Wind power blade defect detection method based on acoustic emission

Abstract

The invention discloses a wind power blade defect detection method based on acoustic emission, and relates to the technical field of wind power blade detection.The wind power blade defect detection method comprises the steps that multiple lasers, a multispectral camera, a thermal imaging sensor and a plurality of acoustic emission sensors are carried on an unmanned aerial vehicle, and multiple multispectral images and thermal imaging images within fixed time are obtained; a multispectral image total discrimination sequence and a thermal imaging image total discrimination sequence are obtained through processing, suspicious pixel points in the two sequences are mapped into actual three-dimensional coordinates of the wind power blade, a light reflection defect area and a thermal distribution defect area are obtained through fitting, the intersection of the two areas is obtained, and a specific area of the wind power blade defect is obtained. And positioning the region by using an acoustic emission sensor to obtain a to-be-detected acoustic signal, denoising by combining LMD with MFSE, extracting a feature vector, and substituting the feature vector into a GRU wind power blade defect judgment model to judge whether an internal fault exists or not. According to the invention, comprehensive perception and correlation judgment of internal and external defects of the wind power blade are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine blade detection, and particularly to a method for detecting wind turbine blade defects based on acoustic emission. Background Art

[0002] As the core load-bearing component of a wind power generation system, wind turbine blades operate in a complex and changeable environment for a long time. Their structural integrity is directly related to the overall operation efficiency of the unit. Existing detection technologies mainly focus on internal and external defects of the blades respectively. However, due to single detection means and insufficient information fusion, the problem of missed detection frequently occurs.

[0003] For internal defects, traditional methods mainly rely on acoustic emission detection technology. However, most wind turbine blades are fiber-reinforced composite materials, and their anisotropy causes acoustic signals to be easily scattered and attenuated during propagation. Moreover, complex environmental noise around acoustic emission sensors will seriously contaminate the original signals, making weak defect signals masked. For external defects, existing technologies generally adopt image recognition methods. However, such methods are often affected by factors such as lighting conditions and surface stains. In addition, external defects often occur concomitantly with internal damages. For example, surface cracks may extend to the inside and cause structural layer failure. However, existing technologies only focus on single-dimensional detection and cannot establish the correlation analysis of internal and external defects, resulting in an increased risk of missed detection of composite defects.

[0004] Therefore, there is an urgent need for a comprehensive detection method that can fuse multi-source detection data and take into account the characteristics of internal and external defects to solve the limitations of single detection means in the existing technology and improve the comprehensiveness and accuracy of defect detection. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a method for detecting wind turbine blade defects based on acoustic emission, which solves the problems of easy missed detection and lack of multi-modal fusion in separately detecting internal and external defects of wind turbine blades in the existing technology.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for detecting wind turbine blade defects based on acoustic emission, including:

[0007] S101. According to multiple collected multi-spectral images, calculate the four-dimensional spectral vector corresponding to each pixel, calculate the angle between it and the reference spectral vector of the normal area. If it is greater than the angle threshold, mark the pixel as a suspicious defect pixel point to obtain a discrimination sequence of a single multi-spectral image. Process multiple discrimination sequences to obtain a total discrimination sequence, map the suspicious pixel points therein to the actual three-dimensional coordinates of the wind turbine blade, and mark them with 0;

[0008] S102. Calculate the temperature values of each pixel in each thermal imaging picture to obtain the temperature value sequence within this period of time, and calculate the total change rate, mean deviation rate, standard deviation coefficient, period coincidence degree, and determination coefficient. If the total change rate is not within the specified range, or the mean deviation rate is greater than the first threshold, or the period coincidence degree is less than the second threshold, initiate a first-level warning and enter the second-level confirmation. If the standard deviation coefficient is greater than the third threshold and the determination coefficient is less than the fourth threshold, determine that this pixel is a suspicious pixel, obtain the total discrimination sequence of the thermal imaging image, map the suspicious pixel to the actual three-dimensional coordinates, and mark it with 1.

[0009] S103. Perform an oriented bounding box fitting on the three-dimensional coordinates marked as 0 and 1 on the wind turbine blade to obtain the optical reflection defect and thermal distribution defect areas respectively. Take the intersection to determine the specific area of the surface defect of the wind turbine blade. Use the unmanned aerial vehicle (UAV) acoustic emission sensor to locate this area and obtain the acoustic signal to be detected. After denoising by LMD combined with MFSE, extract the feature vector and substitute it into the GRU wind turbine blade defect judgment model to further determine whether there is an internal fault in this area.

