Wind turbine generator monitoring method, device and monitoring system

Abstract

The invention discloses a wind turbine generator monitoring method, device and monitoring system. An edge computing device receives sensor data sent by a multi-channel optical fiber demodulator; performing multi-physics coupling load inversion on the sensor data based on a preset wind turbine generator dynamic reduced-order model to obtain a load inversion result; performing feature extraction on the sensor data to obtain feature information; performing wind turbine generator fault prediction based on the prediction model and the feature information to obtain a fault prediction result; under the condition that at least one of the load inversion result and the fault prediction result is abnormal, decoupling calculation is carried out on the load inversion result based on a preset fault transfer matrix, and fault source information is determined; and determining first data according to the sensor data, the load inversion result and the fault source information, and sending the first data to a front-end platform, so that the front-end platform performs three-dimensional visual rendering and state cloud map mapping presentation on the first data, and real-time visual display of the operation state of the wind turbine generator is completed.

Description

Technical Field

[0001] This application relates to the field of wind turbine generator condition monitoring technology, specifically to a wind turbine generator monitoring method, device and monitoring system. Background Technology

[0002] Wind power, as a crucial component of renewable energy, has experienced rapid global development in recent years. With the continuous increase in single-unit capacity of wind turbines (currently reaching 8MW~16MW), blade lengths exceeding 100 meters, and tower heights surpassing 160 meters, the operating loads and fatigue damage risks of various components have also increased. Frequent accidents such as blade breakage, loose tower flanges, and anchor cable failure not only cause significant economic losses (single repair costs exceeding one million yuan) but also threaten the safety of personnel at wind farms. Therefore, establishing a highly reliable and accurate condition monitoring system for key components of wind turbines to achieve early fault warning and intelligent diagnosis has become an urgent need for the industry's development.

[0003] Currently, wind turbine condition monitoring mostly uses resistance strain gauges as sensing elements, which suffer from problems such as poor electromagnetic interference resistance, short lifespan, and severe temperature drift. Common fiber optic monitoring solutions cannot solve the problems of dynamic monitoring of rotating parts and multi-physics coupling analysis. Furthermore, traditional data transmission modes have a contradiction between real-time performance and diagnostic depth, which cannot meet the high-precision, long-life, and intelligent monitoring requirements of large wind turbines. Summary of the Invention

[0004] This application provides a wind turbine monitoring method, device, and monitoring system that can achieve integrated processing of data from multiple components of the wind turbine, taking into account both real-time data processing and in-depth fault diagnosis, and providing a fully intelligent solution for turbine condition monitoring.

[0005] To achieve the above objectives, this application provides the following technical solution: In a first aspect, embodiments of this application provide a wind turbine monitoring method applied to an edge computing device. The edge computing device is connected to a multi-channel fiber optic demodulator of the wind turbine, and the multi-channel fiber optic demodulator is connected to fiber optic sensors located at the blade root, tower flange, and anchor cable anchorage area of ​​the wind turbine. The method includes: Receive sensor data sent by a multi-channel fiber optic demodulator; the sensor data includes dynamic strain signals of the blade at the blade root, displacement signals of the tower flange, and dynamic stress signals of the anchor cable in the anchoring zone. Based on a pre-defined wind turbine dynamics reduced-order model, multi-physics field coupled load inversion is performed on sensor data to obtain load inversion results; Feature extraction is performed on sensor data to obtain feature information; Based on the prediction model and feature information, wind turbine fault prediction is performed to obtain fault prediction results; If at least one of the load inversion results and fault prediction results is abnormal, the load inversion results are decoupled and calculated based on a preset fault transfer matrix to determine the fault source information. Based on sensor data, load inversion results, and fault source information, the first data is determined and sent to the front-end platform so that the front-end platform can perform three-dimensional visualization rendering and status cloud map mapping on the first data, thereby completing the real-time visualization display of the wind turbine's operating status.

[0006] In some embodiments, the method further includes: To obtain the optical power fluctuation value of a multi-channel fiber optic demodulator during data transmission; The dynamic strain signal of the blade is calibrated based on the optical power fluctuation value to obtain the calibrated blade strain signal. Sensor data is determined based on calibrated blade strain signals, tower flange displacement signals, and anchor cable dynamic stress signals.

[0007] In some embodiments, multiphysics coupled load inversion is performed on sensor data based on a preset wind turbine dynamics reduced-order model to obtain load inversion results, including: Based on the preset wind turbine dynamics reduced-order model, the calibrated blade strain signal, tower flange displacement signal, and anchor cable dynamic stress signal in the sensor data are converted into real physical loads to obtain the real-time blade force load, tower flange bending moment load, and anchor cable tension load. The load inversion results are determined based on the real-time stress load on the blades, the bending moment load on the tower flange, and the tension load on the anchor cables.

[0008] In some embodiments, feature extraction is performed on sensor data to obtain feature information, including: Wavelet transform is performed on the sensor data to obtain a multi-scale time-frequency decomposed signal; Extract the peak-to-peak value, spectral energy, and bandwidth amplitude of the multi-scale time-frequency decomposed signal; Characteristic information is determined based on the signal peak-to-peak value, spectral energy, and frequency band amplitude.

[0009] In some embodiments, wind turbine fault prediction is performed based on a prediction model and feature information to obtain fault prediction results, including: Based on the prediction model and feature information, the abnormal conditions of the wind turbine in the future within a preset time period are predicted to obtain the anomaly prediction probability value. Among them, the anomaly prediction probability value includes the probability value of each abnormal condition, such as resonance event, load exceeding standard, flange gap abnormality, and anchor cable loosening. Resonance event represents the abnormal condition of unit resonance caused by blade rotation; load exceeding standard represents the abnormal condition of the load of the blade, tower, and anchor cable exceeding the corresponding threshold; flange gap abnormality represents the abnormal condition of the flange gap not conforming to the preset gap range; and anchor cable loosening represents the abnormal condition of insufficient anchor cable tension leading to anchoring failure.

[0010] The fault prediction result is determined based on the anomaly prediction probability value.

[0011] In some embodiments, the load inversion results are decoupled based on a preset fault transmission matrix to determine fault source information, including: The load inversion results are input into the preset fault transmission matrix for decoupling calculation to obtain the contribution ratio of blade load anomaly, tower flange loosening, and anchor cable slack to the current anomaly. The fault source information is determined based on the contribution percentage.

[0012] In some embodiments, the method further includes: If any load in the load inversion result is greater than the corresponding preset load threshold, it is determined that the load inversion result is abnormal; If any abnormal prediction probability value in the fault prediction result is greater than the corresponding preset probability threshold, the fault prediction result is determined to be abnormal.

[0013] In some embodiments, the method further includes: If there are no abnormalities in the load inversion results and fault prediction results, the second data is determined based on the real-time azimuth angle signal and characteristic information of the wind turbine blade encoder. The second set of data is sent to the front-end platform for display.

[0014] Secondly, embodiments of this application provide an edge computing device, which includes a processor and a memory storing processor-executable instructions; when the executable instructions are executed by the processor, the above-mentioned wind turbine monitoring method is implemented.

