Wind turbine generator blade crack detection method based on acoustics and vibration signal fusion and related device
By fusing acoustic and vibration signals, combined with improved sensor deployment and a lightweight model, the accuracy and adaptability issues of wind turbine blade crack detection were resolved, enabling early identification and localization of microcracks and reducing operation and maintenance costs and time.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-06
- Publication Date
- 2026-03-24
AI Technical Summary
In existing technologies, wind turbine blade crack detection relies on a single vibration signal, which cannot effectively capture weak vibration characteristics. Furthermore, the sensor is prone to failure in high humidity and salt spray environments, making it impossible to achieve full blade coverage monitoring, resulting in a high rate of missed detections and large positioning errors.
An acoustic and vibration signal fusion method is adopted. Acoustic and vibration signals of the blade are acquired through adaptive synchronous acquisition, preprocessed and feature extracted, and crack detection is performed by combining an improved convolutional neural network. Sensors are deployed at key locations using MEMS microphone arrays and micro strain sensors to ensure timestamp alignment. The SMOTE algorithm is used to balance the sample distribution and train a lightweight MobileNetV3 model for detection.
It enables accurate identification and location of microcracks in blades, reduces the missed detection rate and location error, adapts to complex environments, meets the needs of real-time on-site detection, and reduces operation and maintenance costs and time.
Smart Images
Figure CN121721140A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of blade inspection, specifically to a method and related apparatus for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals. Background Technology
[0002] Blades are the core component of wind turbines for capturing wind energy, and their health directly determines the power generation efficiency and operational safety of the unit. Industry statistics show that blade failures account for more than 30% of all wind turbine failures, with cracks being the most typical failure mode. A microcrack of 0.1mm can develop into a dangerous 10mm crack in just 3-6 months. If not detected in time, it can lead to major accidents such as blade breakage and turbine overturning, with downtime losses of 500,000 to 1 million yuan per unit. Therefore, crack detection of wind turbine blades is of great significance for the full life-cycle health management of wind turbine blades and for reducing downtime losses and safety risks caused by blade cracks.
[0003] Current blade crack detection technologies mainly rely on single-type signals and general algorithms. First, at the signal acquisition level, traditional methods rely solely on vibration signals to monitor blade cracks. Aerodynamic noise during normal blade operation can mask the weak vibration characteristics of cracks, leading to a high rate of missed detection of micro-cracks. Second, at the diagnostic algorithm level, traditional methods use general vibration analysis algorithms for other components of wind turbines, such as gearboxes and main shafts, to diagnose blade cracks. This fails to consider the unique operating conditions of the blade's flexible structure and aerodynamic loads, making it unsuitable for blade crack detection and resulting in excessive crack location errors. Third, at the sensor deployment level, offshore wind farm blades must withstand high humidity and salt spray corrosion environments. Traditional sensors are prone to failure in this environment, and wired wiring is difficult to implement, making it impossible to achieve full blade coverage monitoring and unsuitable for the complex application scenarios of offshore wind farms. Summary of the Invention
[0004] The purpose of this invention is to provide a method and related device for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals, so as to solve the problem that traditional blade crack detection methods only use a single vibration signal for general vibration analysis to detect blade cracks, ignore the weak vibration characteristics of cracks, and cannot adapt to the unique working conditions of flexible blade structures and aerodynamic loads.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals includes the following steps: The acoustic and vibration signals of the wind turbine blades are acquired in an adaptive synchronous manner with timestamp alignment. The acoustic signals are then subjected to denoising, bandpass filtering, and aerodynamic noise cancellation in sequence to obtain effective acoustic signals. At the same time, the vibration signals are subjected to filtering and mode decomposition reconstruction in sequence to obtain effective vibration signals. Based on the effective acoustic signal and effective vibration signal, feature extraction is performed to obtain a multi-dimensional fused feature vector composed of acoustic-specific features, vibration-specific features and cross-modal fusion features; The multidimensional fused feature vector is input into a pre-trained improved convolutional neural network to obtain the blade crack detection and localization results.
