A blade detection system and method based on piezoelectric film and microphone array

By acquiring and decoupling the vibration and sound signals of the blades in real time using a piezoelectric film and microphone array system, and using machine learning for detection, the problem of not being able to detect the blade status in real time in existing technologies has been solved, and efficient blade status assessment and damage diagnosis have been achieved.

CN118730519BActive Publication Date: 2024-11-08THREE GORGES NEW ENERGY (XIUYAN MANCHU AUTONOMOUS COUNTY) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411225818.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-08
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

Existing technologies cannot perform real-time, long-term, full life-cycle monitoring during wind turbine operation, nor can they quantitatively assess the operating status and structural design of the blades. Furthermore, the monitoring equipment is large, expensive, and has limited applicability.

Method used

A piezoelectric thin-film sensor is used to acquire blade vibration signals, and a microphone array captures sound signals. By decoupling acoustic-structure interaction algorithm and machine learning, and using a convolutional neural network for detection, real-time analysis of the blade state is achieved.

Benefits of technology

It enables real-time, long-term blade inspection during wind turbine operation, improving inspection efficiency, reducing unplanned downtime and maintenance costs, and detecting blade operational problems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118730519B_ABST
    Figure CN118730519B_ABST
Patent Text Reader

Abstract

The present invention discloses a blade detection system and method based on piezoelectric film and microphone array, and relates to the field of new energy equipment detection technology. A blade detection system based on piezoelectric film and microphone array includes a vibration signal acquisition module, a sound signal acquisition module, a communication module, a signal processing module, a data decoupling module and a detection and diagnosis module. A blade detection method based on piezoelectric film and microphone array includes S100, data acquisition; S200, data preprocessing; S300, feature extraction; S400, data decoupling; S500, detection and diagnosis. The present invention realizes blade detection without stopping, greatly improves detection efficiency, reduces power generation loss, and can detect problems with blades in operation, reduce the number of unplanned shutdowns, and thus reduce maintenance costs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of new energy equipment detection, and in particular to a blade detection method based on a piezoelectric film and a microphone array. Background Art

[0002] Wind turbine blades are the core components of wind turbines (referred to as wind turbines), and their performance directly affects the efficiency and life of the entire wind power system. During the operation of wind turbines, the blades not only have to withstand complex aerodynamic loads, but also have to deal with environmental factors such as temperature changes, humidity, salt spray erosion, lightning strikes, etc. These factors may cause various types of defects in the blades, such as:

[0003] 1. Cracks: During operation, especially under fatigue loads, blades are prone to tiny cracks, fissures and defects, which may eventually lead to blade breakage or failure.

[0004] 2. Delamination: There is poor bonding during the blade manufacturing process. Under the action of long-term alternating loads, delamination may occur on the blade surface, leading and trailing edges, main beam, web and other parts.

[0005] 3. Peeling: The coating or protective layer on the surface of the blade will be exposed to the atmospheric environment due to wear and peeling, causing oxidation and corrosion, thus affecting its safety performance.

[0006] 4. Lightning damage: Blades are very susceptible to lightning strikes during thunderstorms, which can cause cracks, fissures, and ablation on the surface and inside of the blades.

[0007] 5. Wear and shedding: The blade surface coating and protective layer are not wear-resistant enough, or are eroded and worn by natural factors such as wind, rain, and snow during long-term operation, resulting in a decline in blade performance.

[0008] 6. Icing: When supercooled water droplets in the air come into contact with the running wind turbine blades under specific meteorological conditions, they quickly freeze and adhere to the surface of the blades due to the low temperature.

[0009] 7. Corrosion: When the blade surface is corroded by factors such as chemicals, seawater or air pollution, corrosion damage will occur, causing the blade performance to decline or fail.

[0010] 8. Material aging: As the use time increases, the mechanical properties of the blade material gradually decrease, which may lead to blade failure.

[0011] If these defects are not discovered and repaired in time, they may cause the performance of the blades to decline, affecting the efficiency of wind energy utilization; increase the risk of mechanical failure, which in turn affects the stable operation of the wind turbine; increase maintenance costs, which may cause the entire wind power system to shut down; or even cause breakage, posing a threat to personnel and equipment. Therefore, regular and systematic inspection of wind turbine blades is crucial and an indispensable part of wind power maintenance management.