[0015] Thirdly, this application provides a wind turbine monitoring system, including fiber optic sensors installed at the blade root, tower flange and anchor cable anchorage of the wind turbine, a multi-channel fiber optic demodulator connected to the fiber optic sensors, an edge computing device connected to the multi-channel fiber optic demodulator and a front-end platform. Fiber optic sensors are used to collect dynamic strain data of blades, displacement data of tower flanges, and dynamic stress data of anchor cables, respectively. A multi-channel fiber optic demodulator is used to demodulate the data collected by each fiber optic sensor to obtain sensor data, and then send the sensor data to the edge computing device. An edge computing device is used to receive sensor data sent by a multi-channel fiber optic demodulator; perform multi-physics coupled load inversion on the sensor data based on a preset wind turbine dynamics reduced-order model to obtain load inversion results; extract features from the sensor data to obtain feature information; perform wind turbine fault prediction based on a prediction model and feature information to obtain fault prediction results; if at least one of the load inversion results and fault prediction results is abnormal, decouple the load inversion results based on a preset fault transfer matrix to determine fault source information; determine first data based on sensor data, load inversion results, and fault source information, and send the first data to the front-end platform; wherein, the sensor data includes blade dynamic strain signals at the blade root, tower flange displacement signals, and anchor cable dynamic stress signals in the anchor cable anchorage zone; The front-end platform is used to perform 3D visualization rendering and status cloud mapping of the primary data, thereby completing the real-time visualization display of the wind turbine's operating status. Attached Figure Description

[0016] To more intuitively illustrate the prior art and this application, several exemplary figures are provided below. It should be understood that the specific shapes and structures shown in the figures should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary figures, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0017] Figure 1 A schematic diagram of the implementation process of the wind turbine monitoring method provided in this application embodiment. Figure 1 ; Figure 2 A schematic diagram of the implementation process of the wind turbine monitoring method provided in this application embodiment. Figure 2 ; Figure 3 A schematic diagram illustrating the implementation architecture of the wind turbine monitoring method provided in this application embodiment; Figure 4 A schematic diagram showing the deployment location of the fiber optic sensor in a wind turbine according to an embodiment of this application; Figure 5 A schematic diagram illustrating the implementation process of spatiotemporal synchronization and hybrid transmission provided in an embodiment of this application; Figure 6 A schematic diagram illustrating the display function of the front-end platform provided in an embodiment of this application; Figure 7This is a schematic diagram of the composition structure of the edge computing device proposed in this application.

[0018] Figure label: Edge computing device 1, processor 11, memory 12, communication interface 13, bus 14. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Any combination of different embodiments is possible.

[0020] In the description of this application: unless otherwise stated, "a plurality of" means two or more. The terms "first," "second," "third," etc., in this application are intended to distinguish the objects referred to and do not have any special meaning in terms of technical connotation (e.g., they should not be construed as an emphasis on importance or order). Expressions such as "including," "comprising," and "having" also mean "not limited to" (certain units, components, materials, steps, etc.).

[0021] Currently, resistance strain gauges are commonly used as sensing elements for wind turbine condition monitoring. However, these sensors suffer from significant drawbacks in practical applications, including poor electromagnetic interference resistance, short lifespan, severe temperature drift, and complex wiring, making it difficult to meet the monitoring requirements for the designed lifespan of wind turbines. Furthermore, existing monitoring systems are mostly based on a "one machine, one measurement" model, meaning one set of equipment monitors only a single object. This leads to data silos, redundant investment, and complex installation, hindering multi-parameter data fusion analysis and making it difficult to establish coupled damage models between components.

[0022] In recent years, fiber Bragg grating sensors have been introduced into the field of structural monitoring due to their advantages such as resistance to electromagnetic interference and high accuracy. However, existing general-purpose fiber optic monitoring solutions still face the following specific technical challenges when directly applied to wind turbine monitoring: 1. Dynamic monitoring challenges of rotating components: Blades are hundreds of meters long and rotate, traditional fiber optic rotary connectors suffer from signal attenuation, and dynamic strain is difficult to decouple from composite loads such as centrifugal force, gravity, and wind load, making it impossible to achieve high-fidelity monitoring of the entire circumference of the blade; 2. Lack of multi-physics coupling analysis: General solutions only perform independent anomaly detection and lack coupling analysis capabilities based on the physical model of the unit, making it impossible to answer damage tracing questions such as "is the flange gap abnormality caused by blade vibration or anchor cable loosening," making it difficult to guide precise operation and maintenance; 3. The contradiction between real-time performance and diagnostic depth: Continuously transmitting high-fidelity raw data will exhaust bandwidth; transmitting only feature values ​​cannot meet the needs of root cause analysis of faults.

[0023] To address the specific technical challenges of current general-purpose fiber optic monitoring solutions when applied to wind turbines, such as the inability to handle dynamic monitoring of rotating components, multi-physics coupling analysis, and the trade-off between real-time performance and diagnostic depth, this application provides a wind turbine monitoring method. Figure 1 As shown, the wind turbine monitoring method for edge computing devices may include the following steps: Step 101: Receive sensor data sent by the multi-channel fiber optic demodulator; wherein, the sensor data includes the dynamic strain signal of the blade at the blade root, the displacement signal of the tower flange, and the dynamic stress signal of the anchor cable in the anchoring zone.

[0024] In the embodiments of this application, the edge computing device can receive sensor data sent by a multi-channel fiber optic demodulator; wherein, the sensor data includes blade dynamic strain signals at the blade root, tower flange displacement signals, and anchor cable dynamic stress signals in the anchor cable anchorage zone.

[0025] In the embodiments of this application, the edge computing device is a hardware processing unit with an embedded FPGA and equipped with a preset wind turbine dynamics reduction model and prediction model, which is responsible for real-time processing and analysis of local data.

[0026] In the embodiments of this application, the multi-channel fiber optic demodulator is a device that converts the optical signals collected by the fiber optic sensor into digital signals and supports multi-channel parallel acquisition.

[0027] In the embodiments of this application, the sensor data includes blade dynamic strain signals, tower flange displacement signals, and anchor cable dynamic stress signals, which respectively reflect the real-time operating status of the blade, tower, and anchor cable.

[0028] In the embodiments of this application, the edge computing device can acquire the optical power fluctuation value of the multi-channel fiber demodulator during data transmission; calibrate the dynamic strain signal of the blade based on the optical power fluctuation value to obtain the calibrated blade strain signal; and determine the sensor data based on the calibrated blade strain signal, the tower flange displacement signal, and the anchor cable dynamic stress signal.

[0029] In the embodiments of this application, the optical power fluctuation value refers to the change in optical power of the optical fiber transmission signal caused by the rotation of the blade during the rotation process.

[0030] In the embodiments of this application, calibration refers to correcting the dynamic strain signal of the blade by adjusting the optical power fluctuation value, thereby eliminating signal errors caused by rotation.

[0031] Understandably, edge computing devices collect real-time optical power fluctuation values ​​transmitted through optical fibers, use this as a basis to correct the dynamic strain signal of the blade, and integrate the calibrated signal with other signals into standard sensor data. This can eliminate the signal distortion problem caused by rotation, improve the acquisition accuracy of the blade strain signal, and ensure the accuracy of subsequent load inversion.

