A method and system for monitoring the condition of a fan blade

Abstract

This invention discloses a method and system for monitoring the condition of wind turbine blades, belonging to the field of wind turbine blade condition monitoring technology. By using a condition monitoring unit to collect multimodal data of wind turbine blades, stable operation can be achieved in harsh environments. An improved feature fusion algorithm is used to extract and fuse features from the preprocessed multimodal feature data, which can comprehensively reflect the condition of the blades and avoid false detections and missed detections. Subsequently, a pre-trained improved deep learning model is used to identify defects in the fused blade condition feature vectors, which significantly improves the ability to identify early micro-defects. A dynamic threshold early warning strategy is used for comparison to achieve hierarchical early warning, thereby locating the fault point and significantly improving the fault location capability. This method is suitable for wind turbine blade condition monitoring and is easy to apply and promote.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine blade condition monitoring technology, specifically relating to a method and system for monitoring the condition of wind turbine blades. Background Technology

[0002] As the core component of a wind turbine generator, the wind turbine blades are primarily used to capture wind energy and convert it into electrical energy. Their working principle is based on aerodynamics; the blades are shaped like aircraft wings, generating lift through their curved airfoil shape to drive the wind turbine generator's rotation. Because wind turbine generators are installed in harsh environments such as remote mountainous areas or offshore, they face various complex conditions including random wind loads, lightning strikes, temperature variations, rain erosion, and sand erosion. They are among the most vulnerable components of a wind turbine. The health of the blades directly determines the wind energy conversion efficiency, operational safety, and service life of the turbine. If defects such as cracks, delamination, peeling, or corrosion are not detected in time, they can lead to turbine shutdown and power generation losses, or even blade breakage and collapse, causing significant economic losses and safety hazards to the wind farm.

[0003] Currently, the mainstream wind turbine blade condition monitoring technologies on the market mainly include traditional visual inspection, non-destructive testing (NDT) technology, and digital monitoring technology. Traditional visual inspection includes methods such as manual climbing and visual inspection, and telescope observation. NDT technology includes ultrasonic testing, X-ray testing, thermal imaging testing, and acoustic emission testing. Digital monitoring technology includes vibration monitoring, acoustic fingerprint monitoring, fiber optic monitoring, and laser / visual monitoring. However, all of the current mainstream wind turbine blade condition monitoring technologies have corresponding shortcomings. For traditional visual inspection technology, the detection effect heavily depends on the experience of the inspectors, resulting in a high rate of missed detections. For large wind turbine blades, manual inspection is time-consuming, labor-intensive, and has a long inspection cycle. It cannot achieve real-time monitoring and poses safety risks such as falls from heights and electric shocks. Especially in offshore wind farms or under severe weather conditions, it can only detect visible defects on the surface of the wind turbine blades and cannot detect internal structural damage, thus limiting the detection range. For non-destructive testing technologies, ultrasonic testing requires contact operation, which is inefficient for large curved blades and has blind spots, making it difficult to cover the entire area of ​​the wind turbine blades. In high temperature and high humidity environments, the testing accuracy drops significantly. X-ray testing poses radiation risks and is costly. Thermal imaging locates defects by capturing abnormal temperatures on the blade surface, but it is greatly affected by ambient temperature and lighting conditions, resulting in a high false detection rate. Acoustic emission testing is easily interfered with by wind turbine operating noise and wind noise, making it difficult to achieve accurate defect location. For digital monitoring technologies, vibration monitoring has a sampling frequency much lower than the blade vibration frequency, resulting in signal aliasing, insufficient frequency identification accuracy, frequent calibration, and high maintenance costs. It can only monitor vibration anomalies but cannot accurately identify defect types. Acoustic fingerprint monitoring has low defect identification accuracy and is mostly passive monitoring, unable to achieve active identification of early defects and difficult defect localization. Fiber optic monitoring requires additional temperature compensation, otherwise it will lead to a decrease in detection accuracy or failure to identify micro-defects. It also poses risks of lightning strikes or electromagnetic interference, and is difficult and costly to maintain after being embedded in wind turbine blades. Laser / visual monitoring is affected by weather, has low automation, relies on pilot operation, and is affected by lighting and shooting angle images. It has weak ability to identify early micro-defects and limited detection efficiency.

[0004] However, the above monitoring technologies all have common problems, such as (1) insufficient fusion of multi-source data: most detection technologies rely only on single modal data, which cannot fully reflect the blade status and are prone to false detection and missed detection; (2) weak early defect identification capability: most existing technologies are for detecting obvious defects in the middle and late stages, and it is difficult to identify early micro defects and cannot achieve early warning; (3) poor environmental adaptability: most monitoring equipment has poor stability in harsh environments such as extreme temperature, high humidity, and strong wind, and cannot operate stably for a long time; (4) low level of intelligence: most systems still require manual intervention for data analysis and defect judgment, and cannot achieve integrated management of automatic identification, automatic early warning and automatic operation and maintenance.

