Damage monitoring system and method for fan blade protective film based on audio signals

Abstract

The invention discloses a damage monitoring system for a fan blade protective film based on an audio signal, and the system comprises an acoustic sensor array which is formed by more than three microphones and is used for capturing acoustic emission signals at different positions of a blade through a triangulation positioning principle; the signal conditioning module is used for filtering low-frequency mechanical vibration and high-frequency electromagnetic interference in the acoustic emission signal and retaining an acoustic emission characteristic frequency band related to the damage of the blade protective film; the data processing module is used for processing the multi-channel audio streams in parallel and providing time-frequency domain data for subsequent Mel cepstrum analysis; the feature extraction algorithm module is used for converting the linear frequency spectrum into a Mel frequency scale to obtain a damage feature vector; and the SVM classification module is used for carrying out damage type classification based on the feature vectors and outputting damage levels and type labels. The invention further discloses a damage monitoring method for the fan blade protective film based on the audio signals. According to the invention, early warning of corrosion degree and maintenance decision support are realized through real-time audio signal analysis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fan damage monitoring, and particularly relates to a damage monitoring system and method for a fan blade protective film based on an audio signal. BACKGROUND

[0002] As a core component of wind energy conversion, the wind turbine blade is exposed to complex natural environment for a long time. With the increase of single machine capacity and blade length (the longest modern blade has exceeded 100 meters), the problem of leading edge corrosion is increasingly prominent. The International Energy Agency report shows that the leading edge corrosion leads to an annual power generation loss of up to 3%-5%, and the global annual economic loss caused thereby exceeds 1 billion euros. At present, the industry mainly evaluates the blade state through regular manual inspection, unmanned aerial vehicle visual detection or strain sensor monitoring, but these methods have limitations such as response lag, high cost or complex installation.

[0003] The current mainstream protection methods include: ① polyurethane elastomer protective film, which has good impact resistance but cannot quantify the damage degree; ② metal leading edge coating, which has strong wear resistance but changes the aerodynamic characteristics; and ③ regular coating repair, which needs to stop operation and has a long maintenance cycle. The monitoring methods mainly include: a) optical fiber sensor array, which can locate damage but has complex wiring; b) acoustic emission detection, which is sensitive to structural cracks but cannot identify surface corrosion; and c) image recognition algorithm, which is greatly affected by weather conditions and has high computing resource consumption.

[0004] The existing protective film scheme lacks embedded damage evaluation function and cannot realize real-time state feedback; the traditional monitoring technology either relies on physical contact type sensors (which affect the aerodynamic performance) or is significantly affected by environmental interference (such as optical detection failure in rainy and foggy weather); most methods need to stop detection, resulting in power generation loss and rising maintenance cost.

[0005] Therefore, a damage monitoring system and method for a fan blade protective film based on an audio signal are provided. SUMMARY

[0006] To solve the above problems in the prior art, the application provides a damage monitoring system and method for a fan blade protective film based on an audio signal, which realizes early warning of corrosion degree and maintenance decision support through real-time audio signal analysis.

[0007] The technical scheme for realizing the above purpose is as follows: The damage monitoring system for a fan blade protective film based on an audio signal according to one of the application comprises: An acoustic sensor array formed by an array of more than 3 microphones, used to capture acoustic emission signals at different positions of the blade through the triangulation principle; The signal conditioning module is used to filter out low-frequency mechanical vibrations and high-frequency electromagnetic interference in the acoustic emission signal, while retaining the acoustic emission characteristic frequency bands related to blade protective film damage. The data processing module is used to process multi-channel audio streams in parallel, providing time-frequency domain data for subsequent Mel (frequency scale based on human hearing characteristics) cepstral analysis; The feature extraction algorithm module is used to convert the linear spectrum into Mel frequency standard to obtain the damage feature vector; The SVM (Generalized Linear Classifier for Binary Classification) classification module is used to classify damage types based on feature vectors and output damage level and type label. The TDOA (Time Difference of Arrival) positioning module is used to determine the precise location of damage on the protective film of the wind turbine blades. The wireless transmission module is used to transmit damage data to the monitoring center via the LoRaWAN (a network protocol designed for battery-powered terminal devices) protocol. The damage data includes the damage type and damage level. The visualization and alarm module is used to intuitively display monitoring results and provide real-time early warnings.

