Wind power plant equipment centimeter-level three-dimensional reconstruction method and system

Abstract

The invention discloses a centimeter-level three-dimensional reconstruction method and system for wind power plant equipment, and the method comprises the steps: collecting multi-source data of a wind power plant, carrying out the data fusion, and inputting the fused data into a pre-constructed three-dimensional reconstruction model for processing. The model integrates target identification and segmentation of deep learning, a voxel-based three-dimensional reconstruction algorithm and a texture mapping and detail enhancement technology, firstly identifies and segments wind power plant equipment, and finally generates a high-precision three-dimensional model by adjusting the size of voxels and enhancing texture details. According to the method, centimeter-level three-dimensional reconstruction of the wind power plant equipment is realized, and more accurate data support is provided for operation and maintenance management of the wind power plant.

Description

Technical Field

[0001] The invention belongs to the technical field of wind power generation, and relates to a centimeter-level three-dimensional reconstruction method and system for wind farm equipment. Background Art

[0002] The 3D reconstruction technology of wind farm equipment refers to the process of obtaining geometric and visual information of various equipment in the wind farm through various measurement methods and converting this information into an accurate 3D digital model. This technology plays an important role in all stages of wind farm planning, construction, operation and maintenance. Through 3D reconstruction, accurate geometric measurement, status monitoring and performance analysis of wind power equipment can be performed, providing important data support for efficient operation and intelligent management of wind farms.

[0003] At present, the 3D reconstruction of wind farm equipment mainly relies on traditional technologies such as laser scanning and photogrammetry. These methods can achieve good reconstruction results under ideal conditions, but they still face many challenges in practical applications. For example, when dealing with large-scale and complex wind farms, traditional 3D reconstruction technology often has problems such as low data acquisition efficiency, unstable reconstruction accuracy, and poor environmental adaptability. Especially in severe weather conditions or poor lighting conditions, it is difficult for existing technologies to guarantee the accuracy and integrity of reconstruction. In addition, existing technologies also face huge challenges when dealing with the 3D reconstruction of dynamic components (such as rotating wind turbine blades).

[0004] Therefore, how to achieve high-precision, high-efficiency, all-weather 3D reconstruction of wind power equipment in a complex and ever-changing wind farm environment has become a technical problem that needs to be solved urgently. In particular, how to improve the reconstruction accuracy to the centimeter level while ensuring stability and reliability under various environmental conditions is the key to promoting intelligent management and refined operation and maintenance of wind farms. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problem in the prior art of how to achieve high-precision, high-efficiency, all-weather three-dimensional reconstruction of wind power equipment in a complex and changeable wind farm environment, while improving the reconstruction accuracy to the centimeter level, and to provide a centimeter-level three-dimensional reconstruction method and system for wind farm equipment.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: A first aspect of the present invention provides a centimeter-level three-dimensional reconstruction method for wind farm equipment, comprising the following steps: S1, collects multi-source data of wind farms; S2, fusing the collected multi-source data of the wind farm to obtain fused data; S3, inputting the fused data into a pre-established three-dimensional reconstruction model to obtain a three-dimensional model of the wind farm equipment; the pre-established three-dimensional reconstruction model includes: a target recognition and segmentation module based on deep learning, a three-dimensional reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; the fused data is processed by the deep learning target recognition and segmentation module to obtain a first three-dimensional model of the wind farm equipment; the first three-dimensional model is adjusted in voxel size by the voxel-based three-dimensional reconstruction algorithm module and then the details are enhanced by the texture mapping and detail enhancement module to obtain the three-dimensional model of the wind farm equipment.

[0007] A second aspect of the present invention provides a centimeter-level three-dimensional reconstruction system for wind farm equipment, comprising: Data collection layer, used to collect multi-source data of wind farms; The data processing layer is used to fuse the collected multi-source data of the wind farm to obtain fused data; The three-dimensional reconstruction layer is used to input the fused data into a pre-established three-dimensional reconstruction model to obtain a three-dimensional model of the wind farm equipment; the pre-established three-dimensional reconstruction model includes: a target recognition and segmentation module based on deep learning, a three-dimensional reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; the fused data is processed by the deep learning target recognition and segmentation module to obtain a first three-dimensional model of the wind farm equipment; the first three-dimensional model is adjusted in voxel size by the voxel-based three-dimensional reconstruction algorithm module and then the details are enhanced by the texture mapping and detail enhancement module to obtain the three-dimensional model of the wind farm equipment.

[0008] Furthermore, the centimeter-level three-dimensional reconstruction system for wind farm equipment also includes an application layer, and the application layer is used to display and apply the three-dimensional reconstruction results.

