Multi-feature fan blade defect detection method and system based on edge calculation

Through edge computing and multi-feature fusion technology, the problems of insufficient detection accuracy, high cost and low efficiency of fan blades are solved, and efficient and accurate detection of fan blade defects is achieved, reducing detection costs and improving detection efficiency.

CN118918080APending Publication Date: 2024-11-08SHANGHAI UNIVERSITY OF ELECTRIC POWER
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410967807.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing fan blade defect detection methods have problems such as insufficient detection accuracy, high cost and low efficiency, especially in complex backgrounds, and traditional centralized computing methods lead to data transmission delay and bandwidth pressure, which is difficult to meet real-time Testing requirements.

Method used

Using multi-feature fan blade defect detection method based on edge computing, pre-processing and feature extraction by collecting image, vibration, sound wave, temperature and humidity data, edge computing is performed using the improved Yolov8 model, combining multi-modal fusion network and feature selection , dimensionality reduction processing, reduce calculation amount and improve model robustness.

Benefits of technology

It improves the accuracy and efficiency of fan blade defect detection, reduces detection cost, reduces data transmission, and realizes low-latency and high-response detection results, significantly improving the accuracy and reliability of detection results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918080A_ABST
    Figure CN118918080A_ABST
Patent Text Reader

Abstract

The invention relates to a multi-feature fan blade defect detection method and system based on edge calculation, and the method comprises the following steps: S1, collecting and preprocessing multi-feature fan blade data of a to-be-detected fan blade, the multi-feature fan blade data comprising image data, vibration data, sound wave data, temperature data and humidity data; s2, performing feature extraction on the preprocessed multi-feature fan blade data; s3, an improved Yov8 model is adopted, edge calculation is carried out on the features extracted in the step S2, the defect condition of the to-be-detected fan blade is recognized, prompt information is generated, in the improved Yov8 model, an internet module based on partial convolution is used for replacing a Bottleneck module of C2f, and in the improved Yov8 model, the defect condition of the to-be-detected fan blade is recognized; the internet module based on partial convolution comprises a first convolution layer, a segmentation layer, a first bottleneck layer, a second bottleneck layer, a splicing layer and a second convolution layer which are connected in sequence. Compared with the prior art, the detection precision and the detection efficiency can be further improved, and meanwhile the detection cost is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation, and in particular, relates to a multi-feature wind turbine blade defect detection method and system based on edge computing. Background Art

[0002] With the rapid development of the wind power industry, the operating efficiency and reliability of wind turbines have become the focus of attention. As one of the core components of wind turbines, the working state of wind turbine blades directly affects the performance and safety of the entire wind power system. However, since wind turbine blades are exposed to complex natural environments for a long time, such as extreme weather such as wind, sand, rain, snow, and freezing, various defects such as cracks, corrosion, wear, and ice accumulation are prone to appear on the surface of the blades. These defects not only reduce the aerodynamic performance of the blades, but may also cause structural failure of the blades, and in severe cases even cause the entire wind turbine system to shut down or be damaged. Therefore, it is of great significance to conduct effective defect detection on wind turbine blades.

[0003] Traditional methods for detecting defects in fan blades mainly include manual detection and detection methods based on image processing. Manual detection relies on the experience and visual observation of on-site maintenance personnel, and has problems such as low detection efficiency, strong subjectivity, and high cost. Chinese patent application CN202210120588.9 discloses a multimodal-based method for detecting defects in fan blades. It uses the surface image and vibration data of the fan blades to integrate the image features and vibration features as a whole, thereby improving the accuracy of detecting the location and category of defects in the fan blades and improving the detection efficiency. Although the detection method based on image processing has improved the level of automation of detection to a certain extent, the following challenges still exist:

[0004] (1) Insufficient detection accuracy: Detection methods based on image processing are easily affected by noise in complex backgrounds, resulting in insufficient detection accuracy and reliability.

[0005] (2) High detection cost: The detection method based on image processing requires the additional deployment of image acquisition modules and data processing modules. However, wind turbine blades are usually installed at high altitudes or in remote areas, which requires a lot of manual participation and high equipment investment during the construction and maintenance phases.

[0006] (3) Low detection efficiency: The detection method based on image processing needs to transmit a large amount of image data to the central server for processing. However, the traditional centralized computing method has high demand for computing resources, which will cause large data transmission delays and bandwidth pressure, making it difficult to meet the requirements of real-time detection.

