Intelligent monitoring method for blades of wind generating set

Abstract

The invention relates to the technical field of wind turbine generator performance detection, in particular to an intelligent monitoring method for blades of a wind turbine generator, which constructs a system framework with data acquisition, feature extraction and intelligent diagnosis as the core, and utilizes the existing SCADA operation data of the unit to monitor the blades of the wind turbine generator. And performing multi-dimensional feature modeling on the running state of the blade in combination with means such as spectral analysis, power curve deviation analysis and wind deviation modeling. Typical faults such as pneumatic balance abnormity, blade icing, fouling and damage are identified and early warned by introducing a deep learning and probability diagnosis algorithm, so that the real-time monitoring of the health state of the blade and the quantitative evaluation of performance degradation are realized. According to the method, additional sensor equipment is not needed, blade anomaly detection can be completed only through unit operation data, and the method has the advantages of being low in deployment cost, high in applicability and high in fault recognition accuracy.

Description

Technical Field

[0001] This invention relates to the field of wind turbine performance testing technology, and in particular to an intelligent monitoring method for wind turbine blades. Background Technology

[0002] During the operation of wind turbine generators, the aerodynamic state of the blades is affected by factors such as wind deviation, icing, fouling, and damage, leading to abnormal power output and changes in vibration characteristics. These abnormal changes further affect the overall performance and operational stability of the unit. Real-time intelligent monitoring of the blade aerodynamic state can promptly detect problems such as icing, fouling, and damage, accurately assess their impact on the unit's power output, and then implement targeted maintenance measures to ensure the safe and efficient operation of the wind turbine generator.

[0003] To achieve this goal, the intelligent monitoring method first deploys high-precision sensors on key parts of the blades to collect various parameters during operation in real time, such as strain, vibration frequency, and temperature. These sensors can accurately capture subtle parameter fluctuations caused by changes in the blade's aerodynamic state, providing reliable data support for subsequent analysis. The collected data is transmitted in real time to the central processing system via a high-speed and stable wireless transmission network. In the central processing system, advanced algorithms are used to perform in-depth analysis and processing of massive amounts of data. On the one hand, by applying a power density model, a power density curve is fitted, which can intuitively reflect the power distribution of the blades under different operating conditions. On the other hand, based on the fitted power density curve, the boundary clusters of the curve are further calculated. By analyzing the changing characteristics of the boundary clusters, it is possible to accurately determine whether the blades have problems such as icing, fouling, and damage, and the specific impact of these problems on the unit's output. Based on the analysis results, the system can issue early warning information in a timely manner, guiding maintenance personnel to take corresponding maintenance measures, such as de-icing, cleaning, or blade repair, thereby effectively ensuring the safe and efficient operation of the wind turbine generator.

[0004] However, existing technical solutions have the following drawbacks: First, some solutions are not precise enough in capturing subtle parameter fluctuations caused by changes in blade aerodynamic state, leading to insufficient reliability of the data relied upon for subsequent analysis and thus affecting the accurate judgment of the actual blade condition. Second, the wireless transmission networks used in some solutions are unstable, prone to interruptions or delays during real-time data transmission, preventing the central processing system from acquiring complete data in a timely manner and reducing the timeliness of the entire monitoring system. Third, some existing solutions use algorithms that are not advanced enough when analyzing massive amounts of data, resulting in significant errors in fitting power density curves and calculating curve boundary clusters. This makes it difficult to accurately determine whether blades have problems such as icing, fouling, or damage, and to accurately assess the impact of these problems on unit output. Consequently, the early warning information issued by the system is not accurate or timely enough, and cannot effectively guide maintenance personnel to take appropriate maintenance measures. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes an intelligent monitoring method for wind turbine blades. By extracting features and modeling the turbine's operating data, it enables the identification and early warning of blade aerodynamic imbalances, icing, fouling, and damage. This invention utilizes operating data collected by the turbine's SCADA system, including power, wind speed, wind direction, yaw angle, rotational speed, and vibration signals. Features are extracted from this data to obtain spectral features, power curve features, and wind deviation features. The spectral features are used to detect aerodynamic imbalances caused by blade damage or icing; the power curve features are used to identify abnormal turbine output due to blade performance degradation; and the wind deviation features reflect deviations in the blade's aerodynamic state.

[0006] This invention provides an intelligent monitoring method for wind turbine blades, comprising the following steps: S1. Obtain the operating data of the wind turbine generator set; S2. Extract features from the operational data obtained in S1 to obtain spectral features, power curve features, and wind deviation features; S3. Construct a blade condition monitoring model based on the features obtained in S2; S4. The blade condition monitoring model constructed in S3 is trained and inferred by combining deep learning algorithms and probabilistic diagnostic algorithms to realize the identification and early warning of blade aerodynamic balance abnormalities, icing, fouling and damage.