[0008] Preferably, in the feature extraction algorithm module, a fast Fourier transform is performed on each frame of signal to convert the time-domain signal into a linear spectrum. Output the energy distribution of each frequency component: ; In the formula, Index for discrete frequency points; Mel frequency mapping will linear frequency Convert to nonlinear Mel frequency standard : ; By using dense filters in the low-frequency region and sparse filters in the high-frequency region to match human auditory perception, the energy of the filter bank is: ; In the formula, For the first Frequency domain response of a Mel filter; Take the logarithm of the filter bank energy: ; Simulates the nonlinear response of the human ear to sound intensity. right Perform a discrete cosine transform to obtain the Mel frequency cepstral coefficients. : ; In the formula, This represents the total number of Mel filter banks. an index of a current Mel filter; The first 13-dimensional coefficients of the Mel frequency cepstrum coefficients are taken to form a damage feature vector.

[0009] Preferably, in the SVM classification module, the damage level is divided into 0-5 levels, 0 corresponds to no damage, and 5 corresponds to a penetrating damage, and the type label includes crack, corrosion, lightning strike, scratch, icing, pollution, deformation and material aging.

[0010] Preferably, in the TDOA positioning module, the time delay difference of the sound wave received by the sensor array is used to calculate the distance of the sound source from the hub center in combination with the Doppler shift formula: ; In the formula, is the sound speed, is the blade rotation period, and is the whistle frequency shift boundary.

[0011] Preferably, in the visualization and alarm module, the visualization interface presents an integrated spectrum, a damage positioning map, and an energy trend curve. According to the damage severity, a buzzer, a short message or a platform pop-up alarm is triggered.

[0012] The second damage monitoring method of the fan blade protection film based on an audio signal of the application comprises: Step S1, capturing acoustic emission signals at different positions of the blade by arranging three MEMS (Micro Electro Mechanical System) microphones in a triangular shape; Step S2, filtering low-frequency mechanical vibration and high-frequency electromagnetic interference in the acoustic emission signals by using an NLMS (Normalized Least Mean Square) algorithm, and retaining acoustic emission characteristic frequency bands related to the damage of the blade protection film; Step S3, processing multiple-channel audio streams in parallel to provide time-frequency domain data for subsequent Mel cepstrum analysis; Step S4, converting the time-frequency domain data into linear frequency spectrum by using fast Fourier transform, and converting the linear frequency into nonlinear Mel frequency scale by using Mel frequency mapping, thereby obtaining Mel frequency cepstrum coefficients, and taking the first 13-dimensional coefficients of the Mel frequency cepstrum coefficients to form a damage feature vector; Step S5, classifying the damage type based on the feature vector, outputting the damage level and type label, and determining the accurate position of the damage on the fan blade protection film based on TDOA positioning; Step S6, finally visually displaying the monitoring results, i.e., the damage level and position information.

[0013] Preferably, in step S1, if the protective film is damaged, the air flow generates vortex through the crack or hole, exciting 2-8 kHz broadband noise, and the acoustic emission signals at different positions of the blade are captured by three MEMS microphones.

[0014] Preferably, in step S5, a fast Fourier transform is performed on each frame of signal to convert the time domain signal into a linear frequency spectrum , and the energy distribution of each frequency component is output: ; In the formula, is the index of the discrete frequency point; Mel frequency mapping converts the linear frequency to the nonlinear Mel frequency scale : ; The filter bank is dense in the low frequency region and sparse in the high frequency region, matching human ear perception, wherein the filter bank energy is: ; In the formula, is the frequency domain response of the th Mel filter; The filter bank energy is logarithmized: ; Analogous to the nonlinear response of the human ear to sound intensity, the discrete cosine transform is performed on to obtain the Mel frequency cepstral coefficient : ; In the formula, is the total number of Mel filter banks, is the serial number of the current Mel filter; The first 13-dimensional coefficients of the Mel frequency cepstral coefficient are taken to form the damage feature vector.

[0015] Preferably, in step S5, the damage level is divided into 0-5 levels, 0 level corresponds to no damage, and 5 level corresponds to penetrating damage, and the type label includes crack, corrosion, lightning strike, scratch, icing, pollution, deformation and material aging; The time delay difference of the received acoustic waves by the sensor array is used to calculate the distance of the acoustic source from the hub center , and the precise position of the damage on the protective film of the fan blade is determined: ; In the formula, is the speed of sound, is the blade rotation period, and The whistle frequency shift boundary.