[0009] Furthermore, the data acquisition layer includes a high-precision laser radar system, a multi-spectral camera array, a millimeter-wave radar, an oblique photography system carried by a drone, and a ground mobile measurement vehicle; The data processing layer includes a high-performance computing cluster, a point cloud data processing module, an image processing and feature extraction module, and a multi-source data fusion engine; The 3D reconstruction layer includes a target recognition and segmentation module based on deep learning, a 3D reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; The application layer includes a web-based 3D visualization platform, an equipment health monitoring system, a predictive maintenance analysis module, and a virtual reality or augmented reality interface; The data acquisition layer, data processing layer, three-dimensional reconstruction layer and application layer are interconnected through a high-speed data bus and exchange data.

[0010] Furthermore, the multi-source data fusion engine of the data processing layer adopts a data fusion algorithm based on deep learning to perform time synchronization, spatial registration and feature-level fusion of data from different sensors.

[0011] Furthermore, the voxel-based 3D reconstruction algorithm module of the 3D reconstruction layer adopts an adaptive voxel size strategy to dynamically adjust the voxel size according to the target complexity.

[0012] Furthermore, the data acquisition layer also includes an edge computing module, which is deployed on the equipment of the data acquisition layer to perform preliminary processing and compression on the collected data.

[0013] Furthermore, the data acquisition layer also includes an adaptive multi-scale scanning strategy module, which automatically adjusts the scanning density and accuracy according to the complexity and importance of the equipment.

[0014] Furthermore, the three-dimensional reconstruction layer also includes an AI-driven automatic defect detection module, which automatically identifies, locates and labels surface defects of the reconstructed three-dimensional model.

[0015] Furthermore, the three-dimensional reconstruction layer also includes an environmental factor adaptive algorithm module, and the environmental factor adaptive algorithm module automatically adjusts the three-dimensional reconstruction parameters according to environmental factors.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention discloses a centimeter-level three-dimensional reconstruction method for wind farm equipment. By fusing multi-source data (including but not limited to laser radar scanning, drone aerial photography, ground photogrammetry, etc.), the method can achieve high-precision three-dimensional reconstruction of wind farm equipment, achieving centimeter-level accuracy. This high precision not only helps to accurately measure the size and position of the equipment, but also provides a solid data foundation for subsequent wind farm operation and maintenance, asset management and safety assessment; the introduction of a target recognition and segmentation module based on deep learning greatly improves the automation level of data processing. The deep learning algorithm can automatically identify and segment key equipment in the wind farm, reduce manual intervention, improve work efficiency, and reduce the risk of human error; after adjusting the voxel size through the voxel-based three-dimensional reconstruction algorithm module, combined with the texture mapping and detail enhancement module, the generated three-dimensional model is not only accurate in structure, but also has fine surface texture and rich details. This high-quality three-dimensional model not only improves the visual effect, but also makes the model closer to the real scene.

[0017] A centimeter-level 3D reconstruction system for wind farm equipment integrates multi-source data acquisition equipment such as high-precision laser radar system, multi-spectral camera array, millimeter-wave radar, drone tilt photography system and ground mobile measurement vehicle. The system can comprehensively and accurately capture the details of wind farm equipment and ensure the high accuracy and integrity of the 3D reconstruction results; the data processing layer is equipped with a high-performance computing cluster, which cooperates with point cloud data processing, image processing and feature extraction modules and multi-source data fusion engine to achieve rapid processing and seamless fusion of massive data. This not only improves the data processing efficiency, but also ensures the accuracy and consistency of the fused data, providing high-quality input data for subsequent 3D reconstruction work; the 3D reconstruction layer adopts a target recognition and segmentation module based on deep learning, which can automatically identify and segment wind farm equipment, significantly improving the intelligent level of the reconstruction process. At the same time, combined with the voxel-based 3D reconstruction algorithm and texture mapping and detail enhancement technology, the generated 3D model not only maintains a high degree of structural accuracy, but also has rich surface textures and details, greatly improving the visual effect and practical application value of the model.

[0018] Furthermore, the introduction of the application layer enables the system to not only display the 3D reconstruction results, but also apply them to multiple fields such as equipment health monitoring and predictive maintenance analysis. The Web-based 3D visualization platform provides an intuitive and convenient interactive interface, allowing users to easily browse and analyze the 3D model of the wind farm. In addition, the addition of virtual reality or augmented reality interfaces brings an immersive experience to users, further enhancing the application value of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A block diagram of the centimeter-level three-dimensional reconstruction method of wind farm equipment according to the present invention; Figure 2 This is a block diagram of the centimeter-level three-dimensional reconstruction system of wind farm equipment of the present invention; Figure 3 This is a block diagram of an embodiment of a centimeter-level three-dimensional reconstruction system for wind farm equipment according to the present invention.