[0010] As a further solution of the present invention, it further includes S100. The S100 irradiates the wind turbine blade with multiple lasers on the UAV, and respectively uses a multi-spectral camera and a thermal imaging sensor to take a number of multi-spectral images and thermal imaging images within a fixed period of time by carrying multiple lasers, a multi-spectral camera, a thermal imaging sensor, and several acoustic emission sensors on the UAV.

[0011] As a further solution of the present invention, the bands selected by the multiple lasers include the visible band, the near-infrared band, and the short-wave infrared band.

[0012] As a further solution of the present invention, when collecting information on the wind turbine blade, the specific requirements for the surrounding environment include that the ambient temperature needs to be stable at 10 - 30 °C, the humidity needs to be ≤ 80% RH, the wind speed needs to be ≤ 5 m / s, the multi-spectral imaging needs uniform natural light, the thermal imaging needs to avoid direct sunlight, be far away from strong electromagnetic sources, and the temperature difference between the background temperature of the shooting area and the blade needs to be < 5 °C.

[0013] As a further solution of the present invention, process the discrimination sequences of multiple multi-spectral images, calculate the proportions PerDou and PerNor of the suspicious quantity and the normal quantity in each pixel to the total quantity, select max{PerDou, PerNor} as the overall judgment situation of this pixel, and obtain the total discrimination sequence of the multi-spectral image.

[0014] As a further solution of the present invention, if the number of multiple hyperspectral images is even, calculate the proportion PerNei of suspicious pixels in the 3×3 neighborhood around the pixel. If PerNei >= 0.5, the current pixel is determined to be suspicious; otherwise, it is determined to be normal. If the pixel is a corner pixel, the effective neighborhood is three pixels; if the pixel is an edge pixel, the effective neighborhood is five pixels. The corner pixels are the upper left, lower left, upper right, and lower right, and the edge pixels are the left middle, right middle, upper middle, and lower middle.

[0015] As a further solution of the present invention, the specific method for constructing the dynamic threshold of each group of TVR is as follows:

[0016] Collect M groups of data from the same area of the defect-free blade, calculate the TVR of each group of pixels, and construct a normal distribution TVRnormal ~ N(μ,σ 2 );

[0017] Let Crmin = μ - k×σ, Crmax = μ + k×σ, where k is the confidence coefficient and can take 2 or 3.

[0018] As a further solution of the present invention, the specific steps for denoising the acoustic signal using LMD combined with MFSE are as follows:

[0019] Decompose the original acoustic signal into multiple product function components PF and a residual term by LMD;

[0020] Calculate the mean Argmfse of MFSE for each PF. If Argmfse >= Argth, classify the PF component as a noise component; otherwise, classify it as a defect signal component;

[0021] Add the defect signal components and the residual term to obtain the denoised acoustic signal.

[0022] As a further solution of the present invention, extract time-domain features, frequency-domain features, time-frequency domain features, and non-linear features from the denoised acoustic signal. The time-domain features include peak value, root mean square value, rise time, pulse width, and zero-crossing rate. The frequency-domain features include the main frequency of the fast Fourier transform FFT, band energy, skewness and kurtosis of the power spectral density PSD. The time-frequency domain features include wavelet energy entropy and variance of the short-time Fourier transform STFT. The non-linear features include permutation entropy PE and fractal dimension FD.

[0023] As a further solution of the present invention, the initial parameters of the GRU model specifically include an input dimension of 13, 2 GRU layers, 64 neurons in each layer, the activation function is selected as tanh, the dropout rate is 0.2, the output dimension is q + 1, the learning rate is 0.001, the number of training epochs is 50, and the batch size is 32, where q is the number of types of internal defects of the wind turbine blade.

[0024] The present invention provides a method for detecting defects in wind turbine blades based on acoustic emission, which has the following beneficial effects compared with the prior art:

[0025] (1) By integrating multiple lasers, multi-spectral cameras, thermal imaging sensors and acoustic emission sensors, the present invention constructs a collaborative detection system for surface features and internal signals, effectively solves the problem of missed detection of composite defects by a single detection method, and realizes the comprehensive perception and correlation judgment of internal and external defects of the blades;

