Method for failure early warning of dynamic data collection of wind power equipment

Abstract

The application discloses a kind of wind power equipment dynamic data acquisition fault early warning method, it is related to wind power equipment condition monitoring and fault diagnosis technical field, comprising the following steps: S1: in the rotating state of fan blade, electromagnetic induction signal and sound wave reflection signal are collected;S2: electromagnetic induction signal and sound wave reflection signal are treated in time and space synchronization, obtain blade damage signal;S3: the blade damage signal collected is rotated base frequency interference elimination, aerodynamic noise separation, electromagnetic interference suppression and multi-source data time-space registration and signal fusion preprocessing, form the blade damage signal after preprocessing.Fusion electromagnetic induction and sound wave reflection two physical field signals, overcome the limitation of single sensor;Through the blade damage signal collected is rotated base frequency interference elimination, aerodynamic noise separation, electromagnetic interference suppression and multi-source data time-space registration and fusion, accurately eliminate the blade rotating base frequency, aerodynamic noise and frequency converter electromagnetic interference.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment condition monitoring and fault diagnosis technology, and in particular to a fault early warning method for dynamic data acquisition of wind power equipment. Background Technology

[0002] Wind turbine blades are one of the most critical components of wind turbine generators. Operating under complex loads for extended periods, they are prone to structural damage such as cracks, deformation, and delamination. If these damages are not detected and addressed promptly, they can lead to reduced power generation efficiency or, in severe cases, blade breakage and ejection, causing major safety accidents and economic losses.

[0003] Traditional detection methods, relying on a single detection technique, cannot simultaneously address both surface defects and deep damage, limiting their detection coverage and accuracy. In a rotating state, signals are easily affected by the rotational fundamental frequency, aerodynamic noise, and electromagnetic interference from the frequency converter, making it difficult to extract subtle damage features. Fault identification models struggle to capture the spatiotemporal correlation characteristics of blade damage, resulting in insufficient accuracy in damage classification, severity quantification, and life prediction. Summary of the Invention

[0004] The embodiments of the present invention provide a fault early warning method for dynamic data acquisition of wind power equipment, which aims to solve the problems of limited coverage and accuracy of single detection methods, difficulty in extracting weak damage features, difficulty in capturing the spatiotemporal correlation characteristics of blade damage by fault identification models, and insufficient accuracy in damage classification, degree quantification and life prediction.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: A fault early warning method for dynamic data acquisition of wind power equipment includes the following steps: S1: Collect electromagnetic induction signals and acoustic wave reflection signals while the wind turbine blades are rotating; S2: Perform spatiotemporal synchronization processing on the electromagnetic induction signal and the acoustic wave reflection signal to obtain the blade damage signal; S3: Perform rotational fundamental frequency interference cancellation, aerodynamic noise separation, electromagnetic interference suppression, multi-source data spatiotemporal registration and fusion preprocessing on the collected blade damage signal to form the preprocessed blade damage signal; S4: Extract the electromagnetic induction channel fault features and the acoustic wave reflection channel fault features from the preprocessed blade damage signal, and fuse the electromagnetic induction channel fault features and the acoustic wave reflection channel fault features. S5: A multi-scale spatiotemporal graph neural network is used to identify the fused fault features, output the damage type, and perform quantitative analysis of the damage degree and prediction of the remaining service life.

[0006] Furthermore, in S2, the spatiotemporal synchronization process includes: Using the 1PPS signal generated by the GPS timing module as the global time reference, the electromagnetic induction signal and the acoustic wave reflection signal are synchronized to a precision of 100ns. The blade azimuth encoder signal is used as the spatial sampling trigger pulse to establish a two-dimensional coordinate system of blade spanwise and chordwise, with a spanwise resolution of 20mm and a chordwise resolution of 10mm, forming a dual-channel blade damage signal.

[0007] Furthermore, in S3, the rotational fundamental frequency interference cancellation employs an angular domain synchronous averaging method: the time-domain signal is resampled to an equal angular domain using the encoder signal, and the angular domain is used to synchronously average the signal over N consecutive rotations. The specific formula is as follows: in: The target signal after angular domain synchronous averaging is used to characterize the pure blade damage signal after removing the rotating fundamental frequency interference. The blade rotation angle enables resampling mapping of the time-domain signal to the equal-angle domain; To ensure the statistical validity of the averaging process, the number of wind turbine blade rotations is included in the average.