[0021] Among them: 201-data acquisition layer; 202-data processing layer; 203-three-dimensional reconstruction layer. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the 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. The components of the embodiments of the present invention described and marked in the drawings here can be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0025] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0026] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0027] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0028] The present invention is further described in detail below in conjunction with the accompanying drawings: See also Figure 1The present invention provides a centimeter-level three-dimensional reconstruction method for wind farm equipment, comprising the following steps: S1, collects multi-source data of wind farms; Use high-precision LiDAR systems to scan wind farms and obtain high-precision point cloud data. Use multi-spectral camera arrays to take multi-angle photos of wind farm equipment to obtain rich color and texture information. Send drones equipped with oblique photography systems to take aerial photos of wind farms to obtain more comprehensive scene information. Ground mobile survey vehicles are used to supplement detailed data that is difficult to obtain directly through drones or LiDAR.

[0029] S2, fusing the collected multi-source data of the wind farm to obtain fused data; The collected multi-source data is preprocessed, including denoising, correction, registration and other steps to ensure the quality and consistency of the data. The preprocessed point cloud data, image data and other related data are fused using the multi-source data fusion engine. During the fusion process, according to the characteristics of the data and the application scenario, the appropriate fusion algorithm is selected, such as feature-based fusion, model-based fusion, etc., to obtain fused data containing rich information.

[0030] S3, inputting the fused data into a pre-established three-dimensional reconstruction model to obtain a three-dimensional model of the wind farm equipment; the pre-established three-dimensional reconstruction model includes: a target recognition and segmentation module based on deep learning, a three-dimensional reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; the fused data is processed by the deep learning target recognition and segmentation module to obtain a first three-dimensional model of the wind farm equipment; the first three-dimensional model is adjusted in voxel size by the voxel-based three-dimensional reconstruction algorithm module and then the details are enhanced by the texture mapping and detail enhancement module to obtain the three-dimensional model of the wind farm equipment.

[0031] S301, pre-establishing a 3D reconstruction model including a deep learning-based target recognition and segmentation module, a voxel-based 3D reconstruction algorithm module, and a texture mapping and detail enhancement module.

[0032] S302, input the fused data into the target recognition and segmentation module based on deep learning, and use the deep learning algorithm to automatically identify and segment the key equipment in the wind farm. This module can accurately identify the outline and boundary of the equipment by training a large amount of sample data. The target recognition and segmentation module based on deep learning adopts the improved Mask R-CNN network architecture to adjust and optimize the network for specific scenarios of wind farms. By introducing attention mechanisms, adding feature pyramid layers and other strategies, the model's ability to recognize and segment wind power equipment and components is improved. Using the model pre-trained in similar scenarios as the initial weight, fine-tuning is performed through a small amount of annotated wind farm data to quickly adapt to the new environment and improve recognition accuracy and segmentation precision.

[0033] S303, input the segmented device data into the voxel-based 3D reconstruction algorithm module. This module optimizes the grid structure of the 3D model by adjusting the voxel size to make it closer to the shape and size of the real device. At the same time, the color and texture information in the image data is mapped to the 3D model using texture mapping technology to make it more realistic. The voxel-based 3D reconstruction algorithm module adopts an adaptive octree structure, dynamically adjusts the voxel size according to the density of the point cloud data and the complexity of the target, and realizes refined 3D reconstruction. It combines multispectral images and lidar point cloud data, and uses multi-view stereo vision technology to improve the accuracy and completeness of reconstruction. It uses a global optimization algorithm based on image segmentation to correct the errors generated during the reconstruction process to ensure the consistency and geometric correctness of the 3D model.

[0034] S304, finally, the 3D model is further optimized through the detail enhancement module to improve the model's detail expression and visual effect. According to the geometric features of the 3D model, the multi-resolution texture mapping technology is used to accurately map the color, texture and other information in the multispectral image to the model surface. The texture image is enlarged using a super-resolution algorithm based on a generative adversarial network (GAN) to enhance the texture details and make the model more realistic. Combined with normal mapping and displacement mapping technology, the model surface is enhanced in detail to simulate the lighting effect and surface texture of real objects.

[0035] join Figure 2 The present invention discloses a centimeter-level three-dimensional reconstruction system for wind farm equipment, comprising: Data collection layer, used to collect multi-source data of wind farms; The data processing layer is used to fuse the collected multi-source data of the wind farm to obtain fused data; The three-dimensional reconstruction layer is used to input the fused data into a pre-established three-dimensional reconstruction model to obtain a three-dimensional model of the wind farm equipment; the pre-established three-dimensional reconstruction model includes: a target recognition and segmentation module based on deep learning, a three-dimensional reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; the fused data is processed by the deep learning target recognition and segmentation module to obtain a first three-dimensional model of the wind farm equipment; the first three-dimensional model is adjusted in voxel size by the voxel-based three-dimensional reconstruction algorithm module and then the details are enhanced by the texture mapping and detail enhancement module to obtain the three-dimensional model of the wind farm equipment.

