, a method for monitoring a wind turbine pitch control system is provided, the method comprising the following steps:

[0071] Step S301: Collect image data of the wind turbine pitch control system; see Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0072] Step S302: grayscale processing is performed on the image data using an adaptive grayscale algorithm based on parameter adjustment to obtain grayscale image data. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.

[0073] Step S303: Use median filtering, Gaussian filtering, and wavelet filtering to filter the grayscale image data to obtain filtered image data. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0074] Step S304: edge detection and feature and feature point recognition are performed on the filtered image data. Feature point recognition adopts SIFT and SURF fusion algorithm.

[0075] Specifically, the above step S304 includes:

[0076] Step S3041, edge information in the image data is detected based on the Canny edge detection algorithm combined with multi-scale morphological gradient operation and a local adaptive threshold algorithm. The threshold in the local adaptive threshold algorithm is adjusted based on local statistical information, edge strength and texture features of the image data.

[0077] Among them, the traditional Canny edge detection algorithm may cause inaccurate edge detection or edge breakage when processing images of wind turbine pitch systems due to problems such as noise in the image, uneven lighting, and blurred component edges. This embodiment introduces a multi-scale morphological gradient operation when performing edge detection, which can perform gradient calculations on images at different scales to enhance the contrast and coherence of edges. Specifically, when using a multi-scale morphological gradient operation, multiple gradient calculations are performed on the image in combination with structural elements of different scales (such as circular or rectangular structural elements of different sizes). In this way, it can better adapt to edges of different thicknesses and complexities and enhance the coherence of edges.

[0078] The multi-scale morphological gradient operation can be implemented using the following process:

[0079] import cv2

[0080] import numpy as np

[0081] import matplotlib.pyplot as plt

[0082] def multiscale_morphological_gradient(image, scales=[3, 5, 7]):

[0083] gradients = []

[0084] for scale in scales:

[0085] kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (scale, scale))

[0086] gradient = cv2.morphologyEx(image, cv2.MORPH_GRADIENT, kernel)

[0087] gradients.append(gradient)

[0088] return np.max(gradients, axis=0)

[0089] # Read the image

[0090] image = cv2.imread('path_to_image.jpg', cv2.IMREAD_GRAYSCALE)

[0091] # Multi-scale morphological gradient operation

[0092] gradient_image = multiscale_morphological_gradient(image)

[0093] # Display the results

[0094] plt.figure(figsize=(12, 6))

[0095] plt.subplot(1, 2, 1)

[0096] plt.imshow(image, cmap='gray')

[0097] plt.title('Original Image')

[0098] plt.axis('off')

[0099] plt.subplot(1, 2, 2)

[0100] plt.imshow(gradient_image, cmap='gray')

[0101] plt.title('Multiscale Morphological Gradient')

[0102] plt.axis('off')

[0103] plt.show()

[0104] In addition, when using the Canny edge detection algorithm combined with multi-scale morphological gradient operations for edge detection, a local adaptive threshold algorithm can also be used to adjust the parameters of the Canny algorithm (such as the low threshold and high threshold) to further optimize the edge detection results. This local adaptive threshold algorithm automatically adjusts the threshold based on the characteristics of the local area of the image, avoiding edge loss or false detection problems caused by a global fixed threshold. Therefore, through this improved edge detection algorithm, the edge contours of the pitch system components can be detected more accurately and completely. Whether it is the thin edge of the blade or the edge of the complex shape of the pitch mechanism, it can be clearly outlined, providing a highly accurate geometric foundation for the subsequent pitch angle calculation based on edge geometric features.

[0105] Among them, the Canny edge detection algorithm can be implemented by referring to the following process:

[0106] def canny_edge_detection(image, low_threshold=50, high_threshold=150):

[0107] return cv2.Canny(image, low_threshold, high_threshold)

[0108] # Apply Canny edge detection

[0109] edges = canny_edge_detection(gradient_image)

[0110] # Display the results

[0111] plt.figure(figsize=(12, 6))

[0112] plt.subplot(1, 2, 1)

[0113] plt.imshow(gradient_image, cmap='gray')

[0114] plt.title('Gradient Image')

[0115] plt.axis('off')

[0116] plt.subplot(1, 2, 2)

[0117] plt.imshow(edges, cmap='gray')

[0118] plt.title('Canny Edge Detection')

[0119] plt.axis('off')

[0120] plt.show()

[0121] Specifically, when using a local adaptive thresholding algorithm to adjust the threshold, it can incorporate local image statistics (such as mean and variance) as well as image content information such as edge strength and texture features. When adjusting the threshold based on image content, the threshold can be appropriately lowered for areas with high edge strength and increased for areas with complex texture features. This allows for better adaptation to images with varying lighting conditions and noise levels, improving edge detection performance.