[0006] In some embodiments, the step of adaptively and synchronously acquiring timestamp-aligned acoustic and vibration signals of the wind turbine blades specifically includes: Sensor nodes are respectively installed at the root, one-third, two-thirds and tip of the blade along the length of the wind turbine blade. Each sensor node includes an acoustic sensor node and at least two vibration sensor nodes, which are installed along the chord and spanwise directions of the blade. Each of the sensor nodes is equipped with a GPS timing module, the acoustic sensor node is equipped with a MEMS microphone array, and the vibration sensor node is equipped with a miniature strain sensor. Based on the GPS timing module, when the strain value in the vibration signal and the energy ratio of the acoustic signal meet the synchronous acquisition conditions, the MEMS microphone array and the micro strain sensor synchronously acquire the timestamp-aligned acoustic signal and vibration signal of the wind turbine blades, respectively.
[0007] In some implementations, the synchronous acquisition conditions include: The strain value in the vibration signal is greater than a preset reference value, or the energy proportion of the acoustic signal in the 3kHz~8kHz frequency band is greater than 30%.
[0008] In some embodiments, the acoustic signal is subjected to noise reduction, bandpass filtering and aerodynamic noise cancellation in sequence. Specifically, this includes: removing background noise from the acoustic signal using a db4 wavelet basis, then performing bandpass filtering at 3kHz to 8kHz, and finally using the acoustic signal at the blade root to cancel aerodynamic noise at other locations on the blade to obtain an effective acoustic signal. The steps of sequentially filtering and modal decomposition and reconstruction of the vibration signal specifically include: Notch filtering is used to remove grid interference from the vibration signal. Then, empirical mode decomposition is used to decompose the vibration signal into local oscillator mode functions. The high-frequency local oscillator mode functions are removed and reconstructed to obtain the effective vibration signal.
[0009] In some implementations, the improved convolutional neural network is pre-trained through the following steps: Crack samples from wind turbine blades are collected, and the distribution of the crack samples is balanced using the SMOTE algorithm to obtain a crack dataset. The crack dataset is used as the input to the MobileNetV3 model, and the blade crack detection and localization results are used as the output of the MobileNetV3 model for training, resulting in a pre-trained improved convolutional neural network.
[0010] In some implementations, the step of inputting the multidimensional fused feature vector into a pre-trained improved convolutional neural network to obtain the blade crack detection and localization results is as follows: The blade crack detection results include no cracks, microcracks, moderate cracks, and severe cracks; The positioning result is calculated based on the time difference between acoustic impact and strain abrupt change in the multidimensional fused feature vector.
[0011] Secondly, a wind turbine blade crack detection system based on the fusion of acoustic and vibration signals includes: The adaptive synchronous acquisition and preprocessing module is used to adaptively and synchronously acquire acoustic and vibration signals of wind turbine blades with timestamp alignment, and sequentially perform noise reduction, bandpass filtering and aerodynamic noise cancellation on the acoustic signals to obtain effective acoustic signals. At the same time, the vibration signals are sequentially filtered and modal decomposition and reconstruction to obtain effective vibration signals. The multidimensional fusion feature vector extraction module is used to extract features based on the effective acoustic signal and effective vibration signal to obtain a multidimensional fusion feature vector composed of acoustic-specific features, vibration-specific features and cross-modal fusion features; The blade crack detection module is used to input the multi-dimensional fused feature vector into a pre-trained improved convolutional neural network to obtain blade crack detection results and localization results.
[0012] Thirdly, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor executes the computer program to implement the steps of the wind turbine blade crack detection method based on acoustic and vibration signal fusion.
[0013] Fourthly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals.
[0014] Fifthly, a computer program product comprising a computer program that, when executed by a processor, implements the steps of the method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals.
[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals. It adaptively and synchronously acquires timestamp-aligned acoustic and vibration signals from the wind turbine blades, and preprocesses them to obtain effective acoustic and vibration signals. Feature extraction is then performed on these effective acoustic and vibration signals to obtain a multi-dimensional fused feature vector, which avoids the masking of weak crack features by aerodynamic noise and captures the weak vibration characteristics of the cracks. Furthermore, the acoustic signals are sequentially denoised, bandpass filtered, and aerodynamic noise is canceled; the vibration signals are sequentially filtered and modal decomposition reconstructed. The multi-dimensional fused feature vector is then input into a pre-trained improved convolutional neural network to obtain the blade crack detection and localization results. This method can adapt to the unique operating conditions of flexible blade structures and aerodynamic loads, avoiding the limitations of traditional methods that only provide coarse detection, thus improving crack detection accuracy and achieving crack localization.