[0036] See also Figure 3 The present invention discloses a centimeter-level three-dimensional reconstruction system for wind farm equipment, including: a data acquisition layer, a data processing layer, a three-dimensional reconstruction layer and an application layer. These layers are interconnected through a high-speed data bus to achieve efficient data transmission and interaction. The data bus adopts optical fiber communication technology to ensure the fast and stable transmission of a large amount of data.

[0037] The centimeter-level three-dimensional reconstruction system of wind farm equipment of the present invention realizes a comprehensive data flow from data acquisition to final application. In the data acquisition layer, raw information such as LiDAR point cloud data, multispectral images, millimeter-wave radar data, etc. are transmitted to the data processing layer through a high-speed data bus, using standardized data formats such as LAS format and TIFF format, and using efficient real-time transmission protocols such as RTSP. After the data processing layer performs pre-processing such as noise reduction, registration, and segmentation on the raw data, the processing results are transmitted to the three-dimensional reconstruction layer in compressed PLY format (point cloud) and optimized JPEG2000 format (image) through a distributed file system (such as HDFS). After the three-dimensional reconstruction layer completes the reconstruction, the three-dimensional model is transmitted to the application layer in a lightweight glTF format, and a RESTful API interface is provided for the application layer to flexibly call and operate. Finally, the application layer transmits the optimized three-dimensional model and analysis results to the user's browser in real time through Web Socket for visual display, and securely receives the user's operation instructions through the HTTPS protocol. This comprehensive and efficient data flow design ensures the rapid response and data processing capabilities of the entire system, laying the foundation for achieving centimeter-level three-dimensional reconstruction accuracy.

[0038] Data collection layer, which includes high-precision laser radar system, multi-spectral camera array, millimeter-wave radar, oblique photography system carried by drone and ground mobile measurement vehicle; the high-precision laser radar (LiDAR) system adopts the latest phase-type LiDAR technology, with a scanning frequency of up to 100kHz; its measurement accuracy is better than ±2mm@100m, and the maximum measurement distance can reach 300m. It is also equipped with an automatic rotating gimbal to achieve 360° all-round scanning. The multi-spectral camera array consists of 5 high-resolution cameras, which collect images in the visible, near-infrared, mid-infrared, far-infrared and ultraviolet spectral bands respectively; each camera has a resolution of 40 million pixels and a dynamic range of 14 bits; it is equipped with an optical image stabilization system to ensure clear images even when the wind is strong. The operating frequency of the millimeter-wave radar is 77GHz, and the distance resolution can reach 1mm; it adopts MIMO (multiple input multiple output) technology, which can simultaneously measure the distance, azimuth and pitch angle of the target; it has the characteristics of strong penetration and strong anti-interference ability, and is suitable for working in severe weather conditions. The oblique photography system carried by the drone uses a six-rotor drone platform with a maximum flight time of 45 minutes; it is equipped with 5 high-resolution cameras (1 vertical and 4 oblique) to achieve all-round shooting without blind spots; it integrates the RTK-GPS system and the positioning accuracy can reach the centimeter level. The ground mobile survey vehicle uses an all-terrain chassis to adapt to various complex terrains; it integrates LiDAR, panoramic cameras and inertial measurement units (IMU); it is equipped with high-precision GNSS receivers to achieve centimeter-level positioning.

[0039] Data processing layer, which includes a high-performance computing cluster, a point cloud data processing module, an image processing and feature extraction module, and a multi-source data fusion engine; the data processing layer is responsible for preprocessing and preliminary fusion of the collected raw data to prepare for subsequent 3D reconstruction. Hardware configuration: Deploy a distributed computing cluster consisting of multiple high-performance servers, each server is equipped with a multi-core CPU, a large-capacity memory, and a high-speed storage system. Software environment: Install a parallel computing framework (such as MPI, Hadoop) and a task scheduling system to ensure the reasonable allocation and efficient use of computing resources. Point cloud data processing module: Use a point cloud filtering algorithm based on deep learning to denoise the original point cloud data, remove noise points, and retain valid data; use an improved ICP (iterative closest point) algorithm to accurately align the point cloud data scanned by multiple sites to ensure seamless connection between data from different sites, with an alignment accuracy better than 1cm; combine regional growth and deep learning technology to segment the aligned point cloud and accurately identify different equipment components in the wind farm. Image processing and feature extraction module: geometric correction of multispectral images to eliminate image distortion; radiation correction and atmospheric correction of images to improve image authenticity and clarity; use deep learning-based image enhancement algorithms to enhance image details and adjust contrast; use improved SIFT algorithms to extract stable and scale-invariant feature points from enhanced images. The above-mentioned point cloud data processing module and image processing and feature extraction module pre-process LiDAR point cloud, multispectral image and millimeter wave radar data to ensure data consistency and comparability. The multi-source data fusion engine uses a deep learning-based data fusion algorithm to fuse data from different sources at the feature level to improve data complementarity and comprehensive information expression capabilities.