[0032] Step 102: Perform multi-physics coupled load inversion on the sensor data based on the preset wind turbine dynamics reduced-order model to obtain the load inversion results.

[0033] In the embodiments of this application, after receiving sensor data sent by a multi-channel fiber demodulator, the edge computing device can perform multi-physics coupling load inversion on the sensor data based on a preset wind turbine dynamics reduced-order model to obtain the load inversion result.

[0034] In the embodiments of this application, multiphysics coupled load inversion refers to the process of converting sensor signals into real physical loads of blades, towers, and anchor cables based on a preset wind turbine dynamics reduced-order model.

[0035] In the embodiments of this application, the preset wind turbine dynamics reduced-order model is based on the simplified wind turbine physical model after finite element analysis, which is used to quickly realize load conversion.

[0036] In some embodiments of this application, when the edge computing device performs multiphysics coupling load inversion on sensor data based on a preset wind turbine dynamics reduced-order model to obtain load inversion results, it can convert the calibrated blade strain signal, tower flange displacement signal, and anchor cable dynamic stress signal in the sensor data into real physical loads based on the preset wind turbine dynamics reduced-order model, respectively, to obtain the real-time blade stress load, tower flange bending moment load, and anchor cable tension load; and determine the load inversion results based on the real-time blade stress load, tower flange bending moment load, and anchor cable tension load.

[0037] In the embodiments of this application, the real-time force load on the blade can be understood as the combined loads such as aerodynamic and centrifugal loads borne by the blade during operation.

[0038] In the embodiments of this application, the tower flange bending moment load can be understood as the bending load borne at the tower flange.

[0039] In the embodiments of this application, the anchor cable tension load can be understood as the tensile load borne by the anchor cable.

[0040] In this embodiment of the application, the three types of calibrated sensor signals are converted into physical loads of the corresponding components through a dynamic order reduction model, and integrated to form a complete load inversion result. This quantifies the stress state of each key component of the unit, thereby transforming abstract sensor signals into intuitive physical loads, accurately reflecting the stress state of key components of the unit, and providing a quantitative basis for fault diagnosis.

[0041] Step 103: Extract features from the sensor data to obtain feature information.

[0042] In the embodiments of this application, after receiving sensor data sent by a multi-channel fiber optic demodulator, the edge computing device can extract features from the sensor data to obtain feature information.

[0043] In some embodiments of this application, when the edge computing device extracts features from sensor data to obtain feature information, it can perform wavelet transform on the sensor data to obtain a multi-scale time-frequency decomposition signal; extract the signal peak-to-peak value, spectral energy, and frequency band amplitude of the multi-scale time-frequency decomposition signal; and determine the feature information based on the signal peak-to-peak value, spectral energy, and frequency band amplitude.

[0044] In the embodiments of this application, wavelet transform is a signal processing algorithm that performs time-frequency decomposition on sensor data and can be used to extract multi-scale features.

[0045] In the embodiments of this application, the peak-to-peak signal refers to the difference between the maximum and minimum values ​​of the signal, reflecting the amplitude of signal fluctuation.

[0046] In the embodiments of this application, spectral energy refers to the energy distribution of a signal in the frequency domain, characterizing the frequency features of the signal.

[0047] In the embodiments of this application, the frequency band amplitude refers to the signal amplitude within a specific frequency band, reflecting the vibration characteristics of the component.

[0048] In the embodiments of this application, wavelet transform is performed on sensor data to obtain time-frequency decomposition signals, and peak-to-peak value, spectral energy, and frequency band amplitude are extracted as feature information. This can efficiently extract core features of the data, reduce redundant data, provide accurate input for subsequent fault prediction, and improve prediction efficiency.

[0049] Step 104: Based on the prediction model and feature information, perform fault prediction for wind turbine units and obtain fault prediction results.

[0050] In the embodiments of this application, after the edge computing device extracts features from the sensor data and obtains feature information, it can perform wind turbine fault prediction based on the prediction model and the feature information to obtain fault prediction results.

[0051] In the embodiments of this application, fault prediction is to predict the future abnormal state of the wind turbine through a prediction model.

[0052] In the embodiments of this application, the prediction model is a model trained based on a Long Short-Term Memory (LSTM) neural network, which can be used to predict anomalies in wind turbines over a future period of time.

[0053] In some embodiments of this application, when the edge computing device performs wind turbine fault prediction based on a prediction model and feature information and obtains the fault prediction result, it can predict abnormal situations of the wind turbine within a preset time period based on the prediction model and feature information to obtain anomaly prediction probability values. These anomaly prediction probability values ​​include the probability values ​​of various abnormal situations such as resonance events, excessive loads, abnormal flange gaps, and anchor cable slack. A resonance event represents an abnormal situation where blade rotation causes turbine resonance; excessive loads represent an abnormal situation where the loads on the blades, tower, and anchor cables exceed their respective thresholds; abnormal flange gaps represent an abnormal situation where the flange gap does not conform to a preset gap range; and anchor cable slack represents an abnormal situation where insufficient anchor cable tension leads to anchoring failure. The fault prediction result is determined based on the anomaly prediction probability values.

[0054] In the embodiments of this application, the anomaly prediction probability value is a numerical value that characterizes the likelihood of various anomalies occurring; the higher the probability, the greater the anomaly risk.

[0055] In the embodiments of this application, a resonance event can be understood as a severe vibration anomaly caused by the blade rotation frequency coinciding with the unit's natural frequency.

[0056] In the embodiments of this application, overload can be understood as the state in which the load of a component exceeds its corresponding preset safety threshold.

[0057] In the embodiments of this application, flange clearance abnormality can be understood as a fault state in which the flange clearance exceeds the standard preset clearance range.

[0058] In the embodiments of this application, by inputting feature information into the prediction model, the model outputs probability values ​​for four types of anomalies: resonance, excessive load, abnormal flange gap, and loose anchor cable. Based on the probability values, the model generates fault prediction results, enabling early fault prediction, early identification of potential unit faults, allowing time for operation and maintenance, preventing the escalation of faults, and reducing unit downtime losses.

[0059] Step 105: If at least one of the load inversion results and fault prediction results is abnormal, the load inversion results are decoupled based on the preset fault transmission matrix to determine the fault source information.

[0060] In the embodiments of this application, after the edge computing device performs multi-physics coupling load inversion on sensor data based on a preset wind turbine dynamics reduced-order model to obtain load inversion results, and performs wind turbine fault prediction based on prediction models and feature information to obtain fault prediction results, it can perform decoupling calculation on the load inversion results based on a preset fault transmission matrix to determine fault source information if at least one of the load inversion results and fault prediction results is abnormal.