[0005] Therefore, how to provide an effective technical solution to address the problems of insufficient multi-source data fusion, weak early defect identification capability, poor environmental adaptability, and low level of intelligence in existing technologies has become an urgent technical challenge. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for monitoring the condition of wind turbine blades, in order to solve the above-mentioned problems existing in the prior art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for monitoring the condition of wind turbine blades, comprising: The condition monitoring unit acquires multimodal data of the wind turbine blades, including strain data, vibration data, acoustic signature data, wind turbine blade surface images, and environmental parameters. The multimodal data is preprocessed to obtain preprocessed multimodal data; An improved feature fusion algorithm is used to extract features from the preprocessed multimodal data to obtain the blade state feature vector, and then the blade state feature vector is fused to obtain the fused blade state feature vector. The fused blade state feature vector is input into a pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level. Based on the defect identification results and the corresponding defect levels, a pre-built digital twin model is used to assess the condition of the wind turbine blades, and the condition assessment results are obtained. A dynamic threshold early warning strategy is used to compare the defect level and status assessment results to obtain graded early warning signals, and the location of the fault point is determined based on the graded early warning signals.

[0008] In one possible design, the multimodal data is preprocessed to obtain preprocessed multimodal data, including: Missing values ​​are imputed in the multimodal data to obtain the imputed multimodal data; An adaptive thresholding method based on the isolated forest algorithm is used to remove outliers from the filled multimodal data, resulting in the removed multimodal data. The removed multimodal data is downsampled, and the downsampled multimodal data is then standardized to obtain preprocessed multimodal data.

[0009] In one possible design, the condition monitoring unit collects multimodal data of the wind turbine blades, including: Acquire strain and vibration data from a distributed fiber optic MEMS sensing array located in the stress concentration area of ​​the wind turbine blades; Acquire the acoustic signature signals collected by the acoustic signature acquisition modules located in the nacelle and at the root of the wind turbine blades, perform preliminary filtering on the acoustic signature signals, and obtain the acoustic signature data; Acquire images of the wind turbine blade surface captured by the vision acquisition module; The environmental parameters collected by the environmental sensing module include wind data, temperature data, humidity data, and lightning strike signals.

[0010] In one possible design, the blade state feature vector includes acoustic modal features, image modal features, and sensor modal features; The improved feature fusion algorithm is used to extract features from the preprocessed multimodal data to obtain the blade state feature vector, including: MFCC was used to extract features from the preprocessed multimodal data to obtain sound modal features; The improved ResNet-50 network was used to extract features from the preprocessed multimodal data to obtain image modal features; Wavelet decomposition is used to extract features from the preprocessed multimodal data to obtain sensing modal features.

[0011] In one possible design, the pre-trained improved deep learning model includes a fully connected layer, a channel-space attention layer, and an output layer; The step of inputting the fused blade state feature vector into a pre-trained improved deep learning model to obtain defect identification results and corresponding defect levels includes: The fused blade state feature vector is normalized and then input into the fully connected layer. A fully connected layer is used to perform a linear transformation on the normalized blade state feature vector, and the first activation function is used to activate the transformed blade state feature vector to obtain the blade state feature map. The channel spatial attention layer is used to perform global average pooling and global max pooling on the blade state feature map to obtain a first vector and a second vector. The first vector and the second vector are added element by element to obtain a fusion vector. The fusion vector is activated using a second activation function to obtain channel weights. The channel weights are then weighted with the blade state feature map to obtain a weighted blade state feature map. The channel space attention layer is used to perform max pooling and average pooling on the weighted blade state feature map respectively. The blade state feature map after max pooling and the blade state feature map after average pooling are concatenated to obtain the concatenated state feature map. The concatenated state feature map is then convolved to obtain the attention-enhanced blade state feature vector. The output layer is used to activate and classify the attention-enhanced blade state feature vector to obtain the probability distribution of each category. The category with the highest probability distribution is taken as the defect identification result, and the corresponding defect level is obtained based on the defect identification result. The defect identification result includes blade surface cracks, corrosion, degumming, and internal delamination.

[0012] In one possible design, the location of the fault point is determined based on tiered early warning signals, including: The distance to the fault point is obtained by mapping the graded early warning signals based on a preset distance mapping function; Based on the distance to the fault point, the location of the fault point is calculated using the least squares method.

