A fan blade state monitoring and deicing linkage control system and processing method

Abstract

The application discloses a kind of fan blade state monitoring and deicing linkage control system and processing method, it is related to wind power generation technical field, the linkage control system includes front-end sensing device, edge processing module, communication module, deicing linkage interface and remote platform, front-end sensing device integrates sensor, for collecting the original state data of fan blade;Edge processing module carries out multistage digital signal processing to original state data;Edge processing module carries four-dimensional extended Kalman filter, for estimating blade icing parameter, and built-in eight-dimensional nonlinear observation model, constructs the mapping relationship of sensing characteristic quantity and icing state parameter;Communication module is used for information transmission;Deicing linkage interface is set on the outer surface of front-end sensing device, for outputting deicing control signal and receiving deicing equipment state feedback;Remote platform is connected with communication module.The linkage control system can realize the adaptive accurate monitoring and linkage control of fan blade state.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a wind turbine blade condition monitoring and de-icing linkage control system and processing method. Background Technology

[0002] When wind turbines operate in cold, humid, mountainous, or high-altitude environments, the leading edge and windward surface of the blades are prone to icing. Blade icing alters the aerodynamic shape and mass distribution of the blades, deteriorating lift-drag characteristics, reducing power generation efficiency, increasing structural loads, and may cause abnormal vibrations, ice shedding, power-limited operation, or unplanned shutdowns.

[0003] Existing icing monitoring technologies are mainly divided into two categories: direct detection and indirect detection. Direct detection relies on electrical, optical, and acoustic methods to sense changes in ice layer characteristics; indirect detection combines environmental temperature and humidity, vibration data, power curves, speed, load, and SCADA operation data to inversely determine the icing state. For distributed wind farm applications, the front-end monitoring device must also be waterproof, condensation-proof, vibration-resistant, electromagnetic interference-resistant, and have remote communication capabilities; otherwise, even with excellent algorithm model performance, it will be difficult to guarantee long-term stable operation on site.

[0004] However, existing monitoring solutions generally suffer from weak engineering adaptability and poor system coordination. Filtering models with fixed physical parameters rely on manual calibration and cannot fully adapt to different blade materials, lengths, installation locations, and sensor coupling conditions; time-frequency data such as spectrum diagrams are often used as information displayed in the cloud without participating in edge detection; baseline calibration usually stops at the power-on averaging stage, lacking reliable runtime drift compensation; sampling rate, signal processing depth, reporting frequency, and filter sensitivity are usually controlled separately by decentralized conditional judgment logic, making it difficult to form a globally optimal configuration under power-constrained conditions.

[0005] Therefore, the current monitoring scheme suffers from technical problems such as insufficient adaptability and poor coordination, resulting in low detection accuracy. Summary of the Invention

[0006] The purpose of this invention is to overcome the above-mentioned technical deficiencies and propose a wind turbine blade condition monitoring and de-icing linkage control system and processing method to solve the technical problems of insufficient adaptability and poor coordination in the prior art, which leads to low detection accuracy.

[0007] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a wind turbine blade condition monitoring and de-icing linkage control system. This linkage control system includes a front-end sensing device, an edge processing module, a communication module, a de-icing linkage interface, and a remote platform. The front-end sensing device has a receiving cavity and integrates sensors for collecting raw state data of the wind turbine blades. The edge processing module is arranged within the receiving cavity and performs multi-level digital signal processing on the raw state data. The edge processing module is equipped with a four-dimensional extended Kalman filter for estimating blade icing parameters and incorporates an eight-dimensional nonlinear observation model to construct a mapping relationship between sensing features and icing state parameters. The communication module is arranged within the receiving cavity for information transmission. The de-icing linkage interface is located on the outer surface of the front-end sensing device for outputting de-icing control signals to the de-icing equipment and receiving status feedback from the de-icing equipment. The remote platform is communicatively connected to the communication module.

[0008] In some embodiments, the front-end sensing device has an integrated coaxial dual-layer cavity, which includes a main control inner cavity and a sensing outer cavity. The main control inner cavity is used to accommodate various functional modules, and the sensing outer cavity is used to install sensors.

[0009] In some embodiments, the front-end sensing device and the edge processing module transmit and process data signals through a sensing and edge processing link, and the combination of multiple types of sensing features output by the link is an eight-dimensional observation vector: ; Among them, T b For blade temperature; H a For ambient humidity; F p E is the dominant vibration frequency. L Low-frequency energy; A rms M is the root mean square value of the vibration; env R is the envelope modulation depth. axis The energy ratio is the three-axis ratio; D f It is the main frequency drift rate.

[0010] In some embodiments, the four-dimensional extended Kalman filter corrects the physical model parameters online while estimating the icing state, and the state vector is: ; Where, d k This is an estimate of the ice thickness; v k For ice thickness growth rate; Kf k Main frequency influence coefficient; Ke k This is the low-frequency energy influence coefficient.

