A wind turbine vibration monitoring and fault diagnosis method

Abstract

The application discloses a wind turbine vibration monitoring and fault diagnosis method, an edge computing system is additionally arranged between a detection module and a cloud server, the edge computing system collects detection signals in real time, and carries out pre-processing and fault diagnosis on the detection signals, so that the operation burden of the cloud server can be shared. After the edge computing system collects data, the edge computing system carries out screening and reporting, heartbeat uploading is carried out when there is no exception within a threshold, real-time reporting and local early warning are started when the threshold is exceeded, and the burden of the network is reduced. The edge computing system can make feedback nearby, and the business execution capacity is higher, local early warning is carried out in the first time when there is a threshold exceeding danger, and the cloud server is reported at the same time, so that the response time is reduced. The edge computing system adopts a high-speed data acquisition card and a high-precision clock, so that the continuity and synchronism of data acquisition are ensured; a time stamp technology is adopted, accurate time marks are added to each detection signal package, and the synchronism of data processing is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine fault diagnosis, and particularly relates to a wind turbine vibration monitoring and fault diagnosis method. BACKGROUND

[0002] Wind turbine vibration monitoring and fault diagnosis is a key technology for ensuring the normal operation and safety of wind turbines. By monitoring the vibration signals of each component of the wind turbine and analyzing these signals, potential faults can be detected in a timely manner, and appropriate measures can be taken to prevent accidents and ensure equipment reliability.

[0003] Although vibration monitoring and fault diagnosis of wind turbines play an important role in improving equipment operation efficiency and reducing faults, there are still some challenges and deficiencies in practical application. For example, in the presence of multiple vibration sources and environmental noise interference, how to accurately identify the fault location. How to realize uninterrupted monitoring of the system while ensuring the synchronization of each vibration signal collection to ensure the consistency of the signal phase. How to ensure the safety and reliability of the data acquisition system when transmitting data to the host computer, including data storage in the database and transmission to the monitoring and diagnosis system for analysis and prediction. How to realize the timeliness and accuracy of the wind turbine vibration monitoring and fault diagnosis system, especially the timeliness of the signal analysis and fault diagnosis method, and quickly and accurately diagnose the specific fault mode when the system is abnormal. SUMMARY

[0004] The purpose of the present application is to overcome the deficiencies of the prior art and provide a wind turbine vibration monitoring and fault diagnosis method.

[0005] To achieve the above purpose, the technical solution provided by the present application is:

[0006] A wind turbine vibration monitoring and fault diagnosis method, comprising:

[0007] Installing a suitable detection module at a key position of the wind turbine;

[0008] Detecting the detection signal of the key position of the wind turbine through the detection module;

[0009] Based on the detection signal, fault diagnosis is performed through scheme one or scheme two;

[0010] Scheme one:

[0011] An edge computing system is arranged at each wind turbine to collect detection signals in real time, and the detection signals are preprocessed and fault diagnosis is performed;

[0012] Scheme two:

[0013] Real-time detection signals are collected and preprocessed by the edge computing system installed at each wind turbine, and fault diagnosis is performed through the cloud server.

[0014] Further, when installing suitable detection modules at key positions of the wind turbine, a multi-point deployment method is adopted; the key positions of the wind turbine include blades, main shafts, speed increasers, gearboxes, and generators; among them, multiple vibration sensors are installed on the blades to cover the entire blade surface, facilitating comprehensive monitoring of the working state of the blades.

[0015] An analog-to-digital converter is configured in the detection module to collect data at a sampling rate of at least 2 times the rotational speed of the wind turbine.

[0016] Further, in scheme one,

[0017] The edge computing system uses an industrial-grade embedded server and installs an operating system RTLinux or VxWorks optimized for real-time data processing to ensure real-time and reliable data processing, and uses containerization technology to deploy application programs to achieve rapid deployment, isolation, and expansion of applications.

[0018] The edge computing system uses a stream processing framework for real-time data stream preprocessing and analysis.

[0019] During analysis, the edge computing system runs a lightweight machine learning model for fault diagnosis.

[0020] Further, the lightweight machine learning model is trained and adjusted by the cloud server and deployed to the edge computing system, specifically including:

[0021] The cloud server uses historical data and machine learning algorithms to train the machine learning model.

[0022] The trained machine learning model is deployed to the edge computing system, and a lightweight deep learning framework TensorFlow Lite or ONNX Runtime is used for inference.

[0023] An online update mechanism is used to enable the machine learning model to learn and adapt itself according to new operation data.

[0024] Further, encryption technology is used for data transmission between the edge computing system and the cloud server, including SSL / TLS, to encrypt data transmission and prevent data from being stolen or tampered with during transmission; data transmission protocols including MQTT and CoAP are established to ensure correct data transmission in unreliable network environments; and data integrity checking and fault tolerance mechanisms are adopted.

[0025] Further, the data integrity checking and fault tolerance mechanism includes:

[0026] Using a check algorithm to detect any errors or tampering of data during transmission;

[0027] Using sequence numbers and retransmission mechanisms to ensure the order and integrity of data packets, while retransmitting when missing or damaged data packets are found;

[0028] Data redundancy and segmented transmission, dividing data into small pieces and adding redundancy information, so that the original data can still be recovered when part of the data is lost.

[0029] Further, the edge computing system uses high-speed data acquisition cards and high-precision clocks to ensure the continuity and synchronization of data acquisition; uses timestamp technology to add precise time labels to each detection signal packet, ensuring the synchronization of data processing.

