Wind Turbine Blade Monitoring System and Method Based on LiDAR and Multidimensional Data Fusion
By fusing high-frequency lidar and SCADA data, combined with a CNN-SVM model and dual verification, the interference and frequency drift problems in blade balance monitoring were solved, achieving high-precision and reliable blade balance diagnosis and reducing operation and maintenance costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHUHAI GUANGHENG TECH CO LTD
- Filing Date
- 2025-12-26
- Publication Date
- 2026-05-26
AI Technical Summary
Existing blade balance monitoring methods rely on contact sensors, which are susceptible to interference and have insufficient reliability in diagnostic results. Traditional frequency domain analysis is prone to frequency drift, and single data sources lack redundant verification, leading to misjudgments or missed judgments.
High-frequency lidar was used to measure the distance between the blade and the tower. Combined with SCADA data, a three-dimensional spectral image was constructed through moving average filtering, order tracking analysis, and multi-dimensional data fusion. The diagnostic model was then fused using CNN-SVM and double-validated to ensure the accuracy and reliability of the diagnostic results.
It achieves non-contact, interference-resistant blade balance monitoring with high diagnostic accuracy, reduces operation and maintenance costs, adapts to different wind field environments, reduces misjudgments and omissions, and extends the life of key components.
Smart Images
Figure CN121384320B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind power equipment condition monitoring technology, and in particular to a wind turbine blade monitoring system and method based on lidar and multi-dimensional data fusion. Background Technology
[0002] As the core component for capturing wind energy, the operational balance of wind turbine blades directly affects the reliability, safety, and power generation efficiency of the entire turbine. During long-term operation, blades are susceptible to uneven mass distribution due to factors such as wind and sand erosion, snow accumulation, and material fatigue. This leads to unbalanced vibrations, which in turn trigger a chain reaction of problems, including nacelle resonance, accelerated wear of the transmission chain, and shortened bearing life. In severe cases, it can even cause major safety accidents such as blade breakage and tower collapse.
[0003] Existing blade balance monitoring methods mainly rely on indirect signal acquisition using contact sensors, such as measuring vibration acceleration through accelerometers installed in the nacelle, or indirectly judging the imbalance state through fluctuations in generator current and output power. However, these methods have significant technical drawbacks: First, the installation of contact sensors alters the structural dynamics of the blade or nacelle and is susceptible to interference from drivetrain vibration and nacelle swaying, resulting in low signal-to-noise ratios and difficulty in effectively extracting fault characteristics. Second, the rotational speed of wind turbines changes dynamically due to wind speed, leading to frequency drift in traditional frequency domain analysis methods, which masks the characteristic frequencies of imbalance. Third, monitoring methods based on a single data source lack redundant verification, resulting in insufficient reliability of diagnostic results and a high risk of misjudgment or missed diagnosis.
[0004] In existing technologies, some solutions attempt to use lidar for blade clearance monitoring (such as CN118979854A), but this is only used for collision risk warning and does not involve balance state diagnosis. Other solutions achieve imbalance monitoring by fusing wind speed and blade root load (such as CN120402303A), but this relies on contact-based measurements from load sensors and still does not solve the interference problem. Therefore, developing a non-contact, interference-resistant, and highly accurate blade balance monitoring method has become an urgent technical need for the wind power industry. Summary of the Invention
[0005] The purpose of this invention is to provide a non-contact, anti-interference, and high diagnostic accuracy wind turbine blade monitoring system and method based on lidar and multi-dimensional data fusion.
[0006] The technical solution adopted in this invention is a method for monitoring wind turbine blades based on lidar and multi-dimensional data fusion, which includes the following steps:
[0007] S1. Data acquisition: Using high-frequency lidar deployed on the wind turbine tower or nacelle, the dynamic distance data between the blade tip and the tower is measured in real time, and the generator operation data output by the SCADA system is collected simultaneously.
[0008] S2. Data preprocessing: Denoise the lidar distance data collected in step S1 by using a moving average filtering algorithm to eliminate environmental noise interference and synchronizing and aligning the timestamps of the SCADA data to ensure consistency in the time dimension of the two types of data.
