A method of monitoring based on blade rotation sound and a device for assisting in monitoring

Abstract

The application provides a monitoring method based on blade rotating sound and a device for auxiliary monitoring, wherein weak airflow disturbance sound wave signals generated when a wind driven generator blade rotates are captured by a non-contact acoustic sensor array, the captured sound wave signals are standardized and pretreated, ambient noise of the sound wave signals is dynamically filtered, acoustic sensor data from different positions are integrated to form a comprehensive sound image, deep learning is applied to model and predict the health condition of the blade, edge computing is integrated to enable some data processing tasks to be completed close to the acoustic sensor, an Internet of Things platform is adopted to realize interconnection and intercommunication between devices, a cloud-edge collaborative optimization mode is implemented to automatically distribute tasks to the most suitable processing node, sensors do not need to be installed on the blade, physical impact on the device is avoided, ambient noise is effectively filtered out, the signal-to-noise ratio of the target sound wave signal is enhanced, and thus the monitoring precision is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind turbine blade detection, and particularly relates to a monitoring method based on blade rotation sound and a device for auxiliary monitoring. BACKGROUND

[0002] As an important part of renewable energy, the operation efficiency and safety of wind turbines are directly related to the stability and economy of energy production. Wind turbine blades are one of the key components, which are exposed to harsh environments for a long time and are prone to wear, cracks, corrosion and other damages, leading to performance degradation or even failure. Traditional blade monitoring methods mainly rely on periodic manual inspection and maintenance, which is not only time-consuming and labor-intensive, but also difficult to find potential problems in time, which may lead to sudden failures, increasing maintenance costs and downtime.

[0003] Currently, common blade monitoring techniques include vibration analysis, infrared thermal imaging, ultrasonic detection, etc. These methods can reflect the state of the blade to some extent, but have the following limitations:

[0004] 1. The need to install sensors on the blade or conduct close-range detection makes the operation complex and has an impact on the equipment.

[0005] 2. Data processing delay: A large amount of data needs to be transmitted to the cloud for processing, resulting in a long response time and making real-time monitoring impossible.

[0006] 3. Environmental interference: External environmental noise (such as wind noise and mechanical noise) can seriously interfere with the monitoring signal, reducing monitoring accuracy.

[0007] 4. Limited coverage: Single-point or small number of sensors cannot fully cover the entire blade surface, making it difficult to obtain global information. SUMMARY

[0008] In view of the deficiencies in the prior art, the present application provides a monitoring method based on blade rotation sound. To achieve the above purpose, the present application adopts the following technical solutions:

[0009] A monitoring method based on blade rotation sound, comprising the following steps:

[0010] Capturing weak airflow disturbance sound wave signals generated by the rotation of wind turbine blades by a non-contact acoustic sensor array;

[0011] Using a machine learning algorithm to standardize and preprocess the captured sound wave signals;

[0012] Using a least mean square error algorithm or a Kalman filter to dynamically filter the environmental noise of the sound wave signals, and the formula for updating the weight of the least mean square error algorithm is:

[0013] ;

[0014] wherein, is the weight vector of the th iteration, is a step size parameter, is an error signal, is an input signal;

[0015] Using multi-channel signal fusion technology, acoustic sensor data from different positions are integrated to form a comprehensive sound image;

[0016] Applying deep learning to model and predict the health condition of the blade;

[0017] Integrating edge computing, part of the data processing task is completed near the acoustic sensor;

[0018] Using the Internet of Things platform to realize the interconnection between devices, as well as data exchange and service management with the cloud;

[0019] Implementing cloud-edge collaborative optimization, automatically assigning tasks to the most suitable processing node according to network conditions and computing resources.