[0030] Further, during monitoring and fault diagnosis,

[0031] Using shielded cables and optical fiber transmission to reduce electromagnetic interference, while using differential signal transmission technology to reduce common mode noise; dynamically adjusting the sampling rate according to the operating characteristics of the wind turbine and the fault diagnosis requirements, to ensure that the fault characteristic frequency can be accurately captured;

[0032] Effective suppression of noise in a specific frequency range through digital filters; using time-frequency analysis techniques including wavelet transform to decompose signals into sub-bands of different scales and frequencies, from which fault features are extracted; establishing a mathematical model of environmental noise, including wind noise and mechanical noise, to more accurately identify and separate noise; using adaptive filtering technology to dynamically adjust filter parameters based on real-time data, achieving adaptive noise suppression;

[0033] Installing a special environmental noise monitoring device to monitor environmental noise levels in real time; using adaptive filtering-based noise cancellation technology to remove the effects of environmental noise from vibration signals;

[0034] Combining vibration, sound, temperature and other sensor data, using data fusion algorithms to improve signal reliability and accuracy; deploying redundant sensors at key locations to improve fault detection accuracy and robustness through multiple sensor data comparison and fusion.

[0035] Further, the cloud server uses historical data and machine learning algorithms to train machine learning models, the training process includes:

[0036] Collaborating with wind turbine maintenance experts to label historical data of historical fault cases in detail, including fault type, occurrence time, affected components, to provide training data for supervised learning algorithms;

[0037] When applying time-frequency analysis method, key features of vibration signal are extracted, including energy spectrum, frequency components, and waveform characteristics; feature selection techniques are used to reduce noise features and improve the generalization ability of the model;

[0038] CNN is used for feature extraction and classification of vibration signals; LSTM or gated recurrent unit is used to process time series data to capture the dynamic changes of wind turbine operation state and predict potential failure trends;

[0039] Stacking technology is used to combine the prediction results of different models as input to obtain a new machine learning model;

[0040] Unsupervised learning and anomaly detection: K-means or DBSCAN clustering algorithm is applied to cluster normal operation data and identify abnormal data points to discover new or unknown failure patterns; based on auto-encoder anomaly detection model, the model is trained to identify normal operation patterns and monitor and alarm abnormal behavior deviating from normal patterns in real time;

[0041] Online learning and adaptive adjustment: according to the real-time operation state and environmental changes of wind turbine, model parameters and diagnostic thresholds are dynamically adjusted.

[0042] Further, the cloud server is provided with a health management system and a user interface;

[0043] The health management system automatically generates a maintenance plan based on the fault diagnosis result;

[0044] The user interface displays the fault report and the corresponding maintenance plan.

[0045] Compared with the prior art, the present scheme has the following principles and advantages:

[0046] 1. An edge computing system is added between the detection module and the cloud server, which collects detection signals in real time, pre-processes the detection signals, and performs fault diagnosis, thereby sharing the computational burden of the cloud server. The edge computing system filters and reports the data after collecting, and uploads the data within the threshold value, and starts real-time reporting and local early warning when the data exceeds the threshold value, thereby reducing the burden of the network. The edge computing system can provide feedback nearby, and has higher business execution capability. When there is a threshold value exceeding danger, the edge computing system can provide local early warning in the first time, and report to the cloud server, thereby reducing the response time.

[0047] 2. The edge computing system uses signal fusion technology to fuse vibration signals and state signals from multiple sources to realize effective fusion of different types of signals and improve the accuracy and comprehensiveness of monitoring.

[0048] 3、The edge computing system adopts a high-speed data acquisition card and a high-precision clock to ensure the continuity and synchronicity of data acquisition; adopts a timestamp technology to add accurate time labels to each detection signal package, thereby ensuring the synchronicity of data processing.

[0049] 4、The data transmission between the edge computing system and the cloud server adopts encryption technology, including SSL / TLS, to encrypt the data transmission and prevent the data from being stolen or tampered with during transmission; a data transmission protocol including MQTT and CoAP is established to ensure that the data can be correctly transmitted in an unreliable network environment; a check algorithm is used to detect any errors or tampering of the data during transmission; a sequence number and retransmission mechanism are used to ensure the order and integrity of the data package, and retransmission is performed when missing or damaged data packages are found; data redundancy and segmented transmission are performed to divide the data into small blocks and add redundant information so that the original data can still be recovered when part of the data is missing.

[0050] 5、A fault prediction and diagnosis model based on continuous optimization learning is used to diagnose faults of the wind turbine, so that the fault position can be accurately identified even in the presence of multiple vibration sources and environmental noise interference. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, a brief introduction to the services required in the embodiments or the prior art description will be given below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0052] Figure 1 A principle flowchart of a wind turbine vibration monitoring and fault diagnosis method according to an embodiment of the present application;

[0053] Figure 2 A connection block diagram of a wind turbine vibration monitoring and fault diagnosis system according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] The present application will be further described below in conjunction with three specific embodiments:

[0055] Embodiment 1

[0056] As shown in Figure 2 , the wind turbine vibration monitoring and fault diagnosis system according to the present embodiment includes a detection module, an edge computing system, and a cloud server.