[0012] The existing wind turbine blade inspection technology is mainly non-destructive testing technology (NDT), which is specifically divided into:

[0013] 1. Visual inspection: Use a high-definition camera to capture images of the blade surface, and then use image processing technology to detect and analyze defects and wear on the blade surface. When there are interference factors such as dust and stains in the detection environment, it will affect the detection results, and visual inspection can only detect cracks on the blade surface.

[0014] 2. Ultrasonic testing: Utilizing the propagation characteristics of ultrasonic waves in materials, ultrasonic reflection and scattering signals are detected to determine whether there are defects inside the material. Ultrasonic testing can detect cracks, delamination, voids and other defects inside the blades, and has the advantages of fast detection speed and high accuracy. However, this technology is limited by the thickness of the material and the shape of the defect, and has low detection sensitivity for vertical or steep defects. Ultrasonic testing has certain requirements for the flatness of the working surface. For wind turbine blades with complex structures and anisotropic materials, ultrasonic testing has limitations.

[0015] 3. Eddy current testing: Utilize the inductive effect of eddy current in conductive materials to determine surface and near-surface defects of materials by detecting changes in eddy current. Eddy current testing is suitable for detecting cracks, rust and other defects on the blade surface. Eddy current testing is mainly suitable for detecting surface and near-surface defects, but has weaker detection capabilities for deep defects and is easily affected by the conductive properties of the material.

[0016] 4. Infrared thermal imaging detection: Defects are found by detecting the temperature distribution on the blade surface. It is suitable for occasions where factors such as ambient temperature and wind speed have a great influence. When the external ambient temperature fluctuates greatly or the wind speed is unstable, the detection results will be inaccurate. When detecting internal defects of the blade, especially deep defects, it is impossible to detect accurately.

[0017] 5. Fiber Bragg Grating Sensing Technology: Analyze the stress condition by measuring the change of the grating transmission spectrum. Fiber Bragg Grating sensors can detect the displacement, deformation, stress and other information of the blade. The applicability of measuring dynamic signals (such as vibration signals) and certain environmental parameters (such as humidity and wind speed) needs to be improved.

[0018] 6. Microwave detection method: Utilize the propagation characteristics of microwave signals inside dielectric materials to detect internal defects of blades. This method is easily interfered by the external environment, such as electromagnetic interference, which will affect the reliability of the detection results. In addition, there are difficulties in the qualitative and quantitative aspects of defects, especially in the detection of deep and hidden defects.

[0019] 7. X-ray detection: X-rays can penetrate materials and detect the absorption and scattering of X-rays in materials to determine whether there are defects inside the material. X-ray detection can detect cracks, inclusions and other defects inside the blades, and has the advantages of high detection accuracy and visualization. However, the detection equipment of this method is bulky and expensive, which limits its popularity and application to a certain extent. Secondly, there are certain limitations in the detection of common delamination defects in wind turbine blades and cracks parallel to the blade surface.

[0020] 8. Magnetic particle testing: Utilizes the magnetic leakage effect of the magnetic field on the surface and near-surface defects of the material to determine the existence and location of defects through the adsorption and accumulation of magnetic particles. It is suitable for detecting cracks, folds and other defects on the blade surface. This method cannot detect non-magnetic materials such as aluminum, magnesium, titanium, etc., as well as shallow scratches on the surface, deeply buried holes, and delamination and folds with an angle of less than 20° with the workpiece surface.

[0021] In general, most of the existing technologies require shutdown detection, and the detection location is limited to the page or the inside. More importantly, it can only detect blade failures and cannot perform quantitative evaluation on the operating status and structural design of the blades.

[0022] Therefore, those skilled in the art are committed to developing a blade detection system and method based on a piezoelectric film and a microphone array. Summary of the invention

[0023] In view of the above-mentioned defects in the prior art, the technical problem to be solved by the present invention is how to perform real-time, long-term and full life cycle detection on the blades when the wind turbine is running.