[0020] Further, the method of machine learning algorithm comprises the following steps:

[0021] The captured sound wave signal is processed by removing the DC component and normalizing the amplitude;

[0022] Using short-time Fourier transform, wavelet transform or mel frequency cepstral coefficient technology, the time-frequency characteristics reflecting the state of the blade are extracted from the preprocessed sound wave signal, and the formula of short-time Fourier transform is:

[0023] ;

[0024] wherein, is a time domain signal, is a window function, is a time offset, is an angular frequency;

[0025] Using supervised learning, using labeled historical data sets to train one or more deep learning models using regularization techniques and early stopping method;

[0026] Using an adaptive mechanism, the model can automatically adjust the internal parameters according to the changes of wind speed, temperature and external environment;

[0027] After the model training is completed, it is deployed to the actual monitoring system to analyze the real-time collected acoustic signals, judge whether there is a potential problem in the blade, and predict the future possible failure. If an abnormal situation is detected, the system will issue an alarm to notify the maintenance personnel.

[0028] Further, the standardized preprocessing method comprises the following steps:

[0029] The large-scale acoustic data analysis task is divided into several sub-tasks, each corresponding to data of a specific time period or a specific wind turbine generator set;

[0030] A distributed computing cluster is established, including multiple computing nodes. Through a load balancing algorithm, each sub-task is reasonably distributed to different computing nodes for execution;

[0031] The gRPC or ZeroMQ data transmission protocol is used to quickly and accurately transfer the data required by the sub-tasks to the corresponding computing nodes;

[0032] When all sub-tasks are completed, the master node is responsible for collecting the results returned by each computing node and performing merging processing;

[0033] In the task allocation stage, a backup node is prepared for the key sub-task. Once the original node has a problem, the backup node will immediately take over the unfinished task;

[0034] By monitoring the working state of each computing node, the task allocation strategy is dynamically adjusted, and according to the historical running data, the future computing demand is predicted in advance.

[0035] Further, a quantum random number generator is used to initialize the weight values in the machine learning algorithm. The high-entropy random numbers provided by the quantum random number generator are used to improve the selection of initial weights.

[0036] Further, one or more unmanned aerial vehicles are equipped, and the unmanned aerial vehicles are programmed to fly according to a predetermined path to carry acoustic sensors close to the wind turbine for close-range detection, while transmitting the collected data back to the ground station in real time for analysis.

[0037] Further, the method for detecting blades using unmanned aerial vehicles comprises the following steps:

[0038] Genetic algorithm or ant colony algorithm is used to design the optimal flight path for the unmanned aerial vehicle according to factors such as flight distance, energy consumption, safety, and coverage range;

[0039] A wireless communication link is used between the unmanned aerial vehicle and the ground station to support real-time transmission of high-definition audio and video streams;

[0040] The integration of the global positioning system and the inertial measurement unit enables the unmanned aerial vehicle to fly autonomously in complex environments without human intervention.

[0041] The data collected by the unmanned aerial vehicle is transmitted to the ground station through encryption, and the ground station is responsible for data storage, processing and analysis. All data are time-stamped for subsequent comparison and tracking.

[0042] Furthermore, the unmanned aerial vehicle is equipped with a visual recognition module that can identify specific markings or defects on the blades and combine them with acoustic monitoring results to provide a blade health diagnosis report, which includes:

[0043] The unmanned aerial vehicle is equipped with a high-definition camera to capture images of the blade surface from multiple angles.

[0044] The collected images are pre-processed using computer vision techniques, including noise reduction, contrast enhancement and edge detection, and feature extraction algorithms are applied to identify key points and texture information in the images.

[0045] In combination with a pre-trained deep learning model, the image is classified and quantified for blade cracks, corrosion or other damage.

[0046] The results of visual recognition are correlated with acoustic monitoring data for comprehensive evaluation of the overall health of the blade.