[0040] The 3D reconstruction layer includes a deep learning-based target recognition and segmentation module, a voxel-based 3D reconstruction algorithm module, and a texture mapping and detail enhancement module. The 3D reconstruction layer is the core of this platform and is responsible for converting the processed data into a high-precision 3D model. It includes the following key modules: The deep learning-based target recognition and segmentation module adopts an improved Mask R-CNN network architecture to achieve accurate recognition and segmentation of various devices and components in the wind farm. Using transfer learning technology, it can quickly adapt to the new wind farm environment with a small amount of labeled data; the recognition accuracy reaches more than 98%, and the segmentation accuracy (IoU) exceeds 0.95. Voxel-based 3D reconstruction algorithm module: Adopting an adaptive octree structure, dynamically adjusting the voxel size according to the complexity of the target, combining point cloud data and multi-view stereo vision technology to improve the reconstruction accuracy, using a global optimization algorithm based on graph cuts to ensure the consistency and integrity of the reconstructed model, the reconstruction accuracy reaches the centimeter level, which is more than 50% higher than the traditional method. The texture mapping and detail enhancement module adopts multi-resolution texture mapping technology to accurately map multispectral image information onto the three-dimensional model, uses a super-resolution algorithm based on a generative adversarial network (GAN) to enhance texture details, and combines normal mapping and displacement mapping technology to enhance model surface details.

[0041] Application layer, which includes a Web-based 3D visualization platform, an equipment health monitoring system, a predictive maintenance analysis module, and a virtual reality or augmented reality interface; the application layer provides various tools and interfaces to enable users to make full use of the reconstructed 3D model. It includes the following components: Web-based 3D visualization platform: The platform is developed based on WebGL technology, supports mainstream browsers, does not require the installation of plug-ins, uses LOD (multiple levels of detail) technology to achieve smooth display of large-scale 3D models, and supports advanced functions such as model measurement, sectioning, and annotation. The equipment health monitoring system uses 3D models and real-time sensor data to achieve visual monitoring of equipment status; uses anomaly detection algorithms based on deep learning to detect potential problems in a timely manner. Supports historical data playback and trend analysis functions. Predictive maintenance analysis module: Uses machine learning algorithms to predict equipment maintenance needs based on historical data and current status; uses Monte Carlo simulation methods to evaluate the cost-effectiveness of different maintenance strategies, and provides visual maintenance suggestions and optimization solutions. VR / AR interface is set up: Supports mainstream VR / AR devices such as Oculus Quest, HoloLens, etc.

[0042] The data acquisition layer of an embodiment of the centimeter-level three-dimensional reconstruction system for wind farm equipment of the present invention also includes an adaptive multi-scale scanning strategy module, which adopts an adaptive multi-scale scanning strategy to automatically adjust the scanning density and accuracy according to the complexity and importance of the equipment. The specific implementation steps are as follows: S1, pre-scan: quickly scan the entire wind farm with low resolution to obtain rough scene information.

[0043] S2, scene analysis: uses deep learning algorithms to analyze pre-scan data and identify different devices and areas.

[0044] S3, Importance Assessment: Assign an importance coefficient to each area based on equipment type, location, and historical maintenance records.

[0045] S4, complexity calculation: analyze the geometric complexity of each area and calculate the complexity coefficient.

[0046] S5, scanning parameter optimization: dynamically adjust scanning density and accuracy according to importance and complexity.

[0047] Scanning density Determined by the following formula:

[0048] in, is the basic scanning density (dots / m²), is the equipment complexity coefficient (0-1), $I$ is the importance coefficient (0-1), and is the weight factor. By adjusting and The value can balance the requirements of scanning efficiency and accuracy.

[0049] In one embodiment of the centimeter-level 3D reconstruction system for wind farm equipment of the present invention, the 3D reconstruction layer also includes an AI-driven automatic defect detection module, which uses a deep learning algorithm, especially a model based on a convolutional neural network (CNN), to automatically detect defects in the reconstructed 3D model. The specific implementation process is as follows: S1. Data preparation: Collect a large number of wind power equipment defect samples, including normal samples and various defect samples.

[0050] S2. Data augmentation: Expand the training dataset by rotating, scaling, adding noise, etc.

[0051] S3. Model design: Based on the ResNet-50 architecture, a network structure suitable for 3D model defect detection is designed.