[0024] When the blade is in operation, the vibration signal is determined by the internal structure of the blade and the condition of the blade surface, while the sound signal is the sound of the friction between the blade surface and the air, but it is also mixed with the creaking sound of the mechanical structure of the equipment. The blade vibration information is obtained through the piezoelectric film sensor, and the blade is captured through the microphone array. The vibration signal and the sound signal are then decoupled. The acoustic-solid coupling algorithm is used to analyze their mutual influence relationship. The coupling relationship is decoupled using a machine learning algorithm. The output vibration signal reflects the vibration condition inside the blade, and the output sound signal reflects the condition of the blade surface. Finally, a convolutional neural network is used to analyze the decoupled features, determine the monitoring results, and analyze the blade status.

[0025] In one embodiment of the present invention, a blade detection system based on a piezoelectric film and a microphone array is provided, comprising:

[0026] The vibration signal acquisition module uses a piezoelectric film sensor to collect the vibration signal of the blade and sends it to the signal processing module for processing;

[0027] The sound signal acquisition module uses a microphone array to collect the sound signal of the blade, that is, the sound generated by the friction between the blade and the air, and sends it to the signal processing module for processing;

[0028] The communication module is responsible for the communication among the vibration signal acquisition module, the sound signal acquisition module and the signal processing module;

[0029] The signal processing module pre-processes the vibration signal and the sound signal, then extracts the features, outputs the voiceprint information and vibration data in the sound, including frequency and amplitude, and sends them to the data decoupling module;

[0030] The data decoupling module uses the sound-solid coupling algorithm to link the voiceprint information with the vibration signal, eliminate the mutual influence of the voiceprint information and the vibration signal, decouple them, output the sound signal and vibration signal without coupling relationship, and then send them to the detection and diagnosis module;

[0031] The detection and diagnosis module uses a blade detection model based on a convolutional neural network to analyze uncoupled sound and vibration signals, determine monitoring results, and analyze blade status;

[0032] The vibration signal acquisition module and the sound signal acquisition module are respectively connected to the communication module for communication, and the communication module, the signal processing module, the data decoupling module and the detection and diagnosis module are connected to each other for communication in sequence;

[0033] The vibration signal acquisition module and the sound signal acquisition module respectively collect the vibration signal and sound signal of the blade, send them to the signal processing module for preprocessing through the communication module, and then send them to the data decoupling module for feature decoupling. The detection and diagnosis module then completes the blade status analysis and determines the detection results.

[0034] Optionally, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, the piezoelectric film sensor is arranged to be attached to the blade of the wind turbine.

[0035] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, the distribution position is determined by the vibration mode of the blade.

[0036] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, the vibration modes of the blade include torsional deformation around the axis of the blade, flapping vibration (pure bending vibration) at the natural frequency of the low-frequency part, and forced vibration.

[0037] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, the piezoelectric film sensor is attached to the position where the blade deformation is the largest under torsional deformation, pure bending vibration and forced vibration.

[0038] Furthermore, in the blade detection system based on the piezoelectric film and the microphone array in the above embodiment, the position where the deformation amount of the blade is the largest is provided by the blade manufacturer.

[0039] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, the vibration signal collected by the piezoelectric film sensor is sent to the signal processing module through the communication module after analog-to-digital conversion.

[0040] Optionally, in the blade detection system based on piezoelectric film and microphone array in any of the above embodiments, the microphone array adopts an irregular array of array element positions to reduce sound signal distortion caused by aliasing.

[0041] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, the microphone array is based on an elliptical array, and a non-dominated sorting genetic algorithm (NSGA-II) is used to optimize the array to reduce distortion and signal overlap.

[0042] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, NSGA-II uses the objective function as a reference quantity based on the determined constraints, adds the crowding analysis, obtains the optimal number of array elements and array element spacing, and uses the Delay and Sum method as a beamformer. The objective function formula is as follows:

[0043] (1)

[0044] in, Obj_Fuction is the objective function, Band mainlobe is the main lobe bandwidth, Gain mainlobe is the maximum main lobe gain, Gain sidelobe is the maximum sidelobe gain, Number Mic is the number of microphone array elements.

[0045] Optionally, in the blade detection system based on piezoelectric film and microphone array in any of the above embodiments, the communication module adopts wireless communication technology.

[0046] Optionally, in the blade detection system based on piezoelectric film and microphone array in any of the above embodiments, the signal processing module processes the sound signal including preprocessing, wavelet packet analysis, energy feature extraction, normalization processing and construction of feature vectors, and the feature vectors serve as the basis for fault identification.