[0047] The present application also provides a device for assisting monitoring, comprising:

[0048] Microphone array: integrated high-sensitivity microphone array for capturing weak airflow disturbance sound wave signals generated by wind turbine blade rotation;

[0049] Embedded processor: for performing preliminary feature extraction and anomaly detection data processing tasks locally;

[0050] Data transmission module: equipped with a wireless communication module to support sending complete data to a cloud server for in-depth analysis;

[0051] Power management system: solar charging system and intelligent power management system are used;

[0052] Data compression module: for data that needs to be uploaded to the cloud, compression processing is performed first;

[0053] Local storage: built-in large capacity Flash memory for temporary storage of a large amount of monitoring data;

[0054] User interface: provides a user interface for maintenance personnel to view real-time monitoring results, set parameters and receive alarm notifications.

[0055] Further, the embedded processor is built-in with an artificial intelligence chip to accelerate the execution of machine learning algorithms.

[0056] Further, the AR glasses system is also included, and the maintenance personnel can see the historical monitoring data, current state indication and possible problem area virtual information of the blade superimposed on the real world through wearing the AR glasses when checking the wind turbine on site, thereby guiding the maintenance work.

[0057] Compared with the prior art, the present application has the following beneficial effects:

[0058] 1. By using the multi-channel signal fusion technology and the adaptive noise cancellation technology, the present application can effectively filter out environmental noise, enhance the signal-to-noise ratio of the target sound wave signal, thereby improving the monitoring accuracy, especially for small cracks or early damage, which can be timely discovered and warned, and the occurrence of sudden failure is reduced;

[0059] 2. The edge computing capability is integrated, so that part of the data processing task is completed near the acoustic sensor, the delay of data transmission is reduced, real-time monitoring is realized, the maintenance personnel can receive the abnormal alarm in the first time, timely measures are taken, and the expansion of the fault is avoided;

[0060] 3. Through the application of the deep learning model, the present application can model and predict the health status of the blade, identify potential problems in advance, and optimize the maintenance plan, which not only reduces unnecessary regular inspection, but also prolongs the service life of the blade and reduces the maintenance cost;

[0061] 4. Distributed computing resources and fault-tolerant mechanisms are adopted to ensure that the system can still operate normally even in the case of failure of some nodes, and the cloud-edge collaborative optimization strategy automatically allocates tasks according to network conditions and computing resources, ensuring the high efficiency and low power consumption of the system;

[0062] 5. Combined with the unmanned aerial vehicle vision recognition module and the acoustic monitoring result, the present application can provide more accurate blade health diagnosis report, and the maintenance personnel can check the virtual information superimposed on the real world through the AR glasses on site, intuitively understand the historical monitoring data, current state indication and possible problem area of the blade, thereby guiding the maintenance work. DETAILED DESCRIPTION

[0063] The technical solutions in the present application will be further described below in combination with embodiments.

[0064] Embodiment one

[0065] The present application provides a monitoring method based on blade rotating sound, comprising the following steps:

[0066] Standardized preprocessing of captured acoustic signals, including removing DC components, normalizing amplitudes, etc.

[0067] Dynamic filtering of environmental noise in acoustic signals using the least mean square error (LMS) algorithm or Kalman filter, the LMS algorithm updates the weight formula:

[0068] ;

[0069] where, is the weight vector of the th iteration, is the step size parameter, is the error signal, is the input signal;

[0070] Using multi-channel signal fusion technology, integrate acoustic sensor data from different locations to form a comprehensive sound image;

[0071] Use deep learning to model and predict the health of the blade, specifically using convolutional neural networks (CNN) or recurrent neural networks (RNN), especially long short-term memory networks (LSTM), the state update formula of LSTM unit is:

[0072] ;

[0073] ;

[0074] ;

[0075] ;

[0076] ;

[0077] ;

[0078] where, , , are the activation values of the forget gate, input gate and output gate respectively, is the candidate cell state, is the current cell state, is the current hidden state;

[0079] Integrate edge computing capabilities, so that part of the data processing tasks can be completed near the acoustic sensor, reducing the delay and reducing the burden of the cloud server;

[0080] Use the Internet of Things (IoT) platform to realize the interconnection between devices, as well as data exchange and service management with the cloud;

[0081] The cloud-edge collaborative optimization method is implemented, and tasks are automatically allocated to the most suitable processing nodes according to network conditions and computing resources.