[0061] In the embodiments of this application, the preset fault transmission matrix is ​​a standardized mathematical matrix pre-constructed offline based on the reduced-order dynamic model of the wind turbine digital twin. Its core function is to establish a quantitative coupling mapping relationship between the fault source and the multi-physics load anomaly of the unit, providing a core theoretical basis for the accurate determination of fault source information. Its implementation logic is as follows: First, for the wind turbine, based on its reduced-order dynamic model (i.e., the preset reduced-order dynamic model of the wind turbine), all potential fault sources (including blade 3P vibration, tower flange loosening, anchor cable slack, etc.) are simulated and analyzed one by one. The influence weight of each type of fault source on the three core monitoring parameters of real-time blade stress load, tower flange bending moment load, and anchor cable tension load when acting alone is quantified. Then, the influence weight of each fault source on different monitoring parameters is organized into a standardized matrix according to preset rules to form the preset fault transmission matrix. This matrix can indicate the positive correspondence between the fault source action and the load anomaly response, and it is completely matched with the structural characteristics and mechanical laws of the wind turbine.

[0062] Once the edge computing device obtains the actual load anomaly state of the current unit (i.e., the load inversion result) through multiphysics coupling load inversion, it substitutes this load anomaly state as an input vector into the preset fault transmission matrix. Through matrix decoupling calculation (i.e., linear inverse mapping solution), the contribution ratio of each type of potential fault source to the current load anomaly can be derived in reverse. Since the preset fault transmission matrix has pre-quantified the independent influence weight of each fault source, the decoupling calculation result can accurately distinguish the degree of influence of different fault sources on the current anomaly. Finally, the fault source with the highest contribution ratio is selected as the root cause of the current anomaly, thereby achieving accurate determination of fault source information.

[0063] In embodiments of this application, fault source information can indicate the cause of the fault that caused the anomaly, including the core faulty component and the fault type.

[0064] In some embodiments of this application, when the edge computing device performs decoupling calculations on the load inversion results based on a preset fault transmission matrix to determine the fault source information, it can input the load inversion results into the preset fault transmission matrix for decoupling calculations to obtain the contribution ratio of blade load anomaly, tower flange loosening, and anchor cable slack to the current anomaly; and determine the fault source information based on the contribution ratio.

[0065] In the embodiments of this application, the contribution percentage represents the percentage of influence of each fault source on the current anomaly.

[0066] For example, for the current flange abnormal fluctuation (current anomaly), there is a 78% probability (contribution percentage) that it is caused by abnormal blade load, and a 22% probability (contribution percentage) that it is caused by anchor cable slack.

[0067] In some embodiments of this application, when determining fault source information based on contribution percentage, the fault source information can be determined based on the target contribution percentage that is greater than a preset percentage threshold. For example, if the preset percentage threshold is 5%, the target contribution percentage that is greater than 5% includes the contribution percentage of blade load anomaly to the current anomaly and the contribution percentage of anchor cable slack to the current anomaly. Then, the fault source information can be determined based on blade load anomaly and anchor cable slack. The core fault components include blades and anchor cables, and the fault types include blade load anomaly and anchor cable slack.

[0068] In some embodiments of this application, the edge computing device can determine that the load inversion result is abnormal if any load in the load inversion result is greater than the corresponding preset load threshold; and determine that the fault prediction result is abnormal if any abnormal prediction probability value in the fault prediction result is greater than the corresponding preset probability threshold.

[0069] In the embodiments of this application, the preset load threshold refers to the upper limit of the load for the safe operation of each component of the unit, which may include the blade stress load threshold, the tower flange bending moment load threshold, and the anchor cable tension load threshold.

[0070] In the embodiments of this application, the preset probability threshold refers to the critical value of probability for determining that the fault prediction result is abnormal.

[0071] In the embodiments of this application, the load inversion results are compared with the load threshold, and the fault prediction results are compared with the probability threshold. If any condition is met, the corresponding result is determined to be abnormal, thereby achieving standardized judgment of anomalies. This can avoid errors in human judgment and improve the objectivity and accuracy of anomaly identification.

[0072] Step 106: Determine the first data based on sensor data, load inversion results and fault source information, and send the first data to the front-end platform so that the front-end platform can perform three-dimensional visualization rendering and status cloud map mapping on the first data to complete the real-time visualization display of the wind turbine's operating status.

[0073] In the embodiments of this application, when at least one of the load inversion results and fault prediction results is abnormal, the edge computing device performs decoupled calculation on the load inversion results based on a preset fault transmission matrix. After determining the fault source information, it can determine the first data based on sensor data, load inversion results and fault source information, and send the first data to the front-end platform so that the front-end platform can perform three-dimensional visualization rendering and state cloud map mapping on the first data to complete the real-time visualization display of the wind turbine's operating status.

[0074] In some embodiments of this application, the first data may include sensor data, load inversion results, and fault source information within a preset time period before and after the occurrence of a fault or anomaly; for example, if the preset time period is 10 seconds, the sensor data, load inversion results, and fault source information within 10 seconds before and after the occurrence of the anomaly can be packaged and compressed to obtain the first data.

[0075] In the embodiments of this application, data can be converted into visual content such as stress cloud diagrams and state curves of the three-dimensional model of the wind turbine through three-dimensional visualization rendering and state cloud map mapping.

[0076] In some embodiments of this application, 3D visualization rendering and state cloud map mapping can be based on WebG technology to map data to the 3D model of the wind turbine and generate visualization content such as stress cloud map, state curve, and early warning pop-up.

[0077] Understandably, the edge computing device first receives three types of sensor data transmitted by the fiber optic demodulator, completes load inversion through a preset dynamic order reduction model, extracts data features and inputs them into the prediction model to obtain fault prediction results; if the load or prediction results are abnormal, the fault source information is obtained by decoupling through the fault transmission matrix, and finally the data is integrated and sent to the front end for display.

[0078] In some embodiments of this application, the front-end platform, based on a 3D digital twin platform built with WebGL, performs three-dimensional visualization rendering and state cloud map mapping on the first data, including at least one of the following: mapping the real-time stress load of the blade to the three-dimensional model of the rotating blade, generating a three-dimensional dynamic stress distribution cloud map with stress values ​​distinguished by color, and generating a locally magnified stress distribution heat map; converting the tower flange bending moment load into a real-time trend map of flange gap change curve; displaying fault source information in the form of a damage tracing result pop-up window (e.g., labeled "Flange abnormality: 78% blade 3P vibration, 22% anchor cable micro-slack"); and displaying fault prediction results in the form of a warning information bar (e.g., labeled "Resonance warning: probability 92%").

[0079] Furthermore, such as Figure 2As shown, after the edge computing device performs multi-physics coupled load inversion on sensor data based on a preset wind turbine dynamics reduced-order model to obtain the load inversion result, and performs wind turbine fault prediction based on the prediction model and feature information to obtain the fault prediction result, the wind turbine monitoring method may further include the following steps: Step 107: If there are no abnormalities in the load inversion results and fault prediction results, determine the second data based on the real-time azimuth angle signal and characteristic information of the wind turbine blade encoder.

[0080] In the embodiments of this application, after the edge computing device performs multi-physics coupling load inversion on sensor data based on a preset wind turbine dynamics reduced-order model to obtain load inversion results, and performs wind turbine fault prediction based on prediction models and feature information to obtain fault prediction results, it can also determine second data based on the real-time azimuth angle signal and feature information of the wind turbine blade encoder if there are no abnormalities in the load inversion results and fault prediction results.