[0016] Compared with the prior art, the beneficial effects of the present application are: the present application realizes non-contact real-time monitoring of fan blade protective film damage and early identification of millimeter-level precision of protective film damage by capturing acoustic emission signals at different positions of the blade through the triangular layout of the three MEMS microphones, shortens the detection response time to 30 seconds, reduces the false alarm rate caused by environmental interference through multi-sensor fusion, does not require an external power line, the overall weight of the system is small, the influence on the dynamic balance of the blade is negligible, the comprehensive maintenance cost is reduced by 60%, and the service life of the protective film is prolonged by about 1.8 times. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, which together with the embodiments of the present application, serve to explain the present application, and do not constitute a limitation on the present application. In the drawings: Figure 1 is a module diagram of the fan blade protective film damage monitoring system based on audio signals of the present application; Figure 2 is a flowchart of the fan blade protective film damage monitoring method based on audio signals of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0019] As shown in Figure 1 , the fan blade protective film damage monitoring system based on audio signals comprises: an acoustic sensor array 1, a signal conditioning module 2, a data processing module 3, a feature extraction algorithm module 4, an SVM classification module 5, a TDOA positioning module 6, a wireless transmission module 7, and a visualization and alarm module 8.

[0020] The acoustic sensor array 1 is formed by an array of more than three microphones, and is used to capture acoustic emission signals at different positions of the blade through the triangulation principle.

[0021] The signal conditioning module 2 is used to filter out low-frequency mechanical vibrations and high-frequency electromagnetic interferences in the acoustic emission signals, and to retain acoustic emission characteristic frequency bands related to the damage of the blade protective film.

[0022] The data processing module 3 is used to process multiple-channel audio streams in parallel to provide time-frequency domain data for subsequent Mel cepstrum analysis.

[0023] Feature extraction algorithm module 4 is used to convert linear frequency spectrum into Mel frequency scale to obtain damage feature vector.

[0024] In the embodiment, fast Fourier transform is performed on each frame of signal to convert time domain signal into linear frequency spectrum , and energy distribution of each frequency component is output: ; In the formula, is discrete frequency point index; Mel frequency mapping converts linear frequency into nonlinear Mel frequency scale : ; The filter bank is dense in low frequency area and sparse in high frequency area, which matches human ear perception, wherein the filter bank energy is: ; In the formula, is frequency domain response of the th Mel filter; The filter bank energy is logarithmized: ; The Mel frequency cepstral coefficient is obtained by simulating the nonlinear response of human ear to sound intensity, performing discrete cosine transform on : ; In the formula, is total number of Mel filter bank, is serial number of current Mel filter; The first 13-dimensional coefficients of Mel frequency cepstral coefficient are taken to form damage feature vector.

[0025] SVM classification module 5 is used to classify damage type based on feature vector, and output damage level and type label.

[0026] In the embodiment, the damage level is divided into 0-5 levels, 0 level corresponds to no damage, and 5 level corresponds to penetrating damage, and the type label includes crack, corrosion, lightning strike, scratch, icing, pollution, deformation and material aging.

[0027] TDOA positioning module 6 is used to determine the accurate position of damage on the protective film of the fan blade.

[0028] In the embodiment, the time delay difference of sound waves received by the sensor array is used to calculate the distance of the sound source from the hub center in combination with the Doppler shift formula. ; wherein, is the sound speed, is the blade rotation period, and is the whistle frequency shift boundary.

[0029] The wireless transmission module 7 is configured to transmit the damage data to the monitoring center through the LoRaWAN protocol, wherein the damage data comprises a damage type and a damage level.

[0030] The visualization and alarm module 8 is configured to visually display the monitoring results and realize real-time early warning.

[0031] In the embodiment, the visualization interface presents an integrated spectrum graph, a damage positioning map, and an energy trend curve. The buzzer, short message, or platform pop-up alarm is triggered according to the damage severity.

[0032] As shown in Figure 2 , the fan blade protection film damage monitoring method based on an audio signal comprises the following steps: Step S1: capturing acoustic emission signals at different positions of the blade through three MEMS microphones arranged in a triangular shape.

[0033] In the embodiment, if the protection film is damaged, the airflow generates vortexes through cracks or holes, exciting 2-8 kHz broadband noise, and the acoustic emission signals at different positions of the blade are captured through the three MEMS microphones.