[0082] In the embodiment, a quantum random number generator is used to initialize the weight values in the machine learning algorithm, and the high-entropy random numbers provided by the quantum random number generator are used to improve the selection of initial weights.

[0083] Embodiment two

[0084] In embodiment one, the specific steps of the machine learning algorithm are:

[0085] The captured sound wave signal is subjected to DC removal and normalized amplitude processing;

[0086] The time-frequency characteristics reflecting the blade state are extracted from the preprocessed sound wave signal using short-time Fourier transform (STFT), wavelet transform (WT), or mel frequency cepstral coefficient (MFCC) technology. The formula of short-time Fourier transform is:

[0087] ;

[0088] Where, is the time domain signal, is the window function, is the time offset, is the angular frequency;

[0089] A supervised learning method is used to train one or more deep learning models using labeled historical data sets, regularization techniques, and early stopping methods;

[0090] An adaptive mechanism is used to enable the model to automatically adjust internal parameters according to changes in wind speed, temperature, and external environment. For example, through online learning, the model weights are constantly updated to better fit actual working conditions;

[0091] After the model is trained, it is deployed to the actual monitoring system to analyze real-time collected sound wave signals, determine whether the blade has potential problems, and predict future possible failures. If an abnormal situation is detected, the system will issue an alarm to notify maintenance personnel.

[0092] Embodiment three

[0093] In embodiment one, the method for using distributed computing resources is:

[0094] Large-scale sound wave data analysis tasks are divided into several subtasks, each corresponding to a specific time period or data of a specific wind turbine generator set;

[0095] A distributed computing cluster is established, which includes multiple computing nodes. Through a load balancing algorithm, each sub-task is reasonably distributed to different computing nodes for execution. The load balancing algorithm can use a round-robin method or a minimum connection number method;

[0096] The gRPC or ZeroMQ data transmission protocol is used to quickly and accurately transmit the data required by the sub-tasks to the corresponding computing nodes;

[0097] When all sub-tasks are completed, the master node is responsible for collecting the results returned by each computing node and performing merging processing. The merging processing can be realized by a weighted average method;

[0098] In the task allocation stage, backup nodes are prepared for key sub-tasks. Once the original node fails, the backup node can immediately take over the unfinished task;

[0099] By monitoring the working state of each computing node, the task allocation strategy is dynamically adjusted, and according to the historical running data, the future computing demand is predicted in advance.

[0100] Embodiment Four

[0101] On the basis of embodiment one, an unmanned aerial vehicle (UAV) is used for auxiliary monitoring:

[0102] Genetic algorithm or ant colony algorithm is used to design the optimal flight path for the UAV according to factors such as flight distance, energy consumption, safety and coverage range;

[0103] A wireless communication link is used between the UAV and the ground station to support real-time transmission of high-definition audio and video streams;

[0104] Global Positioning System (GPS) and Inertial Measurement Unit (IMU) are integrated to enable the UAV to fly autonomously in complex environments without human intervention;

[0105] The data collected by the UAV is transmitted to the ground station through encryption, and the ground station is responsible for data storage, processing and analysis. All data are marked with time stamps for subsequent comparison and tracking.

[0106] Embodiment Five

[0107] On the basis of embodiment four, the UAV is equipped with a visual recognition module that can identify specific markings or defects on the blades and combine with the acoustic monitoring results to provide a blade health diagnosis report, which includes:

[0108] The UAV is equipped with a high-definition camera to take images of the blade surface from multiple angles;

[0109] The collected images are pre-processed using computer vision techniques, including noise removal, contrast enhancement, and edge detection. Feature extraction algorithms are applied to identify key points and texture information in the images.