[0013] In one possible design, after determining the location of the fault point based on the graded early warning signal, the following is also included: A blade life prediction model was constructed based on LSTM neural network and grey prediction algorithm; Based on the location of the fault point and the pre-defined defect development pattern, the blade life prediction model is used to predict the life of the wind turbine blades, and the wind turbine blade life prediction results are obtained. Based on the predicted lifespan and defect levels of the wind turbine blades, an operation and maintenance strategy is derived.

[0014] Secondly, the present invention provides a wind turbine blade condition monitoring system, comprising: The data acquisition module is used to acquire multimodal data of the wind turbine blades collected by the condition monitoring unit. The multimodal data includes strain data, vibration data, acoustic signature data, wind turbine blade surface images, and environmental parameters. The data preprocessing module is used to preprocess multimodal data to obtain preprocessed multimodal data; The feature extraction and fusion module is used to extract features from the preprocessed multimodal data using an improved feature fusion algorithm to obtain the blade state feature vector, and to fuse the blade state feature vector to obtain the fused blade state feature vector. The defect identification module is used to input the fused blade state feature vector into a pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level. The condition assessment module is used to assess the condition of wind turbine blades using a pre-built digital twin model based on the defect identification results and the corresponding defect levels, and obtain the condition assessment results. The early warning and location module is used to compare the defect level and status assessment results using a dynamic threshold early warning strategy to obtain graded early warning signals, and to determine the location of the fault point based on the graded early warning signals.

[0015] In one possible design, the system further includes a cloud management module and a visualization module. The cloud management module is communicatively connected to the data acquisition module, the data preprocessing module, the feature extraction and fusion module, the defect identification module, the status assessment module, and the early warning and positioning module, respectively. The visualization module is communicatively connected to the cloud management module. The cloud management module is used to store status monitoring data, which includes multimodal data, preprocessed multimodal data, fused blade status feature vector, defect identification results, corresponding defect levels, status assessment results, and the location of the fault point. The visualization module is used to visualize the defect identification results, the corresponding defect level, the status assessment results, and the location of the fault point.

[0016] Thirdly, the present invention provides a wind turbine blade condition monitoring system, comprising a memory, a processor, and a transceiver connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the wind turbine blade condition monitoring method as described in the first aspect above.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the wind turbine blade status monitoring method as described in the first aspect above.

[0018] Fifthly, the present invention provides a computer program product containing instructions that, when the instructions are executed on a computer, cause the computer to perform the wind turbine blade condition monitoring method as described in the first aspect above.

[0019] The beneficial effects of this invention are as follows: This invention discloses a method and system for monitoring the condition of wind turbine blades. It acquires multimodal data of the wind turbine blades through a condition monitoring unit, preprocesses the multimodal data to obtain preprocessed multimodal data, extracts features from the preprocessed multimodal data using an improved feature fusion algorithm to obtain blade condition feature vectors, fuses these feature vectors to obtain fused blade condition feature vectors, inputs the fused blade condition feature vectors into a pre-trained improved deep learning model to obtain defect identification results and corresponding defect levels, assesses the condition of the wind turbine blades using a pre-constructed digital twin model based on the defect identification results and corresponding defect levels, obtains condition assessment results, compares the defect levels and condition assessment results using a dynamic threshold early warning strategy to obtain graded early warning signals, and determines the location of the fault point based on the graded early warning signals. This invention collects multimodal data of wind turbine blades using a condition monitoring unit, enabling stable operation in harsh environments. An improved feature fusion algorithm is employed to extract and fuse features from the preprocessed multimodal data, comprehensively reflecting the blade's condition and avoiding false positives and false negatives. Subsequently, a pre-trained improved deep learning model is used to identify defects in the fused blade condition feature vector, significantly improving the ability to identify early micro-defects. A dynamic threshold warning strategy is used for comparison to achieve tiered warnings, thereby locating the fault point and significantly improving fault location capabilities. This invention is suitable for wind turbine blade condition monitoring and is easy to apply and promote. Attached Figure Description

[0020] Figure 1 A flowchart of a wind turbine blade condition monitoring method provided in an embodiment of the present invention; Figure 2 This is a block diagram of a wind turbine blade condition monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0022] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0023] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0024] Example: like Figure 1 As shown, the first aspect of this embodiment provides a method for monitoring the state of wind turbine blades, which can be executed, but is not limited to, by a computer device or virtual machine with certain computing resources, such as a personal computer or smartphone, or by a virtual machine; the method for monitoring the state of wind turbine blades includes, but is not limited to, the following steps: S1. The condition monitoring unit collects multimodal data of the wind turbine blades, including strain data, vibration data, acoustic signature data, wind turbine blade surface images, and environmental parameters; In specific implementation, the state monitoring unit in this embodiment includes a distributed fiber optic MEMS sensor array, an acoustic fingerprint acquisition module, a visual acquisition module, and an environmental perception module. The distributed fiber optic MEMS sensor array acquires strain and vibration data, the acoustic fingerprint acquisition module acquires acoustic fingerprint data, the visual acquisition module acquires images of the wind turbine blade surface, and the environmental perception module acquires environmental parameters.