[0081] In the embodiments of this application, the real-time azimuth angle signal of the blade encoder represents the current rotation angle of the blade and is used for data spatiotemporal synchronization.

[0082] In the embodiments of this application, the second data represents lightweight feature data when there are no anomalies, including azimuth angle and feature information.

[0083] Step 108: Send the second data to the front-end platform for data display.

[0084] In the embodiments of this application, when there are no abnormalities in the load inversion results and fault prediction results, the edge computing device can send the second data to the front-end platform for data display after determining the second data based on the real-time azimuth angle signal and feature information of the wind turbine blade encoder.

[0085] Understandably, when there are no abnormalities in the wind turbine, the edge computing device integrates the azimuth signal and feature information to form second data and transmits it to the front end, realizing low-bandwidth and high-efficiency data display under normal conditions. It can balance low power consumption in normal monitoring and full data retention in fault conditions, optimize data transmission efficiency, and adapt to remote monitoring scenarios of wind farms.

[0086] In the embodiments of this application, sending the second data to the front-end platform for data display means sending the second data to the front-end platform so that the front-end platform, based on the WebGL 3D digital twin platform, can perform normal visualization rendering of the second data.

[0087] In some embodiments of this application, the front-end platform is based on a WebGL-based 3D digital twin platform. Combined with the real-time azimuth signal of the blade encoder, the feature information is mapped to the three-dimensional model of the rotating blade to generate a three-dimensional dynamic stress distribution cloud map under normal conditions (with real-time stress values ​​mapped by color) and a locally magnified stress distribution heat map. The real-time trend map of the flange gap change curve under stable conditions is displayed simultaneously. Since there are no abnormalities in the unit, the damage tracing result pop-up window and the warning information bar are not displayed. Only a lightweight visualization display of the normal operating status of the wind turbine is completed.

[0088] Based on the above embodiments, in another embodiment of this application, to overcome the specific technical difficulties that current general fiber optic monitoring solutions cannot solve when applied to wind turbines, such as dynamic monitoring of rotating components, multi-physics field coupling analysis, and the contradiction between real-time performance and diagnostic depth, a wind turbine multi-parameter coupling monitoring and damage tracing system (wind turbine monitoring system) and method based on digital twins are provided. The wind turbine multi-parameter coupling monitoring and damage tracing system based on digital twins includes: The fiber optic sensor network is deployed at the blade root, tower flange, and anchor cable anchorage area of ​​the wind turbine to synchronously collect data on dynamic loads on the blades, anchor cable stress, and micro-changes in flange gaps. The fiber optic sensors deployed at the blade root are connected to the demodulator in the nacelle via fiber optic rotary connectors.

[0089] A multi-channel fiber optic demodulator, connected to a fiber optic sensor network, is used to convert optical signals into digital signals. It supports multi-channel parallel acquisition, with a sampling rate ≥500Hz and a strain resolution ≤1με.

[0090] The edge computing device, connected to a multi-channel fiber optic demodulator, incorporates an FPGA and is pre-configured with a wind turbine's overall structural dynamics model based on a reduced-order finite element analysis model. The edge computing device may include: a dynamic optical attenuation compensation module for real-time monitoring and calibration of signal optical power fluctuations caused by blade rotation; a coupled field inversion module for real-time reception of multi-parameter sensor data and outputting load inversion results in conjunction with the structural dynamics model (a pre-configured reduced-order wind turbine dynamics model); a damage tracing engine that, based on a pre-configured fault transmission matrix, calculates the probability contribution of each potential fault source in real-time when abnormal parameters are detected; and a two-stage processing engine: the first-stage processing engine runs a high-speed wavelet transform algorithm in the FPGA to extract the feature values ​​of the vibration signal in real-time; the second-stage processing engine runs an LSTM neural network prediction model to predict future trends based on the feature value sequence.

[0091] The spatiotemporal synchronization and hybrid transmission module is connected to the unit's main control system, receives blade encoder signals in real time, tags the collected data with the real-time azimuth angle of the blades, and adopts a hybrid transmission strategy: under normal conditions, it continuously transmits feature value data extracted by the edge computing unit, and triggers the transmission of original data snapshots when an abnormal event is predicted.

[0092] The front-end 3D digital twin visualization platform has a built-in unit structure dynamics model consistent with the edge computing unit. It is used to receive and render data, display the three-dimensional dynamic stress distribution cloud map and damage tracing results of the rotating blades in real time, and provide early warning information.

[0093] Furthermore, the fiber optic sensing network includes: FBG strain sensors deployed at the root of the blade along the principal stress direction, which are connected to fiber optic rotary connectors via armored optical cables and then introduced into a demodulator at the base of the tower or in the nacelle; micro-displacement fiber optic sensors deployed at the tower flange connection to monitor changes in flange gap with a measurement accuracy of ±10μm; and distributed fiber optic sensors deployed in the anchor cable anchorage area to monitor the static and dynamic stress of the anchor cables. The fiber optic sensors are encapsulated in titanium alloy, weighing ≤500g, with a curvature radius adaptability of ≤5cm, adapting to the curved surface of the blade. A dynamic optical attenuation compensation algorithm monitors the signal optical power fluctuations caused by rotation in real time and performs dynamic calibration to ensure that the static strain accuracy of the data collected during the blade's 0~360° rotation is ≤1με and the dynamic frequency accuracy is ≤0.1Hz.

[0094] Furthermore, the FPGA embedded in the edge computing device supports signal processing with ≥16 channels and up to 32 channels.

[0095] Furthermore, the vibration signal feature values ​​extracted by the first-level processing engine include signal peak-to-peak value, spectral energy, and amplitude in a specific frequency band. The LSTM neural network prediction model run by the second-level processing engine predicts resonance events that may occur within the next 30 minutes based on the feature value sequence, and serves as the basis for triggering the transmission of the original data snapshot.

[0096] Furthermore, based on a preset fault transfer function matrix, the damage tracing engine calculates the coupling relationship between multiple parameters such as blade load, flange clearance, and anchor cable stress in real time when abnormal parameters are detected, outputs the probability contribution of each potential fault source, and sorts them.

[0097] Furthermore, the spatiotemporal synchronization and hybrid transmission module adopts the following hybrid transmission strategy: Under normal circumstances, the feature value data extracted by the edge computing device is transmitted to the front-end platform in parallel via multi-threaded / asynchronous requests at a fixed frequency of 1Hz for continuous monitoring; when the abnormal probability output by the LSTM prediction model exceeds the preset threshold, or when a transient impact event is detected, the original high-fidelity data for 10 seconds before and after the event is rapidly compressed and urgently uploaded through a high-priority channel for in-depth source tracing analysis.

[0098] Furthermore, the front-end 3D digital twin visualization platform is built on WebGL technology and incorporates a wind turbine structural dynamics model consistent with the edge computing device. After receiving data with azimuth labels, it maps strain data from different angles to the corresponding spatial positions of the 3D model, generating a three-dimensional dynamic stress distribution cloud map of the rotating blades under the combined action of gravity, wind load, and centrifugal force in real time. When the system issues an early warning, it simultaneously displays the damage tracing results, including the probability contribution of each fault source and corresponding operation and maintenance suggestions.