[0008] Furthermore, in S3, the aerodynamic noise separation employs an adaptive beamforming spatial filtering method based on the MVDR criterion, specifically formulated as follows: in: This represents the optimal spatial filtering weight vector under the MVDR criterion. This is the inverse of the noise covariance matrix, used for noise power suppression; The spatial steering vector of the blade acoustic wave signal represents the direction and angle characteristics of acoustic wave propagation; The conjugate transpose of the guiding vector; It is the inverse matrix of the received signal covariance matrix, representing the spatial distribution characteristics of the signal.

[0009] Furthermore, in S3, the electromagnetic interference suppression employs a recursive least squares adaptive notch filter, using the inverter control signal as a reference input to adaptively estimate the interference propagation path and reduce interference. The specific formula is as follows: , in: The electromagnetic interference signal is adaptively estimated. It is the transpose of the weight vector of the time-varying filter, and the interference propagation path is updated in real time; It serves as the reference input signal for inverter control and provides a reference for interference suppression; This is the filtered, purified leaf damage signal; The original acquired signal contains electromagnetic interference.

[0010] Furthermore, in S3, the spatiotemporal registration and fusion of multi-source data involves mapping the electromagnetic induction signal and the acoustic wave reflection signal to a unified blade two-dimensional coordinate system. Based on the blade geometric model and measured azimuth / distance parameters, a coordinate transformation matrix for the two channels is established. Bilinear interpolation resampling is then performed on the data from each channel to generate a two-dimensional data frame with uniform resolution. The specific formula is as follows: in: Let be the signal amplitude of the target pixel after bilinear interpolation; The signal amplitude at the original sampling point; The normalized coordinates of the target point in the two-dimensional coordinate system of the blade.

[0011] Furthermore, in S4, the extraction of electromagnetic induction channel fault features specifically involves: S401: Extract the induced voltage signal generated by multi-frequency excitation at each frequency point: Impedance plane trajectory characteristics: The real resistance component and the imaginary reactance component of the induced voltage are plotted on the complex impedance plane, and the trajectory deflection angle and deflection amplitude at the damage point are used as characteristic quantities. Multi-frequency induction pair characteristic value: Defines the ratio of induced voltage at different frequencies. This ratio is sensitive to the depth of damage and can distinguish between surface and deep damage; Differential gradient characteristics: Calculating the induced voltage gradient between adjacent receiving array elements. The location of the gradient peak corresponds to the damage boundary; S402: Feature extraction of the eddy current disturbance wall. The disturbances caused by large-area debonding in the eddy current field are dispersed, and single-point features are not obvious. The local eddy current disturbance arrangement entropy is used, with the specific formula as follows: in: The entropy value of the local eddy current disturbance is used to quantify the disorder of the eddy current signal. The embedding dimension characterizes the phase space dimension of the reconstructed eddy current signal; The time interval for delaying the phase space reconstruction of the eddy current signal; This refers to the modal arrangement and combination of eddy current signals; This represents the probability of occurrence of the corresponding modal permutations and combinations.

[0012] Furthermore, in S4, the specific fault characteristics of the acoustic wave reflection channel are as follows: S403: Take the time-flight characteristics of the echo: Extraction of single-point echo time series after pulse compression: first echo delay time : Corresponds to the reflection from the blade surface, and its changes reflect the distance changes caused by the deformation of the blade surface, with a resolution of ±0.1mm; Extraction of multiple echo time interval sequences: Multiple echo intervals correspond to blade wall thickness in: The wall thickness of the wind turbine blade material; The ultrasonic propagation speed in the blade composite material; The time interval between multiple acoustic echoes; localized thinning of the wall thickness >0.5 mm is a characteristic of damage; Echo amplitude attenuation spectrum: The echo amplitude of healthy fiberglass laminate decreases exponentially with the number of multiple echoes, and an abnormal attenuation coefficient indicates internal defects. S404: Take samples of the damage imaging features from the scan: Extract the peak amplitude from the A-scan data of each spanwise × longitudinal scan point to construct a C-scan amplitude map, and extract the following from the C-scan map: Area of ​​the damaged region; Damage profile shape perturbation, regional splitting, degumming, deformation; Edge gradient intensity: The crack edge gradient intensity is much higher than that in the degummed and deformed zones.