[0007] Therefore, it is necessary to design a fan blade defect detection method to further improve the detection accuracy and efficiency while reducing the detection cost. Summary of the invention

[0008] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a multi-feature wind turbine blade defect detection method and system based on edge computing, so as to further improve the detection accuracy and efficiency while reducing the detection cost.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] The present invention provides a multi-feature wind turbine blade defect detection method based on edge computing, which is applicable to edge devices in a wind power generation system, and includes the following steps:

[0011] S1, collecting multi-feature fan blade data of the fan blade to be detected and preprocessing, wherein the multi-feature fan blade data includes image data, vibration data, sound wave data, temperature data and humidity data;

[0012] S2, extracting features from the preprocessed multi-feature wind turbine blade data;

[0013] S3. Using an improved Yolov8 model in the edge device, performing edge computing on the features extracted in step S2, identifying the defects of the wind turbine blades to be inspected and generating prompt information, wherein a fasternet module based on partial convolution is used to replace the Bottleneck module of C2f in the improved Yolov8 model, and the fasternet module based on partial convolution includes a first convolution layer, a segmentation layer, a first bottleneck layer, a second bottleneck layer, a splicing layer and a second convolution layer connected in sequence.

[0014] Furthermore, in step S1, the preprocessing process is specifically as follows:

[0015] For the image data, performing image denoising and contrast enhancement;

[0016] Performing denoising and filtering on the vibration data and the sound wave data;

[0017] Smoothing processing and outlier detection are performed on the temperature data and the humidity data.

[0018] Furthermore, in step S2, the feature extraction process is specifically as follows:

[0019] For the image data, edge detection algorithms are used to extract edge features in the image, gray-level co-occurrence matrix GLCM or local binary pattern LBP method is used to extract texture features of the image, morphological operators are used to extract shape features of the image, and color feature representation methods are used to extract color space features of the image;

[0020] For the vibration data, calculate the time domain characteristics of the vibration signal, including the mean, variance, peak value and root mean square value, extract the frequency domain characteristics of the vibration signal by fast Fourier transform, including the main frequency, harmonics and frequency band energy, extract the time-frequency characteristics of the vibration signal by wavelet transform, and calculate the statistical characteristics of the vibration signal, including skewness and kurtosis;

[0021] For the sound wave data, the time domain characteristics of the sound wave signal are calculated, including the mean, variance, peak value and energy, the frequency spectrum characteristics of the sound wave signal are extracted by Fourier transform, and the time-frequency characteristics of the sound wave signal are extracted by short-time Fourier transform or wavelet transform;

[0022] For the temperature data and the humidity data, time series features including mean, variance and rate of change are extracted, and long-term change trends and periodic fluctuations are identified through moving average or exponential smoothing methods.

[0023] Furthermore, after feature extraction is completed, the obtained multiple features are fused through a multimodal fusion network.

[0024] Furthermore, in step S2, after feature extraction is completed, feature selection and dimensionality reduction processing are performed.

[0025] Furthermore, the extracted features are selected by principal component analysis or linear discriminant analysis method.

[0026] The present invention also provides a multi-feature wind turbine blade defect detection system based on edge computing, including a sensor module, an edge computing device, a wireless communication module and a monitoring center;

[0027] The sensor module includes multiple types of sensors for collecting multi-feature wind turbine blade data and sending the data to the edge computing device, wherein the multi-feature wind turbine blade data includes image data, vibration data, sound wave data, temperature data and humidity data;

[0028] The edge computing device is used to preprocess and extract features of the data collected by the sensor module, perform edge computing on the feature extraction results through the improved Yolov8 model, and identify the defects of the wind turbine blades to be detected. In the improved Yolov8 model, a fasternet module based on PConv is introduced to replace the Bottleneck module of C2f;

[0029] The wireless communication module is used to send the defect conditions of the wind turbine blades to be detected obtained by the edge computing device to the monitoring center;

[0030] The monitoring center is used to generate prompt information according to the defect conditions of the wind turbine blades to be detected.

[0031] Furthermore, before collecting data, the initial bias of various sensors is eliminated through a standard calibration procedure, and the sensor parameters, including sampling frequency and resolution, are set according to the actual application requirements and the working environment of the wind turbine blades.