[0122] The process of adjusting the threshold using local statistical information can be implemented as follows:

[0123] def adaptive_threshold(image, block_size=11, C=2):

[0124] # Calculate the local mean

[0125] local_mean = cv2.blur(image, (block_size, block_size))

[0126] # Calculate local variance

[0127] local_var = cv2.blur(image^2, (block_size, block_size)) - local_mean^2

[0128] # Calculate the threshold

[0129] threshold = local_mean + C×np.sqrt(local_var)

[0130] # Apply threshold

[0131] binary_image = np.where(image>threshold, 255, 0).astype(np.uint8)

[0132] return binary_image

[0133] # Read the image

[0134] image = cv2.imread('path_to_image.jpg', cv2.IMREAD_GRAYSCALE)

[0135] # Apply local adaptive thresholding

[0136] binary_image = adaptive_threshold(image)

[0137] # Display the results

[0138] plt.figure(figsize=(12, 6))

[0139] plt.subplot(1, 2, 1)

[0140] plt.imshow(image, cmap='gray')

[0141] plt.title('Original Image')

[0142] plt.axis('off')

[0143] plt.subplot(1, 2, 2)

[0144] plt.imshow(binary_image, cmap='gray')

[0145] plt.title('Adaptive Thresholding')

[0146] plt.axis('off')

[0147] plt.show()

[0148] When adjusting the threshold based on the local statistical information of the image (such as mean, variance) and image content information such as edge strength and texture features, the following process can be used:

[0149] def content_based_threshold(image, block_size=11, C=2):

[0150] # Calculate the local mean

[0151] local_mean = cv2.blur(image, (block_size, block_size))

[0152] # Calculate local variance

[0153] local_var = cv2.blur(image ^ 2, (block_size, block_size)) - local_mean ^ 2

[0154] # Calculate edge strength

[0155] edges = cv2.Canny(image, 50, 150)

[0156] edge_strength = cv2.blur(edges, (block_size, block_size))

[0157] # Calculate texture features

[0158] texture = cv2.Laplacian(image, cv2.CV_64F)

[0159] texture_strength = cv2.blur(np.abs(texture), (block_size, block_size))

[0160] # Calculate the threshold

[0161] threshold = local_mean + C × np.sqrt(local_var - edge_strength +texture_strength)

[0162] # Apply threshold

[0163] binary_image = np.where(image>threshold, 255, 0).astype(np.uint8)

[0164] return binary_image

[0165] # Apply thresholding based on image content

[0166] binary_image = content_based_threshold(image)

[0167] # Display the results

[0168] plt.figure(figsize=(12, 6))

[0169] plt.subplot(1, 2, 1)

[0170] plt.imshow(image, cmap='gray')

[0171] plt.title('Original Image')

[0172] plt.axis('off')

[0173] plt.subplot(1, 2, 2)

[0174] plt.imshow(binary_image, cmap='gray')

[0175] plt.title('Content - Based Adaptive Thresholding')

[0176] plt.axis('off')

[0177] plt.show()

[0178] Step S3042: The SIFT algorithm and the SURF algorithm are respectively used to detect and merge feature points in the image data. Specifically, when performing feature point recognition, the SIFT algorithm and the SURF algorithm are respectively used to detect feature points in the image data, and then the detected feature points are merged and duplicate feature points are removed. The process of detecting feature points in the image data using the SIFT algorithm and the SURF algorithm can be implemented by referring to the following process:

[0179] def combined_feature_detection(image):

[0180] keypoints_sift, descriptors_sift = sift_feature_detection(image)

[0181] keypoints_surf, descriptors_surf = surf_feature_detection(image)

[0182] # Merge feature points

[0183] keypoints = keypoints_sift + keypoints_surf

[0184] # Remove duplicate feature points

[0185] keypoints = list(set(keypoints))

[0186] return keypoints, (descriptors_sift, descriptors_surf)

[0187] Traditional SIFT and SURF algorithms may suffer from inaccurate feature point extraction or insufficiently robust feature descriptors in certain situations. This embodiment organically combines the advantages of the SIFT and SURF algorithms. During the feature point detection phase, the two algorithms' ability to detect feature points of different scales and types is comprehensively utilized to improve the number and quality of feature point detection.

[0188] Step S3043, based on the merged feature points, the separately calculated SIFT feature descriptors and SURF feature descriptors are fused to obtain a fused feature descriptor; specifically, for each merged feature point, the SIFT and SURF descriptors are calculated separately, and then the two descriptors are fused. The fusion method can be a simple splicing, or a weighted average or other more complex fusion method. Among them, the descriptor generated by the SIFT or SURF algorithm is usually a vector of a fixed length, for example, the SIFT descriptor length is 128, and the SURF descriptor length is 64 or 128. The fused feature descriptor can be determined by splicing or weighted averaging of individual descriptors, and the length of the fused feature descriptor generated is the sum of the lengths of the SIFT and SURF descriptors (for example, 192), or a descriptor of the same length can be generated by weighted averaging.