[0013] Furthermore, in S4, the fusion of the fault features of the electromagnetic induction channel and the fault features of the acoustic wave reflection channel specifically employs feature-level weighted fusion, which normalizes and weights the electromagnetic induction feature vector and the acoustic wave reflection feature vector, and concatenates them. The core fusion formula is as follows: in: This is the fused fault feature vector; This is the characteristic vector of the electromagnetic induction channel; This represents the characteristic vector of the sound wave reflection channel; , These are adaptive weighting coefficients based on feature sensitivity, and α+β=1.

[0014] Furthermore, in S5, the fault identification employs a multi-scale spatiotemporal graph neural network to complete damage classification and lifetime prediction, with the core formula for quantifying damage severity being: in: Quantify the degree of leaf damage with a score; The number of scales for a multi-scale spatiotemporal graph neural network; The output of the graph neural network at the corresponding scale; This is the activation function for classification.

[0015] Beneficial effects

[0016] By integrating electromagnetic induction and acoustic wave reflection signals, the limitations of a single sensor are overcome. Through the elimination of rotating fundamental frequency interference, aerodynamic noise separation, electromagnetic interference suppression, and spatiotemporal registration and fusion of multi-source data on the collected blade damage signals, the rotating fundamental frequency, aerodynamic noise, and inverter electromagnetic interference are precisely eliminated, enabling the extraction of weak damage signals under complex operating conditions. By extracting fault features from the electromagnetic induction channel and the acoustic wave reflection channel, features with clear physical meaning are designed for different damage types. A multi-scale spatiotemporal graph neural network is used to model the fused features, which can not only identify damage types but also quantify the degree of damage and predict remaining service life, providing a direct decision-making basis for preventative maintenance. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the preprocessing process in this invention; Detailed Implementation