[0026] (2) By adopting a dynamic threshold method in thermal imaging analysis and determining a reasonable threshold range based on defect-free blade data to construct a normal distribution, the present invention reduces the risk of misjudgment caused by a fixed threshold; in acoustic signal processing, the original signal is denoised by LMD decomposition combined with MFSE, and multi-dimensional features are extracted by combining with a GRU model for intelligent judgment, which can accurately capture defect features in complex non-linear acoustic signals and enhance the robustness of the detection system under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flowchart of the steps of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0029] As Figure 1 , the present invention provides a method for detecting defects in wind turbine blades based on acoustic emission, including:

[0030] S100. Mount multiple lasers, multi-spectral cameras, thermal imaging sensors and several acoustic emission sensors on a drone, irradiate the wind turbine blade with the multiple lasers on the drone, and respectively take several multi-spectral images and thermal imaging images within a fixed time period with the multi-spectral camera and the thermal imaging sensor;

[0031] The drone can be a six-axis multi-rotor drone, which can stably hover in a complex airflow environment to achieve close-range high-precision scanning. At the same time, since the drone needs to load multiple acquisition devices, the multi-rotor load redundancy can meet the actual requirements;

[0032] The multi - laser can adopt a four - band laser with wavelengths of 650nm, 532nm, 850nm, and 1064nm respectively. 650nm / 532nm corresponds to the visible band, which is the sensitive area of the human eye and is used for visual identification of surface defects, such as cracks and coating peeling. 850nm corresponds to the near - infrared band, which can penetrate the shallow coating and detect damage to the composite material matrix, such as fiber fracture. 1064nm corresponds to the short - wave infrared band, which is sensitive to the dielectric constant difference between resin and fiber and can identify delamination defects.

[0033] The multi - spectral camera selects a four - channel multi - spectral camera. Since the RGB camera has only three channels and cannot separate the signals of the near - infrared band and the short - wave infrared band, such as 850nm / 1064nm mentioned above, it is easy to lose key defect features. Although the hyperspectral camera covers the full spectrum, the data volume is huge, and it is difficult for the UAV computing power to process it in real time, and the cost is high. The four - channel multi - spectral camera has four customized bands, which can match the laser wavelengths, taking into account both resolution and data efficiency, and supports real - time spectral vector calculation.

[0034] The thermal imaging sensor selects a high - precision thermal imager because the internal defects of the wind turbine blade only cause a temperature difference of 0.5 - 2°C, such as delamination. The high - precision device can capture tiny temperature changes to avoid missed detections. At the same time, it is also for millimeter - level defect positioning to avoid position deviation caused by pixel aliasing.

[0035] The acoustic emission sensor can select a piezoelectric ceramic acoustic emission sensor. It can cover the main frequency band of the defect signal, capture the signals in the whole stage from the initiation to the expansion of micro - cracks, avoid the frequency blind area. At the same time, it can adapt to the rapid attenuation of the acoustic emission signal of the composite material, effectively amplify weak signals, improve the signal - to - noise ratio, and can adapt to the blade surface curve to ensure the acoustic coupling efficiency.

[0036] When using the above - mentioned equipment to collect information about the wind turbine blade, the environmental requirements around the wind turbine blade are as follows:

[0037] The ambient temperature needs to be stable at 10 - 30°C to prevent measurement errors caused by temperature drift of the thermal imager.

[0038] The humidity needs to be ≤80%RH to prevent lens condensation or water vapor from absorbing infrared radiation, and at the same time avoid moisture absorption on the surface of the composite material affecting temperature conduction.

[0039] The wind speed needs to be ≤5m / s to prevent image misalignment caused by UAV jitter and reduce the interference of air flow on the surface temperature distribution of the blade.

[0040] For multi - spectral imaging, uniform natural light is required to avoid shadows or overexposure. For thermal imaging, direct sunlight needs to be avoided to prevent abnormal surface temperature rise.

[0041] Stay away from strong electromagnetic sources, such as substations and frequency converters, to prevent interference with the sensor signal chain and ensure the accuracy of temperature data acquisition;

[0042] The temperature difference between the background temperature of the shooting area and the blade should be < 5°C to avoid background radiation interfering with the thermal imaging contrast.