[0011] In some embodiments, an eight-dimensional nonlinear observation model is used to construct the blade temperature T. b Ambient humidity H a Vibration dominant frequency F p Low-frequency energy E L, Root mean square value of vibration A rms Envelope modulation depth M env Triaxial energy ratio R axis and the main frequency drift rate D f The mapping relationship between the parameters and the freezing state parameters.

[0012] In some embodiments, the edge processing module further includes a main frequency drift rate extraction unit, which extracts the main frequency drift rate D through multi-frame spectrum data calculation. f : ; Where N is the total number of frames stored in the circular spectrum buffer; i is the frame number, f i Let be the peak frequency of the main frequency corresponding to the spectrum of the i-th frame.

[0013] In some embodiments, the edge processing module is further provided with a baseline management unit. The baseline management unit realizes baseline initialization, rapid recovery and dynamic calibration through a three-layer ice-free baseline self-learning and drift compensation mechanism. The three-layer ice-free baseline self-learning and drift compensation mechanism includes three parts in sequence: cold start learning, hot start recovery and runtime drift compensation.

[0014] In some embodiments, the edge processing module further includes a fusion estimation unit and a scheduling unit; the fusion estimation unit generates a risk index by combining ice thickness, ice growth rate, observation residual, main frequency drift rate and historical duration; the scheduling unit dynamically configures operating parameters based on the risk index to perform closed-loop coordinated control.

[0015] In some embodiments, the scheduling unit also integrates de-icing linkage and closed-loop feedback functions, and sets power consumption budget constraints; the power consumption budget constraints are used to evaluate system power consumption and battery life, and to adjust operating parameters in stages.

[0016] Secondly, this invention also provides a method for monitoring the state of wind turbine blades and for de-icing linkage control, applicable to the aforementioned linkage control system. This method includes: data acquisition: acquiring raw state data of the wind turbine blades through sensors in the front-end sensing device; signal preprocessing and feature extraction: filtering and performing multi-level digital signal processing on the raw data through an edge processing module to extract features, calculate the main frequency drift rate, and construct an eight-dimensional observation vector; adaptive state estimation: estimating icing state parameters online using a four-dimensional extended Kalman filter and an eight-dimensional nonlinear observation model mounted on the edge processing module; baseline maintenance: learning, restoring, and compensating for drift in the icing-free baseline through the edge processing module; risk assessment and global scheduling: calculating a risk index based on icing parameters through the edge processing module and dynamically adjusting equipment operating parameters; icing level determination: classifying icing levels according to the risk index through the edge processing module; de-icing linkage and closed-loop verification: outputting control signals and receiving state feedback through the de-icing linkage interface of the front-end sensing device for de-icing linkage and closed-loop verification.

[0017] Compared with existing technologies, the wind turbine blade condition monitoring and de-icing linkage control system provided by this invention adopts an integrated design, combining sensors, edge processing modules, and communication modules. This facilitates installation and allows for stable acquisition of blade operating data. The system utilizes an eight-dimensional nonlinear observation model and a four-dimensional extended Kalman filter to process data and estimate icing parameters, resulting in higher monitoring accuracy. Furthermore, an external de-icing linkage interface enables command output and status feedback, achieving closed-loop control for monitoring and de-icing. The communication module ensures bidirectional communication between local and remote platforms. The entire system effectively solves the problems of poor cross-wind turbine adaptability caused by fixed model parameters, large manual calibration workload, and long-term operational accuracy degradation, achieving adaptive and accurate monitoring and linkage control of wind turbine blade conditions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the framework structure of a wind turbine blade condition monitoring and de-icing linkage control system provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the front-end sensing device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the main control PCB board provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the state estimation and risk assessment results of the EKF algorithm provided in the embodiment of the present invention in the monitoring of icing on wind turbine blades; Figure 5 This is a schematic diagram illustrating the evolution trend of multi-sensor data of wind turbine blades under the "slow icing" condition, provided in an embodiment of the present invention. Figure 6This is a comparison chart of the Fast Fourier Transform (FFT) spectra of wind turbine blades at different icing stages provided in the embodiments of the present invention; Figure 7 This is a schematic diagram of the cepstrum process at an ice thickness of 3.5 mm provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the entire process of envelope analysis when the ice thickness is 3mm, provided in an embodiment of the present invention. Figure 9 This is a time-frequency diagram of the slow icing process of wind turbine blades within 60 minutes, provided in an embodiment of the present invention. Figure 10 This is a comprehensive performance verification and scenario comparison diagram of the entire icing monitoring system provided in this embodiment of the invention under four different actual working conditions; Figure 11 This is a schematic diagram of a wind turbine blade condition monitoring and de-icing linkage control method provided in an embodiment of the present invention.