[0047] Furthermore, in the blade detection system based on piezoelectric film and microphone array in the above embodiment, wavelet packet analysis is based on wavelet analysis to perform detailed decomposition of low-frequency components and high-frequency components, and decompose the spectrum diagram of the sound signal.

[0048] Optionally, in the blade detection system based on piezoelectric film and microphone array in any of the above embodiments, the decoupling of the data decoupling module includes:

[0049] S1000, extract key features, extract the frequency band energy of the pre-processed sound signal and vibration signal respectively through the feature extraction technology of wavelet packet analysis, perform normalization processing, and obtain the voiceprint information and vibration data of the sound, including amplitude and frequency;

[0050] S2000, signal association, using an acoustic-to-sound coupling algorithm, based on a wave equation and an elastic equation, to associate the voiceprint information with the vibration data, thereby eliminating the mutual influence between the voiceprint information and the vibration data. The wave equation describes the propagation characteristics of sound waves in a medium, and the elastic equation describes the stress-strain relationship of a solid material.

[0051] S3000, signal decoupling, uses machine learning generative adversarial network (GAN) algorithm to solve the acoustic-solid coupling model and achieve decoupling of sound signals and vibration signals.

[0052] Based on any of the above embodiments, in another embodiment of the present invention, a leaf detection method based on a piezoelectric film and a microphone array is provided, comprising the following steps:

[0053] S100, data acquisition, using a piezoelectric film sensor and a microphone array sensor to collect original signals of the blade, including vibration signals and sound signals;

[0054] S200, data preprocessing, preprocessing the original signal, including denoising, filtering, and normalization;

[0055] S300, feature extraction, extracting feature information from the preprocessed data, and outputting voiceprint information and vibration data in the sound, including frequency and amplitude;

[0056] S400, data decoupling, uses the acoustic-to-sound coupling algorithm to link the voiceprint information with the vibration data, eliminates the mutual influence of the voiceprint information and the vibration signal, decouples them, and outputs the sound signal and vibration signal without coupling relationship;

[0057] S500, detection and diagnosis, building and training a blade detection model based on a convolutional neural network, parsing uncoupled sound signals and vibration signals, determining monitoring results and analyzing blade status, diagnosing blade damage, determining whether the blade is damaged, and the type and extent of the damage.

[0058] Optionally, in the leaf detection method based on piezoelectric film and microphone array in the above embodiment, step S100 includes:

[0059] S110, setting a piezoelectric film sensor, wherein the piezoelectric film sensor is set and attached to the blade, and attached to the position where the deformation of the blade is the largest;

[0060] S120, setting a microphone array, wherein the microphone array adopts an irregular array of array element positions, is based on an elliptical array, and uses a non-dominated sorting genetic algorithm (NSGA-II) to optimize the array;

[0061] S130, data acquisition, the piezoelectric film sensor and the microphone array respectively collect the original signals of the blade, including vibration signals and sound signals. At this time, the original signals contain noise.

[0062] Optionally, in the leaf detection method based on piezoelectric film and microphone array in the above embodiment, step S200 includes:

[0063] S210, signal denoising, removing noise from the original signal of the blade, that is, the sound signal and the vibration signal, to restore the integrity of the original signal;

[0064] S220, signal filtering, extracting useful information from the original signal of the blade and suppressing noise and interference;

[0065] S230, data normalization, converting and discretizing the data format to standardize the data for easy processing and use.

[0066] Optionally, in the leaf detection method based on piezoelectric film and microphone array in the above embodiment, step S300 includes:

[0067] S310, wavelet packet analysis, based on wavelet analysis, makes detailed decomposition of low-frequency components and high-frequency components, and decomposes the spectrum of the sound signal;

[0068] S320, energy feature extraction, using empirical mode decomposition (EMD) to decompose the signal into intrinsic mode functions (IMFs) for feature extraction;

[0069] S330, normalization processing, mapping the data to the interval [0, 1], converting it into a dimensionless expression, eliminating the dimension effect, and facilitating comparison between different features;

[0070] S340, constructing an eigenvector. Calculate the covariance matrix of the normalized data, perform eigenvalue decomposition, and obtain eigenvalues ​​and corresponding eigenvectors.