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] In the description of this invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0020] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0021] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] Combination Figure 1 and Figure 2 As shown, the present invention provides a fault early warning method for dynamic data acquisition of wind power equipment, comprising the following steps: S1: Collect electromagnetic induction signals and acoustic wave reflection signals while the wind turbine blades are rotating; S2: Perform spatiotemporal synchronization processing on electromagnetic induction signals and acoustic wave reflection signals to obtain blade damage signals; The spatiotemporal synchronization process includes: Using the 1PPS signal generated by the GPS timing module as the global time reference, the electromagnetic induction signal and the acoustic wave reflection signal are synchronized to 100ns accuracy. The blade azimuth encoder signal is used as the spatial sampling trigger pulse to establish a blade spanwise-chordwise two-dimensional coordinate system with a spanwise resolution of 20mm and a chordwise resolution of 10mm, forming a dual-channel blade damage signal. S3: Perform preprocessing on the collected blade damage signal, including rotational fundamental frequency interference cancellation, aerodynamic noise separation, electromagnetic interference suppression, and spatiotemporal registration and signal fusion of multi-source data, to form a preprocessed blade damage signal. The rotational fundamental frequency interference cancellation employs an angular domain synchronous averaging method: the time-domain signal is resampled to an equal angular domain using the encoder signal, and the angular domain is used to synchronously average the signal over N consecutive rotations (N ≥ 50). The specific formula is as follows: in: The target signal after angular domain synchronous averaging is used to characterize the pure blade damage signal after removing the rotating fundamental frequency interference. The blade rotation angle enables resampling mapping of the time-domain signal to the equal-angle domain; To ensure the statistical validity of the averaging process, the number of wind turbine blade rotations is N≥50, accurately eliminating periodic interference from the blade rotation fundamental frequency and preserving the non-periodic characteristics induced by damage. Equal-angle domain resampling avoids the impact of speed fluctuations on signal processing. Multi-rotation averaging improves the signal-to-noise ratio and ensures the detectability of weak damage signals. Aerodynamic noise separation employs an adaptive beamforming spatial filtering method based on the MVDR criterion, with the specific formula as follows: in: This represents the optimal spatial filtering weight vector under the MVDR criterion. This is the inverse of the noise covariance matrix, used for noise power suppression; The spatial steering vector of the blade acoustic wave signal represents the direction and angle characteristics of acoustic wave propagation; The conjugate transpose of the guiding vector; The inverse matrix of the received signal covariance matrix is ​​used to characterize the spatial distribution characteristics of the signal, directionally enhance the target damage acoustic signal, and adaptively suppress aerodynamic noise; spatial filtering improves the signal directionality and resolution, adapting to the complex flow field noise environment of the fan. Electromagnetic interference suppression employs a recursive least squares adaptive notch filter, using the inverter control signal as a reference input to adaptively estimate the interference propagation path and reduce interference. The specific formula is as follows: , in: The electromagnetic interference signal is adaptively estimated. It is the transpose of the weight vector of the time-varying filter, and the interference propagation path is updated in real time; It serves as the reference input signal for inverter control and provides a reference for interference suppression; This is the filtered, purified leaf damage signal; For the original acquisition signal containing electromagnetic interference, the electromagnetic interference characteristics are accurately matched with the frequency converter signal as a reference, and the suppression effect is significant; recursive least squares realizes real-time weight update to adapt to dynamically changing electromagnetic interference; Multi-source data spatiotemporal registration and fusion maps electromagnetic induction signals and acoustic wave reflection signals to a unified blade two-dimensional coordinate system (span × chord). Based on the blade geometric model and measured azimuth / distance parameters, a coordinate transformation matrix for the two channels is established. Bilinear interpolation resampling is then performed on the data from each channel to generate a two-dimensional data frame with uniform resolution. The specific formula is as follows: in: Let be the signal amplitude of the target pixel after bilinear interpolation; The signal amplitude at the original sampling point; The normalized coordinates of the target point in the two-dimensional coordinate system of the blade are used to achieve unified spatial resolution of electromagnetic induction and acoustic wave reflection signals, ensuring accurate registration of multi-source data and improving fusion reliability. S4: Extract the fault features of the electromagnetic induction channel and the acoustic reflection channel from the preprocessed blade damage signal, and fuse the fault features of the electromagnetic induction channel and the acoustic reflection channel. The specific features for extracting fault characteristics of the electromagnetic induction channel are as follows: The induced voltage signal generated by multi-frequency excitation is extracted at each frequency point: Impedance plane trajectory characteristics: the real part of the induced voltage (Resistivity component) and imaginary part (Reactance component) plotted on the complex impedance plane, with the trajectory deflection angle at the damage point. and deflection amplitude As a characteristic quantity; crack: , Significantly increased; Degumming: , Slight increase; Multi-frequency induction pair characteristic value: Defines the ratio of induced voltage at different frequencies. This ratio is sensitive to the depth of damage and can distinguish between surface and deep damage; Differential gradient characteristics: Calculating the induced voltage gradient between adjacent receiving array elements. The location of the gradient peak corresponds to the damage boundary; For feature extraction of the eddy current disturbance wall, the disturbance caused by large-area degumming in the eddy current field is dispersed, and the single-point features are not obvious. Therefore, the local eddy current disturbance arrangement entropy is used, and the specific formula is as follows: in: The entropy value of the local eddy current disturbance is used to quantify the disorder of the eddy current signal. The embedding dimension characterizes the phase space dimension of the reconstructed eddy current signal; The time interval for delaying the phase space reconstruction of the eddy current signal; This refers to the modal arrangement and combination of eddy current signals; To correspond to the probability of occurrence of modal permutations and combinations, it sensitively captures the dispersed eddy current disturbances of large-area debonding, making up for the lack of single-point features; it has strong anti-interference ability and is suitable for complex electromagnetic and mechanical noise environments of wind turbines. The specific characteristics of a sound wave reflection channel fault are as follows: The time-flight characteristics of the echo are analyzed as follows: Extraction of single-point echo time series after pulse compression: first echo delay time : Corresponds to the reflection from the blade surface, and its changes reflect the distance changes caused by the deformation of the blade surface, with a resolution of ±0.1mm; Extraction of multiple echo time interval sequences: Multiple echo intervals correspond to blade wall thickness in: The wall thickness of the wind turbine blade material; The ultrasonic propagation speed in the blade composite material; The time interval between multiple echoes of the sound wave; local thinning of the wall thickness >0.5mm is a damage characteristic. Non-contact and accurate measurement of blade wall thickness with a resolution of up to 0.1mm. Internal damage can be quickly determined based on the wall thickness thinning threshold, and the degree of damage is quantified intuitively and reliably. Echo amplitude attenuation spectrum: The echo amplitude of healthy fiberglass laminate decreases exponentially with the number of multiple echoes, and an abnormal attenuation coefficient indicates internal defects. The characteristics of the scanned damage imaging were taken as follows: Extract the peak amplitude from the A-scan data of each spanwise × longitudinal scan point to construct a C-scan amplitude map, and extract the following from the C-scan map: Area of ​​the damaged region; Damage profile shape perturbation, regional splitting, degumming, deformation; Edge gradient intensity: The crack edge gradient intensity is much higher than that in the debonding and deformation zones; The fusion of fault features from the electromagnetic induction channel and the acoustic wave reflection channel specifically employs feature-level weighted fusion. This involves normalizing and weighting the electromagnetic induction feature vector and the acoustic wave reflection feature vector, and then concatenating them. The core fusion formula is as follows: in: This is the fused fault feature vector; This is the characteristic vector of the electromagnetic induction channel; This represents the characteristic vector of the sound wave reflection channel; , The adaptive weighting coefficients are based on feature sensitivity, and α+β=1, which preserves the complementarity of dual-channel features and improves the robustness and accuracy of damage recognition.