[0099] Furthermore, the system supports local terabyte-level data storage and playback, and is compatible with MODBUS, MQTT, and UDP protocols. The system supports hybrid networking of RS485 / fiber optic ring networks, and a single system can support up to 250 measurement points.

[0100] For example, such as Figure 3As shown, three types of fiber optic sensors are deployed at the wind turbine site. The blade root FBG sensor is connected to the fiber optic sensor network via a fiber optic rotary connector (solving the signal transmission problem for rotating blades). The tower flange micro-displacement sensor and the anchor cable anchorage zone distributed sensor are directly connected, forming a full fiber optic sensor network covering the blades, tower, and anchor cables. This network is used to collect three types of raw optical signals: blade dynamic strain, tower flange micro-displacement, and anchor cable dynamic stress. For the signal demodulation layer, the fiber optic sensor network transmits the collected optical signals to a multi-channel fiber optic demodulator. This demodulator has a sampling rate ≥500Hz and a strain resolution ≤1με, used to demodulate the optical signals into standard... The system converts digital sensor data and transmits it to the edge computing device. The edge computing device is the system's local core processing unit, containing four core functional modules: a dynamic light attenuation compensation module for real-time calibration of light power fluctuations caused by blade rotation, ensuring the accuracy of blade strain signal acquisition; a coupled field inversion module (digital twin model solver) for multi-physics coupled load inversion based on a pre-set wind turbine dynamics reduced-order model, converting sensor signals into real physical loads on the blades, tower, and anchor cables; a damage tracing engine (fault transfer function matrix) for decoupling calculations of abnormal loads and quantifying the contribution ratio of each fault source; and a two-level processing engine (wavelet transform + ...). The system employs an LSTM prediction mechanism. It extracts time-domain and frequency-domain features from the sensor signals using wavelet transform, and then uses an LSTM neural network model to predict future fault trends in the wind turbine. For the spatiotemporal synchronization and transmission layer, the data processed by the edge computing device is input to the spatiotemporal synchronization and hybrid transmission module. This module connects to the wind turbine encoder signal, labeling all data with blade azimuth angles to achieve spatiotemporal synchronization between sensor data and blade rotation position. Simultaneously, a hybrid transmission strategy is implemented: only lightweight feature value data is transmitted during normal turbine operation, while a snapshot of the original data is uploaded when an abnormal event is triggered, balancing data transmission efficiency and fault diagnosis depth. For the visualization layer, the data after hybrid transmission is input to the front-end 3D digital twin visualization platform to generate a three-dimensional dynamic stress cloud map of the rotating blades, display damage tracing analysis results, and push fault warning information, achieving full-dimensional visualization monitoring of the wind turbine's entire lifecycle operation status.

[0101] For example, such as Figure 4 As shown, the fiber optic sensors include FBG sensors deployed at the blade root, micro-displacement fiber optic sensors at the tower flange connection, and distributed fiber optic sensors in the anchor cable anchorage area. All of these sensors are connected to the fiber optic demodulator at the bottom of the tower. The FBG sensors at the blade root are connected to the fiber optic demodulator at the bottom of the tower via fiber optic rotary connectors.

[0102] For example, such as Figure 5As shown, the spatiotemporal synchronization and hybrid transmission based on the blade's spatial position involves a complete process from raw data acquisition, azimuth labeling, edge feature extraction, normal feature transmission, event triggering, to raw data snapshot uploading: First, multi-channel raw data acquisition is performed, simultaneously acquiring raw sensor data from three types of fiber optic sensors: the root of the wind turbine blade, the tower flange, and the anchor cable anchorage area; then, the wind turbine encoder signal is input to obtain the real-time azimuth angle of the blade, and each frame of acquired sensor data is labeled with the corresponding azimuth angle to form a spatiotemporal synchronization data packet, completing the precise spatiotemporal synchronization of sensor data and blade rotation position, providing a unified benchmark for subsequent processing; the spatiotemporal synchronization data packet is input into the edge computing device, where wavelet transform is first performed through the FPGA to extract features from the sensor data. The system generates a feature value sequence. This sequence is processed in two parallel paths: one path enters the normal transmission preparation process, while the other path inputs the LSTM prediction model to calculate the probability of resonance or abnormal events occurring in the wind turbine. Based on the LSTM prediction results, anomaly detection is performed to determine if the anomaly probability exceeds a preset threshold or if an impact event is detected. If the detection result is no anomaly or impact, normal transmission is executed: the feature value data is transmitted to the front-end platform at a frequency of 1Hz through the normal transmission channel. The front-end platform enables real-time monitoring of the wind turbine's operating status, ensuring a first-screen display delay of ≤0.05s, meeting the high real-time requirements of normal monitoring. If the detection result is an anomaly probability exceeding the standard or an impact event is detected, the event transmission process is triggered: first, 10 seconds of raw high-fidelity data are extracted before and after the event occurs, and the data is quickly compressed and uploaded to the front-end platform with high priority. The front-end platform performs in-depth analysis and damage tracing based on the complete raw data to accurately locate the root cause of the fault.

[0103] For example, such as Figure 6As shown, the front-end 3D digital twin platform can display a 3D dynamic stress distribution cloud map, stress distribution heat map, flange gap change curve, and damage tracing result pop-up window of the rotating blade. All visualization content is generated based on data processed by the edge computing device and integrated, rendered, and displayed using a WebGL-based 3D digital twin platform. The 3D model of the rotating blade is a 1:1 digital twin model of the wind turbine blade. Based on the blade load inversion results output by the edge computing device, two types of visualization content are generated: a dynamic stress distribution cloud map and a stress distribution heat map (local magnification). The dynamic stress distribution cloud map uses different colors to intuitively map the real-time stress values ​​at various locations on the blade, realizing the visualization of the 3D dynamic stress distribution during blade rotation. The stress distribution heat map magnifies the key stress-bearing parts of the blade locally, displaying the stress distribution details in detail, which is convenient for maintenance personnel to accurately grasp the local stress state of the blade. The real-time trend chart of the flange gap change curve is generated based on the tower flange load inversion results and gap monitoring data output by the edge computing device, displaying the dynamic change trend of the tower flange gap in real time, intuitively presenting the operating status of the flange, and timely identifying abnormal gap fluctuations. Damage Origin Result Pop-up: Generated based on fault source information output by the edge computing device, this pop-up quantifies the contribution percentage of each fault source causing the current anomaly. For example, the example in the image shows "Flange Anomaly: 78% Blade 3P Vibration, 22% Anchor Cable Micro-Slack," intuitively presenting the influence weight of the root cause of the fault and enabling visualized fault tracing. Warning Information Bar: Generated based on fault prediction results output by the edge computing device, this bar displays potential fault warning information for the wind turbine. For example, the example in the image shows "Resonance Warning: Probability 92%," proactively informing maintenance personnel of the unit's abnormal risks and enabling proactive fault warnings.