[0071] Optionally, in the leaf detection method based on piezoelectric film and microphone array in any of the above embodiments, step S400 includes:

[0072] S410, numerical analysis, using finite element analysis to build a blade model, and simulating the interaction and relationship between the soundprint information and the vibration data through simulation;

[0073] S420, signal correlation, establishing an acoustic-solid coupling algorithm, and using the wave equation and the elastic equation to link the voiceprint information with the vibration data;

[0074] S430, signal decoupling, based on the extracted features, using a machine learning algorithm to solve and classify the acoustic-solid coupling algorithm, to achieve decoupling of the sound signal and the vibration signal, and to obtain a sound signal and a vibration signal without a coupling relationship.

[0075] Optionally, in the leaf detection method based on piezoelectric film and microphone array in any of the above embodiments, step S500 includes:

[0076] S510, constructing a leaf detection model, constructing a leaf detection model based on a convolutional neural network, setting the number of layers of the convolutional neural network, selecting ReLU as an activation function, adding a fully connected layer, and summarizing the extracted features;

[0077] S520, leaf detection model training, collecting training set data to train the leaf detection model, adjusting hyperparameters, including learning rate and batch size, to optimize model performance, using the test set to evaluate the generalization ability of the leaf detection model, checking the final accuracy and confusion matrix, and completing the training of the leaf detection model;

[0078] S530, deploying a blade detection model, deploying the trained blade detection model, determining monitoring results and analyzing blade status, diagnosing blade damage, determining whether the blade is damaged, and the type and extent of the damage;

[0079] S540, updating the blade detection model, regularly checking the performance of the blade detection model, and retraining the blade detection model according to the blade aging condition, that is, executing step S520.

[0080] The present invention collects the vibration signal and sound signal of the blade, and then processes and decouples the vibration signal and the sound signal to perform detection and diagnosis, thereby realizing blade detection without stopping, greatly improving detection efficiency, reducing power generation loss, and being able to detect problems with blades in operation, reducing the number of unplanned shutdowns, thereby reducing maintenance costs.

[0081] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 is a schematic structural diagram illustrating a blade detection system based on a piezoelectric film and a microphone array according to an exemplary embodiment;

[0083] Figure 2 is a flow chart illustrating a blade detection method based on a piezoelectric film and a microphone array according to an exemplary embodiment. DETAILED DESCRIPTION

[0084] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0085] In the drawings, components with the same structure are indicated by the same numerical reference numerals, and components with similar structures or functions are indicated by similar numerical reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the present invention does not limit the size and thickness of each component. In order to make the illustration clearer, the thickness of the components is schematically exaggerated in some places in the drawings.

[0086] The present invention designs a blade detection system based on piezoelectric film and microphone array, such as Figure 1 As shown, including:

[0087] The vibration signal acquisition module attaches the piezoelectric film sensor to the blade of the fan. The distribution position is determined by the vibration mode of the blade. The vibration mode of the blade includes torsional deformation around the axis of the blade, flapping vibration (pure bending vibration) and forced vibration at the natural frequency of the low-frequency part. The piezoelectric film sensor is attached to the position where the blade deformation is the largest under torsional deformation, pure bending vibration and forced vibration. The position where the blade deformation is the largest is provided by the blade manufacturer. The piezoelectric film sensor collects the vibration signal of the blade and sends it to the signal processing module through the communication module after analog-to-digital conversion.

[0088] The sound signal acquisition module uses a microphone array. The microphone array adopts an irregular array of array element positions to reduce the sound signal distortion caused by aliasing, collects the sound signal of the blade, that is, the sound generated by the friction between the blade and the air, and sends it to the signal processing module for processing; the microphone array is based on an elliptical array and uses a non-dominated sorting genetic algorithm (NSGA-II) to optimize the array formation to reduce distortion and signal overlap. Based on the determined constraints, NSGA-II uses the objective function as a reference and adds crowding analysis to obtain the optimal number of array elements and array element spacing. The delay and sum method is used as a beamformer. The objective function formula is as follows:

[0089] (1)

[0090] in, Obj_Fuction is the objective function, Band mainlobe is the main lobe bandwidth, Gain mainlobe is the maximum main lobe gain, Gain sidelobe is the maximum sidelobe gain, Number Mic is the number of microphone array elements.