[0057] As shown in Figure 1 , the working principle of the system is as follows:

[0058] First, suitable detection modules are installed at key locations such as the blades, main shaft, speed-increasing box, gear box, and generator of the wind turbine;

[0059] In this step,

[0060] Blades: Vibration sensors are installed at the root or middle of the blades to monitor the vibration during rotation and the dynamic load caused by wind;

[0061] Main shaft: The main shaft is the core component of the wind turbine, and sensors are installed near the bearings to monitor the rotation and vibration of the main shaft;

[0062] Speed-increasing box and gear box: Sensors are installed at key locations in the speed-increasing box and gear box to monitor the meshing of gears and potential wear or damage;

[0063] Generator: Sensors are installed on the casing or bearings of the generator to monitor the operating state and abnormal vibration of the generator.

[0064] The detection modules include:

[0065] Piezoelectric acceleration sensor: This sensor is based on the piezoelectric effect and can convert mechanical vibration into electrical signals. They have high sensitivity, wide frequency response range, and good environmental adaptability, suitable for capturing vibration data of wind turbines;

[0066] Charge amplification type acceleration sensor: This type of sensor enhances the signal through a charge amplifier, providing high sensitivity and low noise output, suitable for detecting weak vibration signals;

[0067] Speed sensor: Speed sensors measure vibration speed and are usually used to monitor the rotational speed and vibration speed of wind turbines, helping to analyze the dynamic behavior of the equipment;

[0068] Displacement sensor: Displacement sensors are used to measure the relative displacement of mechanical components, which can help identify abnormal movement of components such as bearings and gearboxes;

[0069] Temperature sensor: Temperature sensors are used to monitor the temperature of key components of wind turbines, as changes in temperature may be related to vibrations and are one of the precursors of faults.

[0070] The working principles of the above vibration sensors and temperature sensors are as follows:

[0071] The working principle of the vibration sensor is usually based on the piezoelectric effect or electromagnetic induction. In piezoelectric sensors, mechanical vibrations act on piezoelectric crystals, causing changes in electric charge. These changes in electric charge are amplified and converted to output electrical signals. In sensors based on electromagnetic induction, vibrations cause the magnets inside the sensor to move near the coils, generating an induced current, thereby converting vibrations into electrical signals.

[0072] The working principle of temperature sensors is based on the thermocouple effect or thermistor. Thermocouple sensors are composed of two different metal wires that generate a voltage difference when one end is heated, and this voltage difference is proportional to the temperature change. The resistance value of a thermistor changes with temperature, and this change can be measured and converted into a temperature reading.

[0073] The data collected by these sensors can provide comprehensive monitoring of the operating state of the wind turbine, timely detection of potential faults and performance degradation, and thus preventive maintenance measures to improve the reliability and efficiency of the wind turbine.

[0074] After installing suitable detection modules at key positions such as blades, main shafts, speed increasers, gearboxes, and generators of the wind turbine, then corresponding edge computing systems are arranged at each wind turbine;

[0075] In this embodiment, the edge computing system uses an open-source Linux system, with rich uplink interfaces such as 4G, RJ45, and lora, and can be extended. It has a high-performance processor and large storage space, making it have certain computing power and storage space.

[0076] Next, the detection signals of the key positions of the wind turbine are detected by the detection modules.

[0077] The detection signals include vibration signals and non-vibration form state signals, including speed, temperature, wind speed, and physical and chemical performance parameters of lubricating oil;

[0078] Next, the edge computing system collects detection signals in real time, and pre-processes and diagnoses faults.

[0079] In this step, the detection signals are pre-processed, that is, features are extracted from the original signals, which is a key step in the process of wind turbine vibration monitoring and fault diagnosis. This process involves signal processing techniques aimed at extracting information from raw data that helps identify and classify fault patterns. Here are the basic steps of feature extraction and filtering techniques that can be used to remove environmental noise:

[0080] Feature extraction steps:

[0081] Time domain analysis:

[0082] Mean, variance, and peak value: These basic statistics can provide amplitude and distribution characteristics of the signal.

[0083] Pulse count and pulse width: For impact faults such as bearing damage, pulse count and width can provide information about the severity of the fault.

[0084] Frequency domain analysis:

[0085] Fast Fourier Transform (FFT): Converts time-domain signals to the frequency domain, extracting frequency components and energy distribution.

[0086] Envelope Analysis: By calculating the envelope of the signal, frequency modulation and bearing fault frequencies can be identified.

[0087] Time-Frequency Domain Analysis:

[0088] Short-Time Fourier Transform (STFT): Combines time and frequency information, suitable for analyzing non-stationary signals.

[0089] Wavelet Transform: Provides a multi-scale time-frequency analysis method, suitable for analyzing signals with different frequency components.

[0090] Advanced Feature Extraction Techniques:

[0091] Modal Decomposition: Such as Empirical Mode Decomposition (EMD) and Hilbert-Huang Transform (HHT), used to extract the intrinsic modal functions (IMFs) of the signal.

[0092] Feature Selection and Dimensionality Reduction: Using methods such as Principal Component Analysis (PCA) or Independent Component Analysis (ICA), reducing data dimensions and highlighting important features.

[0093] Filtering Techniques to Remove Environmental Noise:

[0094] Low-Pass Filter:

[0095] Used to remove high-frequency noise, preserving low-frequency components in the signal, suitable for removing wind noise and other high-frequency interference.

[0096] High-Pass Filter:

[0097] Used to remove low-frequency noise, such as low-frequency background noise from mechanical vibrations, preserving high-frequency details in the signal.

[0098] Band-Pass Filter:

[0099] Preserves signals within a specific frequency range, suitable for removing noise outside the wind turbine fault diagnosis frequency range.

[0100] Adaptive Filter:

[0101] Such as the Least Mean Square Error (LMS) algorithm, dynamically adjusts filter parameters based on real-time characteristics of the signal, achieving adaptive noise suppression.