[0019] Explanation of reference numerals in the attached figures: 100. Linkage control system; 110. Front-end sensing device; 120. Edge processing module; 130. Communication module; 140. De-icing linkage interface; 150. Remote platform; 200. De-icing equipment; 300. Handling methods. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] When wind turbines operate in cold, humid, mountainous, or high-altitude environments, the leading edge and windward surface of the blades are prone to icing. Existing icing monitoring technologies are mainly divided into two categories: direct detection and indirect detection. However, current monitoring schemes suffer from technical problems such as insufficient adaptability and poor coordination, leading to low detection accuracy.

[0022] To address the technical problems of low detection accuracy caused by insufficient adaptability and poor coordination, this invention provides a wind turbine blade status monitoring and de-icing linkage control system. This linkage control system can achieve adaptive and accurate monitoring and linkage control of wind turbine blade status.

[0023] It should be noted that the linkage control system provided by the present invention is used in, but not limited to, wind power generation equipment such as onshore wind turbines and offshore wind turbines. For ease of explanation, this invention will only use the application of the linkage control system to onshore wind turbines as an example. The principle of the linkage control system applied to other wind power generation equipment is essentially the same as that applied to onshore wind turbines, and will not be described in detail here.

[0024] This application provides a wind turbine blade condition monitoring and de-icing linkage control system 100, such as... Figures 1 to 3 As shown, the linkage control system 100 includes a front-end sensing device 110, an edge processing module 120, a communication module 130, a de-icing linkage interface 140, and a remote platform 150. The front-end sensing device 110 has a receiving cavity and integrates sensors for collecting raw state data of the wind turbine blades. The edge processing module 120 is arranged inside the receiving cavity and performs multi-level digital signal processing on the raw state data. The edge processing module 120 is equipped with a four-dimensional extended Kalman filter for estimating blade icing parameters and has an built-in eight-dimensional nonlinear observation model to construct a mapping relationship between sensing features and icing state parameters. The communication module 130 is arranged inside the receiving cavity for information transmission. The de-icing linkage interface 140 is located on the outer surface of the front-end sensing device 110 and is used to output de-icing control signals to the de-icing equipment 200 and receive status feedback from the de-icing equipment 200. The remote platform 150 is communicatively connected to the communication module 130.

[0025] The front-end sensing device 110 is used to collect raw state data of the wind turbine blades, including information such as blade surface temperature, ambient temperature and humidity, and triaxial vibration signals.

[0026] For example, such as Figure 2 As shown, the front-end sensing device 110 has a cylindrical structure with an integrated coaxial double-layer cavity inside. The front-end sensing device 110 includes a main control inner cavity, a sensing outer cavity, a shielded signal channel, a communication interface, a power management unit, a potting protective layer, and an installation and fastening structure.

[0027] The main control cavity is used to house the MCU (Microcontroller Unit), DSP (Digital Signal Processing) computing unit, memory, power management circuit, communication module, and edge processing circuit, in order to protect the core electronic units and reduce the effects of condensation, vibration, and electromagnetic interference.

[0028] The sensing cavity primarily houses blade temperature, ambient temperature and humidity, and triaxial vibration sensors, placing them close to the icing environment to enhance sensing sensitivity. Exemplarily, the vibration sensor can be a configurable sampling rate triaxial accelerometer, a broadband MEMS (Micro-Electro-Mechanical Systems) vibration sensor, or other equivalent vibration sensing units; the specific model can be determined based on actual needs. Exemplarily, the vibration sampling rate adaptively switches between three levels: 100Hz, 400Hz, and 800Hz. In embodiments requiring wider bandwidth monitoring, an industrial broadband vibration sensor can be selected, and the above levels can be mapped to its supported output data rate range.

[0029] The shielded signal channel connects the main control area and the sensing area through a shielded wire harness to reduce weak signal crosstalk and sampling drift in a strong electromagnetic environment. The main control cavity and the sensing cavity are isolated from each other and connected through a shielded wire harness or a shielded signal channel to reduce the impact of strong electromagnetic interference in the cabin on weak signal acquisition.

[0030] The communication module 130, such as a 4G / MQTT communication module, is used to upload real-time data, status levels, alarm information, and self-test results, and is suitable for remote operation and maintenance of distributed wind farms.

[0031] The potting protective layer is used for industrial-grade potting of batteries, cable connectors, communication modules and main control circuits to improve waterproofing, condensation prevention, shock resistance, insulation and tamper resistance.

[0032] The de-icing linkage interface 140 is used to output various control signals such as relays, dry contacts, RS485, CAN, and MQTT to realize de-icing triggering and status feedback, and to build a closed loop for the entire monitoring and de-icing process. The dry contacts are passive switch contacts.

[0033] In some embodiments, such as Figures 4 to 6 As shown, the front-end sensing device 110 and the edge processing module 120 transmit and process data signals through a sensing and edge processing link. The system's data link consists of: multi-sensor acquisition, shielded channel transmission, edge preprocessing, multi-level DSP feature extraction, four-dimensional EKF fusion estimation, risk index generation, global scheduling, communication reporting, and de-icing linkage. After the vibration signal is low-pass filtered by FIR, 512-point FFT, 1024-point FFT, envelope analysis, cepstral analysis, spectrogram update, and sliding window statistics are performed according to the scheduling level. Temperature, humidity, and vibration characteristics together form an eight-dimensional observation vector.