[0016] Furthermore, sensor nodes are respectively installed at the root, one-third, two-thirds, and tip positions of the wind turbine blade along its length. Each sensor node includes an acoustic sensor node and at least two vibration sensor nodes. The vibration sensor nodes are installed along the chord and spanwise directions of the blade, respectively, which can cover the signal at key locations on the blade and solve the problems of incomplete traditional sensor deployment and inability to capture crack signals in different areas of the blade. Each sensor node is equipped with a GPS timing module to ensure that the timestamps of the acoustic and vibration signals are aligned, avoiding fusion errors caused by signal asynchrony.
[0017] Furthermore, the synchronous acquisition conditions include: the strain value in the vibration signal is greater than a preset reference value, and the energy proportion of the acoustic signal in the 3kHz~8kHz frequency band is greater than 30%. This can avoid invalid acquisition in non-crack scenarios, reduce data redundancy, reduce the computing power consumption of subsequent signal processing, and avoid missing crack signals.
[0018] Furthermore, the SMOTE algorithm is used to balance the distribution of the crack samples, which can solve the detection bias problem caused by uneven sample distribution in traditional models and ensure that the model has a high recognition rate for various types of cracks. The crack dataset is used as the input of the MobileNetV3 model, and the blade crack detection results and localization results are used as the output of the MobileNetV3 model for training, resulting in a pre-trained improved convolutional neural network. This model is lightweight, which can reduce the number of parameters and inference time, and solve the problems of large size, slow inference and inability to be deployed at the edge of traditional large convolutional neural networks, thus meeting the needs of real-time detection on site. Attached Figure Description
[0019] Figure 1 A flowchart of a wind turbine blade crack detection method based on acoustic and vibration signal fusion provided in an embodiment of the present invention; Figure 2 This is a structural diagram of a wind turbine blade crack detection system based on the fusion of acoustic and vibration signals, provided in an embodiment of the present invention. Detailed Implementation
[0020] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. The content described herein is for explanation rather than limitation of the present invention.
[0021] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of this invention are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, systems, products, or devices.
[0022] like Figure 1 As shown in the figure, this embodiment provides a method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals, including the following steps: S1, adaptively and synchronously acquire the acoustic and vibration signals of the wind turbine blades with timestamp alignment, and sequentially perform noise reduction, bandpass filtering and aerodynamic noise cancellation on the acoustic signals to obtain effective acoustic signals. At the same time, sequentially perform filtering and mode decomposition reconstruction on the vibration signals to obtain effective vibration signals. To accurately acquire acoustic and vibration signals from blade cracks, sensors are deployed as follows: The acoustic sensor node employs a MEMS microphone array with four channels, a frequency range of 1kHz to 20kHz, a sensitivity of -42dBV / Pa, and a signal-to-noise ratio of 60dB. It focuses on capturing high-frequency impact noise in the 3kHz to 8kHz frequency band during crack propagation, avoiding aerodynamic noise interference. The vibration sensor node uses a miniature strain sensor with a range of ±2000με, an accuracy of ±1με, and a sampling frequency of 500Hz. It monitors strain changes during blade bending deformation and captures sudden changes in local stiffness caused by cracks.
[0023] Deployment locations: Sensor nodes are installed along the length of the wind turbine blade at the root, one-third, two-thirds, and tip of the blade, for a total of four sets of sensor nodes. Each set includes one acoustic sensor node and two vibration sensor nodes. The vibration sensor nodes are installed along the chord and spanwise directions of the blade, respectively. The sensor nodes are fixed by both magnetic attraction and high-strength adhesive, with an IP68 protection rating, capable of withstanding high and low temperatures and resisting gust impacts.
[0024] All sensor nodes have built-in GPS timing modules with a time synchronization accuracy of less than 1ms, ensuring that the timestamps of acoustic signals and vibration signals are aligned and avoiding fusion errors caused by signal asynchrony. LoRaWAN wireless communication is adopted, with a communication distance of 5km and a transmission current of less than 100mA. Combined with a flexible solar power supply with a power of 100mW, it can achieve at least 4 years of maintenance-free operation.