[0009] S3. Feature Extraction: Perform order tracking analysis on the preprocessed lidar range data, converting the frequency domain signal into an order domain signal, and extracting the feature order f corresponding to blade imbalance. o =k×n / 60, where k is the order coefficient, k=1, 2, 3, and n is the real-time blade rotation speed in r / min. Simultaneously, power spectral density analysis is performed on the synchronous data to calculate the peak spectral density P. s ;
[0010] S4. Data fusion, based on the feature order and spectral peak value P extracted in step S3. s Combined with SCADA operation data, a three-dimensional multivariate spectrum image is constructed, wherein the horizontal axis of the image represents the order value, the vertical axis represents the power spectral density value, and the depth axis represents the normalized rotational speed value;
[0011] S5. Balance state diagnosis: Input the multivariate spectrum image constructed in step S4 into the trained CNN-SVM fusion diagnostic model, and output the blade imbalance state identification result. The model is trained by historical fault data and normal operation data. The identification result includes four levels: normal balance, slight imbalance, moderate imbalance and severe imbalance.
[0012] S6. Dual verification: Calculation of the theoretical unbalance M of the blade based on the principle of rigid body rotational dynamics. th The consistency of the diagnosis results in step S5 is verified. When the deviation between the theoretical calculation value and the model output result is less than the set percentage, the diagnosis result is confirmed to be valid and output.
[0013] Furthermore, in step S1, the measurement accuracy of the high-frequency lidar is not less than ±0.1mm, the measurement distance range is 0.5~50m, the laser emission frequency is 5~20kHz, and the laser wavelength is selected as 905nm or 1550nm.
[0014] Furthermore, in step S2, the window size N of the moving average filtering algorithm satisfies N = 2 × floor[f s / (20×f max )]+1, where f s f is the sampling frequency.max This is the highest operating frequency of the blades.
[0015] Furthermore, in step S3, the order tracking analysis adopts a resampling algorithm based on instantaneous rotational speed. Specifically, based on the rotational speed n(t) collected in real time by the SCADA system, the instantaneous angular frequency ω(t) = 2πn(t) / 60 is calculated, and the distance signal is resampled at equal angular intervals based on ω(t) to eliminate frequency drift caused by variable rotational speed.
[0016] Furthermore, in step S3, the power spectral density analysis uses the Welch method to process the signal segments by overlapping, with an overlap rate of 50%. The window function is the Hanning window, and the spectral resolution Δf ≤ 0.1 Hz.
[0017] Further, in step S5, the construction process of the CNN-SVM fusion diagnostic model is as follows: a CNN network is used to extract deep feature vectors from multivariate spectral images. The CNN network contains 3 convolutional layers, 2 pooling layers, and 1 fully connected layer. The 128-dimensional feature vector output by the fully connected layer is input into the SVM classifier, and the radial basis function is used to achieve state classification.
[0018] Furthermore, in step S6, the theoretical imbalance quantity M th The calculation model is: M th =(J×Δω) / (r×t), where J is the blade's moment of inertia, Δω is the change in angular acceleration, r is the distance from the blade tip to the center of rotation, and t is the length of the monitoring time window.
[0019] A wind turbine blade monitoring system based on lidar and multidimensional data fusion for implementing the method described above, the system comprising:
[0020] The lidar acquisition module is used to acquire dynamic distance data between the blade tip and the tower. The lidar acquisition module includes a high-frequency lidar sensor, a data transmission unit, and a mounting bracket.
[0021] The data synchronization module is used to receive data from the SCADA system and align it with the lidar data using timestamps to achieve multi-source data synchronization.
[0022] The signal processing module is used to perform data preprocessing, order tracking analysis and power spectral density calculation, and output characteristic parameters;
[0023] The data fusion module is used to construct multivariate spectral images and realize the fusion of lidar feature data and SCADA operation data;
[0024] The intelligent diagnostic module, with a built-in CNN-SVM fusion diagnostic model, outputs the blade balance state identification results; and
[0025] The dual verification module calculates the theoretical imbalance based on the rigid body rotational dynamics model to verify the diagnostic results.
[0026] Furthermore, the mounting bracket of the lidar acquisition module adopts an adjustable angle structure with an adjustment range of -30° to 30°. The surface of the bracket is provided with an anti-electromagnetic interference coating, and the protection level is not lower than IP65.