[0034] ; Among them, T b This refers to the blade temperature or the temperature of the casing near the blade side; H a This refers to the combined amount of ambient humidity or ambient temperature and humidity; F pE is the dominant vibration frequency. L Low-frequency energy; A rms M is the root mean square value of the vibration; env R is the envelope modulation depth. axis The energy ratio is the three-axis ratio; D f It is the main frequency drift rate.

[0035] In some embodiments, a four-dimensional extended Kalman filter joint estimation is employed, extending the traditional two-dimensional filter state that only estimates ice thickness and growth rate to a four-dimensional state-parameter joint estimation. This allows the filter to simultaneously estimate the icing state and correct key physical model parameters online. The state vector is defined as follows: ; Where, d k This is an estimate of the ice thickness; v k For ice thickness growth rate; Kf k Main frequency influence coefficient; Ke k These are the low-frequency energy influence coefficients. Kf describes the impact of increased icing mass on the blade's natural frequency, while Ke describes the impact of icing on low-frequency vibration energy. Both are highly correlated with blade material, length, installation location, and sensor coupling state, making them suitable as online estimation parameters.

[0036] ; ; ; ; For example, σ d² Take 0.01, σ v² Take 0.001, σ Kf² and σ Ke² Take 1×10 -6 The process noise corresponding to Kf and Ke is 3 to 4 orders of magnitude lower than that of ice thickness and growth rate, which allows the parameters to be essentially locked after convergence, but can still slowly track blade aging, changes in installation status, and long-term environmental drift. The initial value of Kf can be 0.04, and the initial value of Ke can be 0.3; after each filter update, Kf is limited to the range of 0.005 to 0.5, and Ke is limited to the range of 0.05 to 1.0 to avoid numerical divergence or physically unreasonable results.

[0037] In some embodiments, an eight-dimensional nonlinear observation model is used to construct the blade temperature T. b Ambient humidity H a Vibration dominant frequency F p Low-frequency energy E L , Root mean square value of vibration A rms Envelope modulation depth M env Triaxial energy ratio Raxis and the main frequency drift rate D f The mapping relationship between the parameters and the icing state can be established using the following set of nonlinear observation equations, with specific coefficients determined experimentally or based on engineering experience. See the table below for details.

[0038]

[0039] Wherein, F0 and E0 are the dominant frequency baseline and low-frequency energy baseline under ice-free conditions, respectively. h3 uses the online estimated Kf to replace the fixed dominant frequency influence coefficient, and h4 uses the online estimated Ke to replace the fixed energy influence coefficient; h8 is obtained by differentiating the mass-spring model with respect to time, which can simultaneously couple the ice thickness d, the growth rate v, and the dominant frequency influence coefficient Kf, thus enhancing the observability of the filtering system.

[0040] ; ; ; ; ; ; Therefore, the Jacobian matrix H k The dimension is 8×4, and the observation noise covariance matrix R k The dimension is 8×8, and the innovation covariance S k The dimension is 8×8. Compared to seven-dimensional observation, the computational complexity of matrix inversion increases, but it can still meet the requirements of real-time fusion computing on an STM32F4-level or equivalent MCU with hardware FPU and DSP instruction set.

[0041] In some embodiments, the edge processing module further includes a main frequency drift rate extraction unit for extracting the main frequency drift rate from the spectrum. This main frequency drift rate extraction unit extracts the main frequency drift rate D through multi-frame spectrum data calculations. f Specifically, the system maintains a circular spectrogram buffer consisting of N frames of spectrum, for example, N is 48, and each frame contains several frequency bins. The system finds the position of the dominant frequency peak for each frame of spectrum, obtaining the peak frequency sequence f. i For f i A linear regression is performed on frame number i, and the slope of the regression is the main frequency drift rate D. f : ; Where N is the total number of frames stored in the circular spectrum buffer; i is the frame number, f i This represents the peak frequency of the main frequency corresponding to the spectrum of the i-th frame. When the number of effective frames is less than a preset number, such as 8 frames, D... fIt can be set to 0 or retain the previous confidence value. D f A negative value indicates a continuous decrease in the main frequency, which usually corresponds to the development of icing; D f A positive value indicates a recovery in the main frequency, typically corresponding to the recovery process after icing or de-icing. This feature utilizes the time dimension of the spectrogram, which can reduce the probability of misjudgment caused by single-frame FFT being affected by gusts, instantaneous load disturbances, or accidental peak changes.

[0042] In some embodiments, the edge processing module is further provided with a baseline management unit. The baseline management unit realizes baseline initialization, rapid recovery and dynamic calibration through a three-layer ice-free baseline self-learning and drift compensation mechanism. The three-layer ice-free baseline self-learning and drift compensation mechanism includes three parts in sequence: cold start learning, hot start recovery and runtime drift compensation.