[0104] For example, this system is deployed on a 5MW onshore wind turbine in a wind farm. Three FBG strain sensors are installed at the blade root along the principal stress direction. The sensors are connected to the fiber optic rotary connector in the nacelle via armored optical cables, and then introduced to the demodulator at the bottom of the tower. Two micro-displacement fiber optic sensors are installed at the tower flange connection to monitor the flange gap change with a measurement accuracy of ±10μm. Distributed fiber optic sensors are installed in the anchor cable anchorage area to monitor the static and dynamic stress of the anchor cable. The fiber optic demodulator (32 channels, 500Hz sampling rate) is installed at the bottom of the tower and connected to the sensors via armored optical cables. The edge computing device (integrated FPGA, with a built-in digital twin model of the unit) is integrated with the demodulator and connected to the unit's main control system to acquire blade encoder signals in real time. Multi-channel data is acquired in real time and synchronously marked with blade azimuth information; the edge computing device continuously extracts feature values ​​(1Hz) and transmits them to the remote monitoring center; during operation, the LSTM model predicts that the flange gap will reach the resonance threshold after 30 minutes, and the system automatically triggers the transmission of raw data snapshots, compressing and uploading high-fidelity data 10 seconds before and after the event. The front-end platform displays a real-time three-dimensional dynamic stress cloud map of the blade, clearly showing the changes in stress distribution during one rotation of the blade; the system issues a resonance warning and simultaneously displays the damage tracing results: "The current abnormal flange fluctuation has a 78% probability of being caused by the blade's 3P vibration, and a 22% probability of being caused by the slight relaxation of anchor cable No. 3." Based on the tracing results, the maintenance personnel prioritized adjusting the preload of anchor cable No. 3, successfully avoiding a potential downtime failure.

[0105] For example, this system was deployed on an 8MW offshore wind turbine in a certain offshore wind farm, with the following optimizations made for the marine environment: the sensor packaging was upgraded to IP68 protection level, using a titanium alloy shell resistant to salt spray corrosion; the fiber optic demodulator was built into the nacelle and communicated with the edge computing unit at the bottom of the tower through a fiber optic ring network; the dynamic optical attenuation compensation algorithm was optimized for the larger swing amplitude of the offshore turbine; and a wave load coupling analysis module was added to the front-end platform to realize the coupled display of aerodynamic-structural-wave multiphysics fields. After deployment, the system ran continuously for 6 months, successfully identifying 3 abnormal blade load patterns, and accurately distinguishing two overloads caused by gusts and one stress concentration caused by internal blade defects through the damage tracing function, verifying the reliability and advanced nature of the system in harsh environments.

[0106] In summary, this application, by integrating all-fiber sensing, edge computing, digital twin, and intelligent diagnostic technologies, achieves a comprehensive technological upgrade compared to existing wind turbine monitoring solutions, offering high precision, high reliability, intelligence, and low bandwidth consumption. The core technological advantages are as follows: The use of all-fiber sensing combined with dynamic optical attenuation compensation calibration effectively overcomes the signal attenuation and distortion problems caused by blade rotation, achieving high-fidelity monitoring of the rotating blades in the entire circumference. The sensor is resistant to electromagnetic interference, adaptable to extreme environments, and its service life matches that of wind turbines (20 years). With a design lifespan of 10 years, it completely solves the shortcomings of traditional sensors, such as short lifespan and severe temperature drift. Based on a reduced-order model of wind turbine dynamics, it achieves multi-physics coupled load inversion, transforming abstract sensor signals into intuitive physical load data, accurately quantifying the actual stress state of blades, towers, and anchor cables, and providing quantitative basis for fault diagnosis. By combining wavelet transform feature extraction with LSTM neural network prediction, it can predict potential faults such as unit resonance, overload, and abnormal flange clearance in advance, achieving proactive early warning and significantly reducing the risk of unit downtime and operation and maintenance costs. Relying on a preset fault transmission matrix to complete fault source decoupling calculation, it achieves accurate fault tracing under multi-parameter coupled conditions, directly locating the root cause of the fault, and upgrading the traditional "phenomenon alarm" to "cause diagnosis". The system provides precise maintenance guidance for operators; it employs a hybrid transmission strategy that combines normal transmission of characteristic values ​​with abnormal triggering of original data snapshots, balancing real-time monitoring with in-depth fault diagnosis, significantly reducing network bandwidth and central computing resource pressure, and resolving the efficiency bottleneck of traditional transmission modes; based on a 3D digital twin platform, it completes three-dimensional visualization rendering and stress cloud mapping, intuitively displaying the unit's operating status, stress distribution, and fault tracing results, with clear and intuitive visualization, greatly improving the convenience of wind farm operation and maintenance and decision-making efficiency; it achieves synchronous integrated monitoring of multiple components and parameters, including blades, tower flanges, and anchor cables, breaking the traditional "one machine, one test" data silo problem, completing multi-component collaborative monitoring and coupled analysis, and fully adapting to the intelligent monitoring needs of large wind turbine units.

[0107] Based on the above embodiments, another embodiment of this application provides a wind turbine monitoring system, including fiber optic sensors installed at the blade root, tower flange and anchor cable anchorage area of ​​the wind turbine, a multi-channel fiber optic demodulator connected to the fiber optic sensors, an edge computing device connected to the multi-channel fiber optic demodulator, and a front-end platform. Fiber optic sensors are used to collect dynamic strain data of blades, displacement data of tower flanges, and dynamic stress data of anchor cables, respectively. A multi-channel fiber optic demodulator is used to demodulate the data collected by each fiber optic sensor to obtain sensor data, and then send the sensor data to the edge computing device. An edge computing device is used to receive sensor data sent by a multi-channel fiber optic demodulator; perform multi-physics coupled load inversion on the sensor data based on a preset wind turbine dynamics reduced-order model to obtain load inversion results; extract features from the sensor data to obtain feature information; perform wind turbine fault prediction based on a prediction model and feature information to obtain fault prediction results; if at least one of the load inversion results and fault prediction results is abnormal, decouple the load inversion results based on a preset fault transfer matrix to determine fault source information; determine first data based on sensor data, load inversion results, and fault source information, and send the first data to the front-end platform; wherein, the sensor data includes blade dynamic strain signals at the blade root, tower flange displacement signals, and anchor cable dynamic stress signals in the anchor cable anchorage zone; The front-end platform is used to perform 3D visualization rendering and status cloud mapping of the primary data, thereby completing the real-time visualization display of the wind turbine's operating status.

[0108] Based on the above embodiments, another embodiment of this application provides an edge computing device, such as... Figure 7 As shown, the edge computing device 1 proposed in this application embodiment may include a processor 11 and a memory 12 storing instructions executable by the processor 11; further, the edge computing device 1 may also include a communication interface 13 and a bus 14 for connecting the processor 11, the memory 12 and the communication interface 13.

[0109] In the embodiments of this application, the processor 11 can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that for different devices, the electronic device used to implement the above-mentioned processor function can also be other, and this application embodiment does not specifically limit it. The memory 12 can be connected to the processor 11, wherein the memory 12 is used to store executable program code, which includes computer operation instructions. The memory 12 may include high-speed RAM memory, and may also include non-volatile memory, such as at least two disk drives.

[0110] In the embodiments of this application, bus 14 is used to connect communication interface 13, processor 11 and memory 12 to enable communication between these devices.