[0189] Among them, the generation of the fused feature descriptor can be achieved by referring to the following process:

[0190] def combined_feature_descriptor(keypoints, descriptors_sift,descriptors_surf):

[0191] combined_descriptors = []

[0192] for kp in keypoints:

[0193] sift_desc = descriptors_sift[keypoints.index(kp)]

[0194] surf_desc = descriptors_surf[keypoints.index(kp)]

[0195] # Splicing descriptors

[0196] combined_desc = np.concatenate((sift_desc, surf_desc))

[0197] # or weighted average

[0198] # combined_desc = 0.5 × sift_desc + 0.5 × surf_desc

[0199] combined_descriptors.append(combined_desc)

[0200] return np.array(combined_descriptors)

[0201] In this embodiment, the fused feature descriptor combines the scale invariance of the SIFT algorithm and the speed and robustness of the SURF algorithm, thereby improving the robustness and recognition of the feature descriptor.

[0202] Step S3044, the feature points are screened according to the local geometric features and texture features of the wind turbine pitch system, and the identified feature points are obtained based on the screened feature points and feature descriptors; specifically, for the identified feature points, a local feature constraint mechanism can be introduced to screen and optimize the feature points according to the local geometric structure and texture features of the pitch system components, to ensure that the extracted feature points have stronger recognition and stability at different pitch angles.

[0203] Among them, the local feature constraint mechanism sets constraints based on the local geometric structure and texture characteristics of the pitch system components, such as the distribution density of feature points, the distance between feature points and edges, etc., to screen feature points. For example, when screening based on edge distance, the distance between feature points and edges should be less than a certain threshold to ensure that feature points are located near the edge. When screening based on texture features, the texture intensity of feature points should be less than a certain threshold to ensure that feature points are located in areas with weaker textures and reduce the impact of noise. When screening based on distribution density, the distribution density of feature points should be moderate to avoid overly dense or sparse feature points and ensure a uniform distribution of feature points.

[0204] The local feature constraint mechanism can be implemented by referring to the following process:

[0205] def filter_keypoints(keypoints, image, edge_threshold=10, texture_threshold= 10):

[0206] filtered_keypoints = []

[0207] edges = cv2.Canny(image, 50, 150)

[0208] texture = cv2.Laplacian(image, cv2.CV_64F)

[0209] for kp in keypoints:

[0210] x, y = kp.pt

[0211] # Check the distance between feature points and edges

[0212] if edges[int(y), int(x)]>edge_threshold:

[0213] continue

[0214] # Check the distance between feature points and texture

[0215] if np.abs(texture[int(y), int(x)])>texture_threshold:

[0216] continue

[0217] filtered_keypoints.append(kp)

[0218] return filtered_keypoints

[0219] Specifically, the feature point extraction process can be implemented by referring to the following process:

[0220] def combined_feature_extraction(image):

[0221] # Feature point detection

[0222] keypoints, (descriptors_sift, descriptors_surf) = combined_feature_

[0223] detection(image)

[0224] # Feature descriptor generation

[0225] combined_descriptors = combined_feature_descriptor(keypoints,

[0226] descriptors_sift, descriptors_surf)

[0227] # Local feature constraint screening

[0228] filtered_keypoints = filter_keypoints(keypoints, image)

[0229] return filtered_keypoints, combined_descriptors

[0230] # Read the image

[0231] image = cv2.imread('path_to_image.jpg', cv2.IMREAD_GRAYSCALE)

[0232] # Apply fusion feature point extraction algorithm

[0233] keypoints, descriptors = combined_feature_extraction(image)

[0234] # Display the results

[0235] image_with_keypoints = cv2.drawKeypoints(image, keypoints, None,color=(0, 255, 0))

[0236] cv2.imshow('Keypoints', image_with_keypoints)

[0237] cv2.waitKey(0)

[0238] cv2.destroyAllWindows()

[0239] Step S3045, a deep convolutional neural network is used to extract features of the image data. Specifically, a deep convolutional neural network model is used to extract features of the image. These features can be high-level semantic features of the image, rather than simple pixel information. For example, a pre-trained CNN model (such as ResNet, VGG, etc.) can be used to extract features. Among them, for the training process of the CNN model, learning can be carried out through a large amount of image data (including images with different lighting, noise and deformation), so that the model can automatically learn feature representations that are insensitive to these changes. This learning process enables the model to better recognize and match image content when facing new images, even if the image has noise, deformation or lighting changes.

[0240] When using a pre-trained CNN model for feature extraction, deep-level image features are gradually extracted through structures such as convolutional layers, pooling layers, and activation functions. These features include not only low-level information such as edges and textures, but also high-level information such as object shape and structure. High-level semantic features are insensitive to image noise and lighting variations, and therefore can more accurately represent image content.