[0043] Specifically, the icing-free baseline is used to determine reference values ​​for temperature, humidity, and vibration characteristics under icing-free conditions, directly affecting all subsequent observation equations. To balance power-on speed, long-term stability, and fault protection, this invention employs a three-layer mechanism: cold start learning, hot start recovery, and runtime drift compensation. Details are shown in the table below.

[0044]

[0045] The baseline drift compensation process requires a three-level cascaded verification. If any level fails, baseline updates are prohibited. The first level is hardware fault diagnosis, detecting fault indicators in the temperature, humidity, vibration, communication, and power modules. The second level is single-channel statistical verification, determining whether the variances of data channels such as main frequency, root mean square, and temperature are within the normal range, eliminating anomalies such as data jams, sudden changes, and poor line contact. The third level is multi-dimensional joint distribution verification, using an autoencoder with 8-dimensional input, 4-dimensional bottleneck layer, and 8-dimensional output to calculate the reconstruction mean square error. When the error exceeds the 99th percentile threshold of normal samples in the training set, it is determined to be a multi-dimensional data joint anomaly, blocking the baseline update operation.

[0046] ; The autoencoder is only invoked in the baseline drift compensation process and does not intervene in the main loop of the Extended Kalman Filter (EKF). The model parameters can be quantized using INT8 and deployed to the microcontroller (MCU), with approximately 76 parameters in total, making it suitable for resource-constrained edge devices. This does not replace the physical model with an artificial intelligence (AI) algorithm, but rather serves as a third-level data quality gating mechanism, in addition to hardware fault diagnosis and single-channel variance verification, to identify multi-channel collaborative anomalies that cannot be detected by channel statistical methods.

[0047] In some embodiments, the edge processing module further includes a fusion estimation unit and a scheduling unit; the fusion estimation unit generates a risk index by combining ice thickness, ice growth rate, observation residual, main frequency drift rate and historical duration; the scheduling unit dynamically configures operating parameters based on the risk index to perform closed-loop coordinated control.

[0048] Specifically, the fusion estimation unit calculates a risk index R ranging from 0 to 100 based on ice thickness d, growth rate v, observation residuals, dominant frequency drift rate, and historical duration. The scheduling unit uses R as the core input to uniformly determine the vibration sampling rate, DSP processing depth, communication reporting cycle, and EKF process noise scaling factor, ensuring that monitoring accuracy, communication power consumption, and filtering sensitivity change in a coordinated manner within the same closed loop.

[0049] ; ; ; ;

[0050] For discrete parameters such as sampling rate and DSP processing depth, a separate threshold hysteresis control is adopted, with the hysteresis intervals corresponding to seven risk levels, and a minimum hold duration of 30 seconds set. During the hold period, only the reporting cycle and q are allowed to be reported. scale For continuous parameter updates, failure to execute discrete switching that would cause hardware reconfiguration or buffer interruption will result in such events.

[0051] In some embodiments, the scheduling unit also integrates de-icing linkage and closed-loop feedback functions, and sets power consumption budget constraints; the power consumption budget constraints are used to evaluate system power consumption and battery life, and to adjust operating parameters in stages.

[0052] Specifically, regarding power consumption budget constraints, the system estimates the remaining battery life of the device based on the current operating configuration, rather than relying solely on the traditional method of hard-switching based on battery percentage. The total power consumption of the device can be calculated using the following formula: ; ; When T remain If the data usage falls below the set target value, such as 24 hours, the system will gradually reduce its load according to the priority of decreasing DSP processing depth, lowering the sampling rate, and extending the data reporting cycle. In this scheme, the quantization coefficient q... scale It does not adjust synchronously with power consumption reduction, ensuring that the estimation sensitivity of the extended Kalman filter (EKF) is not affected by the energy-saving strategy under high-risk operating conditions, and prioritizing the core reliability of icing state determination in scenarios with limited power.

[0053] Specifically, regarding the de-icing linkage and closed-loop feedback, the front-end sensing device of this invention can not only output icing alarm signals, but also undertake the command triggering and status feedback functions of the de-icing system. When the risk index reaches the warning threshold, the system increases the sampling frequency and data reporting frequency, and uploads warning information to the remote platform; when it is determined that the blades are iced, it can issue start control commands to the heating film, hot air equipment, resistance heating device, ultrasonic transducer, vibration exciter, electromagnetic pulse de-icing equipment, and wind turbine main control system. If severe icing is detected and de-icing operations cannot be carried out on site, the system will output power limit, shutdown protection, and manual inspection prompts.

[0054] During de-icing operations, the system monitors in real time characteristics such as main frequency recovery, frequency band energy decline, RMS value returning to normal, and time-frequency drift rate turning positive and approaching zero, thereby verifying the de-icing effect. When the risk index remains below the de-icing threshold and meets the minimum holding time requirement, the system issues a de-icing stop command to avoid prolonged ineffective heating, repeated excitation, and accidental unit shutdowns.