[0027] Furthermore, the intelligent diagnostic module also includes a model self-updating unit, which uses incremental learning algorithms to optimize the parameters of the CNN-SVM fusion diagnostic model by adding new running data and manually labeled results, thereby improving its adaptability under different wind field environments.
[0028] The beneficial effects of this invention are:
[0029] Significantly improved monitoring accuracy: The non-contact measurement method of lidar directly reflects the dynamic displacement changes of the blade, greatly improving the measurement accuracy. Compared with the indirect measurement of traditional accelerometers, the signal transmission path is shorter and the distortion is smaller. Order tracking analysis decouples the feature order from the rotational speed. Even under conditions where the rotational speed fluctuates greatly, the feature recognition accuracy remains at a high level. In principle, it solves the frequency drift problem caused by variable rotational speed and ensures the accurate extraction of unbalanced features.
[0030] The anti-interference capability is greatly enhanced: the moving average filtering algorithm adaptively adjusts the window size according to the sampling frequency and the blade operating frequency, effectively filtering external interference such as environmental airflow and electromagnetic radiation; multi-dimensional data fusion integrates two types of heterogeneous data, geometric dynamics and operating status. The interference factors from different data sources are independent of each other. By complementing the data, the influence of single signal interference is reduced, which significantly improves the adaptability of the diagnostic model to complex working conditions.
[0031] High reliability of diagnostic results: The CNN-SVM fusion model combines the deep feature extraction capability of deep learning with the small sample classification advantage of support vector machines, resulting in high discrimination of different degrees of imbalance. Dual verification based on physical principles: The calculation of theoretical imbalance and the data-driven diagnostic results mutually verify each other. When the deviation between the two is less than a set percentage, the result is confirmed to be valid. In principle, this avoids overfitting and misjudgment by pure data models, ensuring the reliability of diagnostic results.
[0032] Significantly reduced operation and maintenance costs: The online monitoring mode enables real-time early warning of blade balance status, avoiding power generation loss caused by traditional periodic shutdowns for inspection; accurate identification of imbalance status can guide targeted maintenance, avoiding the waste of manpower and resources caused by blind repairs; early warning of blade imbalance problems reduces wear on related components such as transmission chains and bearings, extends the service life of key components of the whole machine, and reduces the operation and maintenance costs of wind farms from the perspective of the entire life cycle.
[0033] High adaptability and scalability: The lidar mounting bracket is angle-adjustable to fit the structural dimensions of different wind turbine models; the model self-updating unit optimizes model parameters according to the environmental characteristics of different wind fields (such as wind speed distribution and terrain conditions) through incremental learning algorithms, and can adapt to new scenarios without retraining; the system architecture supports the access of new sensor data (such as blade temperature and wind direction data), and has good functional scalability. Attached Figure Description
[0034] Figure 1 This is a simplified structural diagram of the system of the present invention;
[0035] Figure 2 This is a simplified flowchart of the method of the present invention;
[0036] Figure 3 It is the frequency domain diagram of the original aerodynamic torque signal of a traditional wind turbine under blade imbalance fault after being processed by FFT (Fast Fourier Transform);
[0037] Figure 4 This is the aerodynamic torque signal spectrum diagram reconstructed by the method of this invention after processing a wind turbine under blade imbalance fault.
[0038] Figure 5 It is the aerodynamic torque signal spectrum reconstructed by the method of this invention after processing a wind turbine under a pitch angle fault (fault angle of 1°);
[0039] Figure 6 It is the aerodynamic torque signal spectrum reconstructed by the method of this invention after processing a wind turbine under an installation angle fault (fault angle of 1°). Detailed Implementation
[0040] like Figure 1 , Figures 3-6 As shown, a method for monitoring wind turbine blades based on lidar and multidimensional data fusion includes the following steps:
[0041] S1. Data acquisition: Using high-frequency lidar deployed on the wind turbine tower or nacelle, the dynamic distance data between the blade tip and the tower is measured in real time. Simultaneously, the operating data such as generator speed n, output power P, and drive shaft torque T output by the SCADA system are collected. The data sampling frequency is not less than 1kHz.
[0042] S2. Data preprocessing: Denoise the lidar distance data collected in step S1 by using a moving average filtering algorithm to eliminate environmental noise interference and synchronizing and aligning the timestamps of the SCADA data to ensure consistency in the time dimension of the two types of data.