[0007] Furthermore, in S1, operating data for each wind turbine is read from its SCADA system. This data includes power, wind speed, wind direction, yaw angle, and engine speed. During the data reading process, the integrity and accuracy of the data are ensured; missing or abnormal data is marked and processed to guarantee the reliability of subsequent feature extraction. The read power data reflects the actual power generation capacity of the turbine, while wind speed and direction data are key environmental factors affecting turbine operation. Yaw angle and engine speed data also play a crucial role in analyzing the turbine's operating status and performance. This operating data lays a solid foundation for the accurate extraction of spectral features, power curve features, and wind deviation features.

[0008] Furthermore, in S2, the spectral characteristics include the frequency domain components of the vibration signal during unit operation. These frequency domain components are used to detect aerodynamic imbalances caused by blade damage or icing. The power curve characteristics include the deviation between the actual power curve and the theoretical power curve of the unit, used to identify abnormal unit output caused by blade performance degradation. The wind deviation characteristics cover the deviation relationship between wind speed and yaw angle. This deviation reflects the efficiency changes of the blades in the process of capturing wind energy, thus helping to determine whether there are problems with blade fouling or aerodynamic performance degradation. Through detailed extraction and analysis of these characteristics, subtle changes in blade condition can be captured more accurately, providing strong data support for the subsequent construction of an efficient blade condition monitoring model.

[0009] Furthermore, in S4, the deep learning algorithm includes a neural network model, and the probabilistic diagnostic algorithm includes Bayesian inference, used to improve the accuracy and robustness of blade fault identification. The neural network model, through learning from a large amount of historical operating data and fault samples, can automatically extract complex features and patterns from the data, thereby accurately classifying and predicting the types of faults that may occur in the blades. Bayesian inference, based on the principles of probability theory, combines prior knowledge and observational data to infer and update the probability of blade fault occurrence, effectively reducing the risk of false positives and false negatives. Combining these two algorithms can fully leverage their respective advantages, significantly improving the accuracy and reliability of blade fault identification, and providing strong support for the safe and stable operation of wind turbine generators.

[0010] Compared with existing technologies, this invention has unique advantages: 1. It monitors blade condition based on existing on-site operating data, eliminating the need for additional sensor equipment and reducing engineering costs; 2. It organically combines the unit's power characteristics, vibration spectrum characteristics, and wind deviation characteristics to construct a blade condition monitoring model, enabling the identification of complex fault modes; 3. By combining deep learning and probabilistic diagnostic algorithms, it improves the accuracy and robustness of blade fault identification, providing reliable technical support for the intelligent operation and maintenance and performance optimization of wind turbine generators. Attached Figure Description

[0011] Figure 1 This is a diagram of the system network topology.

[0012] Figure 2 This is a system functional architecture diagram.

[0013] Figure 3 This is a diagram for early warning analysis of impeller balance.

[0014] Figure 4 This is a conclusion based on the early warning timeline.

[0015] Figure 5 The wind speed distribution is based on the baseline value.

[0016] Figure 6 The results are from the analysis of the early warning model for the pitch motor. Detailed Implementation

[0017] The present invention will be described below with reference to examples. These examples are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0018] A smart monitoring method for wind turbine blades, such as Figure 1 and Figure 2 As shown, this intelligent monitoring method utilizes sensors deployed on the surface of wind turbine blades and key stress-bearing components, combined with SCADA system operational data, to achieve real-time monitoring and intelligent analysis of blade operating status. The system deploys an intelligent operation and maintenance platform at the booster station side. This platform incorporates data acquisition, preprocessing, model training, status recognition, and early warning push modules, enabling dynamic monitoring and fault prediction of blade operating status with limited hardware investment.

[0019] The intelligent monitoring method for wind turbine blades of this invention includes the following specific steps: 1. Read the operating data of the wind turbine generator set from its SCADA system, including parameters such as power, wind speed, wind direction, yaw angle, speed, and vibration signal.

[0020] Second, the operational data is denoised and normalized, and key features are extracted, including: obtaining the blade vibration spectrum components by analyzing the frequency domain characteristics of the vibration signal, which is used to detect aerodynamic imbalance caused by blade damage or icing; calculating the deviation between the actual power curve and the theoretical power curve, which is used to identify abnormal output caused by blade performance degradation; and reflecting the deviation of the blade aerodynamic state by combining wind speed, wind direction and yaw angle data.

[0021] Third, a blade condition monitoring model is constructed based on the extracted features. The spectral features, power curve features, and wind deviation features are used as inputs to form a multi-dimensional feature space, which is used to characterize the aerodynamic balance state and potential failure modes of the blade.

[0022] Fourth, the blade condition monitoring model is trained using deep learning algorithms such as neural networks to learn the feature mapping relationship under different fault modes; at the same time, probabilistic diagnostic algorithms such as Bayesian inference are combined to infer the model output, thereby improving the accuracy and robustness of fault identification.