[0110] A pre-trained deep learning model is combined to classify and quantify leaf cracks, corrosion, or other damages in the images.

[0111] The results from visual recognition are correlated with acoustic monitoring data for comprehensive evaluation of the overall health of the blades.

[0112] Embodiment Six

[0113] The present application also provides a device for auxiliary monitoring, comprising:

[0114] Microphone array: integrated high-sensitivity microphone array for capturing weak airflow disturbance sound wave signals generated by wind turbine blade rotation;

[0115] Embedded processor: built-in high-performance embedded processor capable of performing preliminary data processing tasks such as feature extraction and anomaly detection locally;

[0116] Data transmission module: equipped with wireless communication modules such as Wi-Fi, Bluetooth, or LoRa, supporting complete data transmission to cloud servers for in-depth analysis;

[0117] Power management system: solar charging system and intelligent power management system to ensure long-term stable operation of the device, while having a low-power mode that automatically enters sleep state when no activity is detected, further extending battery life;

[0118] Data compression module: for data to be uploaded to the cloud, compression is performed first, using lossless or lossy compression algorithms (such as FLAC or MP3) to reduce data volume, saving transmission time and bandwidth costs;

[0119] Local storage: built-in large-capacity Flash memory for temporary storage of large amounts of monitoring data to deal with unstable or interrupted network conditions;

[0120] User interface: provides a user interface for maintenance personnel to view real-time monitoring results, set parameters, and receive alarm notifications.

[0121] In this embodiment, to accelerate inference, an artificial intelligence chip is built into the embedded processor to accelerate the execution of machine learning algorithms.

[0122] In the embodiment, the maintenance personnel wear AR glasses, and when checking the wind turbine on site, can directly see the historical monitoring data, current state indication and possible problem area virtual information of the blade superimposed on the real world, thereby guiding the maintenance work.

[0123] The application can realize real-time, accurate and comprehensive monitoring of the wind turbine blade, and improve the reliability and maintenance efficiency of the system.

[0124] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the application and are not limiting. Although the application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the application can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the application, and they should be covered in the scope of the claims of the application.

Claims

1. A monitoring method based on blade rotation sound, characterized by, The method comprises the following steps: Capturing weak airflow disturbance acoustic wave signals generated by wind turbine blade rotation through a non-contact acoustic sensor array; Standardizing and preprocessing the captured acoustic wave signals using a machine learning algorithm; Using a least mean square error algorithm or a Kalman filter to dynamically filter environmental noise from the acoustic wave signals, with the least mean square error algorithm updating the weight formula as follows: ； wherein is the weight vector of the th iteration, is a step size parameter, is an error signal, is an input signal; Using multi-channel signal fusion technology to integrate acoustic sensor data from different locations to form a comprehensive sound image; Applying deep learning to model and predict the health of the blade; Integrating edge computing to complete part of the data processing task near the acoustic sensor; Using an Internet of Things platform to realize the interconnection between devices, as well as data exchange and service management with the cloud; Implementing cloud-edge collaborative optimization to automatically assign tasks to the most suitable processing nodes according to network conditions and computing resources.

2. A monitoring method based on the sound of the rotation of the blades according to claim 1, characterized in that: The method of the machine learning algorithm comprises the following steps: Removing the DC component and normalizing the amplitude of the captured acoustic wave signals; Using short-time Fourier transform, wavelet transform, or mel-frequency cepstral coefficient technology to extract time-frequency features that reflect the state of the blade from the preprocessed acoustic wave signals, with the short-time Fourier transform formula as follows: ； wherein is a time domain signal, is a window function, is a time offset, is an angular frequency; Using supervised learning to train one or more deep learning models using annotated historical data sets, regularization techniques, and early stopping methods; Using an adaptive mechanism to allow the model to automatically adjust internal parameters according to changes in wind speed and temperature in the external environment; After the model is trained, it is deployed to the actual monitoring system to analyze real-time acoustic wave signals, determine whether the blade has potential problems, and predict future possible failures. If an abnormal situation is detected, the system will issue an alarm to notify maintenance personnel.