[0241] Furthermore, when training models with large amounts of image data, data augmentation and regularization techniques can be used to process the image data, further improving the model's robustness to noise, deformation, and illumination changes. Data augmentation involves performing random transformations on image data (such as rotation, scaling, cropping, and color adjustment) to generate more diverse training samples, enabling the model to learn features under more varied conditions. For example, by randomly adjusting the brightness and contrast of an image, the model can better adapt to varying lighting conditions. Regularization techniques (such as dropout and L2 regularization) applied to image data can prevent overfitting and improve the model's generalization capabilities. Regularization techniques help the model maintain good performance even when presented with unseen images.

[0242] Step S305: Based on the identified feature points, the angle of the wind turbine pitch system is calculated using a geometric relationship algorithm, or based on the identified features, the angle of the wind turbine pitch system is calculated using a template matching algorithm. It should be noted that when calculating the angle, either the geometric relationship algorithm or the template matching algorithm can be selected based on the specific application scenario. Alternatively, the angles can be calculated using both algorithms, and the angles calculated by the two algorithms can be averaged or weighted averaged to obtain the final angle calculation result.

[0243] Specifically, when a geometric algorithm is used to calculate the angle, step S305 includes:

[0244] Step S3051: Train a deep neural network model based on the calibration images and camera parameters to obtain a camera parameter prediction model. Methods for calculating pitch angles based on geometric relationships require precise knowledge of the camera's internal parameters (such as focal length, principal point coordinates, etc.) and external parameters (such as installation position and posture). These parameters are often difficult to accurately measure and calibrate in practice, and as the wind turbine's operating time increases, the camera may experience slight displacement or posture changes, resulting in inaccurate parameters. In this embodiment, a deep neural network model is trained by collecting a large number of calibration images and corresponding real-world camera parameter data. This model can automatically predict the camera's internal and external parameters based on image features and optimize and adjust them in real time.

[0245] Specifically, the calibration image is an image used to calibrate camera parameters, and usually contains a calibration plate of known geometric shape and position (such as a chessboard or dot array). In this embodiment, the calibration image for training can be collected in the following manner: use a camera to capture images of the calibration plate at different angles and positions of the wind turbine pitch system. Ensure that the calibration plate covers the entire field of view and capture it under different lighting conditions to obtain a variety of calibration images. The real camera parameter data includes the camera's internal parameters (such as focal length, principal point coordinates, distortion coefficient) and external parameters (such as the camera's installation position and posture). In this embodiment, the initial camera parameters for the real camera parameters used for training can be obtained through traditional camera calibration methods (such as Zhang Zhengyou's calibration method). Then, these parameters are optimized and calibrated using a deep learning model.

[0246] When training the camera parameter prediction model, we use calibrated images and real camera parameter data as the sample set for training the deep learning model. That is, by training a deep neural network model with a large number of calibrated images and the corresponding real camera parameter data, the trained camera parameter prediction model can automatically predict the camera's internal and external parameters based on the features in the image.

[0247] Step S3052 matches the identified feature points with the pre-built geometric model of the wind turbine pitch system to obtain a matching result. Specifically, when using a geometric algorithm to calculate angles, a geometric model of the wind turbine pitch system must first be constructed. This model includes information such as the geometric shape, installation position, and attitude of the wind turbine blades. When constructing the geometric model, the following geometric parameters are defined: blade length L, blade width W, distance R from the blade root to the center, pitch axis position (x0, y0), and pitch angle θ. It is then assumed that the blade rotates within a two-dimensional plane, with the pitch axis located at the blade root. The blade's geometric shape can be simplified to a rectangle, with vertex coordinates: (x0, y0), (x0 + Lcos(θ), y0 + Lsin(θ)), (x0 + Lcos(θ) − Wsin(θ), y0 + Lsin(θ) + Wcos(θ)), (x0 − Wsin(θ), y0 + Lsin(θ) + Wcos(θ)), (x0 − Wsin(θ), y0 + Wcos(θ)). This results in a geometric model of the wind turbine pitch system. The FLANN matcher is then used to match the identified feature points with points in the geometric model. For example, the feature points on the blade edge are matched with points on the blade edge in the geometric model. The specific matching process can be implemented as follows:

[0248] def match_keypoints(keypoints1, descriptors1, keypoints2,descriptors2):

[0249] flann = cv2.FlannBasedMatcher_create()

[0250] matches = flann.knnMatch(descriptors1, descriptors2, k=2)

[0251] good_matches = []

[0252] for m, n in matches:

[0253] if m.distance<0.7 × n.distance:

[0254] good_matches.append(m)

[0255] return good_matches

[0256] In step S3053, based on the matching results, the 2D feature points are converted into 3D geometric points using a perspective transformation algorithm and camera parameters predicted by a camera parameter prediction model. Specifically, the 2D feature points to be converted can be those that have been successfully matched after matching the identified feature points with the geometric model. During the conversion, the camera parameter prediction model is used to predict the intrinsic and extrinsic parameters of the camera (i.e., the image acquisition device). The camera intrinsic parameters are then used to convert the pixel coordinates of the 2D feature points into normalized coordinates in the camera coordinate system. Finally, the perspective transformation algorithm and the camera extrinsic parameters are combined to convert the normalized coordinates into 3D geometric points.