[0043] S3. Feature Extraction: Perform order tracking analysis on the preprocessed lidar range data, converting the frequency domain signal into an order domain signal, and extracting the feature order f corresponding to blade imbalance. o =k×n / 60, where k is the order coefficient, k=1, 2, 3, and n is the real-time blade rotation speed in r / min. Simultaneously, power spectral density analysis is performed on the synchronous data to calculate the peak spectral density P. s ;
[0044] S4. Data fusion, based on the feature order and spectral peak value P extracted in step S3. s Combined with SCADA operation data, a three-dimensional multivariate spectrum image is constructed, wherein the horizontal axis of the image represents the order value, the vertical axis represents the power spectral density value, and the depth axis represents the normalized rotational speed value;
[0045] S5. Balance state diagnosis: Input the multivariate spectrum image constructed in step S4 into the trained CNN-SVM fusion diagnostic model, and output the blade imbalance state identification result. The model is trained by historical fault data and normal operation data. The identification result includes four levels: normal balance, slight imbalance, moderate imbalance and severe imbalance.
[0046] S6. Dual verification: Calculation of the theoretical unbalance M of the blade based on the principle of rigid body rotational dynamics. th The consistency of the diagnosis results in step S5 is verified. When the deviation between the theoretical calculation value and the model output result is less than 5%, the diagnosis result is confirmed to be valid and output.
[0047] Specifically, in step S1, the measurement accuracy of the high-frequency lidar is no less than ±0.1mm, the measurement distance range is 0.5~50m, the laser emission frequency is 5~20kHz, and the laser wavelength is selected as 905nm or 1550nm. In step S2, the window size N of the moving average filtering algorithm satisfies N=2×floor[f s / (20×f max )]+1, where f s f is the sampling frequency. max The highest operating frequency of the blade is denoted by 'floor', which is a floor function, i.e., a Gaussian function. In step S3, the order tracking analysis employs a resampling algorithm based on instantaneous rotational speed. Specifically, based on the real-time rotational speed n(t) acquired by the SCADA system, the instantaneous angular frequency ω(t) = 2πn(t) / 60 is calculated. Based on ω(t), the distance signal is resampled at equal angular intervals to eliminate frequency drift caused by variable rotational speed. Power spectral density (PSD) analysis uses the Welch method, segmenting and overlapping the signal with an overlap rate of 50%. The Hanning window is selected as the window function, with a spectral resolution Δf ≤ 0.1Hz.
[0048] In step S5, the construction process of the CNN-SVM fusion diagnostic model is as follows: a CNN network is used to extract deep feature vectors from multivariate spectral images. The CNN network contains 3 convolutional layers, 2 pooling layers, and 1 fully connected layer. The 128-dimensional feature vector output by the fully connected layer is input into the SVM classifier, and the radial basis function kernel function is used to achieve state classification. In step S6, the theoretical imbalance quantity M... th The calculation model is: M th =(J×Δω) / (r×t), where J is the blade's moment of inertia, Δω is the change in angular acceleration, r is the distance from the blade tip to the center of rotation, and t is the length of the monitoring time window.
[0049] A wind turbine blade monitoring system based on lidar and multidimensional data fusion for implementing the above method includes:
[0050] The lidar acquisition module 1 is used to acquire dynamic distance data between the blade tip and the tower. The lidar acquisition module 1 includes a high-frequency lidar sensor, a data transmission unit, and a mounting bracket.
[0051] Data synchronization module 2 is used to receive SCADA system data and align it with lidar data using timestamps to achieve multi-source data synchronization;
[0052] Signal processing module 3 is used to perform data preprocessing, order tracking analysis and power spectral density calculation, and output characteristic parameters;
[0053] Data fusion module 4 is used to construct multivariate spectral images and realize the fusion of lidar feature data and SCADA operation data;
[0054] Intelligent diagnostic module 5, with a built-in CNN-SVM fusion diagnostic model, outputs the blade balance state identification results; and
[0055] The dual verification module 6 calculates the theoretical imbalance based on the rigid body rotation dynamics model to verify the diagnostic results.