[0034] Step S2: filtering low-frequency mechanical vibrations and high-frequency electromagnetic interferences in the acoustic emission signals using the NLMS algorithm, and retaining acoustic emission characteristic frequency bands related to the blade protection film damage.

[0035] Step S3: parallel processing of multi-channel audio streams to provide time-frequency domain data for subsequent Mel cepstrum analysis; Step S4: converting the time-frequency domain data into linear frequency spectrum through fast Fourier transform, and converting the linear frequency into nonlinear Mel frequency scale through Mel frequency mapping, thereby obtaining Mel frequency cepstrum coefficients, taking the first 13-dimensional coefficients of the Mel frequency cepstrum coefficients, and constructing a damage feature vector.

[0036] In the embodiment, fast Fourier transform is performed on each frame of signal to convert the time domain signal into linear frequency spectrum , and the energy distribution of each frequency component is output: ; wherein, is the discrete frequency point index; Mel frequency mapping converts the linear frequency Converting to a nonlinear Mel frequency scale : ; By filtering the low-frequency area densely and the high-frequency area sparsely, the human ear perception is matched, wherein the filter bank energy is: ; In the formula, is the frequency domain response of the Mth Mel filter; Taking the logarithm of the filter bank energy: ; Simulating the nonlinear response of the human ear to sound intensity, Discrete cosine transform is performed on to obtain Mel frequency cepstral coefficients : ; In the formula, is the total number of Mel filter banks, is the serial number of the current Mel filter; Take the first 13-dimensional coefficients of the Mel frequency cepstral coefficients to form a damage feature vector.

[0037] Step S5, based on the feature vector, the damage type classification is carried out, the damage level and type label are output, and the accurate position of the damage on the protective film of the fan blade is determined based on TDOA positioning.

[0038] In the embodiment, the damage level is divided into 0-5 levels, 0 corresponds to no damage, and 5 corresponds to a penetrating damage, and the type label includes cracks, corrosion, lightning strikes, scratches, icing, pollution, deformation and material aging; The time delay difference of the sound wave received by the sensor array is used to calculate the distance of the sound source from the hub center , and then the accurate position of the damage on the protective film of the fan blade is determined: ; In the formula, is the sound speed, is the blade rotation period, and are the whistle frequency shift boundaries.

[0039] Step S6, finally, the monitoring results, i.e. the damage level and position information, are visually displayed.

[0040] ​Finally, it should be noted that the above is only the preferred embodiment of the present application, and is not intended to limit the present application, although the foregoing embodiments of the present application are described in detail, for those skilled in the art, it still can be modified to the technical solution recorded in the foregoing embodiments, or equivalent replacement of some technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A damage monitoring system for wind turbine blade protective films based on audio signals, characterized in that, include: An acoustic sensor array, consisting of three or more microphones, is used to capture acoustic emission signals from different positions on the blade using the principle of triangulation. The signal conditioning module is used to filter out low-frequency mechanical vibrations and high-frequency electromagnetic interference in the acoustic emission signal, while retaining the acoustic emission characteristic frequency bands related to blade protective film damage. The data processing module is used to process multi-channel audio streams in parallel, providing time-frequency domain data for subsequent Mel cepstral analysis; The feature extraction algorithm module is used to convert the linear spectrum into Mel frequency standard to obtain the damage feature vector; The SVM classification module is used to classify damage types based on feature vectors and output damage level and type label. The TDOA positioning module is used to determine the precise location of damage to the protective film on the wind turbine blades. The wireless transmission module is used to transmit damage data to the monitoring center via the LoRaWAN protocol. The damage data includes the damage type and damage level. The visualization and alarm module is used to intuitively display monitoring results and provide real-time early warnings.

2. The damage monitoring system for wind turbine blade protective film based on audio signals according to claim 1, characterized in that, In the feature extraction algorithm module, a Fast Fourier Transform is performed on each frame of signal to convert the time-domain signal into a linear spectrum. Output the energy distribution of each frequency component: ； In the formula, Index for discrete frequency points; Mel frequency mapping will linear frequency Convert to nonlinear Mel frequency standard : ； By using dense filters in the low-frequency region and sparse filters in the high-frequency region to match human auditory perception, the energy of the filter bank is: ； In the formula, For the first Frequency domain response of a Mel filter; Take the logarithm of the filter bank energy: ； Simulates the nonlinear response of the human ear to sound intensity. right Perform a discrete cosine transform to obtain the Mel frequency cepstral coefficients. : ； In the formula, This represents the total number of Mel filter banks. This is the current Mel filter number; The first 13 dimensions of the Mel frequency cepstral coefficients are used to construct the damage feature vector.