[0055] To better understand this invention, the following is combined with... Figures 1 to 10 The technical solution of the present invention will be described in detail below. Figure 4 This is a schematic diagram of the state estimation and risk assessment results of the EKF algorithm in wind turbine blade icing monitoring; Figure 5 This is a schematic diagram illustrating the evolution trend of multi-sensor data for wind turbine blades under the "slow icing" condition. Figure 6 This is a comparison chart of the Fast Fourier Transform (FFT) spectra of wind turbine blades at different icing stages; Figure 7 This is a schematic diagram of the cepstrum process at an ice thickness of 3.5 mm; Figure 8 This is a schematic diagram of the entire process of envelope analysis when the ice thickness is 3mm; Figure 9 It is a time-frequency graph of the slow icing process of the wind turbine blades over 60 minutes; Figure 10 This is a comprehensive performance verification and scenario comparison chart of the entire icing monitoring system under four different actual working conditions.

[0056] In some embodiments, the wind turbine blade condition monitoring and de-icing linkage control system 100 includes a front-end sensing device 110, an edge processing module 120, a communication module 130, a de-icing linkage interface 140, and a remote platform 150. The front-end sensing device 110 collects blade surface temperature, ambient temperature and humidity, and triaxial vibration signals; the edge processing module 120 performs multi-level DSP processing on the vibration signals to extract time-domain, frequency-domain, and time-frequency features; the fusion estimation unit estimates ice thickness, growth rate, main frequency influence coefficient, and energy influence coefficient online based on a four-dimensional extended Kalman filter; the scheduling unit uniformly adjusts the sampling rate, DSP processing depth, reporting cycle, and filter process noise scaling factor according to the risk index; the de-icing linkage interface 140 outputs linkage signals to the heating film, ultrasonic / vibration de-icing device, shutdown protection, or remote platform 150 according to the risk level, and continuously monitors the feature recovery status after de-icing.

[0057] For example, the front-end sensing device 110 adopts a dual-cavity sealed housing with dimensions of 300mm × 300mm × 70mm. The main control cavity houses an STM32F407VET6 or equivalent MCU, a W25Q64 SPI Flash (serial flash memory chip), power management circuitry, a 4G communication module, and interface protection circuitry. The sensing outer cavity houses a blade temperature sensor, an ambient temperature and humidity sensor, and a triaxial vibration sensor. Each sensor is connected to the main control board via a 6-core shielded cable or an equivalent shielded channel. Cable through-holes and connector locations are sealed. The main control board is secured with threaded fasteners, and core electronic components are encapsulated with flame-retardant potting compound, effectively improving the device's waterproof, anti-condensation, vibration-resistant, insulating, and anti-disassembly performance.

[0058] Regarding edge DSP processing, for example, triaxial vibration data is entered into a circular buffer via DMA. The system first performs a 32nd-order FIR low-pass filter, and then performs different processing according to the DSP depth level: L0 calculates a 512-point FFT and extracts the dominant frequency and fundamental band energy; L1 adds a 1024-point FFT, spectral centroid, and spectral entropy; L2 adds bandpass envelope analysis and cepstral analysis from 20 to 80 Hz; L3 updates 48 frames of spectrograms and calculates the mean, variance, slope, rate of change, and cross-channel correlation within a 5-minute sliding window. The above processing can be implemented by calling CMSIS-DSP or an equivalent digital signal processing library.

[0059] Regarding four-dimensional EKF fusion, for example, the system assembles an eight-dimensional observation vector z in each fusion cycle. kThe system performs prediction, observation updates, parameter constraints, and risk index calculations. Kf and Ke gradually converge to the physical mapping parameters corresponding to the current blade and installation location during the ice-free or low-risk phase. During the icing development phase, the dominant frequency drift rate, along with observations of the dominant frequency and low-frequency energy, causes rapid changes in ice thickness and growth rate estimates. If an anomaly occurs in a certain channel, the system can mitigate its impact through increased observation noise, fault flags, or data quality gating.

[0060] Regarding baseline management, for example, when the device is first powered on, if the temperature is above 5°C and the humidity is below 60%, 30 sets of valid samples are continuously collected to calculate the eight-dimensional baseline; if the conditions are not met, the process waits until the ice-free learning conditions are met. After learning is complete, the baseline, version number, and checksum are written to the RTC backup register or non-volatile memory. Upon subsequent restarts, valid baselines are read first, achieving second-level recovery. During operation, exponential moving average compensation with α=0.001 is only performed when the ice thickness estimate is below 0.1mm for 15 consecutive minutes and all three levels of verification pass.

[0061] Regarding baseline management, for example, the system classifies operating states into four risk levels: normal, warning, icing, and severe icing. When the equipment is in normal condition, monitoring is conducted in low-power mode; upon entering the warning state, the sampling frequency and data reporting frequency are simultaneously increased; when icing is detected, a de-icing trigger command is output; when severe icing is identified, emergency alarms, shutdown protection, and power limiting control suggestions are issued. During de-icing operations, the system upgrades the digital signal processing level and continuously monitors the recovery trend of the main frequency and bandwidth energy; once the risk level is resolved, a stop command is promptly output or the system switches to a lower-level operating mode.