[0102] Notch Filter:

[0103] Designed to eliminate specific frequency interference, such as power line interference (50 / 60Hz), to eliminate noise at these specific frequencies.

[0104] Digital signal processing technology:

[0105] Using digital filters such as Finite Impulse Response (FIR) or Infinite Impulse Response (IIR) filters, the filter can be designed more accurately to meet specific frequency response requirements.

[0106] Through the above feature extraction and filtering technology, the features that help fault diagnosis can be extracted from the original vibration signal of the wind turbine, and the environmental noise can be removed, improving the accuracy and reliability of fault detection.

[0107] In the above, when installing suitable detection modules at key positions of the wind turbine, a multi-point deployment method is adopted; the key positions of the wind turbine include blades, main shafts, speed increasers, gearboxes and generators; among them, multiple vibration sensors are installed on the blades to cover the entire blade surface, facilitating comprehensive monitoring of the working state of the blades;

[0108] Analog-to-digital converters are configured in the detection module to collect data at a sampling rate of at least 2 times the speed of the wind turbine;

[0109] The edge computing system uses an industrial-grade embedded server, installs an operating system RTLinux or VxWorks optimized for real-time data processing, ensures the real-time and reliability of data processing, and uses containerization technology (such as Docker) to deploy applications, realizes the rapid deployment, isolation and expansion of applications;

[0110] The edge computing system uses a stream processing framework (such as Apache Kafka Streams or Apache Flink) for real-time data stream preprocessing and analysis;

[0111] During analysis, the edge computing system runs a lightweight machine learning model for fault diagnosis.

[0112] During fault diagnosis, the safe and reliable transmission of detection signals from each sensor of the detection module to the edge computing system involves several key links, ensuring the safety and integrity of the detection signal data during collection, transmission and processing:

[0113] Data encryption: Use encryption protocols such as SSL / TLS to encrypt sensor data end-to-end, ensuring data security during transmission. This means that even if the data is intercepted during transmission, unauthorized third parties cannot interpret the data content.

[0114] Data transmission protocol: Establish data transmission protocols including MQTT, CoAP, etc. to ensure correct data transmission in unreliable network environments. These protocols aim to provide low-power, high-efficiency data exchange for device communication in the Internet of Things environment.

[0115] Data integrity check and fault tolerance mechanism: including using check algorithm to detect any errors or tampering of data during transmission, using sequence number and retransmission mechanism to ensure the order and integrity of data packets, and retransmitting when missing or damaged data packets are found. In addition, data redundancy and segmented transmission are carried out, which divides data into small pieces and adds redundant information, so that the original data can still be recovered when part of the data is lost. These measures together improve the reliability and robustness of data transmission.

[0116] Data preprocessing of edge computing: the edge computing system filters and reports the data after collecting, and uploads the heartbeat if there is no exception within the threshold, and starts real-time reporting and local warning if it is out of limit, reducing the burden of the network and making feedback nearby. This not only reduces the burden of the network, but also improves the timeliness and efficiency of data processing.

[0117] These measures work together to provide a safe, reliable and efficient transmission path for data transmission from sensors to processing units, ensuring the security and integrity of data during collection, transmission and processing, while also optimizing the use of network resources to ensure high performance and high reliability of the system.

[0118] Further, the lightweight machine learning model is trained and adjusted by the cloud server and deployed to the edge computing system, including:

[0119] The cloud server uses historical data and machine learning algorithms to train the machine learning model;

[0120] Deploy the trained machine learning model to the edge computing system, use lightweight deep learning framework TensorFlow Lite or ONNX Runtime for inference;

[0121] Use online update mechanism to enable the machine learning model to learn and adapt to new operation data.

[0122] Data transmission between edge computing system and cloud server uses encryption technology, including SSL / TLS, to encrypt data transmission and prevent data from being stolen or tampered with during transmission; establish data transmission protocols including MQTT, CoAP, to ensure correct transmission of data in unreliable network environment; and use data integrity check and fault tolerance mechanism.

[0123] Data integrity check and fault tolerance mechanism includes:

[0124] Use check algorithm to detect any errors or tampering of data during transmission;

[0125] Use sequence number and retransmission mechanism to ensure the order and integrity of data packets, and retransmit when missing or damaged data packets are found.

[0126] Data redundancy and segmented transmission are performed, dividing data into small pieces and adding redundant information to recover the original data even if part of the data is lost.

[0127] The edge computing system uses high-speed data acquisition cards and high-precision clocks to ensure the continuity and synchronization of data acquisition; it uses timestamp technology to add precise time markers to each detection signal package, ensuring the synchronization of data processing.

[0128] During monitoring and fault diagnosis,

[0129] Shielded cables and optical fiber transmission are used to reduce electromagnetic interference, while differential signal transmission technology is used to reduce common-mode noise; based on the operating characteristics of wind turbines and fault diagnosis requirements, the sampling rate is dynamically adjusted to ensure accurate capture of fault feature frequencies;

[0130] Condition Monitoring and Fault Diagnosis: Explain how to identify and classify bearing and gearbox faults through analyzed and processed data, which may involve specific diagnostic algorithms. (Identify and classify bearing and gearbox faults through analyzed and processed data, which usually involves a series of signal processing techniques and specific diagnostic algorithms. The following is a detailed process, including data acquisition, preprocessing, feature extraction, application of fault diagnosis algorithms, and final fault classification.