[0025] Specifically, in step S1, the multimodal data of the wind turbine blades collected by the condition monitoring unit includes: S11. Acquire strain and vibration data from a distributed fiber optic MEMS sensor array located in the stress concentration area of ​​the wind turbine blades; S12. Acquire the acoustic signature signal collected by the acoustic signature acquisition module set in the nacelle and the root of the wind turbine blades, perform preliminary filtering on the acoustic signature signal, and obtain the acoustic signature data; S13. Acquire the image of the wind turbine blade surface captured by the vision acquisition module; S14. Obtain environmental parameters collected by the environmental sensing module, including wind data, temperature data, humidity data, and lightning strike signals.

[0026] It should be noted that the distributed fiber optic MEMS sensor array adopts a passive design, embedding it into stress-concentrated areas such as the main beam and / or web of the wind turbine blade. Each wind turbine blade is equipped with 20 sensing nodes, enabling the acquisition of strain and vibration data in extreme environments ranging from -40℃ to 85℃. The sampling frequency is set to 100Hz, with the specific frequency dynamically adjusted according to the blade's operating status, ranging from 10Hz to 1000Hz, without a specific limitation. It has the ability to resist lightning strikes and electromagnetic interference, with a service life of no less than 20 years and no need for frequent calibration. The acoustic signature acquisition module includes a high-sensitivity microphone array, consisting of 4 microphones, which are installed in the nacelle and on the blades. The root of the device is used to collect acoustic fingerprint signals during propeller operation and to initially filter noise from the collected acoustic fingerprint signals. The visual acquisition module includes a gimbal high-definition camera and a drone inspection device. The gimbal high-definition camera has a resolution of 1080P, and the drone inspection device has a flight time of 60 minutes. By fixing the gimbal camera to the top of the drone inspection device's nacelle, it captures images of the propeller surface in real time. The drone inspection device adopts a fully automatic flight mode and performs a comprehensive scan of the propellers periodically, such as a comprehensive scan of 3 propellers per week. It also performs preliminary real-time preprocessing of the propeller images based on edge computing to improve the real-time response rate of the images. The environmental perception module is used to collect wind data, temperature data, humidity data, and lightning strike signals to facilitate subsequent data preprocessing.

[0027] S2. Preprocess the multimodal data to obtain preprocessed multimodal data; Specifically, in step S2, the multimodal data is preprocessed to obtain preprocessed multimodal data, including: S21. Impute missing values ​​in the multimodal data to obtain the imputed multimodal data; S22. Use an adaptive thresholding method based on the isolated forest algorithm to remove outliers from the filled multimodal data to obtain the removed multimodal data; S23. Downsample the removed multimodal data and standardize the downsampled multimodal data to obtain preprocessed multimodal data.

[0028] It should be noted that this embodiment uses linear interpolation to fill missing values ​​in multimodal data. The principle of linear interpolation is to assume that the function change between two known points is linear, and to estimate the value at any position in the interval by using a straight line connecting the two points, thereby obtaining the missing value. The principle of the isolated forest algorithm is to utilize the characteristics of outlier data, and to construct multiple isolation trees by randomly selecting features and randomly partitioning values, repeatedly splitting the data points until each point is isolated. Therefore, since normal data is densely distributed, multiple splits are required for isolation, while outlier data is less numerous and has obvious feature differences, requiring only a few splits to separate them. The shorter the average path, the easier it is for the point to be isolated, and the higher its outlier score, thus achieving the removal of outliers from the filled multimodal data. Subsequently, the removed multimodal data is downsampled to 50Hz. Through the above preprocessing, irrelevant data such as wind noise and electromagnetic interference can be effectively filtered, significantly improving data quality.

[0029] S3. An improved feature fusion algorithm is used to extract features from the preprocessed multimodal data to obtain the blade state feature vector, and the blade state feature vector is fused to obtain the fused blade state feature vector. The blade state feature vector includes acoustic modal features, image modal features, and sensor modal features; It should be noted that monitoring data of normal blade operation under 32 typical natural environmental conditions was obtained. This data was used as training data to construct an initial improved feature fusion algorithm based on a hard sample mining strategy. Subsequently, the initial improved feature fusion algorithm was trained using the training data and a paired loss function to obtain an improved feature fusion algorithm. This enables supervised learning of multimodal feature domains, which can solve the problem of heterogeneity of multimodal data, significantly improve fusion accuracy, and avoid the limitations caused by single-modal data.