3. The damage monitoring system for wind turbine blade protective film based on audio signals according to claim 1, characterized in that, In the SVM classification module, the damage level is divided into 0-5 levels, with level 0 corresponding to no damage and level 5 corresponding to penetrating damage. The type labels include cracks, corrosion, lightning strikes, scratches, icing, contamination, deformation, and material aging.

4. The damage monitoring system for wind turbine blade protective film based on audio signals according to claim 1, characterized in that, In the TDOA positioning module, the time delay difference of the sound waves received by the sensor array is used to calculate the distance between the sound source and the center of the wheel hub using the Doppler frequency shift formula: ； In the formula, For the speed of sound, The blade rotation period, and This is the boundary of the whistle frequency shift.

5. The damage monitoring system for wind turbine blade protective film based on audio signals according to claim 1, characterized in that, The visualization and alarm module displays an integrated spectrum diagram, a damage location map, and an energy trend curve in its visualization interface. The alarm will be triggered by a buzzer, SMS, or platform pop-up depending on the severity of the damage.

6. A method for damage monitoring of wind turbine blade protective film based on audio signals, characterized in that, include: Step S1: Acoustic emission signals from different positions on the blade are captured using three MEMS microphones arranged in a triangular layout. Step S2: The NLMS algorithm is used to filter out low-frequency mechanical vibration and high-frequency electromagnetic interference in the acoustic emission signal, while retaining the acoustic emission characteristic frequency band related to blade protective film damage. Step S3: Parallel processing of multi-channel audio streams to provide time-frequency domain data for subsequent Mel cepstral analysis; Step S4: Convert the time-frequency domain data into a linear spectrum through Fast Fourier Transform, and convert the linear frequency into a nonlinear Mel frequency standard through Mel frequency mapping, thereby obtaining the Mel frequency cepstral coefficients. Take the first 13 dimensions of the Mel frequency cepstral coefficients to form the damage feature vector. Step S5: Classify the damage type based on the feature vector, output the damage level and type label, and determine the precise location of the damage in the wind turbine blade protective film based on TDOA positioning. Step S6 finally displays the monitoring results visually, namely the damage level and location information.

7. The method for damage monitoring of wind turbine blade protective film based on audio signals according to claim 6, characterized in that, In step S1, if the protective film is damaged, the airflow generates eddies through cracks or holes, which excite 2-8kHz broadband noise. The acoustic emission signals at different positions of the blade are captured by three MEMS microphones.

8. The method for damage monitoring of wind turbine blade protective film based on audio signals according to claim 6, characterized in that, In step S4, a Fast Fourier Transform is performed on each frame of signal to convert the time-domain signal into a linear spectrum. Output the energy distribution of each frequency component: ； In the formula, Index for discrete frequency points; Mel frequency mapping will linear frequency Convert to nonlinear Mel frequency standard : ； By using dense filters in the low-frequency region and sparse filters in the high-frequency region to match human auditory perception, the energy of the filter bank is: ； In the formula, For the first Frequency domain response of a Mel filter; Take the logarithm of the filter bank energy: ； Simulates the nonlinear response of the human ear to sound intensity. right Perform a discrete cosine transform to obtain the Mel frequency cepstral coefficients. : ； In the formula, This represents the total number of Mel filter banks. This is the current Mel filter number; The first 13 dimensions of the Mel frequency cepstral coefficients are used to construct the damage feature vector.

9. The method for damage monitoring of wind turbine blade protective film based on audio signals according to claim 6, characterized in that, In step S5, the damage level is divided into 0-5 levels, with level 0 corresponding to no damage and level 5 corresponding to penetrating damage. The type labels include cracks, corrosion, lightning strikes, scratches, icing, contamination, deformation, and material aging. The distance from the sound source to the center of the wheel hub is calculated by using the time delay difference of the sound waves received by the sensor array and combining it with the Doppler frequency shift formula. This allows for the precise determination of the location of the damage on the protective film of the wind turbine blades. ； In the formula, For the speed of sound, The blade rotation period, and This is the boundary of the whistle frequency shift.

Images