[0131] 1. Data Acquisition

[0132] First, appropriate sensors need to be installed at key locations of the wind turbine, such as bearings and gearboxes. These sensors can be vibration sensors, temperature sensors, or sound sensors, used to capture the running state of the equipment. Through high-speed data acquisition cards and high-precision clocks, ensure the continuity and synchronization of data acquisition, while using timestamp technology to add precise time markers to each detection signal package.

[0133] 2. Data Preprocessing

[0134] In the edge computing system, the collected raw data is preprocessed, including removing environmental noise, filtering, normalization, etc. This stage may use digital filters such as Finite Impulse Response (FIR) or Infinite Impulse Response (IIR) filters to effectively suppress noise in specific frequency ranges. At the same time, time-frequency analysis techniques such as wavelet transform can be used to decompose the signal into sub-bands of different scales and frequencies, preparing for subsequent feature extraction.

[0135] 3. Feature Extraction

[0136] Feature extraction is the process of extracting useful information for fault diagnosis from raw data. Frequency features can be extracted using Fast Fourier Transform (FFT), or time-frequency features can be extracted using wavelet transform. In addition, machine learning algorithms such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA) can be applied to extract and select features useful for fault diagnosis.

[0137] 4. Fault diagnosis algorithm

[0138] In the cloud server, deep learning frameworks such as TensorFlow or PyTorch are used to train fault prediction and diagnosis models. These models can be image recognition models based on Convolutional Neural Networks (CNN) for identifying fault patterns from vibration signal spectrograms or time-domain graphs; or sequence models based on Recurrent Neural Networks (RNN) for processing time series data and capturing dynamic changes in device operating conditions.

[0139] 5. Fault classification

[0140] The pre-processed data is classified by the trained model to identify specific fault types of bearings and gearboxes. For example, bearing faults may exhibit abnormal frequency components, while gearbox faults may be associated with specific vibration patterns or sound characteristics. The model will distinguish between different types of faults based on these features.

[0141] 6. Result verification and optimization

[0142] Finally, the diagnostic results need to be verified to ensure accuracy. The performance of the model can be evaluated by comparing it with actual maintenance records and expert knowledge. Based on the verification results, model parameters and diagnostic thresholds are continuously optimized to improve the accuracy and reliability of fault diagnosis.

[0143] Through the above process, the faults of bearings and gearboxes in wind turbines can be effectively identified and classified, thereby improving the efficiency and reliability of the equipment, reducing potential downtime and maintenance costs.

[0144] Effective suppression of noise in specific frequency ranges through digital filters such as Finite Impulse Response (FIR) or Infinite Impulse Response (IIR) filters; use of time-frequency analysis techniques including wavelet transform to decompose signals into sub-bands of different scales and frequencies, from which fault features are extracted; establishment of mathematical models of environmental noise, including wind noise and mechanical noise, to more accurately identify and separate noise; use of adaptive filtering techniques such as Least Mean Square Error (LMS) algorithm to dynamically adjust filter parameters based on real-time data, achieving adaptive noise suppression;

[0145] Install special environmental noise monitoring devices to monitor environmental noise levels in real time; use adaptive filtering-based noise cancellation technology to remove the influence of environmental noise from vibration signals;

[0146] Combine vibration, sound, temperature, and other sensor data, use data fusion algorithms to improve signal reliability and accuracy; deploy redundant sensors at key locations, compare and fuse data from multiple sensors to improve fault detection accuracy and robustness.

[0147] The cloud server uses historical data and machine learning algorithms to train machine learning models, and the training process includes:

[0148] Collaborate with wind turbine maintenance experts to label historical data from historical fault cases in detail, including fault type, occurrence time, and affected components, to provide training data for supervised learning algorithms;

[0149] Apply time-frequency analysis methods to extract key features of vibration signals, including energy spectrum, frequency components, and waveform characteristics; use feature selection techniques to reduce noise features and improve model generalization ability;

[0150] Use CNN to extract features and classify vibration signals; use LSTM or gated recurrent units to process time series data and capture dynamic changes in wind turbine operating conditions to predict potential failure trends;

[0151] Use stacking techniques to use the prediction results of different models as input to obtain a new machine learning model;

[0152] Unsupervised learning and anomaly detection: apply K-means or DBSCAN clustering algorithms to cluster normal operating data and identify abnormal data points to discover new or unknown failure patterns; use an autoencoder-based anomaly detection model to train the model to identify normal operating patterns, monitor and alert abnormal behavior that deviates from normal patterns in real time;

[0153] Online learning and adaptive adjustment: dynamically adjust model parameters and diagnostic thresholds based on real-time operating conditions and environmental changes of wind turbines.

[0154] The cloud server has a health management system and a user interface;

[0155] The health management system automatically generates maintenance plans based on fault diagnosis results;

[0156] The user interface displays fault reports and corresponding maintenance plans.

[0157] In this embodiment,

[0158] An edge computing system is added between the detection module and the cloud server, which collects detection signals in real time, pre-processes the detection signals, and performs fault diagnosis, which can share the computational burden of the cloud server. The edge computing system filters and reports the data after collecting it. If there is no abnormality within the threshold, it will upload the heartbeat. If it exceeds the limit, it will start real-time reporting and local early warning, reducing the burden on the network. The edge computing system can provide feedback locally, improving the system's operational capacity. When there is a threshold risk, the system will provide local early warning and report to the cloud server, reducing response time.