[0023] Fifth, based on the completed monitoring model, reason about the real-time data during the operation of the unit, identify the aerodynamic balance of the blades, icing, fouling and damage, and generate corresponding early warning information to provide a basis for the operation and maintenance decision of the wind turbine.

[0024] In the specific implementation process, accelerometers and strain gauges arranged on the blade surface first acquire blade vibration response and stress-strain data. Simultaneously, temperature, wind speed, and air pressure sensors at the blade root collect environmental parameters and blade operating conditions. The acquired raw signals are transmitted to the intelligent operation and maintenance host at the booster station via the SCADA system. The host preprocesses the data, including normalization, noise filtering, outlier removal, and data interpolation, to ensure data integrity and accuracy. The processed data is then input into a deep neural network model. The model uses the Dropout method to suppress overfitting, enabling pattern recognition and health assessment of the blade's structural state.

[0025] This embodiment describes a blade condition monitoring model based on extracted features. This model employs an architecture combining convolutional neural networks and long short-term memory networks. This hybrid model structure can simultaneously capture the spatial and temporal features of blade vibration data. Transfer learning techniques are introduced during model training, using pre-trained model parameters under similar operating conditions for initialization, effectively shortening the training convergence time. A multi-scale feature fusion mechanism is used to stitch together feature maps from different levels, enhancing the model's ability to identify early, minor faults. The final monitoring model achieves continuous monitoring and trend prediction of progressive faults such as blade aerodynamic performance degradation and structural damage development.

[0026] The intelligent monitoring technology described in this embodiment utilizes a Bayesian network based on TAN to model the blade's operating characteristics, enabling it to characterize the correlation between blade aerodynamic loads, rotational speed, and vibration amplitude, thereby identifying potential blade imbalance conditions. When the model detects that the blade energy is significantly higher than the historical baseline threshold, the system automatically determines it as an aerodynamic imbalance trend and generates an early warning, such as... Figure 3 Impeller balance early warning analysis diagram and Figure 4 The early warning sequence conclusions for the turbine generator set are shown below. The system further integrates feedback from the unit's yaw angle, wind direction, and pitch angle to trace the cause of the imbalance, determining whether it is due to blade icing, surface contamination, or structural damage. Figure 5 The wind speed distribution of the baseline value is shown. Figure 6 The analysis results of the pitch motor early warning model are presented.

[0027] In terms of blade fatigue monitoring, this embodiment constructs a rainflow counting method cyclic distribution using long-term collected stress time series data, and calculates the remaining blade life by combining it with a mining-based fatigue damage accumulation model. The model's results are displayed in a visual format on the intelligent operation and maintenance platform, allowing maintenance personnel to intuitively obtain the blade life decay trend. When the cumulative damage factor at a certain monitoring point on the blade approaches 1, the system automatically pushes maintenance suggestions, thereby avoiding serious accidents caused by fatigue fracture.

[0028] The technology described in this embodiment has the advantages of flexible deployment, high monitoring accuracy, and long lead time for early warning. This intelligent monitoring system not only enables real-time perception of blade status but also allows for early prediction of potential blade failures, providing reliable assurance for the safe and stable operation of wind turbine units, while reducing maintenance costs and extending blade service life.

[0029] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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 intelligent monitoring of wind turbine blades, characterized in that, The method comprises the following steps: S1, obtaining operation data of a wind turbine generator set; S2, performing feature extraction on the operation data obtained in S1 to obtain spectral features, power curve features, and wind deviation features; S3, constructing a blade state monitoring model based on the features obtained in S2; S4, training and reasoning the blade state monitoring model constructed in S3 in combination with a deep learning algorithm and a probabilistic diagnosis algorithm to realize identification and early warning of blade aerodynamic balance abnormalities, icing, fouling, and damage.

2. The method for intelligent monitoring of wind turbine blade as claimed in claim 1 wherein, In S1, operation data of each unit is read from a SCADA system of the wind turbine generator set.

3. The method for intelligent monitoring of wind turbine blade according to claim 2, wherein, The operation data includes power, wind speed, wind direction, yaw angle, and rotating speed.

4. The method for intelligent monitoring of wind turbine blade as claimed in claim 1 wherein, In S2, the spectral features include frequency domain components of vibration signals in the operation process of the unit, and the frequency domain components are used to detect aerodynamic imbalance caused by blade damage or icing.

5. The method for intelligent monitoring of wind turbine blade as claimed in claim 1 wherein, In S2, the power curve features include deviations between actual power curves and theoretical power curves of the unit, and are used to identify abnormal output of the unit caused by performance degradation of the blades.

6. The method for intelligent monitoring of wind turbine blade as claimed in claim 1 wherein, In S4, the deep learning algorithm includes a neural network model, and the probabilistic diagnosis algorithm includes Bayesian reasoning, which are used to improve the accuracy and robustness of blade fault identification.

Images