[0023] S5: A multi-scale spatiotemporal graph neural network is used to identify faults based on the fused fault features, output the damage type, and perform quantitative analysis of the damage degree and prediction of the remaining service life.

[0024] The fault identification employs a multi-scale spatiotemporal graph neural network to perform damage classification and lifetime prediction. The core formula for quantifying damage severity is: in: Quantify the degree of leaf damage with a score; The number of scales for a multi-scale spatiotemporal graph neural network; The output of the graph neural network at the corresponding scale; As a classification activation function, multi-scale feature fusion improves the accuracy of damage identification and enables precise quantification of damage degree and remaining lifespan.

[0025] Example 1: Early warning of surface cracks in the blades of a small 1.5MW wind turbine Step S1: Dynamic signal acquisition With the blades of a 1.5MW wind turbine rotating, an electromagnetic induction sensor array and an acoustic wave reflection sensor array are installed. The rated blade speed is 18 r / min, and the acquisition frequencies are: 10 kHz for the electromagnetic induction channel and 200 kHz for the acoustic wave reflection channel.

[0026] Step S2: Spatiotemporal synchronization processing Using GPS timing 1PPS as the global time reference, and the synchronization accuracy of the two signal clocks is 100ns; a two-dimensional coordinate system of blade spanwise and chordwise directions is established using the blade azimuth encoder as the spatial trigger pulse. Transverse resolution: 20mm Chord resolution: 10mm to form dual-channel spatiotemporally aligned blade damage signals.

[0027] Step S3: Multi-stage preprocessing Formula for rotating fundamental frequency interference cancellation (angular domain synchronization averaging method): By taking the number of rotations N=60, the time-domain signal is resampled to the equiangular domain to eliminate the 1PPS rotating fundamental frequency interference and retain the non-periodic characteristics of the damage.

[0028] Aerodynamic noise separation (MVDR adaptive beamforming) formula: Aerodynamic noise is suppressed by using the inverse of the noise covariance matrix, thereby directionally enhancing the blade damage echo signal.

[0029] Electromagnetic interference suppression (RLS adaptive notch filter) formula: , The two signals are mapped to a two-dimensional grid with a span of 20mm and a chord of 10mm to generate data frames with consistent resolution.

[0030] Step S4: Fault Feature Extraction and Fusion Electromagnetic induction channel characteristics Impedance plane trajectory: Surface crack satisfies , Significantly increased; Multi-frequency induction pair eigenvalues: Pick =100kHz, =300kHz, an abnormal ratio indicates surface damage; Differential gradient features: Gradient peak value is used to locate the crack boundary.

[0031] Sound wave reflection channel characteristics First echo delay Damage is determined by a resolution of ±0.1mm and a surface deformation distance offset >0.3mm. Wall thickness calculation: Take the speed of sound wave propagation =3200m / s, wall thickness reduction > 0.5mm, crack propagation is determined.

[0032] Feature-level weighted fusion formula: For surface cracks, α=0.7 and β=0.3 are used to highlight the advantages of electromagnetic induction characteristics.

[0033] Step S5: Multi-scale spatiotemporal graph neural network diagnosis Formula for quantifying the degree of damage: With scale number Z=3, output: Damage type: surface crack, damage severity score: =0.82, remaining service life: 128 days.

[0034] Example 2: Early warning of large-area degumming fault inside 3.0MW wind turbine blades 3.0MW wind turbine blades, rated speed 15r / min, electromagnetic induction channel sampling rate 15kHz, acoustic wave reflection channel sampling rate 250kHz, with a focus on monitoring the bonding layer area in the middle of the blades.