[0091] The communication module adopts wireless communication technology and is responsible for the communication among the vibration signal acquisition module, the sound signal acquisition module and the signal processing module.

[0092] The signal processing module preprocesses the vibration signal and the sound signal, and then extracts the features, outputs the voiceprint information and vibration data in the sound, including frequency and amplitude, and sends them to the data decoupling module; the signal processing module processes the sound signal including preprocessing, wavelet packet analysis, energy feature extraction, normalization and construction of feature vectors. The feature vectors serve as the basis for fault identification. The wavelet packet analysis is based on wavelet analysis to make a detailed decomposition of the low-frequency and high-frequency components, and decompose the spectrum of the sound signal.

[0093] The data decoupling module uses the acoustic-solid coupling algorithm to link the voiceprint information with the vibration signal, eliminate the mutual influence of the voiceprint information and the vibration signal, decouple, output the sound signal and vibration signal without coupling relationship, and then send them to the detection and diagnosis module; the decoupling of the data decoupling module includes:

[0094] S1000, extract key features, extract the frequency band energy of the pre-processed sound signal and vibration signal respectively through the feature extraction technology of wavelet packet analysis, perform normalization processing, and obtain the voiceprint information and vibration data of the sound, including amplitude and frequency;

[0095] S2000, signal association, using an acoustic-to-sound coupling algorithm, based on a wave equation and an elastic equation, to associate the voiceprint information with the vibration data, thereby eliminating the mutual influence between the voiceprint information and the vibration data. The wave equation describes the propagation characteristics of sound waves in a medium, and the elastic equation describes the stress-strain relationship of a solid material.

[0096] S3000, signal decoupling, uses machine learning generative adversarial network (GAN) algorithm to solve the acoustic-solid coupling model and achieve decoupling of sound signals and vibration signals.

[0097] The detection and diagnosis module uses a blade detection model based on a convolutional neural network to parse uncoupled sound and vibration signals, determine monitoring results, and analyze blade status.

[0098] The vibration signal acquisition module and the sound signal acquisition module are respectively connected to the communication module for communication, and the communication module, the signal processing module, the data decoupling module and the detection and diagnosis module are connected to each other for communication in sequence.

[0099] The vibration signal acquisition module and the sound signal acquisition module respectively collect the vibration signal and sound signal of the blade, send them to the signal processing module for preprocessing through the communication module, and then send them to the data decoupling module for feature decoupling. The detection and diagnosis module then completes the blade status analysis and determines the detection results.

[0100] Based on the above embodiments, the present invention provides a blade detection method based on a piezoelectric film and a microphone array, such as Figure 2 As shown, the following steps are included:

[0101] S100, data acquisition, using a piezoelectric film sensor and a microphone array sensor to collect the original signal of the blade, including vibration signals and sound signals; specifically including:

[0102] S110, setting a piezoelectric film sensor, wherein the piezoelectric film sensor is set and attached to the blade, and attached to the position where the deformation of the blade is the largest;

[0103] S120, setting a microphone array, wherein the microphone array adopts an irregular array of array element positions, is based on an elliptical array, and uses a non-dominated sorting genetic algorithm (NSGA-II) to optimize the array;

[0104] S130, data acquisition, the piezoelectric film sensor and the microphone array respectively collect the original signals of the blade, including vibration signals and sound signals. At this time, the original signals contain noise.

[0105] S200, data preprocessing, preprocessing the original signal, including denoising, filtering, and normalization; specifically including:

[0106] S210, signal denoising, removing noise from the original signal of the blade, that is, the sound signal and the vibration signal, to restore the integrity of the original signal;

[0107] S220, signal filtering, extracting useful information from the original signal of the blade and suppressing noise and interference;

[0108] S230, data normalization, converting and discretizing the data format to standardize the data for easy processing and use.