[0111] In embodiments of this application, memory 12 is used to store instructions and data.

[0112] In practical applications, the aforementioned memory 12 can be volatile memory, such as random-access memory (RAM), or non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 11.

[0113] Furthermore, in this embodiment, the functional modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.

[0114] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment.

[0115] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0116] This application is described with reference to schematic diagrams and / or block diagrams of implementations of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each step and / or block in the schematic diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more steps of the schematic diagrams and / or one or more blocks in the block diagrams.

[0117] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in the implementation flow diagram. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0118] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0119] The above embodiments are merely preferred embodiments provided to fully illustrate this application, and the scope of protection of this application is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on this application are all within the scope of protection of this application.

Claims

1. A method for monitoring wind turbine generators, characterized in that, The method is applied to an edge computing device, which is connected to a multi-channel fiber optic demodulator of a wind turbine, and the multi-channel fiber optic demodulator is connected to fiber optic sensors located at the blade root, tower flange, and anchor cable anchorage area of ​​the wind turbine; the method includes: The system receives sensor data transmitted by the multi-channel fiber optic demodulator; wherein the sensor data includes the dynamic strain signal of the blade at the blade root, the displacement signal of the tower flange, and the dynamic stress signal of the anchor cable in the anchoring zone. Based on a preset wind turbine dynamics reduced-order model, the sensor data is subjected to multi-physics field coupled load inversion to obtain the load inversion result. Feature extraction is performed on the sensor data to obtain feature information; Based on the prediction model and the aforementioned feature information, wind turbine fault prediction is performed to obtain fault prediction results. If at least one of the load inversion results and the fault prediction results is abnormal, the load inversion results are decoupled and calculated based on a preset fault transmission matrix to determine the fault source information. Based on the sensor data, the load inversion results, and the fault source information, the first data is determined and sent to the front-end platform so that the front-end platform can perform three-dimensional visualization rendering and status cloud map mapping on the first data to complete the real-time visualization display of the wind turbine's operating status.

2. The wind turbine monitoring method according to claim 1, characterized in that, The method further includes: Obtain the optical power fluctuation value of the multi-channel fiber optic demodulator during data transmission; The dynamic strain signal of the blade is calibrated based on the optical power fluctuation value to obtain the calibrated blade strain signal. The sensor data is determined based on the calibrated blade strain signal, the tower flange displacement signal, and the anchor cable dynamic stress signal.

3. The wind turbine monitoring method according to claim 2, characterized in that, The multiphysics coupled load inversion of the sensor data based on the preset wind turbine dynamics reduced-order model yields the load inversion results, including: Based on the preset wind turbine dynamics reduced-order model, the calibrated blade strain signal, tower flange displacement signal, and anchor cable dynamic stress signal in the sensor data are converted into real physical loads to obtain the real-time blade stress load, tower flange bending moment load, and anchor cable tension load. The load inversion result is determined based on the real-time stress load on the blade, the bending moment load on the tower flange, and the tension load on the anchor cable.

4. The wind turbine monitoring method according to claim 1, characterized in that, The step of extracting features from the sensor data to obtain feature information includes: Wavelet transform is performed on the sensor data to obtain a multi-scale time-frequency decomposed signal; Extract the signal peak-to-peak value, spectral energy, and bandwidth amplitude of the multi-scale time-frequency decomposition signal; The characteristic information is determined based on the signal peak-to-peak value, the spectral energy, and the frequency band amplitude.

5. The wind turbine monitoring method according to claim 1, characterized in that, The method of predicting wind turbine faults based on the prediction model and the feature information, and obtaining fault prediction results, includes: Based on the prediction model and the feature information, abnormal situations of the wind turbine unit within a preset time period are predicted to obtain abnormal prediction probability values. These abnormal prediction probability values ​​include the probability values ​​for each abnormal situation: resonance event, excessive load, abnormal flange gap, and anchor cable loosening. The resonance event represents an abnormal situation where blade rotation causes unit resonance; the excessive load represents an abnormal situation where the load on the blade, tower, and anchor cable exceeds the corresponding threshold; the abnormal flange gap represents an abnormal situation where the flange gap does not conform to the preset gap range; and the anchor cable loosening represents an abnormal situation where insufficient anchor cable tension leads to anchoring failure. The fault prediction result is determined based on the anomaly prediction probability value.

6. The wind turbine monitoring method according to claim 1, characterized in that, The step of decoupling the load inversion results based on a preset fault transmission matrix to determine fault source information includes: The load inversion results are input into a preset fault transmission matrix for decoupling calculation to obtain the contribution ratio of blade load anomaly, tower flange loosening, and anchor cable slack to the current anomaly. The fault source information is determined based on the contribution ratio.

7. The wind turbine monitoring method according to claim 1, characterized in that, The method further includes: If any load in the load inversion results is greater than the corresponding preset load threshold, it is determined that the load inversion results are abnormal. If any abnormal prediction probability value in the fault prediction results is greater than the corresponding preset probability threshold, it is determined that the fault prediction results are abnormal.

8. The wind turbine monitoring method according to claim 1, characterized in that, The method further includes: If there are no abnormalities in the load inversion results and the fault prediction results, the second data is determined based on the real-time azimuth angle signal of the wind turbine blade encoder and the feature information. The second data is sent to the front-end platform for display.

9. An edge computing device, characterized in that, The edge computing device includes a processor and a memory storing executable instructions of the processor; when the executable instructions are executed by the processor, the wind turbine monitoring method as described in any one of claims 1 to 8 is implemented.

10. A wind turbine monitoring system, characterized in that, It includes fiber optic sensors installed at the blade root, tower flange, and anchor cable anchorage of the wind turbine, a multi-channel fiber optic demodulator connected to the fiber optic sensors, an edge computing device connected to the multi-channel fiber optic demodulator, and a front-end platform. The fiber optic sensors are used to collect dynamic strain data of the blades, displacement data of the tower flanges, and dynamic stress data of the anchor cables, respectively. The multi-channel fiber optic demodulator is used to demodulate the data collected by each of the fiber optic sensors to obtain sensor data, and then send the sensor data to the edge computing device. The edge computing device is used to receive sensor data sent by the multi-channel fiber optic demodulator; Based on a preset wind turbine dynamics reduced-order model, the sensor data is subjected to multi-physics field coupled load inversion to obtain the load inversion result. Feature extraction is performed on the sensor data to obtain feature information; Based on the prediction model and the aforementioned feature information, wind turbine fault prediction is performed to obtain fault prediction results. If at least one of the load inversion results and the fault prediction results is abnormal, the load inversion results are decoupled and calculated based on a preset fault transmission matrix to determine the fault source information; first data is determined based on the sensor data, the load inversion results, and the fault source information, and the first data is sent to the front-end platform; wherein, the sensor data includes the blade dynamic strain signal at the blade root, the tower flange displacement signal of the tower flange, and the anchor cable dynamic stress signal of the anchor cable anchorage zone; The front-end platform is used to perform three-dimensional visualization rendering and status cloud map mapping on the first data, so as to complete the real-time visualization display of the wind turbine's operating status.

Images