[0052] S4. Model training: Use GPU clusters for large-scale parallel training and use cross-validation to ensure model generalization ability.

[0053] S5. Model optimization: Use model pruning and quantization techniques to reduce model size and improve inference speed.

[0054] S6. Defect detection: Convert the 3D model into a 2D projection image and input it into the trained CNN model for defect detection.

[0055] The confidence level $P$ of defect detection is calculated as follows:

[0056] in, is the output value of the CNN model, The value range of is 0-1, indicating the probability of detecting a defect. When it is greater than a preset threshold (such as 0.8), the system will mark the area as a potential defect and remind humans to review it.

[0057] In one embodiment of the centimeter-level three-dimensional reconstruction system for wind farm equipment of the present invention, the three-dimensional reconstruction layer uses millimeter-wave radar and high-speed camera to capture the tiny deformation of wind turbine blades during operation to realize dynamic deformation analysis; the specific implementation steps are as follows: S1. Data acquisition: Use a high frame rate (1000fps) camera and a high-precision millimeter-wave radar to synchronously collect blade motion data.

[0058] S2. Data preprocessing: noise reduction and filtering of image and radar data.

[0059] S3. Feature point tracking: The improved KLT (Kanade-Lucas-Tomasi) algorithm is used to track the feature points on the blade.

[0060] S4. 3D reconstruction: Combine camera and radar data to reconstruct the 3D motion trajectory of feature points.

[0061] S5. Deformation calculation: Based on the reconstructed trajectory, the deformation of each part of the blade is calculated.

[0062] Deformation range The calculation formula is as follows:

[0063] in, , and Respectively represent the displacement changes in the x, y, and z directions. The time series changes of the blades can be used to evaluate the dynamic characteristics and potential structural problems of the blades.

[0064] In one embodiment of the present invention, the 3D reconstruction layer further includes an environmental factor adaptive algorithm module, which automatically adjusts the 3D reconstruction parameters according to environmental factors. The environmental factor adaptive algorithm automatically adjusts the 3D reconstruction parameters according to environmental factors such as lighting conditions and weather conditions. The main implementation steps include: S1. Environmental Sensing: Real-time monitoring of environmental conditions using multispectral cameras and meteorological sensors.

[0065] S2. Parameter mapping: Establish the mapping relationship between environmental factors and reconstruction parameters.

[0066] S3. Automatic adjustment: Dynamically adjust the parameters of the 3D reconstruction algorithm according to the current environmental conditions.

[0067] For example, in low light conditions, the system increases the camera’s exposure time or ISO sensitivity; adjusts the LiDAR’s transmit power and receive sensitivity; and increases the intensity of denoising and smoothing during 3D reconstruction. Through this adaptive approach, the system is able to maintain stable reconstruction accuracy under a variety of environmental conditions.

[0068] The present invention deploys an edge computing module on the data acquisition device to perform preliminary processing and compression of the data. The specific implementation includes: S1. Edge device configuration: Integrate a high-performance embedded processor on each data acquisition device (such as LiDAR, multispectral camera); equip it with a high-speed solid-state drive for temporary storage and fast reading and writing of data.

[0069] S2. Edge processing algorithm: Real-time noise reduction and compression of LiDAR point cloud data, using octree encoding method, with a compression ratio of up to 10:1. Preprocess image data, including noise reduction, color balancing and preliminary feature extraction. Implement a simplified version of the target detection algorithm to preliminarily identify the main components of wind power equipment.

[0070] S3. Data prioritization: Based on the importance and timeliness of the data, the processed data is divided into three priorities: high, medium, and low. High-priority data (such as anomaly detection results) is transmitted to the cloud in real time. Medium and low-priority data are gradually transmitted as the network bandwidth permits.

[0071] The present invention adopts a cloud-based collaborative processing method, deploys a distributed computing framework such as Apache Spark on the cloud, and implements large-scale parallel processing; a dynamic task scheduling algorithm is designed to allocate computing resources according to the current computing load and data priority. The MQTT protocol is used to achieve real-time communication between edge devices and the cloud. The cloud can send instructions to edge devices based on the processing results and dynamically adjust the data collection strategy.

[0072] Through this collaborative approach of edge computing and the cloud, the system can significantly reduce the amount of data transmitted to the cloud and improve the overall system response speed and efficiency. Preliminary tests show that compared with traditional pure cloud processing methods, data processing speed has increased by about 50% and network bandwidth usage has been reduced by 60%.

[0073] The present invention adopts a data fusion algorithm based on deep learning to fuse data from different sensors with high precision. The specific implementation steps are as follows: S1. Data preprocessing: Time synchronization and spatial registration of each source data. Wavelet transform is used to reduce noise and enhance features of the data.