[0056] The mounting bracket of the lidar acquisition module 1 adopts an adjustable angle structure with an adjustment range of -30° to 30°. The surface of the bracket is coated with an anti-electromagnetic interference coating, and the protection level is not lower than IP65. The intelligent diagnostic module 5 also includes a model self-updating unit. By using newly added running data and manually labeled results, it employs an incremental learning algorithm to optimize the parameters of the CNN-SVM fusion diagnostic model, thereby improving its adaptability to different wind field environments.
[0057] The invention will be further described below with more specific field embodiments.
[0058] I. System Hardware Configuration
[0059] LiDAR acquisition module: A pulsed LiDAR with a wavelength of 1550nm is selected, with a measurement accuracy of ±0.08mm and a sampling frequency of 1kHz~5kHz (specifically 2kHz) to meet the dynamic capture requirements under high blade speed. The measurement distance range is 1m~30m, and the protection level is IP67. It is installed on the upper part of the tower or the side of the nacelle, 5m away from the bottom of the nacelle, to ensure that the measurement line of sight is unobstructed. The bracket angle is adjusted to 15° to ensure that the blade tip running trajectory completely covers the measurement field of view.
[0060] Data synchronization module: It adopts an industrial-grade Ethernet switch for real-time transmission and uses the NTP protocol to achieve timestamp synchronization. The synchronization error is controlled within 1ms. It receives speed (sampling frequency 100Hz), power (sampling frequency 50Hz), and torque (sampling frequency 50Hz) data output from the SCADA system.
[0061] Signal processing and diagnostics: It adopts an embedded industrial computer, equipped with an Intel Core i7 processor, 16GB of memory, 512GB solid-state drive, pre-installed with Linux operating system and Python development environment, and integrates signal processing algorithms and CNN-SVM fusion model;
[0062] Communication module: Adopts 4G / 5G / 6G dual-mode communication to realize real-time transmission of diagnostic results to the wind farm monitoring center, and supports local data storage and remote access.
[0063] II. Specific Implementation Steps
[0064] Data Acquisition: The lidar and data synchronization module are activated. The lidar acquires dynamic distance data d(i) between the blade tip and the tower at a frequency of 2kHz (i=1,2,...,N, where N is the number of sampling points). The SCADA system synchronously outputs rotational speed n(t), power P(t), and torque T(t). The data acquisition lasts for 10 minutes, acquiring 6×10⁻⁶ data. 6 Group of lidar data and 3×10 4 Group SCADA data;
[0065] Data preprocessing: Moving average filtering is used to denoise d(i). According to the formula N=2×floor[2000 / (20×10)]+1=21, the window size is determined to be 21. The filtered data is then calculated.
[0066] d'(i)=(d(i-10)+d(i-9)+...+d(i+10)) / 21;
[0067] The SCADA data was timestamped and two missing speed data points were supplemented using linear interpolation.
[0068] Feature extraction: Based on the real-time rotational speed n(t) acquired by the SCADA system, the instantaneous angular frequency ω(t) = 2πn(t) / 60 is calculated. d'(i) is resampled at equal angular intervals to obtain the order domain signal. The first-order feature order f1 = n / 60 is extracted, and its corresponding power spectral density peak value P is calculated. S1 Similarly, extract the second and third order features and their corresponding peak values P. S2 P S3 ;
[0069] Data fusion: combining feature order (f1, f2, f3) and peak power spectral density (P) S1 ,P S2 P S3 ) and the normalized value of rotational speed n' = n / n nom (n) nom A three-dimensional multivariable spectrum image was constructed for the rated speed, and the image resolution was set to 256×256×256.
[0070] Balance state diagnosis: Input the three-dimensional spectral image into the trained CNN-SVM fusion model, and the model outputs a "moderate imbalance" diagnosis result;
[0071] Double verification: Based on the blade parameters, the moment of inertia J = 1200 kg·m² was calculated; the change in angular acceleration Δω = 0.2 rad / s² was monitored within the time window t = 60 s; and the blade tip rotation radius r = 45 m. Substituting these values into the formula M... th =(1200×0.2) / (45×60)=0.089kg・m; The moderate imbalance output of the model corresponds to an imbalance range of 0.05~0.1kg・m. The deviation between the theoretical calculation value and the model output is 1.1%, which is less than 5%, confirming the validity of the diagnostic result;
[0072] Output: The system sends a "moderate imbalance of blades" warning message to the monitoring center via the communication module, along with the characteristic order, peak power spectral density, and theoretical verification data, to guide maintenance personnel in performing dynamic balance correction.