[0257] Among them, the perspective transformation matrix H in the perspective transformation algorithm is calculated as follows:

[0258] import cv2

[0259] import numpy as np

[0260] def compute_perspective_transform(src_pts, dst_pts):

[0261] H, _ = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)

[0262] return H

[0263] In addition, the process of converting using the perspective transformation algorithm and the camera internal parameter matrix K and external parameter matrix [R|t] is as follows:

[0264] def perspective_transform(points, camera_matrix, dist_coeffs, rvec,tvec):

[0265] points_3d = cv2.undistortPoints(points, camera_matrix, dist_coeffs)

[0266] points_3d = cv2.perspectiveTransform(points_3d, np.hstack((rvec,tvec)))

[0267] return points_3d

[0268] Step S3054: Calculate the angle of the wind turbine pitch system based on the three-dimensional geometric points and trigonometric functions. Specifically, after converting the three-dimensional geometric points, the angle between the two three-dimensional geometric points is calculated using their coordinates in three-dimensional space, thereby obtaining the wind turbine pitch angle. The angle calculation process can be implemented as follows:

[0269] def compute_pitch_angle_optimized(p1, p2, camera_matrix, dist_coeffs,rvec, tvec):

[0270] # 3D reconstruction

[0271] p1_3d = perspective_transform(p1, camera_matrix, dist_coeffs, rvec,tvec)

[0272] p2_3d = perspective_transform(p2, camera_matrix, dist_coeffs, rvec,tvec)

[0273] # Calculate vector

[0274] v = p2_3d - p1_3d

[0275] # Calculate pitch angle

[0276] angle = np.arctan2(v[1], v[0]) × 180 / np.pi

[0277] return angle

[0278] In addition, in other implementations, the angle calculation may also be performed according to the following process:

[0279] def compute_pitch_angle(p1, p2):

[0280] # Calculate vector

[0281] v = p2 - p1

[0282] # Calculate pitch angle

[0283] angle = np.arctan2(v[1], v[0]) × 180 / np.pi

[0284] return angle

[0285] Specifically, feature point extraction, matching, perspective transformation, 3D reconstruction, and angle calculation can be implemented as follows:

[0286] import cv2

[0287] import numpy as np

[0288] # Read the image

[0289] image = cv2.imread('path_to_image.jpg', cv2.IMREAD_GRAYSCALE)

[0290] # Feature point extraction

[0291] keypoints, descriptors = combined_feature_extraction(image)

[0292] # Feature point matching

[0293] template_keypoints, template_descriptors = combined_feature_extraction (template_image)

[0294] matches = match_keypoints(keypoints, descriptors, template_keypoints,template_descriptors)

[0295] Perspective Transformation

[0296] src_pts = np.float32([keypoints[m.queryIdx].pt for m in matches]).reshape(-1, 1, 2)

[0297] dst_pts = np.float32([template_keypoints[m.trainIdx].pt for m inmatches]).reshape(-1, 1, 2)

[0298] H = compute_perspective_transform(src_pts, dst_pts)

[0299] # 3D reconstruction

[0300] camera_matrix, dist_coeffs, rvec, tvec = load_camera_parameters()

[0301] p1_3d = perspective_transform(src_pts[0], camera_matrix, dist_coeffs,rvec, tvec)

[0302] p2_3d = perspective_transform(src_pts[1], camera_matrix, dist_coeffs,rvec, tvec)

[0303] # Calculate pitch angle

[0304] angle = compute_pitch_angle_optimized(p1_3d, p2_3d, camera_matrix,dist_coeffs, rvec, tvec)

[0305] print(f"pitch angle: {angle} degrees"

[0306] Specifically, when the template matching algorithm is used to calculate the angle, the above step S305 includes:

[0307] Step S3055: Match the features with the image features of the template images in the preset template library. The preset template library includes multiple template images, image features and corresponding pitch angles. The preset template library is dynamically updated.

[0308] Step S3056, determine the angle of the wind turbine pitch system based on the matching results, the matching results include a single template image or multiple template images that are successfully matched. When there are multiple template images, the angle of the wind turbine pitch system is the fusion value of the pitch angles corresponding to the multiple template images.

[0309] Specifically, before employing the template matching algorithm, a template library must be constructed. This template image library contains multiple template images and their corresponding pitch angles. These template images were captured at different pitch angles, and their pitch angles are known. For example: Template image 1, pitch angle 1 degree; Template image 2, pitch angle 2 degrees; Template image 3, pitch angle 3 degrees; Template image 10, pitch angle 10 degrees.