[0032] Furthermore, the edge computing device classifies, stores and manages the multi-feature wind turbine blade data through a data indexing mechanism, and the establishment process of the data indexing mechanism is as follows:

[0033] Design data models and metadata schemas based on data characteristics and access patterns, and configure corresponding databases and storage systems;

[0034] Design data import and indexing processes, and develop data access APIs and retrieval interfaces;

[0035] Design data backup and recovery strategies, and optimize indexes and queries in real time based on system performance.

[0036] Furthermore, the improved Yolov8 model is updated in the following manner:

[0037] The detection results of the model are recorded, and the number of failures in identifying each type of defect is counted. If the number of failures in identifying a certain type of defect exceeds the set threshold, the corresponding collected data is sent from the edge computing device to the monitoring center for retraining. After the monitoring center completes the training, the model is sent back to the edge computing device.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The present invention provides a multi-feature fan blade defect detection method based on edge computing. First, multi-feature fan blade data of a fan blade to be detected is collected and preprocessed; then, feature extraction is performed on the preprocessed multi-feature fan blade data; finally, an improved Yolov8 model is used in the edge device to perform edge computing on the extracted features, identify the defect conditions of the fan blade to be detected and generate prompt information. Edge computing, as a computing paradigm that pushes data processing, analysis and storage tasks to the edge of the network, has the advantages of low latency, high response, reduced data transmission volume, etc., and can further improve the detection efficiency of fan blade defects; In the improved Yolov8 model, the fasternet module based on partial convolution is used to replace the Bottleneck module of C2f. The fasternet module based on partial convolution includes a first convolution layer, a segmentation layer, a first bottleneck layer, a second bottleneck layer, a splicing layer and a second convolution layer connected in sequence. It can only perform feature extraction on a part of the channels of the input feature map, thereby reducing the amount of calculation and having higher computational efficiency. It can effectively increase the diversity of learning features of the convolution layer, reduce the influence of unnecessary redundant information on the model, improve the robustness and generalization ability of the model, and further improve the detection accuracy of fan blade defects.

[0040] 2. The present invention performs feature extraction on multi-feature fan blade data, which includes image data, vibration data, sound wave data, temperature data and humidity data. The multiple features obtained are fused through a multimodal fusion network, which can reduce the impact of noise interference and significantly improve the accuracy and reliability of fan blade defect detection results.

[0041] 3. The present invention provides a multi-feature wind turbine blade defect detection system based on edge computing, which makes full use of existing sensors and edge computing equipment, does not require a large amount of manual participation and expensive fixed monitoring equipment investment, and can further reduce the cost of wind turbine blade defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of the method of the present invention;

[0043] Figure 2 It is a schematic diagram of the structure of the improved Yolov8 model;

[0044] Figure 3 It is a schematic diagram of the structure of the FasterNet module based on partial convolution;

[0045] Figure 4 is a schematic diagram of the structure of the system of the present invention,

[0046] Explanation of the accompanying drawings: 1. Sensor module; 101. Image sensor; 102. Vibration sensor; 103. Acoustic sensor; 104. Temperature sensor; 105. Humidity sensor; 2. Edge computing device; 201. High-performance processor; 202. Storage unit; 3. Wireless communication module; 4. Monitoring center; 401. Visual interface; 402. Cloud server; 403. Database. DETAILED DESCRIPTION

[0047] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0048] Example 1

[0049] This embodiment provides a multi-feature fan blade defect detection method based on edge computing, such as Figure 1 As shown, the following steps are included:

[0050] S1. Collect multi-feature fan blade data of the fan blade to be detected and perform preprocessing, the multi-feature fan blade data including image data, vibration data, sound wave data, temperature data and humidity data.

[0051] The data collection and processing are directly related to the accuracy and reliability of subsequent defect detection and analysis. This embodiment realizes comprehensive data collection of the surface of the wind turbine blade and its working state through the collaborative work of multiple sensors. The specific steps are as follows:

[0052] 1) Sensor layout and installation

[0053] Various types of sensors are rationally arranged on the wind turbine blades and their surrounding areas to ensure comprehensive and accurate data collection. The main sensors include but are not limited to:

[0054] Image sensor: used to capture high-resolution images of the blade surface.

[0055] Vibration sensor: used to detect the vibration of the blade to determine whether there are structural defects.

[0056] Acoustic sensor: used to capture the acoustic wave signals generated by the blades under the action of wind and analyze possible cracks or other internal defects.