[0159] Data transmission between the edge computing system and the cloud server uses encryption technology, including SSL / TLS, to encrypt data transmission and prevent data from being stolen or tampered with during transmission. Data transmission protocols such as MQTT and CoAP are established to ensure correct data transmission in unreliable network environments. Checksum algorithms are used to detect any errors or tampering during data transmission. Sequence numbers and retransmission mechanisms are used to ensure data packet order and integrity, and retransmission is performed when missing or damaged data packets are found. Data redundancy and segmented transmission are used to divide data into small blocks and add redundant information to recover the original data when some data is missing.

[0160] A fault prediction and diagnosis model based on continuous optimization learning is used to diagnose faults in wind turbines, allowing accurate identification of fault locations even in the presence of multiple vibration sources and environmental noise interference.

[0161] Edge computing applications: This section explains how edge computing improves system real-time performance, reduces costs, and simplifies maintenance. The application of edge computing in wind turbine vibration monitoring and fault diagnosis systems can significantly improve system real-time performance, reduce operating costs, and simplify system maintenance. The following are several key aspects that detail how edge computing achieves these advantages:

[0162] 1. Improve system real-time performance

[0163] Data preprocessing and preliminary analysis: By preprocessing and preliminary analysis of data at edge computing nodes, the amount of data that needs to be transmitted to the cloud server can be significantly reduced, thereby reducing network latency. This is crucial for real-time monitoring and fault detection of wind turbines, as it ensures that potential problems can be quickly identified and responded to even in poor network conditions.

[0164] Local decision-making and response: Edge computing nodes can analyze and make decisions at the data generation site, which means that for some simple fault warnings or operational optimizations, the system does not need to wait for cloud processing and return instructions, allowing for fast response.

[0165] 2. Reduce operating costs

[0166] Reduced data transmission volume: By performing data filtering and compression at the edge, only necessary data is transmitted to the cloud for in-depth analysis. This not only reduces the burden on the cloud processing, but also reduces data transmission costs.

[0167] Optimized resource usage: The deployment of edge computing nodes can be dynamically adjusted according to actual needs, enabling the system to effectively utilize computing resources and reduce unnecessary resource consumption, thereby reducing energy consumption and operation and maintenance costs.

[0168] 3. Simplified system maintenance

[0169] Modular deployment: The modular design of edge computing nodes makes the system easy to expand and maintain. When upgrading the system or increasing computing capacity, more edge nodes can be simply added instead of replacing the entire system.

[0170] Fault tolerance and resilience: Edge computing enhances the fault tolerance and resilience of the system through a distributed architecture. Even if a node fails, other nodes can continue to work, ensuring the continuous operation of the system and the reliability of services.

[0171] Localized processing reduces dependency: For wind farms located in remote or unstable network connection areas, edge computing can reduce dependence on central cloud servers, enabling the system to continue to operate even in offline or semi-offline states.

[0172] By utilizing edge computing, the wind turbine vibration monitoring and fault diagnosis system can provide more efficient, reliable and economical operation and maintenance solutions, while also providing strong flexibility for future expansion and upgrading of the system.

[0173] Example Two

[0174] In this embodiment, the wind turbine vibration monitoring and fault diagnosis system includes system hardware components and system software components.

[0175] System hardware components include:

[0176] Detection module:

[0177] Function: Installed at key positions of the wind turbine, such as blades, main shafts, speed increasers and generators, etc., for real-time detection of vibration signals and other status signals.

[0178] Interaction: Real-time transmission of detected signals to the corresponding edge computing system for preprocessing.

[0179] Edge computing system:

[0180] Function:

[0181] Real-time acquisition and preprocessing of detection module signals, including data filtering and preliminary calculations.

[0182] Signal fusion techniques such as wavelet transform and Kalman filtering are used to fuse vibration signals and state signals.

[0183] High-speed data acquisition cards and high-precision clocks are used in data acquisition to ensure data continuity and synchronization.

[0184] The collected data is screened and reported, realizing heartbeat upload, real-time reporting when exceeding limits, and local early warning.

[0185] Interaction: The data processed by the edge computing system is sent to the cloud server, while local feedback and early warning can be performed.

[0186] Cloud Server:

[0187] Function: Receive detection signals preprocessed by the edge computing system, and use fault prediction and diagnosis models based on continuous optimization learning to diagnose faults of wind turbines.

[0188] Interaction: The analysis and diagnosis results of the server can be used to guide maintenance decisions, improve the operation efficiency and reliability of wind turbines, and may be fed back to the edge computing system to optimize local processing strategies.

[0189] System Software Components

[0190] Data Transmission Protocol: Including MQTT, CoAP, etc., to ensure correct and secure data transmission in unreliable network environments.

[0191] Encryption Technology: Use SSL / TLS and other encryption technologies for end-to-end encryption of data to protect data security during transmission.

[0192] Data Processing and Analysis Software:

[0193] In the edge computing system, software is responsible for real-time data stream processing, implementation of signal fusion technology, and execution of early warning mechanisms.

[0194] In the cloud server, software uses deep learning and machine learning models for fault prediction and diagnosis.

[0195] Monitoring and Optimization Tools: Prometheus, Grafana, etc. are used to monitor system performance and health status, and ELKStack (Elasticsearch, Logstash, Kibana) is used for log management and analysis to support system maintenance and performance optimization.

[0196] Through the close cooperation of these components, the system can achieve efficient and accurate wind turbine vibration monitoring and fault diagnosis, improve the real-time performance and reliability of the system, and reduce the operation and maintenance costs.