[0030] In a preferred embodiment, step S3, which involves using an improved feature fusion algorithm to extract features from the preprocessed multimodal data to obtain a blade state feature vector, includes: S31. MFCC is used to extract features from the preprocessed multimodal data to obtain sound modal features; S32. Use an improved ResNet-50 network to extract features from the preprocessed multimodal data to obtain image modal features; S33. Wavelet decomposition is used to extract features from the preprocessed multimodal data to obtain sensing modal features.

[0031] It should be noted that MFCC (Mel-frequency cepstral coefficients) is an audio feature extraction method that simulates the auditory characteristics of the human ear. Its principle is to transform the speech signal from the time domain to a Mel-scale frequency domain that is closer to human auditory perception, and then extract the spectral envelope through cepstral analysis to obtain the feature representation. In this embodiment, MFCC technology is used to extract 36 frames of second-order feature vectors, i.e., sound modal features. The ResNet-50 network is a deep convolutional neural network. Its core principle lies in using the network to learn residual mappings and using three-layer residual blocks as basic units. Through skip connections, the input is directly superimposed onto the output of the convolutional layer to form an identity mapping, effectively... To address the issues of gradient vanishing and degradation, an improved ResNet-50 network can be implemented by adjusting its downsampling strategy to achieve higher resolution in the extracted image modal features. In this embodiment, the improved ResNet-50 network extracts and aggregates 68-dimensional feature vectors to obtain image modal features. Wavelet decomposition is a signal analysis method based on wavelet transform, which separates signals according to frequency components through multi-scale decomposition to obtain sensing modal features. In this embodiment, wavelet decomposition is performed on the preprocessed multimodal data to obtain an 8-dimensional feature representation, which is then concatenated into a 72-dimensional feature vector.

[0032] S4. Input the fused blade state feature vector into the pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level; The pre-trained improved deep learning model includes a fully connected layer, a channel space attention layer, and an output layer. In a preferred embodiment, step S4 involves inputting the fused blade state feature vector into a pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level, including: S41. Normalize the fused blade state feature vector and input the normalized blade state feature vector into the fully connected layer; S42. A fully connected layer is used to perform a linear transformation on the normalized blade state feature vector, and the first activation function is used to activate the transformed blade state feature vector to obtain the blade state feature map. S43. Use the channel spatial attention layer to perform global average pooling and global max pooling on the blade state feature map to obtain the first vector and the second vector. Add the first vector and the second vector element by element to obtain the fusion vector. Use the second activation function to activate the fusion vector to obtain the channel weights. Add the channel weights to the blade state feature map to obtain the weighted blade state feature map. S44. Use the channel space attention layer to perform max pooling and average pooling on the weighted blade state feature map respectively, and then concatenate the blade state feature map after max pooling and the blade state feature map after average pooling to obtain the concatenated state feature map. Then, perform convolution on the concatenated state feature map to obtain the attention-enhanced blade state feature vector. S45. Use the output layer to activate and classify the attention-enhanced blade state feature vector to obtain the probability distribution of each category. Take the category with the highest probability distribution as the defect identification result, and obtain the corresponding defect level based on the defect identification result. The defect identification result includes blade surface cracks, corrosion, degumming, and internal delamination.

[0033] It should be noted that the corresponding defect levels include Level 1 minor defect, Level 2 moderate defect, and Level 3 severe defect. In this embodiment, the improved deep learning model is built based on the ResNet-50 network. Specifically, the last fully connected layer of the ResNet-50 network is replaced with a fully connected layer containing 128 neurons, and an attention mechanism is introduced to focus on areas of stress concentration and vulnerable parts of the blade, significantly improving the ability and efficiency of early micro-defect identification. In specific implementation, when the defect identification result is a minor crack, its corresponding defect level is Level 1 minor defect.

[0034] S5. Based on the defect identification results and the corresponding defect levels, the state of the wind turbine blades is assessed using a pre-built digital twin model to obtain the state assessment results; In practice, blade design parameters, manufacturing data, and real-time monitoring data are acquired. A digital twin model is constructed based on these data. The digital twin model enables visualization and simulation of blade condition and defect development trends, providing an intuitive reflection of blade health. Based on defect identification results and corresponding defect levels, a pre-constructed digital twin model is used for condition assessment. If the defect identification result is a minor crack and the corresponding defect level is a level 1 minor defect, the condition assessment result shows that the minor crack is in a stable state and does not show any expansion trend, indicating that the overall health of the blade is good.

[0035] S6. Use a dynamic threshold early warning strategy to compare the defect level and status assessment results to obtain graded early warning signals, and determine the location of the fault point based on the graded early warning signals.