[0073] (III) Verification of Implementation Results
[0074] An experiment was conducted on a 3MW wind turbine at a wind farm, comparing the monitoring results of the method of this invention with those of the traditional accelerometer method. The method of this invention accurately identified a moderate imbalance state with a diagnostic error of 1.1%; the traditional method, due to interference from transmission chain vibration, misjudged it as a slight imbalance, with an error of 15.3%. The experimental results show that the monitoring accuracy and reliability of the method of this invention are significantly better than those of the traditional method, and can effectively meet the actual needs of wind farm blade balance monitoring.
[0075] This invention directly acquires dynamic distance data between the blade tip and the tower using high-frequency lidar, simultaneously obtaining the operating parameters of the SCADA system. After preprocessing, feature extraction, and data fusion, the system outputs the equilibrium state results using an intelligent diagnostic model and performs dual verification through physical principles, forming a closed-loop monitoring process of "acquisition-processing-fusion-diagnosis-verification".
[0076] Specifically, the LiDAR is deployed on the upper part of the tower or the side of the nacelle to ensure an unobstructed line of sight for measurement. The sampling frequency is set to 1kHz to 5kHz to meet the dynamic capture requirements under high blade speeds. SCADA data is transmitted in real time via industrial Ethernet, using the NTP protocol for timestamp synchronization, with synchronization errors controlled within 1ms. LiDAR distance data uses a moving average filter, and the window size is adaptively adjusted according to the sampling frequency and the highest operating frequency of the blades, effectively eliminating noise caused by environmental airflow and electromagnetic interference. Outliers in the SCADA data are removed, and missing data is supplemented using linear interpolation. Order tracking analysis converts the traditional frequency domain to the order domain through instantaneous speed resampling, decoupling the unbalanced characteristic orders (1st, 2nd, and 3rd orders) from the speed and avoiding frequency drift. Power spectral density analysis uses the Welch method to enhance the identification of characteristic frequencies, with the peak power (PS) as a quantitative indicator of the degree of imbalance. A three-dimensional multivariate spectral image is constructed, integrating characteristic orders, power spectral density, and normalized speed values to achieve deep fusion of LiDAR and SCADA data, providing rich feature inputs for intelligent diagnostics. The CNN-SVM fusion model combines the image feature extraction capabilities of convolutional neural networks with the strong classification capabilities of support vector machines. Trained on over 1000 sets of historical data, the model achieves an accuracy of no less than 95%. The unbalance quantity is calculated based on rigid body rotational dynamics theory, and its consistency with the model's diagnostic results is verified to ensure interpretability and avoid black-box misjudgments inherent in purely data-driven models.
[0077] This invention employs a non-contact lidar as the core sensing unit to directly measure the dynamic distance at the blade tip, avoiding the impact of contact sensors on the equipment structure and acquiring high signal-to-noise ratio raw data, thus improving monitoring accuracy from the source. It introduces order tracking analysis technology to solve the frequency drift problem under variable speed conditions of wind turbines, clarifying the order of blade imbalance characteristics and overcoming the application limitations of traditional frequency domain analysis. It constructs a multivariate spectral image to achieve deep fusion of LiDAR and SCADA data, integrating geometric dynamic data and operational status data to provide more comprehensive feature support for intelligent diagnosis. Finally, it establishes a dual mechanism of "data-driven diagnosis + physical principle verification," ensuring both diagnostic efficiency and the interpretability and credibility of the results, overcoming the limitations of a single diagnostic mode.