[0074] S2. Feature extraction: Extract geometric features from LiDAR point cloud data, such as normal vectors, curvature, etc. Extract texture features and spectral features from multispectral images. Extract dynamic features of targets from millimeter wave radar data.

[0075] S3. Feature-level fusion: Design a multimodal deep learning network, such as the improved PointNet++, to process point cloud and image features simultaneously. Use the attention mechanism to automatically learn the importance weights of different features.

[0076] S4. Decision-level fusion: Detect and classify each data source separately. Adopt Dempster-Shafer evidence theory to combine the results of each detector to get the final fusion result.

[0077] S5. Time series data fusion: Use long short-term memory (LSTM) networks to capture the temporal correlation of data. Implement a sliding window mechanism to balance real-time performance and accuracy.

[0078] S6. Quality evaluation and feedback: Design fusion quality evaluation indicators, such as spatial consistency, temporal coherence, etc. Dynamically adjust fusion strategies and parameters based on the evaluation results.

[0079] By fusing multi-source data, the system can take full advantage of the advantages of various sensors to improve the accuracy and integrity of reconstruction. Experimental results show that compared with a single data source, the fused 3D reconstruction accuracy is improved by 30% and the integrity is improved by 25%.

[0080] The adaptive voxel 3D reconstruction layer of the present invention adopts an adaptive voxel size strategy during the 3D reconstruction process. For areas with high complexity, a smaller voxel size is used; for relatively simple areas, a larger voxel size is used. The specific implementation is as follows: S1. Initial voxelization: The entire space is divided using voxels of uniform size, with an initial voxel size of 10 cm.

[0081] S2. Complexity Assessment: The point cloud density and curvature variation are calculated for each voxel.

[0082] Deep learning based methods evaluate geometric complexity within voxels.

[0083] S3. Adaptive Segmentation: Set a complexity threshold and subdivide when the complexity of a voxel exceeds the threshold.

[0084] The octree structure is used to realize the adaptive subdivision of voxels, and the minimum voxel size can reach 1cm.

[0085] S4. Surface reconstruction: An implicit function is used for each voxel to represent the local surface.

[0086] An improved Poisson surface reconstruction algorithm is used to ensure smooth transition between voxels.

[0087] S5. Detail Enhancement: Using high-resolution image information, sub-voxel level details are restored via a belief propagation algorithm.

[0088] S6. Global Optimization: The graph cut algorithm is used to globally optimize the reconstruction results and eliminate local inconsistencies. Laplace smoothing is applied to improve the visual quality of the model.

[0089] This adaptive voxel strategy can optimize the use of computing resources while ensuring accuracy. Test results show that the computing time is reduced by 40% compared to a fixed voxel size while maintaining centimeter-level reconstruction accuracy.

[0090] Through the implementation of the above technical solution, the centimeter-level 3D reconstruction platform for wind farm equipment of the present invention can achieve the following effects: 1. Reconstruction accuracy: The reconstruction accuracy of large equipment (such as wind turbine towers) reaches ±1.5cm. The reconstruction accuracy of small parts (such as bolts and flanges) reaches ±0.5cm. The absolute positioning accuracy of the overall model is better than 2cm.

[0091] 2. Environmental adaptability: It can maintain stable operation in the temperature range of -20℃ to 50℃. It can collect data normally when the wind speed is less than 15m / s. It can still maintain more than 80% reconstruction quality in the case of insufficient light (such as cloudy days and dusk).

[0092] 3. Data processing efficiency: Compared with traditional methods, data processing speed is increased by 60%. The complete reconstruction time of a 1 square kilometer wind farm is shortened from 5 days to 2 days. 90% of the reconstruction process is automated, greatly reducing manual intervention.

[0093] 4. Defect detection accuracy: For common defects such as surface cracks and corrosion, the detection accuracy reaches 98%. The false alarm rate is controlled below 5%, which greatly improves the maintenance efficiency. It can detect surface defects as small as 1mm.

[0094] 5. Dynamic characteristic analysis: It can capture tiny deformations of 0.1mm. The blade vibration frequency measurement accuracy reaches 0.01Hz. It provides reliable data support for fatigue damage assessment.

[0095] 6. Economic benefits: Wind farm operation and maintenance costs are reduced by about 25%. Equipment failure warning time is advanced by 48 hours, effectively reducing downtime losses. Equipment service life is extended by 8-12%.

[0096] 7. Digital management: Achieve 100% digital management of wind farm assets. Support remote collaboration and improve multi-party collaboration efficiency by 30%. Lay a data foundation for the intelligent upgrade of wind farms.