[0036] In practice, when the defect level is a minor defect (Level 1), a Level 1 warning signal is obtained. This signal can be pushed to the mobile phone of maintenance personnel to remind them to conduct regular inspections. The coordinates of the fault point are determined by the graded warning signals and the location information collected by the status monitoring unit. In this embodiment, the dynamic threshold warning strategy can dynamically adjust the warning threshold according to environmental parameters and blade operating status, enabling graded warnings for sudden and gradual faults, improving warning response time, reducing false alarm rate, and ensuring that faults can be detected and handled quickly.

[0037] Furthermore, the tiered early warning signals include Level 1, Level 2, and Level 3 early warning signals. When the tiered early warning signal is Level 1, indicating a minor defect, it prompts maintenance personnel to conduct regular inspections. When the tiered early warning signal is Level 2, indicating a moderate defect, it prompts maintenance personnel to shut down the system for inspection. When the tiered early warning signal is Level 1, indicating a severe defect, it triggers an emergency shutdown operation.

[0038] Specifically, in step S6, determining the location of the fault point based on the graded early warning signal includes: S61. Map the graded early warning signals based on a preset distance mapping function to obtain the distance to the fault point; S62. Based on the distance to the fault point, calculate the location of the fault point using the least squares method.

[0039] In practice, the positioning information collected by the status monitoring unit is obtained, and the positioning information and the graded early warning signal are mapped using a preset distance mapping function to obtain the fault point distance. At this time, the obtained fault point distance is the distance of multiple fault points mapped by the positioning information of multiple sensors. Therefore, the least squares method is used to locate the multiple fault point distances to obtain the final fault point location.

[0040] Furthermore, after determining the location of the fault point based on the graded early warning signal, it also includes: A blade life prediction model is constructed based on LSTM neural network and grey prediction algorithm; based on the location of the fault point and the preset defect development law, the blade life prediction model is used to predict the life of the wind turbine blades, and the wind turbine blade life prediction results are obtained; based on the wind turbine blade life prediction results and defect level, the operation and maintenance strategy is obtained.

[0041] In this embodiment, based on the predicted lifespan and defect level of the wind turbine blades, a personalized operation and maintenance strategy is obtained, such as maintenance time, maintenance method or spare parts replacement plan, so as to realize intelligent operation and maintenance management of the entire life cycle of wind turbine blade design, manufacturing, operation and maintenance.

[0042] The first aspect of this embodiment discloses a method for monitoring the condition of wind turbine blades. This method utilizes a passive design with a distributed fiber optic MEMS sensor array, offering advantages such as resistance to lightning strikes, electromagnetic interference, high and low temperatures, and high humidity. It exhibits strong environmental adaptability, requires no frequent calibration, and has a long service life. By analyzing multimodal data, it can capture early microscopic defects in the blades and, combined with a dynamic threshold early warning strategy, achieve tiered early warning, shifting from "post-incident maintenance" to "pre-incident early warning," preventing defects from escalating and causing serious accidents. Feature extraction and feature fusion of the multimodal data significantly improve the accuracy of defect identification. By reducing the rates of missed and false detections and combining location point information to pinpoint fault locations, defects can be quickly located, reducing the troubleshooting time for maintenance personnel and improving maintenance efficiency. Simultaneously, by integrating data from the entire blade lifecycle, blade lifespan prediction can be achieved, reducing maintenance costs and increasing the return on investment for wind farms. The overall condition monitoring process is fully automated, requiring no manual intervention, which reduces the workload of maintenance personnel. It is suitable for the intelligent management of large-scale wind farms. The condition monitoring unit equipment reduces equipment replacement and calibration costs, minimizing investment in single high-precision devices, and combines detection accuracy with cost-effectiveness, making it widely applicable in small and medium-sized wind farms.

[0043] like Figure 2 As shown, the second aspect of this embodiment provides a wind turbine blade condition monitoring system, including: The data acquisition module is used to acquire multimodal data of the wind turbine blades collected by the condition monitoring unit. The multimodal data includes strain data, vibration data, acoustic signature data, wind turbine blade surface images, and environmental parameters. The data preprocessing module is used to preprocess multimodal data to obtain preprocessed multimodal data; The feature extraction and fusion module is used to extract features from the preprocessed multimodal data using an improved feature fusion algorithm to obtain the blade state feature vector, and to fuse the blade state feature vector to obtain the fused blade state feature vector. The defect identification module is used to input the fused blade state feature vector into a pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level. The condition assessment module is used to assess the condition of wind turbine blades using a pre-built digital twin model based on the defect identification results and the corresponding defect levels, and obtain the condition assessment results. The early warning and location module is used to compare the defect level and status assessment results using a dynamic threshold early warning strategy to obtain graded early warning signals, and to determine the location of the fault point based on the graded early warning signals.