[0310] After building the template library, the acquired image features are matched with the features in the template image library. Cosine similarity or other similarity metrics can be used to calculate the similarity between features. Based on the matching results, the most similar template image is found and its corresponding pitch angle is obtained. For some complex pitch states, a multi-template fusion matching strategy can be used to determine the fusion value of the pitch angle. For example, multiple similar template images can be found and their corresponding angles can be averaged, or fused using methods such as weighted averaging to reduce the error caused by single template matching and improve the accuracy and reliability of angle calculation.

[0311] Specifically, when using weighted averaging, the angles of each template can be weighted averaged based on similarity. The weight can be a function of similarity, with higher similarity giving higher weights. This weighted averaging method can better utilize information from multiple templates and improve the accuracy of angle calculation.

[0312] Furthermore, to enhance the system's adaptability and robustness, the template image library can be dynamically updated. During wind turbine operation, the template image library is continuously updated based on real-time image features and known pitch angles, enriching its content. This dynamic update mechanism allows the template image to learn image features from a wider range of changing conditions, further improving robustness to noise, deformation, and illumination variations. This allows the system to adapt to varying operating states and environmental conditions, thereby enhancing matching accuracy.

[0313] When the template library is dynamically updated, if the wind turbine operates under different lighting conditions or in different seasons (such as summer and winter), the template image library will include image features from these different conditions, thereby improving matching accuracy and robustness. Furthermore, if a template performs poorly in multiple matches, it can be adjusted or replaced to improve overall performance. Furthermore, during dynamic updates, the contents of the template image library can be dynamically adjusted in real time based on collected image features and pitch angle data. This real-time feedback mechanism ensures that the template image library always contains the latest and most relevant image features, thereby improving matching accuracy and robustness.

[0314] This invention utilizes industrial-grade CCD or CMOS cameras and multiple observation points for image acquisition. These cameras boast ultra-high resolution, excellent low-light performance, and fast frame rates, enabling precise capture of subtle structural and component details while adapting to complex lighting and dynamic operating conditions. Multiple optimal observation points, determined through simulation and field testing, enable comprehensive, in-depth, and intuitive monitoring of the variable pitch system's operating status. This allows for more accurate identification of potential faults, such as minor blade wear and loose connections. This effectively addresses the single-dimensional nature of traditional sensors, which struggle to fully reflect the overall system's operating status. This provides sufficient and accurate data support and decision-making basis for refined wind turbine operation and maintenance.

[0315] The present invention automatically adjusts the light source based on real-time lighting conditions, ensuring that the image acquisition area is always optimally illuminated and avoiding shadow interference. This significantly improves image quality, thereby enhancing the accuracy and reliability of subsequent pitch angle recognition. This overcomes the impact of unstable lighting in the wind turbine operating environment on monitoring, ensuring the high-precision operation of the monitoring system, avoiding shadow interference, and improving image quality, resolving the issue of traditional image acquisition being significantly affected by lighting. This provides a strong guarantee for subsequent pitch angle recognition.

[0316] In the present invention, an adaptive grayscale algorithm is adopted: compared with the traditional fixed parameter grayscale method, it can automatically adjust the grayscale conversion parameters according to the image content, greatly compress the data volume while retaining key information, reduce the computational complexity, improve the processing speed and real-time performance, and solve the problems of large data volume and low processing efficiency of traditional image analysis methods, so that the entire monitoring system can respond to changes in the state of the variable pitch system more quickly.

[0317] In the present invention, the advantages of median filtering, Gaussian filtering and wavelet filtering are combined to accurately remove various types of noise generated by the operation of the fan, improve image clarity and signal-to-noise ratio, and provide reliable image data for subsequent feature extraction and angle calculation. This avoids the problem that traditional single filtering methods cannot effectively remove complex noise and affect the accuracy of subsequent analysis, thereby ensuring the stability and reliability of the monitoring system.

[0318] In the present invention, when using the Canny edge detection algorithm for edge detection, multi-scale morphological gradient operations and adaptive threshold selection strategies are introduced to enhance edge contrast and continuity, avoid edge inaccuracy or breakage, and provide a highly accurate geometric basis for pitch angle calculation based on edge geometric features. This solves the problem of inaccurate edge detection caused by factors such as noise and uneven illumination when processing wind turbine pitch system images using traditional edge detection algorithms, thereby improving the accuracy of pitch angle calculation.

[0319] In the present invention, by fusing the feature point extraction algorithm, organically combining the advantages of SIFT and SURF algorithms, optimizing the feature point detection and descriptor generation stages, introducing a local feature constraint mechanism, the quantity and quality of feature point detection, as well as the recognition and stability at different pitch angles, can be improved, the pitch angle can be determined more accurately, and the accuracy and reliability of the monitoring system can be improved, thus overcoming the problems of inaccurate feature point extraction or insufficiently robust feature descriptors existing in traditional feature point extraction algorithms.