[0197] The working principle of this embodiment is as follows:

[0198] S1. Data acquisition

[0199] First, appropriate sensors need to be installed at key locations of the wind turbine, such as bearings and gearboxes. These sensors can be vibration sensors, temperature sensors, or sound sensors, used to capture the running state of the equipment. Through high-speed data acquisition cards and high-precision clocks, the continuity and synchronization of data acquisition are ensured, and the timestamp technology is used to add accurate time labels to each detection signal package.

[0200] S2. Data preprocessing

[0201] In the edge computing system, the collected raw data is preprocessed, including removing environmental noise, filtering, normalization, etc. This stage may use digital filters such as Finite Impulse Response (FIR) or Infinite Impulse Response (IIR) filters to effectively suppress noise in specific frequency ranges. At the same time, time-frequency analysis techniques such as wavelet transform can be used to decompose the signal into sub-bands of different scales and frequencies, preparing for subsequent feature extraction.

[0202] S3. Feature extraction

[0203] Feature extraction is to extract useful information for fault diagnosis from raw data. Fast Fourier Transform (FFT) can be used to extract frequency features, or wavelet transform can be used to extract time-frequency features. In addition, machine learning algorithms such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA) can be applied to extract and select features useful for fault diagnosis.

[0204] S4. Fault diagnosis algorithm

[0205] In the cloud server, deep learning frameworks such as TensorFlow or PyTorch are used to train fault prediction and diagnosis models. These models can be image recognition models based on Convolutional Neural Networks (CNN), used to identify fault patterns from vibration signal spectrograms or time-domain graphs; or sequence models based on Recurrent Neural Networks (RNN), used to process time series data and capture the dynamic changes of equipment running state.

[0206] S5. Fault classification

[0207] The pre-processed data is classified by the trained model to identify specific fault types of bearings and gearboxes. For example, bearing faults may exhibit abnormal frequency components, while gearbox faults may be associated with specific vibration patterns or sound characteristics. The model distinguishes between different types of faults based on these characteristics.

[0208] S6. Result verification and optimization

[0209] Finally, the diagnostic results need to be verified to ensure accuracy. The performance of the model can be evaluated by comparing it with actual maintenance records and expert knowledge. Based on the verification results, the model parameters and diagnostic thresholds are continuously optimized to improve the accuracy and reliability of fault diagnosis.

[0210] Through the above process, the faults of bearings and gearboxes in wind turbines can be effectively identified and classified, thereby improving the operation efficiency and reliability of the equipment, reducing potential downtime and maintenance costs.

[0211] Example Three

[0212] Compared with Example Two (directly deploying a deep learning model on a cloud server for fault prediction and diagnosis), this embodiment first deploys an integrated health management system, and then deploys a deep learning model inside the health management system to assist in fault prediction and diagnosis. This method provides a more comprehensive and modular solution, which can manage and coordinate different monitoring and diagnosis activities through the health management system, and can also flexibly adjust and optimize the functions of each module according to different needs and situations.

[0213] Specifically, in this embodiment,

[0214] Data acquisition and transmission:

[0215] Use wired or wireless networks to transmit sensor data to the integrated health management system. Choose appropriate communication methods such as Ethernet, Wi-Fi, LoRaWAN, etc. according to actual conditions.

[0216] Deploy data acquisition modules to convert analog quantities collected by sensors into digital quantities and transmit them to the health management system through the network.

[0217] Consider network stability and real-time performance to ensure that data can be transmitted to the cloud server in a timely and accurate manner.

[0218] Cloud storage and processing:

[0219] Build a database system on the cloud server to store the data collected by the sensors. Choose a database suitable for large-scale data storage and high-concurrency access, such as MySQL.

[0220] A data processing platform is configured for real-time processing of sensor data, feature extraction, and modeling. Technologies such as Apache Spark, AWS Lambda, etc. are used to realize real-time processing and analysis of data.

[0221] Data backup and disaster recovery mechanisms are implemented to ensure data security and reliability. Regular data backups are taken, and multiple copy storage methods are used to improve data availability.

[0222] State monitoring and fault diagnosis:

[0223] A state monitoring module is established in the health management system to track the running state of the equipment in real time. According to the preset threshold, the health status of the equipment is judged, and an alarm notification is generated.

[0224] Integrate fault diagnosis engine, combined with expert knowledge and historical fault cases, analyze monitoring data, provide root cause analysis of faults. Use artificial intelligence technology to build fault prediction and diagnosis model, improve the accuracy and reliability of fault diagnosis.

[0225] Maintenance decision support:

[0226] Maintenance decision support is provided in the health management system, and maintenance work orders are automatically generated based on fault diagnosis results. Combine real-time monitoring data and historical maintenance records to arrange necessary maintenance resources and personnel.

[0227] Design user interface, display fault report and maintenance guide, including fault diagnosis steps, preventive measures and repair suggestions. Provide a user-friendly operation interface through the Web interface, making it easy for maintenance personnel to check and operate.

[0228] Through the above detailed implementation, a complete wind turbine monitoring and intelligent diagnosis system can be established, realizing comprehensive monitoring, fault prediction and remote maintenance of equipment, improving the reliability and operation efficiency of the wind power system. At the same time, such a system can also reduce the cost of manual maintenance, reduce downtime, and maximize the stability of the wind power system.

[0229] The above-mentioned embodiments are only the preferred embodiments of the present application, and do not limit the scope of the present application. Any changes made in accordance with the shape and principle of the present application shall be covered within the scope of protection of the present application.