[0025] The synchronous acquisition of acoustic and vibration signals is triggered by an adaptive threshold. When the fluctuation amplitude of the strain signal, i.e. the strain value in the vibration signal, exceeds the preset reference value when the blade is free of cracks, or when the energy proportion of the acoustic signal in the 3kHz~8kHz frequency band is greater than 30%, synchronous acquisition is triggered. In the non-triggered state, the sensor node is in sleep mode to reduce power consumption.
[0026] The acoustic signal is sampled at a frequency of 20kHz, with a sampling duration of 2s and 40,000 sampling points per channel; the vibration signal is sampled at a frequency of 500Hz, with a sampling duration of 2s and 1,000 sampling points per channel; a routine sampling is performed every 30 minutes, and if an abnormality is triggered, the sampling time is shortened to 5 minutes per sampling to ensure that no crack signals are missed.
[0027] Acoustic signal denoising employs a combination of wavelet threshold filtering and bandpass filtering: first, background noise is removed by five-layer decomposition of the db4 wavelet basis, then 3kHz~8kHz bandpass filtering is used to retain crack impact noise, and finally adaptive noise cancellation technology is used, taking the acoustic signal at the blade root as a reference. Since there are no cracks at the blade root, the noise is mainly aerodynamic noise, which cancels the aerodynamic noise interference from other position sensor nodes, improving the signal-to-noise ratio to over 45dB and obtaining an effective acoustic signal. The vibration signal noise reduction first uses a 50Hz notch filter to remove power grid interference, and then uses Empirical Mode Decomposition (EMD) to decompose the vibration signal into 8 Intrinsic Mode Functions (IMFs). Noisy high-frequency IMFs (not less than the 6th order) are removed, and the effective signal is reconstructed to obtain the effective vibration signal.
[0028] S2, based on the effective acoustic signal and effective vibration signal, feature extraction is performed to obtain a multi-dimensional fused feature vector composed of acoustic-specific features, vibration-specific features and cross-modal fusion features; Specifically, multi-dimensional fused feature vectors are extracted from the preprocessed effective acoustic and vibration signals and divided into three categories: Acoustic-specific characteristics (6 dimensions): energy proportion in the 3kHz~8kHz frequency band, number of impulse pulses (number of pulses greater than 50dB), peak acoustic signal, spectral centroid, bandwidth, and the first three average values of Mel frequency cepstral coefficients (MFCC).
[0029] Vibration-specific characteristics (6 dimensions): peak strain, kurtosis (pulse characteristics caused by cracks), number of strain abrupt changes (abrupt changes greater than 10με), peak power spectral density (PSD) in the 10Hz~50Hz frequency band, strain variance, and strain difference between adjacent nodes.
[0030] Cross-modal fusion characteristics (6 dimensions): cross-correlation coefficient of acoustic-strain signals (greater than 0.8 when there is no crack, less than 0.5 when there is a crack), time difference between acoustic impact and strain abrupt change, spectral coherence coefficient of the two, joint entropy, covariance and fusion entropy.
[0031] S3, input the multidimensional fused feature vector into the pre-trained improved convolutional neural network to obtain the blade crack detection result and the localization result.
[0032] An improved lightweight convolutional neural network, MobileNetV3, was used as the classifier. The input was a multi-dimensional fused feature vector, and the output consisted of four categories: "no crack," "micro-crack (0.1mm~2mm)," "moderate crack (2mm~10mm)," and "severe crack (greater than 10mm)." The training and optimization of the improved lightweight MobileNetV3 model are as follows: A sample dataset was constructed by collecting 1000 sets of blade crack samples (including different crack lengths, wind speeds, and temperature conditions). The crack dataset was obtained by balancing the sample distribution using the SMOTE algorithm. The crack dataset was used as the input of the MobileNetV3 model, and the blade crack detection and localization results were used as the output of the MobileNetV3 model for training, resulting in a pre-trained improved convolutional neural network.
[0033] The model optimization employs knowledge distillation technology to transfer knowledge from the pre-trained large model to MobileNetV3. Generally, when the number of parameters is reduced by 70%, the inference time is less than 100ms, meeting the real-time processing requirements of the edge. The test set accuracy and microcrack recognition rate are no less than 95%, and the false alarm rate is less than 2%.