[0097] The implementation of this invention not only significantly improves the operation and maintenance efficiency of wind farms and reduces costs, but also lays the foundation for the intelligent and digital management of wind farms. Through high-precision three-dimensional reconstruction and real-time monitoring, the reliability and safety of wind power equipment are greatly improved. At the same time, the application of this platform also provides a rich data source for big data analysis and artificial intelligence applications in the wind power industry, which is expected to promote technological progress and innovative development in the entire industry.

[0098] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A centimeter-level three-dimensional reconstruction method for wind farm equipment, characterized in that: The following steps are involved: S1, collects multi-source data of wind farms; S2, fusing the collected multi-source data of the wind farm to obtain fused data; S3, inputting the fused data into a pre-established three-dimensional reconstruction model to obtain a three-dimensional model of the wind farm equipment; The pre-established 3D reconstruction model includes: a target recognition and segmentation module based on deep learning, a 3D reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; The fused data is processed by a deep learning target recognition and segmentation module to obtain a first three-dimensional model of the wind farm equipment; the first three-dimensional model is adjusted in voxel size by a voxel-based three-dimensional reconstruction algorithm module and then the details are enhanced by a texture mapping and detail enhancement module to obtain a three-dimensional model of the wind farm equipment.

2. A centimeter-level 3D reconstruction system for wind farm equipment, based on the centimeter-level 3D reconstruction method for wind farm equipment according to claim 1, characterized in that: include: Data collection layer, used to collect multi-source data of wind farms; The data processing layer is used to fuse the collected multi-source data of the wind farm to obtain fused data; A three-dimensional reconstruction layer, used for inputting the fused data into a pre-established three-dimensional reconstruction model to obtain a three-dimensional model of the wind farm equipment; the pre-established three-dimensional reconstruction model includes: a target recognition and segmentation module based on deep learning, a three-dimensional reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; The fused data is processed by a deep learning target recognition and segmentation module to obtain a first three-dimensional model of the wind farm equipment; the first three-dimensional model is adjusted in voxel size by a voxel-based three-dimensional reconstruction algorithm module and then the details are enhanced by a texture mapping and detail enhancement module to obtain a three-dimensional model of the wind farm equipment.

3. The centimeter-level three-dimensional reconstruction system for wind farm equipment according to claim 2, characterized in that: The centimeter-level three-dimensional reconstruction system for wind farm equipment also includes an application layer, and the application layer is used to display and apply the three-dimensional reconstruction result.

4. The centimeter-level three-dimensional reconstruction system for wind farm equipment according to claim 3 is characterized in that: The data acquisition layer includes a high-precision laser radar system, a multi-spectral camera array, a millimeter-wave radar, an oblique photography system carried by a drone, and a ground mobile measurement vehicle; The data processing layer includes a high-performance computing cluster, a point cloud data processing module, an image processing and feature extraction module, and a multi-source data fusion engine; The 3D reconstruction layer includes a target recognition and segmentation module based on deep learning, a 3D reconstruction algorithm module based on voxels, and a texture mapping and detail enhancement module; The application layer includes a web-based 3D visualization platform, an equipment health monitoring system, a predictive maintenance analysis module, and a virtual reality or augmented reality interface; The data acquisition layer, data processing layer, three-dimensional reconstruction layer and application layer are interconnected through a high-speed data bus and exchange data.

5. The centimeter-level three-dimensional reconstruction system for wind farm equipment according to claim 4, characterized in that: The multi-source data fusion engine of the data processing layer adopts a data fusion algorithm based on deep learning to perform time synchronization, spatial registration and feature-level fusion on data from different sensors.

6. The centimeter-level three-dimensional reconstruction system for wind farm equipment according to claim 4, characterized in that: The voxel-based 3D reconstruction algorithm module of the 3D reconstruction layer adopts an adaptive voxel size strategy to dynamically adjust the voxel size according to the target complexity.

7. The centimeter-level three-dimensional reconstruction system for wind farm equipment according to claim 4, characterized in that: The data collection layer also includes an edge computing module, which is deployed on the equipment of the data collection layer to perform preliminary processing and compression on the collected data.

8. The centimeter-level three-dimensional reconstruction method of wind farm equipment according to claim 4, characterized in that: The data acquisition layer also includes an adaptive multi-scale scanning strategy module, which automatically adjusts the scanning density and accuracy according to the complexity and importance of the equipment.

9. The centimeter-level three-dimensional reconstruction method of wind farm equipment according to claim 4, characterized in that: The 3D reconstruction layer also includes an AI-driven automatic defect detection module, which automatically identifies, locates and labels surface defects of the reconstructed 3D model.

10. The centimeter-level three-dimensional reconstruction method of wind farm equipment according to claim 4, characterized in that: The three-dimensional reconstruction layer also includes an environmental factor adaptive algorithm module, and the environmental factor adaptive algorithm module automatically adjusts the three-dimensional reconstruction parameters according to environmental factors.

Images