[0320] In the present invention, the camera parameters are automatically predicted and adjusted in real time through a deep learning model, and the pitch angle is calculated by combining an improved perspective transformation algorithm and optimized trigonometric functions. This overcomes the problems of inaccurate parameters and poor environmental adaptability in traditional geometric relationship calculation methods, achieves high-precision and adaptive pitch angle calculation, improves the accuracy and stability of angle calculation, and adapts to the displacement and posture changes that may occur in the camera during long-term operation of the wind turbine.

[0321] In the present invention, a dynamically updated template library is constructed, a deep learning image feature matching algorithm is adopted and a multi-template fusion matching strategy is introduced to effectively overcome the influence of image noise, deformation and illumination changes, and more accurately find matching templates to determine the pitch angle, providing an efficient and accurate solution for the angle calculation of the wind turbine pitch system, solving the problem that traditional template matching methods are difficult to adapt to the image changes of the wind turbine pitch system under different operating states and environmental conditions, and improving the accuracy and reliability of angle calculation.

[0322] In the present invention, through the innovative application of the above-mentioned series of technical means, non-contact, high-precision, multi-dimensional monitoring of the wind turbine pitch system is achieved, overcoming many disadvantages of traditional sensor monitoring methods, and being able to timely and accurately grasp the operating status of the pitch system, discover potential faults in advance and take corresponding measures, effectively reducing the probability of wind turbine shutdown due to pitch system failure, ensuring the safe and stable operation of the wind turbine, improving the reliability and power generation efficiency of the wind power generation system, reducing operation and maintenance costs, and having significant economic and social benefits.

[0323] This embodiment also provides a monitoring device for a wind turbine pitch control system, which is used to implement the above-mentioned embodiments and preferred embodiments. Details that have already been described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0324] This embodiment provides a monitoring device for a wind turbine pitch control system, such as Figure 2 Shown, including:

[0325] Image acquisition module 21, used to collect image data of the wind turbine pitch control system;

[0326] A grayscale processing module 22 is configured to perform grayscale processing on the image data based on an adaptive grayscale algorithm with parameter adjustment to obtain grayscale image data;

[0327] The filtering module 23 is used to filter the grayscale image data using median filtering, Gaussian filtering, and wavelet filtering to obtain filtered image data;

[0328] The feature recognition module 24 is used to perform edge detection and feature and feature point recognition on the filtered image data. Feature point recognition adopts SIFT and SURF fusion algorithm;

[0329] The angle calculation module 25 is used to calculate the angle of the wind turbine pitch system based on the identified feature points according to a geometric relationship algorithm, or to calculate the angle of the wind turbine pitch system based on the identified features according to a template matching algorithm.

[0330] The further functional description of each of the above modules is the same as that of the above corresponding embodiments and will not be repeated here.

[0331] The embodiment of the present invention also provides a computer device having the above Figure 2 The monitoring device of the wind turbine pitch system is shown.

[0332] See also Figure 3 , Figure 3 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 3 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 3In the example, a processor 10 is used. The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof. The memory 20 stores instructions executable by at least one processor 10, so as to enable the at least one processor 10 to implement the method described in the above embodiment.

[0333] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created based on the use of a computer device to display a small program landing page, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned networks include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof. The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 may also include a combination of the above-mentioned types of memory. The computer device also includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0334] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0335] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0336] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A monitoring method for a wind turbine pitch control system, characterized in that: The method comprises: Collect image data of wind turbine pitch control system; Performing grayscale processing on the image data based on an adaptive grayscale algorithm with parameter adjustment to obtain grayscale image data; The grayscale image data is filtered using median filtering, Gaussian filtering and wavelet filtering to obtain filtered image data; Performing edge detection and feature and feature point recognition on the filtered image data, wherein the feature point recognition adopts SIFT and SURF fusion algorithm; Based on the identified feature points, the angle of the wind turbine pitch system is calculated according to a geometric relationship algorithm, or based on the identified features, the angle of the wind turbine pitch system is calculated according to a template matching algorithm; Before collecting image data of the wind turbine pitch control system, the method further includes: Setting multiple groups of observation points in a pre-established three-dimensional model of a wind turbine pitch system; Determining evaluation parameters corresponding to each group of observation points based on image data observed at each group of observation points, wherein the image data includes image data observed when the wind turbine is running under different operating conditions; Screening the plurality of observation points according to the evaluation parameters to obtain preliminary candidate observation points; Arranging an image acquisition device in the wind turbine pitch control system according to the preliminary candidate observation points; Adjusting the position of the preliminary candidate observation point based on a comparison result between the image data collected by the image acquisition device and the image data observed at the preliminary candidate observation point; The final observation point is determined based on the adjusted preliminary candidate observation point to arrange the image acquisition device, which is used to acquire image data of the wind turbine pitch system. The image acquisition device is an industrial-grade CCD or industrial-grade CMOS.