[0034] The localization results were obtained using a triangulation method based on the time difference between acoustic impact and strain abrupt change, calculating the time difference between the arrival of crack impact noise at two adjacent acoustic sensor nodes. Δt Given the distance between the nodes of the two acoustic sensors L Speed of sound v The distance from the crack to one of the acoustic sensor nodes is calculated, and the positioning result is corrected by combining the location of the local stiffness change of the vibration sensor node, so that the final positioning error is less than 0.5m.
[0035] like Figure 2 As shown, this embodiment provides a wind turbine blade crack detection system based on the fusion of acoustic and vibration signals, including: The adaptive synchronous acquisition and preprocessing module is used to adaptively and synchronously acquire acoustic and vibration signals of wind turbine blades with timestamp alignment, and sequentially perform noise reduction, bandpass filtering and aerodynamic noise cancellation on the acoustic signals to obtain effective acoustic signals. At the same time, the vibration signals are sequentially filtered and modal decomposition and reconstruction to obtain effective vibration signals. The multidimensional fusion feature vector extraction module is used to extract features based on the effective acoustic signal and effective vibration signal to obtain a multidimensional fusion feature vector composed of acoustic-specific features, vibration-specific features and cross-modal fusion features; The blade crack detection module is used to input the multi-dimensional fused feature vector into a pre-trained improved convolutional neural network to obtain blade crack detection results and localization results.
[0036] This embodiment aims to overcome the shortcomings of existing blade crack detection technologies, such as single signal, generalized algorithms, and lack of prediction. Through customized sensor deployment, simultaneous acquisition of multi-modal signals, and fusion diagnostic algorithms, it achieves accurate identification and location of blade microcracks larger than or equal to 0.1 mm, with an identification accuracy greater than 95% and a location error less than 0.5 m, meeting the low-cost, high-efficiency blade maintenance requirements of wind farms. This embodiment addresses the high-frequency acoustic impact and local strain abrupt changes in blade cracks, overcoming the limitations of traditional single vibration signals and significantly improving the microcrack identification rate. It employs an improved MobileNetV3 model, balancing identification accuracy and real-time performance, and can be deployed at the edge, such as in the nacelle controller, without relying on cloud computing power. The sensor nodes utilize adaptive triggering and solar power, with an annual power consumption of less than 5 Wh, enabling at least four years of maintenance-free operation and adapting to the complex environment of offshore wind farms.
[0037] Therefore, this embodiment can achieve early identification and early warning of microcracks in blades, avoiding blade breakage accidents caused by crack propagation, reducing the downtime rate of wind farm blade failure by at least half; the operation and maintenance cost of a single unit blade can be reduced to 5,000 yuan / year, and it is estimated that a 150MW wind farm will save two to three million yuan in operation and maintenance costs over five years; the crack location error is reduced, and the maintenance time can be shortened from several days to several hours, greatly improving the efficiency of blade maintenance; it is applicable to different types of blades on land / offshore, and can be connected to the existing wind farm CMS (Condition Monitoring System) without modifying the original equipment, with low deployment cost and strong compatibility.
[0038] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.
[0039] This embodiment also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program (in this embodiment, the computer program includes a computing component and an iterative component, capable of model calculation and model updating). The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function. The processor described in this embodiment can be used for the operation of a wind turbine blade crack detection method based on acoustic and vibration signal fusion.
[0040] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the wind turbine blade crack detection method based on acoustic and vibration signal fusion in the above embodiment.
[0041] This embodiment also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the corresponding steps of the wind turbine blade crack detection method based on acoustic and vibration signal fusion in the above embodiment.
[0042] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0043] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0044] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0045] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals, characterized in that, Includes the following steps: The acoustic and vibration signals of the wind turbine blades are acquired in an adaptive synchronous manner with timestamp alignment. The acoustic signals are then subjected to denoising, bandpass filtering, and aerodynamic noise cancellation in sequence to obtain effective acoustic signals. At the same time, the vibration signals are subjected to filtering and mode decomposition reconstruction in sequence to obtain effective vibration signals. Based on the effective acoustic signal and effective vibration signal, feature extraction is performed to obtain a multi-dimensional fused feature vector composed of acoustic-specific features, vibration-specific features and cross-modal fusion features; The multidimensional fused feature vector is input into a pre-trained improved convolutional neural network to obtain the blade crack detection and localization results.