[0109] S300, feature extraction, extracting feature information from the preprocessed data, and outputting voiceprint information and vibration data in the sound, including frequency and amplitude; specifically including:

[0110] S310, wavelet packet analysis, based on wavelet analysis, makes detailed decomposition of low-frequency components and high-frequency components, and decomposes the spectrum of the sound signal;

[0111] S320, energy feature extraction, using empirical mode decomposition (EMD) to decompose the signal into intrinsic mode functions (IMFs) for feature extraction;

[0112] S330, normalization processing, mapping the data to the interval [0, 1], converting it into a dimensionless expression, eliminating the dimension effect, and facilitating comparison between different features;

[0113] S340, constructing an eigenvector. Calculate the covariance matrix of the normalized data, perform eigenvalue decomposition, and obtain eigenvalues ​​and corresponding eigenvectors.

[0114] S400, data decoupling, uses the acoustic-to-sound coupling algorithm to link the voiceprint information with the vibration data, eliminates the mutual influence of the voiceprint information and the vibration signal, decouples them, and outputs the sound signal and vibration signal without coupling relationship; specifically includes:

[0115] S410, numerical analysis, using finite element analysis to build a blade model, and simulating the interaction and relationship between the soundprint information and the vibration data through simulation;

[0116] S420, signal correlation, establishing an acoustic-solid coupling algorithm, and using the wave equation and the elastic equation to link the voiceprint information with the vibration data;

[0117] S430, signal decoupling, based on the extracted features, using a machine learning algorithm to solve and classify the acoustic-solid coupling algorithm, to achieve decoupling of the sound signal and the vibration signal, and to obtain a sound signal and a vibration signal without a coupling relationship.

[0118] S500, detection and diagnosis, building and training a blade detection model based on a convolutional neural network, parsing uncoupled sound signals and vibration signals, determining monitoring results and analyzing blade status, diagnosing blade damage, determining whether the blade is damaged, and the type and extent of the damage; specifically including:

[0119] S510, constructing a leaf detection model, constructing a leaf detection model based on a convolutional neural network, setting the number of layers of the convolutional neural network, selecting ReLU as an activation function, adding a fully connected layer, and summarizing the extracted features;

[0120] S520, leaf detection model training, collecting training set data to train the leaf detection model, adjusting hyperparameters, including learning rate and batch size, to optimize model performance, using the test set to evaluate the generalization ability of the leaf detection model, checking the final accuracy and confusion matrix, and completing the training of the leaf detection model;

[0121] S530, deploying a blade detection model, deploying the trained blade detection model, determining monitoring results and analyzing blade status, diagnosing blade damage, determining whether the blade is damaged, and the type and extent of the damage;

[0122] S540, updating the blade detection model, regularly checking the performance of the blade detection model, and retraining the blade detection model according to the blade aging condition, that is, executing step S520.

[0123] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A blade detection system based on piezoelectric film and microphone array, characterized in that: include: The vibration signal acquisition module uses a piezoelectric film sensor to collect the vibration signal of the blade and sends it to the signal processing module for processing; The sound signal collection module uses a microphone array to collect the sound signal of the blade, that is, the sound generated by the friction between the blade and the air, and sends it to the signal processing module for processing; A communication module, responsible for the communication among the vibration signal acquisition module, the sound signal acquisition module and the signal processing module; A signal processing module pre-processes the vibration signal and the sound signal, then performs feature extraction, outputs the voiceprint information and vibration data in the sound, including frequency and amplitude, and sends them to a data decoupling module; A data decoupling module uses an acoustic-solid coupling algorithm to link the voiceprint information with the vibration data, eliminate the mutual influence of the voiceprint information and the vibration data, perform decoupling, output a sound signal and a vibration signal without coupling relationship, and then send them to the detection and diagnosis module; A detection and diagnosis module uses a blade detection model based on a convolutional neural network to analyze the uncoupled sound signal and vibration signal, determine the monitoring results and analyze the blade status; The vibration signal acquisition module and the sound signal acquisition module are respectively connected to the communication module for communication, and the communication module, the signal processing module, the data decoupling module and the detection and diagnosis module are connected to each other for communication in sequence.

2. The blade detection system based on piezoelectric film and microphone array as claimed in claim 1, characterized in that: The piezoelectric film sensor is arranged and attached to the blades of the fan.

3. The blade detection system based on piezoelectric film and microphone array as claimed in claim 2, characterized in that: The piezoelectric film sensor is attached to the position where the blade deformation amount is the largest under torsional deformation, pure bending vibration and forced vibration.