[0057] Temperature sensor: monitors temperature changes on the blade surface to identify possible local overheating or freezing.

[0058] Humidity sensor: records the ambient humidity to analyze possible corrosion on the blade surface.

[0059] 2) Data collection initialization

[0060] Before data collection begins, perform the following initialization steps:

[0061] Sensor calibration: Ensure that all sensors are working in the best condition and eliminate initial deviations through standard calibration procedures.

[0062] Parameter setting: According to the actual application requirements and the blade working environment, set the sensor's sampling frequency, resolution and other parameters to ensure the validity and accuracy of the data.

[0063] 3) Multi-feature data collection

[0064] Image data acquisition: The camera takes high-resolution images of the blade surface at a set frequency. If necessary, an infrared camera can be used to capture images at night or in low-light environments.

[0065] Vibration data collection: The vibration sensor records the vibration signals of the blades in real time during operation, and obtains subtle vibration changes through high-frequency sampling.

[0066] Acoustic wave data acquisition: The acoustic wave sensor continuously captures the acoustic wave signals on the blade surface and identifies abnormal acoustic wave characteristics through spectrum analysis.

[0067] Temperature and humidity data collection: Temperature and humidity sensors regularly record the temperature and humidity changes on the leaf surface and the surrounding environment to assist in determining the status of the leaf surface.

[0068] 4) Data preprocessing

[0069] Perform preliminary processing on the collected data on the edge device to reduce the computational burden of subsequent analysis and improve data quality:

[0070] Image preprocessing: including image denoising, contrast enhancement, etc., to highlight the detailed features of the leaf surface.

[0071] Preprocessing of vibration signals and sound wave signals: denoising, filtering and other processing.

[0072] Temperature data and humidity data preprocessing: smoothing and outlier detection are performed to ensure data continuity and accuracy.

[0073] The preprocessed data is stored in the local memory of the edge device and transmitted to the cloud server for further analysis and storage on a scheduled or real-time basis as needed:

[0074] The above steps achieve comprehensive and accurate monitoring of the status of wind turbine blades through the collaborative work of multiple sensors and efficient processing of edge devices. The pre-processed multi-feature data provides a solid foundation for subsequent defect detection and analysis, ensuring the real-time and reliability of the system.

[0075] S2. Extract features from the preprocessed multi-feature wind turbine blade data.

[0076] Feature extraction can provide important information support for subsequent defect identification and classification by extracting meaningful features from multiple sensor data. The specific steps are as follows:

[0077] 1) Image feature extraction

[0078] Image feature extraction mainly includes the processing and analysis of blade surface images, aiming to identify possible defect areas and features:

[0079] Edge detection: Use edge detection algorithms such as Sobel and Canny to extract edge information in the image and highlight the cracks and damage features on the blade surface.

[0080] Texture analysis: Use gray-level co-occurrence matrix (GLCM), local binary pattern (LBP) and other methods to extract the texture features of the image and distinguish between normal areas and defective areas.

[0081] Morphological processing: Use morphological operators (such as corrosion, expansion, opening and closing operations) to process the image to enhance the contrast and shape characteristics of the defect area.

[0082] Color feature extraction: Extract features of color spaces such as RGB and HSV from color images to identify defects with obvious color changes such as corrosion and coating peeling.

[0083] 2) Vibration feature extraction

[0084] Vibration feature extraction identifies possible structural defects by analyzing the vibration signals generated by the blade during operation:

[0085] Time domain characteristics: Calculate the mean, variance, peak value, root mean square value and other time domain characteristics of the vibration signal to preliminarily determine the intensity and change trend of the vibration.

[0086] Frequency domain features: The vibration signal is converted from the time domain to the frequency domain through fast Fourier transform (FFT), and the frequency domain features such as main frequency, harmonics, and frequency band energy are extracted to identify periodic vibration patterns.

[0087] Time-frequency characteristics: Use wavelet transform and other methods to perform time-frequency analysis on vibration signals, extract frequency characteristics at different time scales, and identify instantaneous vibration changes and local anomalies.

[0088] Statistical features: Calculate the skewness, kurtosis and other statistical features of the vibration signal to describe the shape and distribution characteristics of the signal.

[0089] 3) Acoustic wave feature extraction

[0090] Acoustic feature extraction identifies possible internal defects by analyzing the acoustic wave signals generated by the blade under the action of wind:

[0091] Time domain features: Calculate the mean, variance, peak value, energy and other time domain features of the sound wave signal to evaluate the intensity and changes of the sound wave signal.