Claims

1. A method for wind turbine vibration monitoring and fault diagnosis, characterized in that: include: Install appropriate detection modules at key locations of wind turbines; Detection signals of key positions of wind turbines through detection modules; Based on the detection signals, fault diagnosis is performed in the following ways: Deploy an edge computing system at each wind turbine to collect detection signals in real time, pre-process the detection signals, and perform fault diagnosis; The edge computing system uses industrial-grade embedded servers and is installed with RTLinux or VxWorks, operating systems optimized for real-time data processing, to ensure real-time and reliable data processing. It also utilizes containerization technology to deploy applications, enabling rapid deployment, isolation, and scalability. Edge computing systems use stream processing frameworks to preprocess and analyze real-time data streams; During analysis, the edge computing system runs a lightweight machine learning model for fault diagnosis; The lightweight machine learning model is trained and adjusted through a cloud server and deployed to an edge computing system, specifically including: The cloud server uses historical data and machine learning algorithms to train the machine learning model; Deploy the trained machine learning model to the edge computing system and use the lightweight deep learning framework TensorFlow Lite or ONNX Runtime for inference; Use online update mechanisms to enable machine learning models to self-learn and adapt based on new operational data; During monitoring and troubleshooting, Shielded cables and optical fiber transmission are used to reduce electromagnetic interference, while differential signal transmission technology is used to reduce common-mode noise. The sampling rate is dynamically adjusted according to the operating characteristics of the wind turbine and fault diagnosis requirements to ensure that the fault characteristic frequency can be accurately captured. Effectively suppress noise within a specific frequency range through digital filters. Utilize time-frequency analysis techniques, including wavelet transform, to decompose the signal into sub-bands of varying scales and frequencies, from which fault features can be extracted. Establish a mathematical model for environmental noise, including wind noise and mechanical noise, to more accurately identify and separate it. Adaptive filtering technology is employed to dynamically adjust filter parameters based on real-time data, enabling adaptive noise suppression. Install a dedicated environmental noise monitoring device to monitor the environmental noise level in real time; use noise elimination technology based on adaptive filtering to remove the impact of environmental noise from the vibration signal; Combining multiple sensor data such as vibration, sound, and temperature, and using data fusion algorithms, improves signal reliability and accuracy; deploying redundant sensors at key locations, and improving the accuracy and robustness of fault detection through data comparison and fusion of multiple sensors.

2. A wind turbine vibration monitoring and fault diagnosis method according to claim 1, characterized in that: When installing appropriate detection modules at key locations on wind turbines, a multi-point deployment approach is adopted. Key locations on wind turbines include blades, main shafts, speed increasers, gearboxes, and generators. Multiple vibration sensors are installed on the blades to cover the entire blade surface, facilitating comprehensive monitoring of the blade's operating status. An analog-to-digital converter is configured in the detection module to collect data at a sampling rate that is at least twice the speed of the wind turbine.

3. A wind turbine vibration monitoring and fault diagnosis method according to claim 1, characterized in that: Data transmission between the edge computing system and the cloud server uses encryption technology, including SSL / TLS, to encrypt data transmission to prevent data from being stolen or tampered with during transmission; data transmission protocols including MQTT and CoAP are established to ensure that data can be correctly transmitted even in unreliable network environments; and data integrity verification and fault tolerance mechanisms are adopted.

4. A wind turbine vibration monitoring and fault diagnosis method according to claim 3, characterized in that: Data integrity checking and fault tolerance mechanisms include: Use verification algorithms to detect any errors or tampering of data during transmission; Use sequence numbers and retransmission mechanisms to ensure the order and integrity of data packets, and retransmit when lost or damaged packets are found; Perform data redundancy and segmented transmission, dividing data into small blocks and adding redundant information so that the original data can still be restored when part of the data is lost.

5. A wind turbine vibration monitoring and fault diagnosis method according to claim 1, characterized in that: The edge computing system uses a high-speed data acquisition card and a high-precision clock to ensure the continuity and synchronization of data acquisition; and uses timestamp technology to add an accurate time mark to each detection signal packet to ensure the synchronization of data processing.

6. A wind turbine vibration monitoring and fault diagnosis method according to any one of claims 1, 3, and 4, characterized in that: The cloud server uses historical data and machine learning algorithms to train the machine learning model. The training process includes: Collaborate with wind turbine maintenance experts to annotate historical data on fault cases, including fault type, occurrence time, and affected components, to provide training data for supervised learning algorithms; Apply time-frequency analysis methods to extract key features of vibration signals, including energy spectrum, frequency components, and waveform characteristics; use feature selection techniques to reduce noise characteristics and improve the generalization ability of the model; Use CNN to extract and classify vibration signals; use LSTM or gated recurrent units to process time series data, capture dynamic changes in wind turbine operating status, and predict potential failure trends; Using stacking technology, the prediction results of CNN, SVM and random forest are used as input to obtain a new machine learning model; Perform unsupervised learning and anomaly detection: Apply K-means or DBSCAN clustering algorithms to cluster normal operating data, identify abnormal data points, and thus discover new or unknown failure modes. Use an autoencoder-based anomaly detection model to train the model to identify normal operating modes and monitor and alert abnormal behaviors that deviate from normal modes in real time. Online learning and adaptive adjustment dynamically adjust model parameters and diagnostic thresholds according to the real-time operating status of the wind turbine and environmental changes.

7. A wind turbine vibration monitoring and fault diagnosis method according to claim 6, characterized in that: The cloud server is equipped with a health management system and user interface; The health management system automatically generates maintenance plans based on fault diagnosis results; The user interface displays fault reports and corresponding maintenance plans.

Images