2. The method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals according to claim 1, characterized in that, The steps for adaptively and synchronously acquiring timestamp-aligned acoustic and vibration signals from wind turbine blades specifically include: Sensor nodes are respectively installed at the root, one-third, two-thirds and tip of the blade along the length of the wind turbine blade. Each sensor node includes an acoustic sensor node and at least two vibration sensor nodes, which are installed along the chord and spanwise directions of the blade. Each of the sensor nodes is equipped with a GPS timing module, the acoustic sensor node is equipped with a MEMS microphone array, and the vibration sensor node is equipped with a miniature strain sensor. Based on the GPS timing module, when the strain value in the vibration signal and the energy ratio of the acoustic signal meet the synchronous acquisition conditions, the MEMS microphone array and the micro strain sensor synchronously acquire the timestamp-aligned acoustic signal and vibration signal of the wind turbine blades, respectively.
3. The method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals according to claim 2, characterized in that, The synchronous acquisition conditions include: The strain value in the vibration signal is greater than a preset reference value, or the energy proportion of the acoustic signal in the 3kHz~8kHz frequency band is greater than 30%.
4. The method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals according to claim 1, characterized in that, The steps of performing noise reduction, bandpass filtering and aerodynamic noise cancellation on the acoustic signal in sequence include: removing background noise of the acoustic signal using the db4 wavelet basis, then performing bandpass filtering of 3kHz to 8kHz, and finally using the acoustic signal at the blade root to cancel aerodynamic noise at other positions of the blade to obtain an effective acoustic signal. The steps of sequentially filtering and modal decomposition and reconstruction of the vibration signal specifically include: Notch filtering is used to remove grid interference from the vibration signal. Then, empirical mode decomposition is used to decompose the vibration signal into local oscillator mode functions. The high-frequency local oscillator mode functions are removed and reconstructed to obtain the effective vibration signal.
5. The method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals according to claim 1, characterized in that, The improved convolutional neural network is pre-trained through the following steps: Crack samples from wind turbine blades are collected, and the distribution of the crack samples is balanced using the SMOTE algorithm to obtain a crack dataset. The crack dataset is used as the input to the MobileNetV3 model, and the blade crack detection and localization results are used as the output of the MobileNetV3 model for training, resulting in a pre-trained improved convolutional neural network.
6. The method for detecting cracks in wind turbine blades based on the fusion of acoustic and vibration signals according to claim 1, characterized in that, In the step of inputting the multidimensional fused feature vector into a pre-trained improved convolutional neural network to obtain the blade crack detection and localization results, The blade crack detection results include no cracks, microcracks, moderate cracks, and severe cracks; The positioning result is calculated based on the time difference between acoustic impact and strain abrupt change in the multidimensional fused feature vector.
7. A wind turbine blade crack detection system based on acoustic and vibration signal fusion, characterized in that, include: The adaptive synchronous acquisition and preprocessing module is used to adaptively and synchronously acquire acoustic and vibration signals of wind turbine blades with timestamp alignment, and sequentially perform noise reduction, bandpass filtering and aerodynamic noise cancellation on the acoustic signals to obtain effective acoustic signals. At the same time, the vibration signals are sequentially filtered and modal decomposition and reconstruction to obtain effective vibration signals. The multidimensional fusion feature vector extraction module is used to extract features based on the effective acoustic signal and effective vibration signal to obtain a multidimensional fusion feature vector composed of acoustic-specific features, vibration-specific features and cross-modal fusion features; The blade crack detection module is used to input the multi-dimensional fused feature vector into a pre-trained improved convolutional neural network to obtain the blade crack detection result and the localization result.
8. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable in the processor. When the processor executes the computer program, it implements the steps of the wind turbine blade crack detection method based on the fusion of acoustic and vibration signals as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the wind turbine blade crack detection method based on the fusion of acoustic and vibration signals as described in any one of claims 1 to 6.
10. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the wind turbine blade crack detection method based on the fusion of acoustic and vibration signals as described in any one of claims 1 to 6.