[0092] Frequency domain features: Extract the spectrum features of the sound wave signal through Fourier transform, analyze the frequency distribution and energy concentration area, and identify abnormal sound wave patterns.

[0093] Time-frequency characteristics: Time-frequency analysis methods such as short-time Fourier transform (STFT) and wavelet transform are used to extract the characteristics of sound wave signals at different times and frequencies and locate instantaneous abnormal sound wave signals.

[0094] Feature fusion: Multiple sound wave features are combined to form a comprehensive feature vector to improve the robustness and descriptive ability of the features.

[0095] 4) Temperature and humidity feature extraction

[0096] Temperature and humidity feature extraction identifies possible environmental impact defects by analyzing the temperature and humidity data of the blade surface and the surrounding environment:

[0097] Time series characteristics: Analyze the time series characteristics of temperature and humidity data, calculate the mean, variance, rate of change, etc., and evaluate the stability and volatility of temperature and humidity.

[0098] Trend analysis: Use moving average, exponential smoothing and other methods to perform trend analysis on temperature and humidity data to identify long-term trends and periodic fluctuations.

[0099] Anomaly detection: Statistical methods and machine learning algorithms are used to detect outliers in temperature and humidity data to identify possible mutations and extreme situations.

[0100] Feature combination: Combine temperature and humidity features to form joint features, which are used to analyze the comprehensive impact of environmental factors on leaf status.

[0101] 5) Feature selection and dimensionality reduction

[0102] After completing the preliminary feature extraction, feature selection and dimensionality reduction are performed to reduce redundant features and improve the effectiveness of features:

[0103] Feature selection: Use methods such as principal component analysis (PCA) and linear discriminant analysis (LDA) to screen the extracted features and retain the most discriminative features.

[0104] Dimensionality reduction processing: Apply dimensionality reduction technology to compress high-dimensional feature space into low-dimensional space, reduce computational complexity, and enhance the interpretability of features.

[0105] The feature extraction step extracts discriminative multi-dimensional features through processing and analysis of multiple sensor data. These features provide an important basis for subsequent defect identification and classification, ensuring the efficiency and accuracy of wind turbine blade defect detection results.

[0106] S3. Using the improved Yolov8 model, edge computing is performed on the features extracted in step S2 to identify the defects of the wind turbine blades to be inspected and generate prompt information.

[0107] 1) Edge device deployment and configuration

[0108] Hardware deployment: Deploy high-performance edge devices near wind turbine blades and configure necessary computing resources (such as CPU, GPU) and storage space to meet real-time computing and data storage requirements.

[0109] Software configuration: Install and configure the necessary software environment on the edge device, including the operating system, edge computing platform, database system, and defect detection algorithm module.

[0110] 2) Data synchronization and management

[0111] Data synchronization: A high-speed data transmission channel is established between edge devices and sensors to achieve real-time data synchronization and ensure that the timestamps of each sensor data are consistent for subsequent analysis.

[0112] Data management: Classify, store and manage the collected data, and establish an efficient data indexing mechanism to facilitate rapid retrieval and analysis. The implementation steps of the data indexing mechanism are as follows: ① Analyze data characteristics and access patterns, and select appropriate storage and indexing solutions; ② Design data models and metadata schemas; ③ Select and configure appropriate databases and storage systems; Implement data import and indexing processes; ④ Develop data access APIs and retrieval interfaces; ⑤ Implement data backup and recovery strategies; ⑥ Monitor system performance and optimize indexes and queries as needed.

[0113] 3) Real-time defect detection

[0114] Load the pre-trained defect detection model on the edge device, that is, the improved Yolov8 model. The structural diagram of the improved Yolov8 model is as follows: Figure 2As shown, LWDetect represents the output of the network. In order to detect surface defects of wind turbine blades at the edge, it is necessary to ensure high accuracy while lightweighting the network structure to enhance the model's ability to recognize different defects on the blades and speed up network reasoning. The number of traditional convolution parameters is usually large, which will increase the calculation and storage costs of the model. Therefore, this embodiment uses a fasternet module based on partial convolution (PConv) to replace the Bottleneck module of C2f. The structural diagram of the fasternet module based on partial convolution is shown in Figure 1. Figure 3 As shown, it includes a first convolutional layer (Conv), a segmentation layer (split), a first bottleneck layer (FBottleneck), a second bottleneck layer (FBottleneck), a concatenation layer (Concat) and a second convolutional layer (Conv) which are connected in sequence.