[0044] In one possible design, the system further includes a cloud management module and a visualization module. The cloud management module is communicatively connected to the data acquisition module, the data preprocessing module, the feature extraction and fusion module, the defect identification module, the status assessment module, and the early warning and positioning module, respectively. The visualization module is communicatively connected to the cloud management module. The cloud management module is used to store status monitoring data, which includes multimodal data, preprocessed multimodal data, fused blade status feature vector, defect identification results, corresponding defect levels, status assessment results, and the location of the fault point. The visualization module is used to visualize the defect identification results, the corresponding defect level, the status assessment results, and the location of the fault point.

[0045] Furthermore, the cloud management module supports linkage with the wind turbine main control PLC to achieve automatic shutdown control in case of failure, which can minimize safety hazards. The cloud management module adopts an edge-cloud collaborative structure. The edge performs data preprocessing, real-time defect identification and rapid early warning, while the cloud performs data storage, model training, life prediction and global operation and maintenance management, thereby reducing the pressure of data transmission and improving the system response speed.

[0046] The working process, working details and technical effects of the wind turbine blade condition monitoring system provided in the second aspect of this embodiment can be found in the wind turbine blade condition monitoring method described in the first aspect, and will not be repeated here.

[0047] This embodiment provides a wind turbine blade condition monitoring system, including a memory, a processor, and a transceiver connected in sequence. The memory stores a computer program, the transceiver sends and receives messages, and the processor reads the computer program and executes the wind turbine blade condition monitoring method described in the first aspect. Specifically, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-input first-output (FIFO) memory, and / or first-input last-output (FILO) memory, etc.; the processor may include, but is not limited to, an STM32F105 series microprocessor. Furthermore, the computer device may also include, but is not limited to, a power supply module, a display screen, and other necessary components.

[0048] The working process, working details and technical effects of the aforementioned wind turbine blade condition monitoring system provided in the third aspect of this embodiment can be found in the wind turbine blade condition monitoring method described in the first aspect, and will not be repeated here.

[0049] The fourth aspect of this embodiment provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the wind turbine blade state monitoring method described in the first aspect is performed. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0050] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be found in the wind turbine blade condition monitoring method described in the first aspect, and will not be repeated here.

[0051] The fifth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, are used to implement the wind turbine blade condition monitoring method as described in the first aspect.

[0052] The working process, working details and technical effects of the aforementioned computer program product provided in this embodiment can be found in the wind turbine blade condition monitoring method described in the first aspect, and will not be repeated here.

[0053] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring the condition of wind turbine blades, characterized in that, include: The condition monitoring unit acquires multimodal data of the wind turbine blades, including strain data, vibration data, acoustic signature data, wind turbine blade surface images, and environmental parameters. The multimodal data is preprocessed to obtain preprocessed multimodal data; An improved feature fusion algorithm is used to extract features from the preprocessed multimodal data to obtain the blade state feature vector, and then the blade state feature vector is fused to obtain the fused blade state feature vector. The fused blade state feature vector is input into a pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level. Based on the defect identification results and the corresponding defect levels, a pre-built digital twin model is used to assess the condition of the wind turbine blades, and the condition assessment results are obtained. A dynamic threshold early warning strategy is used to compare the defect level and status assessment results to obtain graded early warning signals, and the location of the fault point is determined based on the graded early warning signals.

2. The method for monitoring the condition of a wind turbine blade according to claim 1, characterized in that, The multimodal data is preprocessed to obtain preprocessed multimodal data, including: Missing values ​​are imputed in the multimodal data to obtain the imputed multimodal data; An adaptive thresholding method based on the isolated forest algorithm is used to remove outliers from the filled multimodal data, resulting in the removed multimodal data. The removed multimodal data is downsampled, and the downsampled multimodal data is then standardized to obtain preprocessed multimodal data.

3. The method for monitoring the condition of wind turbine blades according to claim 1, characterized in that, The acquisition status monitoring unit collects multimodal data of the wind turbine blades, including: Acquire strain and vibration data from a distributed fiber optic MEMS sensing array located in the stress concentration area of ​​the wind turbine blades; Acquire the acoustic signature signals collected by the acoustic signature acquisition modules located in the nacelle and at the root of the wind turbine blades, perform preliminary filtering on the acoustic signature signals, and obtain the acoustic signature data; Acquire images of the wind turbine blade surface captured by the vision acquisition module; The environmental parameters collected by the environmental sensing module include wind data, temperature data, humidity data, and lightning strike signals.