[0062] In this embodiment, the linkage control system 100 addresses several shortcomings of existing technologies: First, it improves the poor adaptability of fixed parameter models, eliminating the need for manual calibration across aircraft models and the accuracy degradation caused by blade aging and changes in operating conditions. Second, it fully leverages the value of frequency spectrum characteristics, transforming frequency domain data into effective edge-end observations to accurately distinguish between instantaneous frequency fluctuations and inherent frequency shifts caused by icing. Third, it adds multi-level data quality gating for baseline drift compensation, preventing incorrect baseline updates due to sensor failures and data anomalies. Fourth, it achieves global linkage scheduling of sampling, signal processing, communication, and filtering strategies, balancing equipment power consumption and algorithm estimation sensitivity. Fifth, it optimizes the overall protection structure of the device, improving engineering reliability under complex on-site conditions. The entire solution relies on highly reliable front-end sensing devices to construct a complete closed-loop system integrating monitoring, judgment, communication, de-icing linkage, and effect feedback.

[0063] In this embodiment, the linkage control system 100 has the following beneficial effects.

[0064] (1) Enhance cross-model adaptability. By estimating Kf and Ke online, the system can automatically adapt to the current blade and installation conditions after deployment, reducing the workload of manual calibration, and can slowly track long-term environmental changes and sensor coupling state changes.

[0065] (2) Improve the reliability of early icing identification. The dominant frequency drift rate in the spectrum is converted into the eighth dimension observation of EKF, which enables the system to identify the continuous downward trend of the dominant frequency and reduces the risk of misjudgment caused by wind speed disturbance or instantaneous load changes.

[0066] (3) Improve the reliability of baseline management. The combination of cold start condition learning, RTC hot start recovery and runtime drift compensation under three-level quality gating can reduce the probability of erroneous baseline contamination and maintain long-term operational accuracy.

[0067] (4) Optimize power consumption and monitoring accuracy. The risk index uniformly drives the sampling rate, DSP depth, reporting cycle, and q. scale It enables low-power operation in low-risk scenarios and high-precision monitoring in high-risk scenarios; combined with a hysteresis mechanism and minimum hold time, it avoids frequent parameter switching.

[0068] (5) Achieve closed-loop control for monitoring and de-icing. The system can not only detect icing risks, but also provide data for de-icing initiation, de-icing effect confirmation, de-icing cessation and shutdown protection, enabling the de-icing strategy to shift from manual experience judgment to closed-loop control based on edge data.

[0069] (6) Adaptable to embedded real-time deployment. The four-dimensional state extension mainly adds a small amount of matrix storage. The autoencoder is only called in the low-frequency drift compensation path. The overall computation and storage overhead is suitable for STM32F4 level or equivalent embedded platforms.

[0070] (7) Improve engineering reliability. The device adopts a dual-cavity structure, shielded signal channel, threaded fixing, potting protection and remote communication design, which can improve the long-term stability of the fan in low temperature, condensation, vibration and electromagnetic interference environment.

[0071] This invention also provides a wind turbine blade condition monitoring and de-icing linkage control method 300, applicable to the aforementioned linkage control system 100, such as... Figure 11 As shown, the processing method 300 includes: Step S310, Data Acquisition: Collect raw state data of the wind turbine blades through the sensors of the front-end sensing device.

[0072] Step S320, Signal preprocessing and feature extraction: The original data is filtered and processed through multi-level digital signal processing by the edge processing module to extract features, calculate the main frequency drift rate, and construct an eight-dimensional observation vector.

[0073] Step S330, Adaptive state estimation: The icing state parameters are estimated online using the four-dimensional extended Kalman filter and eight-dimensional nonlinear observation model mounted on the edge processing module.

[0074] Step S340, Baseline Maintenance: The edge processing module completes the learning, recovery, and drift compensation of the ice-free baseline.

[0075] Step S350, Risk Assessment and Global Scheduling: The risk index is calculated based on the icing parameters by the edge processing module, and the equipment operating parameters are dynamically adjusted.

[0076] Step S360, Icing Level Determination: The icing level is determined by the edge processing module according to the risk index.

[0077] Step S370, De-icing linkage and closed-loop verification: Output control signals and receive status feedback through the de-icing linkage interface of the front-end sensing device for de-icing linkage and closed-loop verification.

[0078] In this embodiment, the processing method 300 relies on a front-end device to complete data acquisition, feature extraction, and state estimation, and in conjunction with a standardized baseline maintenance mechanism, ensures the reliability of monitoring data. The system performs level determination and global scheduling based on a risk index, dynamically balancing equipment power consumption and detection sensitivity; simultaneously, it implements de-icing command issuance, state feedback, and closed-loop verification through an interface. This method is highly automated, adaptable to complex operating conditions at wind turbine sites, can accurately identify early icing, reduce misjudgments, lower manual calibration and maintenance costs, and achieve closed-loop management of the entire process of blade monitoring, early warning, de-icing, and safety protection.