[0035] Step S2: Spatiotemporal synchronization processing Time synchronization accuracy of 100ns, spatial coordinate system: span 20mm, chord 10mm, realizing full blade scanning sampling.

[0036] Step S3: Multi-stage preprocessing Rotational fundamental frequency interference elimination: Take N=70 cycles as the average to completely filter out rotational period interference; Aerodynamic noise separation: MVDR spatial filtering suppresses wind field turbulence noise; Electromagnetic interference suppression: The RLS filter tracks time-varying electromagnetic interference and cancels it in real time; Spatiotemporal registration: Bilinear interpolation unifies two signals to the same two-dimensional grid.

[0037] Step S4: Fault Feature Extraction and Fusion Electromagnetic induction channel characteristics Impedance plane trajectory: Debonding satisfies <30°, Slight increase; Local eddy current perturbation arrangement entropy: Embedding dimension =5, delay time =3, a significant decrease in entropy value indicates large-area degumming.

[0038] Sound wave reflection channel characteristics Multiple echo intervals are abnormal, and the calculated wall thickness fluctuates locally by more than 1.0 mm; C-scan imaging: Damage area > 150 cm², low edge gradient intensity, consistent with degumming characteristics.

[0039] Feature-weighted fusion formula: For internal degumming, α=0.3 and β=0.7 are used to enhance the ability to detect deep layers of sound wave reflection.

[0040] Step S5: Fault Identification and Lifespan Prediction Output of the multi-scale spatiotemporal graph neural network: Damage type: Large-area internal degumming Damage severity score: =0.91 Remaining service life: 47 days. The system will automatically trigger a Level 1 warning, and it is recommended to shut down for maintenance within 7 days.

[0041] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fault early warning method for dynamic data acquisition of wind power equipment, characterized in that, Includes the following steps: S1: Collect electromagnetic induction signals and acoustic wave reflection signals while the wind turbine blades are rotating; S2: Perform spatiotemporal synchronization processing on the electromagnetic induction signal and the acoustic wave reflection signal to obtain the blade damage signal; S3: Perform rotational fundamental frequency interference cancellation, aerodynamic noise separation, electromagnetic interference suppression, multi-source data spatiotemporal registration and fusion preprocessing on the collected blade damage signal to form the preprocessed blade damage signal; S4: Extract the electromagnetic induction channel fault features and the acoustic wave reflection channel fault features from the preprocessed blade damage signal, and fuse the electromagnetic induction channel fault features and the acoustic wave reflection channel fault features. S5: A multi-scale spatiotemporal graph neural network is used to identify the fused fault features, output the damage type, and perform quantitative analysis of the damage degree and prediction of the remaining service life.

2. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S2, the spatiotemporal synchronization process includes: Using the 1PPS signal generated by the GPS timing module as the global time reference, the electromagnetic induction signal and the acoustic wave reflection signal are synchronized to a precision of 100ns. The blade azimuth encoder signal is used as the spatial sampling trigger pulse to establish a two-dimensional coordinate system of blade spanwise and chordwise, with a spanwise resolution of 20mm and a chordwise resolution of 10mm, forming a dual-channel blade damage signal.

3. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S3, the rotational fundamental frequency interference cancellation employs an angular domain synchronous averaging method: the time-domain signal is resampled to an equal angular domain using the encoder signal, and the angular domain is used to synchronously average the signal over N consecutive revolutions. The specific formula is as follows: in: The target signal after angular domain synchronous averaging is used to characterize the pure blade damage signal after removing the rotating fundamental frequency interference. The blade rotation angle enables resampling mapping of the time-domain signal to the equal-angle domain; To ensure the statistical validity of the averaging process, the number of wind turbine blade rotations is included in the average.

4. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S3, the aerodynamic noise separation employs an adaptive beamforming spatial filtering method based on the MVDR criterion, with the specific formula being: in: This represents the optimal spatial filtering weight vector under the MVDR criterion. This is the inverse of the noise covariance matrix, used for noise power suppression; The spatial steering vector of the blade acoustic wave signal represents the direction and angle characteristics of acoustic wave propagation; The conjugate transpose of the guiding vector; It is the inverse matrix of the received signal covariance matrix, representing the spatial distribution characteristics of the signal.

5. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S3, the electromagnetic interference suppression employs a recursive least squares adaptive notch filter, using the inverter control signal as a reference input to adaptively estimate the interference propagation path and reduce interference. The specific formula is as follows: , in: The electromagnetic interference signal is adaptively estimated. It is the transpose of the weight vector of the time-varying filter, and the interference propagation path is updated in real time; It serves as the reference input signal for inverter control and provides a reference for interference suppression; This is the filtered, purified leaf damage signal; The original acquired signal contains electromagnetic interference.

6. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S3, the multi-source data spatiotemporal registration and fusion involves mapping the electromagnetic induction signal and the acoustic wave reflection signal to a unified blade two-dimensional coordinate system. Based on the blade geometric model and measured azimuth / distance parameters, a coordinate transformation matrix for the two channels is established. Bilinear interpolation resampling is then performed on the data from each channel to generate a two-dimensional data frame with uniform resolution. The specific formula is as follows: in: Let be the signal amplitude of the target pixel after bilinear interpolation; The signal amplitude at the original sampling point; The normalized coordinates of the target point in the two-dimensional coordinate system of the blade.

7. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S4, the extraction of electromagnetic induction channel fault features specifically refers to: S401: Extract the induced voltage signal generated by multi-frequency excitation at each frequency point: Impedance plane trajectory characteristics: The real resistance component and the imaginary reactance component of the induced voltage are plotted on the complex impedance plane, and the trajectory deflection angle and deflection amplitude at the damage point are used as characteristic quantities. Multi-frequency induction pair characteristic value: Defines the ratio of induced voltage at different frequencies. This ratio is sensitive to the depth of damage and can distinguish between surface and deep damage; Differential gradient Feature: Calculate the induced voltage gradient of adjacent receiver elements. The location of the gradient peak corresponds to the damage boundary; S402: Feature extraction of the eddy current disturbance wall. The disturbances caused by large-area debonding in the eddy current field are dispersed, and single-point features are not obvious. The local eddy current disturbance arrangement entropy is used, with the specific formula as follows: in: The entropy value of the local eddy current disturbance is used to quantify the disorder of the eddy current signal. The embedding dimension characterizes the phase space dimension of the reconstructed eddy current signal; The time interval for delaying the phase space reconstruction of the eddy current signal; This refers to the modal arrangement and combination of eddy current signals; This represents the probability of occurrence of the corresponding modal permutations and combinations.

8. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S4, the specific fault characteristics of the acoustic wave reflection channel are as follows: S403: Take the time-flight characteristics of the echo: Extraction of single-point echo time series after pulse compression: The delay time of the first echo corresponds to the reflection on the blade surface, and its change reflects the distance change caused by the deformation of the blade surface, with a resolution of ±0.1mm; Extraction of multiple echo time interval sequences: Multiple echo intervals correspond to blade wall thickness in: The wall thickness of the wind turbine blade material; The ultrasonic propagation speed in the blade composite material; The time interval between multiple acoustic echoes; localized thinning of the wall thickness >0.5 mm is a characteristic of damage; Echo amplitude attenuation spectrum: The echo amplitude of healthy fiberglass laminate decreases exponentially with the number of multiple echoes, and an abnormal attenuation coefficient indicates internal defects. S404: Take samples of the scanned damage imaging features: Extract the peak amplitude from the A-scan data of each spanwise × longitudinal scan point to construct a C-scan amplitude map, and extract the following from the C-scan map: Area of ​​the damaged region; Damage profile shape perturbation, regional splitting, degumming, deformation; Edge gradient intensity: The gradient intensity at the crack edge is much higher than that in the degummed and deformed zones.

9. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S4, the fusion of the fault features of the electromagnetic induction channel and the fault features of the acoustic wave reflection channel specifically employs feature-level weighted fusion, which normalizes and weights the electromagnetic induction feature vector and the acoustic wave reflection feature vector, and the core fusion formula is as follows: in: This is the fused fault feature vector; This is the characteristic vector of the electromagnetic induction channel; This represents the characteristic vector of the sound wave reflection channel; , These are adaptive weighting coefficients based on feature sensitivity, and α+β=1.

10. The fault early warning method for dynamic data acquisition of wind power equipment according to claim 1, characterized in that, In S5, the fault identification uses a multi-scale spatiotemporal graph neural network to complete damage classification and lifetime prediction. The core formula for quantifying the degree of damage is: in: Quantify the degree of leaf damage with a score; The number of scales for a multi-scale spatiotemporal graph neural network; The output of the graph neural network at the corresponding scale; This is the activation function for classification.

Images