4. The method for monitoring the condition of a wind turbine blade according to claim 1, characterized in that, The blade state feature vector includes acoustic modal features, image modal features, and sensor modal features; The improved feature fusion algorithm is used to extract features from the preprocessed multimodal data to obtain the blade state feature vector, including: MFCC was used to extract features from the preprocessed multimodal data to obtain sound modal features; The improved ResNet-50 network was used to extract features from the preprocessed multimodal data to obtain image modal features; Wavelet decomposition is used to extract features from the preprocessed multimodal data to obtain sensing modal features.

5. The method for monitoring the condition of a wind turbine blade according to claim 1, characterized in that, The pre-trained improved deep learning model includes a fully connected layer, a channel space attention layer, and an output layer; The step of inputting the fused blade state feature vector into a pre-trained improved deep learning model to obtain defect identification results and corresponding defect levels includes: The fused blade state feature vector is normalized and then input into the fully connected layer. A fully connected layer is used to perform a linear transformation on the normalized blade state feature vector, and the first activation function is used to activate the transformed blade state feature vector to obtain the blade state feature map. The channel spatial attention layer is used to perform global average pooling and global max pooling on the blade state feature map to obtain a first vector and a second vector. The first vector and the second vector are added element by element to obtain a fusion vector. The fusion vector is activated using a second activation function to obtain channel weights. The channel weights are then weighted with the blade state feature map to obtain a weighted blade state feature map. The channel space attention layer is used to perform max pooling and average pooling on the weighted blade state feature map respectively. The blade state feature map after max pooling and the blade state feature map after average pooling are concatenated to obtain the concatenated state feature map. The concatenated state feature map is then convolved to obtain the attention-enhanced blade state feature vector. The output layer is used to activate and classify the attention-enhanced blade state feature vector to obtain the probability distribution of each category. The category with the highest probability distribution is taken as the defect identification result, and the corresponding defect level is obtained based on the defect identification result. The defect identification result includes blade surface cracks, corrosion, degumming, and internal delamination.

6. The method for monitoring the condition of a wind turbine blade according to claim 1, characterized in that, Determining the location of the fault point based on tiered early warning signals includes: The distance to the fault point is obtained by mapping the graded early warning signals based on a preset distance mapping function; Based on the distance to the fault point, the location of the fault point is calculated using the least squares method.

7. The method for monitoring the condition of a wind turbine blade according to claim 1, characterized in that, After determining the location of the fault point based on the graded early warning signals, the process also includes: A blade life prediction model was constructed based on LSTM neural network and grey prediction algorithm; Based on the location of the fault point and the pre-defined defect development pattern, the blade life prediction model is used to predict the life of the wind turbine blades, and the wind turbine blade life prediction results are obtained. Based on the predicted lifespan and defect levels of the wind turbine blades, an operation and maintenance strategy is derived.

8. A condition monitoring system for wind turbine blades, used to implement the method according to any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire multimodal data of the wind turbine blades collected by the condition monitoring unit. The multimodal data includes strain data, vibration data, acoustic signature data, wind turbine blade surface images, and environmental parameters. The data preprocessing module is used to preprocess multimodal data to obtain preprocessed multimodal data; The feature extraction and fusion module is used to extract features from the preprocessed multimodal data using an improved feature fusion algorithm to obtain the blade state feature vector, and to fuse the blade state feature vector to obtain the fused blade state feature vector. The defect identification module is used to input the fused blade state feature vector into a pre-trained improved deep learning model to obtain the defect identification result and the corresponding defect level. The condition assessment module is used to assess the condition of wind turbine blades using a pre-built digital twin model based on the defect identification results and the corresponding defect levels, and obtain the condition assessment results. The early warning and location module is used to compare the defect level and status assessment results using a dynamic threshold early warning strategy to obtain graded early warning signals, and to determine the location of the fault point based on the graded early warning signals.

9. A wind turbine blade condition monitoring system according to claim 8, characterized in that, The system also includes a cloud management module and a visualization module. The cloud management module is communicatively connected to the data acquisition module, data preprocessing module, feature extraction and fusion module, defect identification module, status assessment module, and early warning and positioning module, respectively. The visualization module is communicatively connected to the cloud management module. The cloud management module is used to store status monitoring data, which includes multimodal data, preprocessed multimodal data, fused blade status feature vector, defect identification results, corresponding defect levels, status assessment results, and the location of the fault point. The visualization module is used to visualize the defect identification results, the corresponding defect level, the status assessment results, and the location of the fault point.

10. A condition monitoring system for wind turbine blades, characterized in that, The device includes a memory, a processor, and a transceiver that are connected in series and communicate with each other. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the wind turbine blade condition monitoring method as described in any one of claims 1 to 7.

Images