3. The method of claim 1, wherein: The method of standardizing and preprocessing comprises the following steps: Breaking down large-scale acoustic wave data analysis tasks into several subtasks, each corresponding to data from a specific time period or specific wind turbine generator; Establishing a distributed computing cluster containing multiple computing nodes, and reasonably distributing each subtask to different computing nodes for execution through a load balancing algorithm; Using gRPC or ZeroMQ data transmission protocols to quickly and accurately transfer data required by subtasks to corresponding computing nodes; When all subtasks are completed, the master node is responsible for collecting the results returned by each computing node and performing merging processing; Preparing backup nodes for critical subtasks during task allocation. If the original node fails, the backup node will immediately take over the unfinished task; By monitoring the working state of each computing node, dynamically adjusting the task allocation strategy, and predicting future computing needs based on historical running data.

4. The method of claim 1, wherein: A quantum random number generator is used to initialize the weight values in the machine learning algorithm. The high-entropy random numbers provided by the quantum random number generator are used to improve the selection of initial weights.

5. The method of claim 1, wherein: It also includes one or more drones programmed to fly along a predetermined path to carry acoustic sensors close to wind turbines for close-range detection, while transmitting the collected data back to the ground station in real time for analysis.

6. A monitoring method based on the sound of the rotation of the blades according to claim 5, characterized in that: The method for detecting blades using drones includes the following steps: Using genetic algorithms or ant colony algorithms to design optimal flight paths for drones based on factors such as flight distance, energy consumption, safety, and coverage range; Using wireless communication links between drones and ground stations to support real-time transmission of high-definition audio and video streams; Integrating global positioning systems and inertial measurement units to enable autonomous flight of drones in complex environments without human intervention; The data collected by the drone is transmitted to the ground station through encryption, and the ground station is responsible for data storage, processing, and analysis; all data are time-stamped for subsequent comparison and tracking.

7. A monitoring method based on the sound of the rotation of the blades according to claim 5, characterized in that: The drone is equipped with a visual recognition module that can identify specific markings or defects on the blades and combine them with acoustic monitoring results to provide a blade health diagnosis report, which includes: The drone carries a high-definition camera to capture images of the blade surface from multiple angles; Using computer vision techniques to preprocess the collected images, including denoising, contrast enhancement, and edge detection, and applying feature extraction algorithms to identify key points and texture information in the images; Combining pre-trained deep learning models to classify and quantify blade cracks, corrosion, or other damage in the images; Correlating the results of visual recognition with acoustic monitoring data to comprehensively assess the overall health of the blades.

8. An apparatus for assisted monitoring, the apparatus comprising: To provide hardware support for the monitoring method of any one of claims 1-7, including: Microphone array: integrating a high-sensitivity microphone array to capture weak airflow disturbance sound wave signals generated by the rotation of wind turbine blades; Embedded processor: for performing preliminary feature extraction and anomaly detection data processing tasks locally; Data transmission module: equipped with a wireless communication module to support sending complete data to a cloud server for in-depth analysis; Power management system: using a solar charging system and intelligent power management system; Data compression module: compressing data that needs to be uploaded to the cloud; Local storage: built-in large-capacity Flash memory for temporary storage of a large amount of monitoring data; User interface: providing a user interface for maintenance personnel to view real-time monitoring results, set parameters, and receive alarm notifications.

9. A device for assisted monitoring according to claim 8, characterized in that: The embedded processor has an artificial intelligence chip built-in to accelerate the execution of machine learning algorithms.

10. The device for assisted monitoring according to claim 8, characterized in that: It also includes an AR glasses system, which allows maintenance personnel to see historical monitoring data, current status indicators, and possible problem areas virtually overlaid on the real world when viewing wind turbines on site, thereby guiding maintenance work.

Images