[0115] The computational complexity of PConv is shown in formula (1), and the FLOPs is only 1 / 16 of that of conventional convolution.

[0116]

[0117] In addition, the number of visits to PConv is also extremely small as shown in formula (2), which is only 1 / 4 of that of conventional Conv.

[0118]

[0119] In the above formula, h and w represent the height and width of the feature map, k is the size of the convolution kernel, and c p is the number of channels used for convolution, which is equal to 1 / 4 of the normal number of channels.

[0120] Through this reconstruction, the structure effectively reduces the impact of unnecessary redundant information on the model, improves the learning ability of the model, enables the model to better adapt to different types of defects, and improves the robustness and generalization ability of the model.

[0121] The defect detection model infers based on the input features, identifies the possible defect types and locations, and generates prompt information based on this, indicating the possible defects of the wind turbine blades and their severity.

[0122] The defect detection model is continuously optimized and adjusted based on real-time data and test results to improve detection accuracy. The specific process is as follows:

[0123] The detection results of the model are recorded, and the number of failed identifications of each type of defect is counted. If the number of failed identifications of a certain type of defect exceeds a set threshold (such as 50) or the number of all types of defects that are not correctly identified exceeds a set threshold (such as 100), the corresponding collected data is sent from the edge computing device to the cloud server for retraining, and the model is sent back to the edge computing device after the training is completed.

[0124] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0125] Example 2

[0126] This embodiment provides a multi-feature fan blade defect detection system based on edge computing, which can be used to implement the multi-feature fan blade defect detection method based on edge computing described in Example 1. The system structure is as follows Figure 4 As shown, it includes a sensor module 1, an edge computing device 2, a wireless communication module 3 and a monitoring center 4.

[0127] The sensor module 1 includes an image sensor 101 , a vibration sensor 102 , an acoustic wave sensor 103 , a temperature sensor 104 , and a humidity sensor 105 , which are used to collect image data, vibration data, acoustic wave data, temperature data, and humidity data, respectively.

[0128] The edge computing device 2 includes a high-performance processor 201 and a storage unit 202. The high-performance processor 201 is used to preprocess and extract features of the data collected by the sensor module 1, and perform edge computing on the feature extraction results through a defect detection model, that is, an improved Yolov8 model, to identify the defect of the wind turbine blade to be detected. The storage unit 202 is used to store the data collected by the sensor module 1.

[0129] The wireless communication module 3 is used to send the defect conditions of the wind turbine blades to be inspected obtained by the edge computing device 2 to the monitoring center 4. The wireless communication module 3 supports multiple communication methods such as 4G / 5G and Wi-Fi, and adopts a secure encryption algorithm to ensure the security and privacy protection of data transmission.

[0130] The monitoring center 4 includes a visualization interface 401, a cloud server 402 and a database 403. The visualization interface 401 is used to display the inspection report generated according to the defect conditions of the wind turbine blades to be inspected, including the defect type, location, severity and recommended maintenance measures. The cloud server 402 is used to regularly update the defect detection model in the edge computing device 2, and the database 403 is used to record the inspection results of the model.

[0131] The above description of the embodiments is to facilitate the understanding and use of the invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to these embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the present invention should be within the scope of protection of the present invention.

Claims

1. A multi-feature wind turbine blade defect detection method based on edge computing, suitable for edge devices in wind power generation systems, characterized in that: The following steps are involved: S1. Collecting and preprocessing multi-feature fan blade data of a fan blade to be detected, wherein the multi-feature fan blade data includes image data, vibration data, sound wave data, temperature data and humidity data; S2, extracting features from the preprocessed multi-feature wind turbine blade data; S3. Using an improved Yolov8 model in the edge device, performing edge computing on the features extracted in step S2, identifying the defects of the wind turbine blades to be inspected and generating prompt information, wherein a fasternet module based on partial convolution is used to replace the Bottleneck module of C2f in the improved Yolov8 model, and the fasternet module based on partial convolution includes a first convolution layer, a segmentation layer, a first bottleneck layer, a second bottleneck layer, a splicing layer and a second convolution layer connected in sequence.