2. The method according to claim 1, characterized in that Collect image data of the wind turbine pitch system, including: Collect image data of the wind turbine pitch system based on a pre-set image acquisition device and light source; Determining the distribution of shadows in the image data using an image processing algorithm; Adjusting the luminous intensity and angle of the light source according to the ambient light intensity and shadow distribution; Image data of the wind turbine pitch control system is collected based on a preset image acquisition device and an adjusted light source.

3. The method according to claim 1, characterized in that Grayscale processing is performed on the image data using an adaptive grayscale algorithm based on parameter adjustment to obtain grayscale image data, including: Detecting edge information in the image data based on the Canny edge detection algorithm combined with multi-scale morphological gradient operation; Calculating texture features of the image data; The image data is grayscaled based on an adaptive grayscale conversion algorithm and an adaptive threshold algorithm with dynamic adjustment of grayscale conversion parameters to obtain grayscale image data. The grayscale conversion parameters are dynamically adjusted based on the edge information and texture features.

4. The method according to claim 1, wherein Performing edge detection and feature and feature point recognition on the filtered image data, including: Detecting edge information in the image data based on a Canny edge detection algorithm combined with a multi-scale morphological gradient operation and a local adaptive threshold algorithm, wherein the threshold in the local adaptive threshold algorithm is adjusted based on local statistical information, edge strength, and texture features of the image data; SIFT algorithm and SURF algorithm are used to detect feature points in image data and merge them; Based on the merged feature points, the separately calculated SIFT feature descriptors and SURF feature descriptors are fused to obtain the fused feature descriptor; Feature points are screened according to the local geometric features and texture features of the wind turbine pitch system, and the identified feature points are obtained based on the screened feature points and feature descriptors; A deep convolutional neural network is used to extract features from image data.

5. The method according to claim 1, characterized in that Based on the identified feature points, the angle of the wind turbine pitch system is calculated using a geometric relationship algorithm, including: The deep neural network model is trained based on the calibration image and camera parameters to obtain the camera parameter prediction model; Match the identified feature points with the pre-built geometric model of the wind turbine pitch system to obtain a matching result; Based on the matching results, the perspective transformation algorithm and the camera parameters predicted by the camera parameter prediction model are used to convert the two-dimensional feature points into three-dimensional geometric points; Calculate the angle of the wind turbine pitch system based on 3D geometric points and trigonometric functions.

6. The method according to claim 1, characterized in that Based on the identified features, the angle of the wind turbine pitch system is calculated using a template matching algorithm, including: Matching the feature with an image feature of a template image in a preset template library, wherein the preset template library includes a plurality of template images, image features, and corresponding pitch angles, and the preset template library is dynamically updated; The angle of the wind turbine pitch system is determined based on the matching results, which include a single template image or multiple template images that are successfully matched. When there are multiple template images, the angle of the wind turbine pitch system is the fusion value of the pitch angles corresponding to the multiple template images.

7. A monitoring device for a wind turbine pitch control system, characterized in that: The device comprises: Image acquisition module, used to collect image data of the wind turbine pitch control system; A grayscale processing module, configured to perform grayscale processing on the image data based on an adaptive grayscale algorithm with parameter adjustment to obtain grayscale image data; A filtering module is used to filter the grayscale image data using median filtering, Gaussian filtering, and wavelet filtering to obtain filtered image data; A feature recognition module is used to perform edge detection and feature and feature point recognition on the filtered image data, wherein the feature point recognition adopts SIFT and SURF fusion algorithm; An angle calculation module, used to calculate the angle of the wind turbine pitch system based on the identified feature points according to a geometric relationship algorithm, or to calculate the angle of the wind turbine pitch system based on the identified features according to a template matching algorithm; The device also includes: An image acquisition device arrangement module is used to set multiple groups of observation points in a pre-established three-dimensional model of a wind turbine pitch control system; determine evaluation parameters corresponding to each group of observation points based on image data observed at each group of observation points, the image data including image data observed when the wind turbine is operating under different working conditions; screen preliminary candidate observation points from multiple groups of observation points based on the evaluation parameters; arrange image acquisition devices in the wind turbine pitch control system based on the preliminary candidate observation points; adjust the position of the preliminary candidate observation points based on a comparison result between the image data collected by the image acquisition device and the image data observed at the preliminary candidate observation points; determine the final observation points based on the adjusted preliminary candidate observation points to arrange the image acquisition device, the image acquisition device is used to collect image data of the wind turbine pitch control system, and the image acquisition device is an industrial-grade CCD or industrial-grade CMOS.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the monitoring method of the wind turbine pitch system according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the wind turbine pitch control system monitoring method according to any one of claims 1 to 6.

Images