4. The blade detection system based on piezoelectric film and microphone array as claimed in claim 3, characterized in that: The microphone array adopts an irregular array of array element positions to reduce sound signal distortion caused by aliasing.

5. The blade detection system based on piezoelectric film and microphone array as claimed in claim 4, characterized in that: The microphone array is based on an elliptical array and uses a non-dominated sorting genetic algorithm (NSGA-II) to optimize the array to reduce distortion and signal overlap.

6. The blade detection system based on piezoelectric film and microphone array as claimed in claim 1, characterized in that: The signal processing module processes the sound signal including preprocessing, wavelet packet analysis, energy feature extraction, normalization processing and constructing feature vectors, and the feature vectors serve as the basis for fault identification.

7. The blade detection system based on piezoelectric film and microphone array according to claim 6, wherein the decoupling of the data decoupling module comprises: S1000, extract key features, extract the frequency band energy of the pre-processed sound signal and vibration signal respectively through the feature extraction technology of wavelet packet analysis, perform normalization processing, and obtain the voiceprint information and vibration data of the sound, including amplitude and frequency; S2000, signal association, using an acoustic-to-sound coupling algorithm, based on a wave equation and an elastic equation, to associate the voiceprint information with the vibration data, thereby eliminating the mutual influence between the voiceprint information and the vibration data, wherein the wave equation describes the propagation characteristics of sound waves in a medium, and the elastic equation describes the stress-strain relationship of a solid material; S3000, signal decoupling, using a machine learning generative adversarial network (GAN) algorithm to solve the acoustic-solid coupling model to achieve decoupling of the sound signal and the vibration signal.

8. A leaf detection method based on piezoelectric film and microphone array, using the leaf detection system based on piezoelectric film and microphone array as claimed in any one of claims 1 to 7, characterized in that: The steps include: S100, data acquisition, using a piezoelectric film sensor and a microphone array sensor to collect original signals of the blade, including vibration signals and sound signals; S200, data preprocessing, preprocessing the original signal, including denoising, filtering, and normalization; S300, feature extraction, extracting feature information from the preprocessed data, and outputting voiceprint information and vibration data in the sound, including frequency and amplitude; S400, data decoupling, using an acoustic-to-sound coupling algorithm to link the voiceprint information with the vibration data, eliminating the mutual influence between the voiceprint information and the vibration data, performing decoupling, and outputting a sound signal and a vibration signal without coupling relationship; S500, detection and diagnosis, building and training a blade detection model based on a convolutional neural network, parsing the uncoupled sound signal and vibration signal, determining monitoring results and analyzing blade status, performing damage diagnosis on the blade, determining whether the blade is damaged, and the type and extent of the damage.

9. The leaf detection method based on piezoelectric film and microphone array as claimed in claim 8, characterized in that: The step S400 includes: S410, numerical analysis, using finite element analysis to build a blade model, and simulating the interaction and relationship between the soundprint information and the vibration data through simulation; S420, signal correlation, establishing an acoustic-solid coupling algorithm, and using the wave equation and the elastic equation to link the voiceprint information with the vibration data; S430, signal decoupling, based on the extracted features, using a machine learning algorithm to solve and classify the acoustic-solid coupling algorithm, to achieve decoupling of the sound signal and the vibration signal, and to obtain a sound signal and a vibration signal without a coupling relationship.

10. The blade detection method based on piezoelectric film and microphone array according to claim 8 or 9, characterized in that: The step S500 includes: S510, constructing a leaf detection model, constructing a leaf detection model based on a convolutional neural network, setting the number of layers of the convolutional neural network, selecting ReLU as an activation function, adding a fully connected layer, and summarizing the extracted features; S520, leaf detection model training, collecting training set data to train the leaf detection model, adjusting hyperparameters, including learning rate and batch size, to optimize model performance, using a test set to evaluate the generalization ability of the leaf detection model, checking the final accuracy and confusion matrix, and completing the training of the leaf detection model; S530, deploying a blade detection model, deploying the trained blade detection model, determining monitoring results and analyzing blade status, performing damage diagnosis on the blade, and determining whether the blade is damaged, as well as the type and extent of the damage; S540, updating the blade detection model, regularly checking the performance of the blade detection model, and retraining the blade detection model according to the blade aging condition, that is, executing step S520.