2. The multi-feature wind turbine blade defect detection method based on edge computing according to claim 1 is characterized in that: In step S1, the preprocessing process is specifically as follows: For the image data, performing image denoising and contrast enhancement; Performing denoising and filtering on the vibration data and the sound wave data; Smoothing processing and outlier detection are performed on the temperature data and the humidity data.

3. The multi-feature wind turbine blade defect detection method based on edge computing according to claim 1 is characterized in that: In step S2, the feature extraction process is as follows: For the image data, edge detection algorithms are used to extract edge features in the image, gray-level co-occurrence matrix GLCM or local binary pattern LBP method is used to extract texture features of the image, morphological operators are used to extract shape features of the image, and color feature representation methods are used to extract color space features of the image; For the vibration data, calculate the time domain characteristics of the vibration signal, including the mean, variance, peak value and root mean square value, extract the frequency domain characteristics of the vibration signal by fast Fourier transform, including the main frequency, harmonics and frequency band energy, extract the time-frequency characteristics of the vibration signal by wavelet transform, and calculate the statistical characteristics of the vibration signal, including skewness and kurtosis; For the sound wave data, the time domain characteristics of the sound wave signal are calculated, including the mean, variance, peak value and energy, the frequency spectrum characteristics of the sound wave signal are extracted by Fourier transform, and the time-frequency characteristics of the sound wave signal are extracted by short-time Fourier transform or wavelet transform; For the temperature data and the humidity data, time series features including mean, variance and rate of change are extracted, and long-term change trends and periodic fluctuations are identified through moving average or exponential smoothing methods.

4. The multi-feature wind turbine blade defect detection method based on edge computing according to claim 3 is characterized in that: After feature extraction, the various features obtained are fused through a multimodal fusion network.

5. The multi-feature wind turbine blade defect detection method based on edge computing according to claim 1 is characterized in that: In step S2, after feature extraction is completed, feature selection and dimensionality reduction are performed.

6. The multi-feature wind turbine blade defect detection method based on edge computing according to claim 5 is characterized in that: The extracted features are selected by principal component analysis or linear discriminant analysis method.

7. A multi-feature fan blade defect detection system based on edge computing, characterized in that: Includes sensor modules, edge computing devices, wireless communication modules and monitoring centers; The sensor module includes multiple types of sensors for collecting multi-feature wind turbine blade data and sending the data to the edge computing device, wherein the multi-feature wind turbine blade data includes image data, vibration data, sound wave data, temperature data and humidity data; The edge computing device is used to preprocess and extract features of the data collected by the sensor module, perform edge computing on the feature extraction results through the improved Yolov8 model, and identify the defects of the wind turbine blades to be detected. In the improved Yolov8 model, a fasternet module based on PConv is introduced to replace the Bottleneck module of C2f; The wireless communication module is used to send the defect conditions of the wind turbine blades to be detected obtained by the edge computing device to the monitoring center; The monitoring center is used to generate prompt information according to the defect conditions of the wind turbine blades to be detected.

8. The multi-feature wind turbine blade defect detection system based on edge computing according to claim 7 is characterized in that: Before collecting data, the initial deviations of various sensors are eliminated through standard calibration procedures, and the sensor parameters, including sampling frequency and resolution, are set according to the actual application requirements and the working environment of the wind turbine blades.

9. The multi-feature wind turbine blade defect detection system based on edge computing according to claim 7, characterized in that: The edge computing device classifies, stores and manages the multi-feature wind turbine blade data through a data indexing mechanism, and the establishment process of the data indexing mechanism is as follows: Design data models and metadata schemas, and configure corresponding databases and storage systems based on data characteristics and access patterns; Design data import and indexing processes, and develop data access APIs and retrieval interfaces; Design data backup and recovery strategies, and optimize indexes and queries in real time based on system performance.

10. The multi-feature wind turbine blade defect detection system based on edge computing according to claim 7, characterized in that: The improved Yolov8 model is updated in the following way: The detection results of the model are recorded, and the number of failures in identifying each type of defect is counted. If the number of failures in identifying a certain type of defect exceeds the set threshold, the corresponding collected data is sent from the edge computing device to the monitoring center for retraining. After the monitoring center completes the training, the model is sent back to the edge computing device.

Citation Information

Cited By

  • Wind turbine blade fault identification method and system based on turbulent boundary layer trailing edge noise Mel frequency spectrum

    CN120876389A