[0079] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A wind turbine blade condition monitoring and de-icing linkage control system, characterized in that, include: A front-end sensing device, which has a receiving cavity and integrates sensors for collecting raw state data of the wind turbine blades; An edge processing module is arranged within the receiving cavity to perform multi-level digital signal processing on the original state data. The edge processing module is equipped with a four-dimensional extended Kalman filter for estimating blade icing parameters and has an built-in eight-dimensional nonlinear observation model to construct a mapping relationship between sensing features and icing state parameters. A communication module, disposed within the receiving cavity, is used for information transmission; The de-icing linkage interface is located on the outer surface of the front-end sensing device and is used to output de-icing control signals and receive de-icing equipment status feedback. A remote platform, which is communicatively connected to the communication module.

2. The linkage control system according to claim 1, characterized in that, The front-end sensing device has an integrated coaxial double-layer cavity, which includes a main control inner cavity and a sensing outer cavity. The main control inner cavity is used to accommodate various functional modules, and the sensing outer cavity is used to install sensors.

3. The linkage control system according to claim 1, characterized in that, The front-end sensing device and the edge processing module transmit and process data signals through a sensing and edge processing link. The combination of multiple types of sensing features output by the link forms an eight-dimensional observation vector. ; Among them, T b For blade temperature; H a For ambient humidity; F p E is the dominant vibration frequency. L Low-frequency energy; A rms M is the root mean square value of the vibration; env R is the envelope modulation depth. axis The energy ratio is the three-axis ratio; D f It is the main frequency drift rate.

4. The linkage control system according to claim 3, characterized in that, The four-dimensional extended Kalman filter simultaneously estimates the icing state and corrects the physical model parameters online. The state vector is: ; Where, d k This is an estimate of the ice thickness; v k For ice thickness growth rate; Kf k Main frequency influence coefficient; Ke k This is the low-frequency energy influence coefficient.

5. The linkage control system according to claim 4, characterized in that, The eight-dimensional nonlinear observation model is used to construct the blade temperature T. b Ambient humidity H a Vibration dominant frequency F p Low-frequency energy E L , Root mean square value of vibration A rms Envelope modulation depth M env Triaxial energy ratio R axis and the main frequency drift rate D f The mapping relationship between the parameters and the freezing state parameters.

6. The linkage control system according to claim 5, characterized in that, The edge processing module also includes a main frequency drift rate extraction unit, which extracts the main frequency drift rate D through multi-frame spectrum data calculation. f : ; Where N is the total number of frames stored in the circular spectrum buffer; i is the frame number, f i Let be the peak frequency of the main frequency corresponding to the spectrum of the i-th frame.

7. The linkage control system according to claim 1, characterized in that, The edge processing module is also equipped with a baseline management unit. The baseline management unit realizes baseline initialization, rapid recovery and dynamic calibration through a three-layer ice-free baseline self-learning and drift compensation mechanism. The three-layer ice-free baseline self-learning and drift compensation mechanism includes three parts in sequence: cold start learning, hot start recovery and runtime drift compensation.

8. The linkage control system according to claim 1, characterized in that, The edge processing module also includes a fusion estimation unit and a scheduling unit; the fusion estimation unit combines ice thickness, ice growth rate, observation residual, dominant frequency drift rate and historical duration to generate a risk index; The scheduling unit dynamically configures operating parameters based on the risk index to perform closed-loop coordinated control.

9. The linkage control system according to claim 8, characterized in that, The scheduling unit also integrates de-icing linkage and closed-loop feedback functions, and sets power consumption budget constraints; the power consumption budget constraints are used to evaluate the system power consumption and battery life, and adjust the operating parameters in stages.

10. A method for monitoring the condition of wind turbine blades and for coordinated control of de-icing, applicable to the coordinated control system described in any one of claims 1-9, characterized in that, The processing method includes: Data acquisition: The raw state data of the wind turbine blades are collected through sensors in the front-end sensing device; Signal preprocessing and feature extraction: The edge processing module filters the raw data and performs multi-level digital signal processing to extract features, calculate the main frequency drift rate, and construct an eight-dimensional observation vector; Adaptive state estimation: The four-dimensional extended Kalman filter and eight-dimensional nonlinear observation model mounted on the edge processing module are used to estimate the icing state parameters online; Baseline maintenance: The edge processing module is used to learn, restore, and compensate for the drift of the ice-free baseline. Risk assessment and global scheduling: The edge processing module calculates the risk index based on icing parameters and dynamically adjusts the equipment operating parameters. Icing level determination: Icing levels are determined by the edge processing module based on the risk index; De-icing linkage and closed-loop verification: The de-icing linkage interface of the front-end sensing device outputs control signals and receives status feedback for de-icing linkage and closed-loop verification.

Images