[0078] Finally, it should be emphasized that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for monitoring wind turbine blades based on lidar and multidimensional data fusion, characterized in that, The monitoring method includes the following steps: S1. Data acquisition: The high-frequency lidar deployed on the wind turbine tower measures the dynamic distance between the blade tip and the tower in real time, and simultaneously acquires the generator operation data output by the SCADA system. S2. Data preprocessing: Denoise the lidar distance data collected in step S1 by using a moving average filtering algorithm to eliminate environmental noise interference and synchronizing and aligning the timestamps of the SCADA operation data to ensure the consistency of the time dimension of the two types of data. S3. Feature Extraction: Order tracking analysis is performed on the preprocessed LiDAR range data, converting the time-domain signal into an order-domain signal. The order tracking analysis employs a resampling algorithm based on instantaneous rotational speed. Specifically: based on the rotational speed n(t) collected in real-time by the SCADA system, the instantaneous angular velocity ω(t) = 2πn(t) / 60 is calculated. The LiDAR range data is then resampled at equal angular intervals based on ω(t) to obtain the order-domain signal, eliminating frequency drift caused by variable rotational speed. The first-order feature order is extracted, and its corresponding power spectral density peak value is calculated. Similarly, the second and third-order feature orders and their corresponding power spectral density peak values are extracted. S4. Data fusion: Based on the feature order, power spectral density peak value and SCADA operation data extracted in step S3, a three-dimensional multivariate spectrum image is constructed. The horizontal axis of the image is the feature order, the vertical axis is the power spectral density peak value, and the depth axis is the normalized value of the rotational speed. S5. Balance state diagnosis: Input the multivariate spectrum image constructed in step S4 into the trained CNN-SVM fusion diagnostic model, and output the blade imbalance state identification result. The diagnostic model is trained by historical fault data and normal operation data. The identification result includes four levels: normal balance, slight imbalance, moderate imbalance and severe imbalance. S6. Dual verification: Calculation of the theoretical unbalance M of the blade based on a rigid body rotational dynamics model. th The consistency of the diagnostic identification results from step S5 is verified. When the deviation between the theoretically calculated value and the output result of the diagnostic model is less than the set percentage, the diagnostic identification results are confirmed to be valid and output.
2. The monitoring method according to claim 1, characterized in that, In step S1, the measurement accuracy of the high-frequency lidar is not less than ±0.1mm, the measurement distance range is 0.5 to 50m, the laser emission frequency is 5 to 20kHz, and the laser wavelength is 905nm or 1550nm.
3. The monitoring method according to claim 1, characterized in that, In step S2, the window size N of the moving average filtering algorithm satisfies N = 2 × floor[f s / (20×f max )]+1, where f s f is the sampling frequency. max is the highest operating frequency of the blade, where floor is the floor function.
4. The monitoring method according to claim 1, characterized in that, In step S5, the construction process of the CNN-SVM fusion diagnostic model is as follows: a CNN network is used to extract deep feature vectors from multivariate spectral images. The CNN network contains 3 convolutional layers, 2 pooling layers, and 1 fully connected layer. The 128-dimensional feature vector output by the fully connected layer is input into the SVM classifier, and the radial basis function is used to achieve state classification.
5. A wind turbine blade monitoring system based on lidar and multidimensional data fusion for implementing the monitoring method as described in any one of claims 1 to 4, characterized in that, The monitoring system includes: The lidar acquisition module (1) is used to acquire dynamic distance data between the blade tip and the tower. The lidar acquisition module (1) includes a high-frequency lidar, a data transmission unit and a mounting bracket. The data synchronization module (2) is used to receive SCADA operation data and align it with the lidar distance data using timestamps to achieve multi-source data synchronization. The signal processing module (3) is used to perform data preprocessing, order tracking analysis and power spectral density calculation, and output characteristic parameters; The data fusion module (4) is used to construct a multivariate spectrum image and realize the fusion of lidar feature parameters and SCADA operation data; The intelligent diagnostic module (5) has a built-in CNN-SVM fusion diagnostic model and outputs the blade balance state identification results; and The dual verification module (6) calculates the theoretical imbalance based on the rigid body rotation dynamics model and verifies the diagnostic identification results.
6. The monitoring system according to claim 5, characterized in that, The mounting bracket of the laser radar acquisition module (1) adopts an adjustable angle structure with an adjustment range of -30° to 30°. The bracket surface is provided with an anti-electromagnetic interference coating with a protection level of not less than IP65.
7. The monitoring system according to claim 5, characterized in that, The intelligent diagnostic module (5) also includes a model self-updating unit, which uses incremental learning algorithm to optimize the parameters of the CNN-SVM fusion diagnostic model by adding new running data and manual annotation results, thereby improving its adaptability under different wind field environments.