[0043] S101. Based on a number of collected multispectral images, calculate the four-dimensional spectral vector corresponding to each pixel in each multispectral image, calculate the angle θ between the spectral vector of each pixel and the reference spectral vector of the normal area. If θ is greater than η, mark the pixel as a suspicious defect pixel; otherwise, mark the pixel as a normal pixel;

[0044] The dimensionality of the spectral vector corresponding to each pixel should be the same as the number of multi-laser emitters, and each component represents the reflection intensity at that wavelength;

[0045] Defects in wind turbine blades will cause changes in surface material, roughness, and reflectivity, which will in turn cause specific changes in spectral characteristics, such as cracks, coating peeling, and fiber fracture. By normalizing the vector angle, the influence of light intensity fluctuations on the absolute value is eliminated, and only the spectral shape differences are concerned. At the same time, the multi-spectral information provided by multiple lasers forms a high-dimensional spectral vector, and defects will produce abnormal responses at specific wavelengths;

[0046] η needs to be dynamically adjusted according to blade material, coating type, and laser wavelength combination. Usually, it is taken as 15°. The spectral angle range under normal working conditions is:

[0047] For the defect-free area, the pixel spectrum is highly consistent with the reference spectrum, and the angle is usually less than or equal to 10, and ideally close to 0°, but there are certain fluctuations due to the slight roughness of the blade surface and the illumination uniformity;

[0048] For slightly abnormal areas, such as surface stains and slight wear, the angle may be between 10° and 15°;

[0049] For the defect area, when the angle is greater than 15°, the spectral difference is significant, and the more serious the defect, the larger the angle, such as deep cracks and large-area peeling;

[0050] Judge whether all pixel points of each multispectral image are suspicious or normal to obtain the multispectral image discrimination sequence. Then, process the multispectral image discrimination sequences of several multispectral images during this period, calculate the proportions PerDou and PerNor of the suspicious number and the normal number in the total number for each pixel point, and select max{PerDou, PerNor} as the overall judgment of this pixel point to obtain the total multispectral image discrimination sequence;

[0051] If the number of multiple spectral images within this period is even, there may be a situation where PerDou equals PerNer for a pixel point, and simply relying on max{PerDou, PerNor} cannot make a judgment. At this time, a priority determination rule needs to be added based on the original method. The specific processing steps are as follows:

[0052] Calculate the proportion PerNei of suspicious pixel points in the 3×3 neighborhood around this pixel point. If PerNei >= 0.5, the current pixel point is determined to be suspicious; otherwise, it is determined to be normal;

[0053] Defects usually have spatial aggregation in the image. For example, the area around a crack may have abnormal spectral reflection due to surface morphology changes. Neighborhood consistency checking can use spatial correlation to reduce the influence of isolated noise;

[0054] For corner pixels, such as the upper left, lower left, upper right, and lower right, the effective neighborhood is three pixels;

[0055] For edge pixels, such as the left middle, right middle, upper middle, and lower middle, the effective neighborhood is five pixels;

[0056] For example, there are 3 multiple spectral images, each image has 5 pixel points. The discrimination sequence of the first multiple spectral image is {suspicious, normal, normal, suspicious, normal}, the discrimination sequence of the second multiple spectral image is {normal, suspicious, normal, normal, suspicious}, and the discrimination sequence of the third multiple spectral image is {suspicious, suspicious, normal, suspicious, normal}. For the first pixel point, the number of suspicious is 2, the number of normal is 1, PerDou = 2 / 3, PerNor = 1 / 3, max{PerDou, PerNor} = PerDou, so the first pixel point is determined to be suspicious. And so on, the total discrimination sequence of the multiple spectral image is {suspicious, suspicious, normal, suspicious, normal};

[0057] Extract the pixel points marked as suspicious in the total discrimination sequence of the multiple spectral image, and locate them to the actual three-dimensional coordinates of the wind turbine blade, and mark this three-dimensional coordinate with 0.

[0058] S102. Calculate the temperature value of each pixel point in each thermal imaging picture. For each pixel point, form a temperature value sequence within this period, and calculate the total change rate TVR, mean offset rate MOR, coefficient of variation CoV, period coincidence degree PC, and determination coefficient R2 of each pixel point temperature value sequence. If or MOR > miu1 or PC < miu2, a first-level warning is initiated and a second-level confirmation is entered. If CoV > miu3 and R2 < miu4, then this pixel point is determined to be a suspicious pixel point, where miu1, miu2, miu3, and miu4 are all set by the user themselves;

[0059] When determining the TVR for each group, the upper limit Crmin and the lower limit Crmax of the threshold interval are dynamically changing. Since the TVR of each group of pixels is affected by independent random factors such as environmental noise and material uniformity fluctuations, according to the central limit theorem, its distribution approaches a normal distribution, providing a theoretical basis for threshold statistics modeling. The specific method for constructing dynamic thresholds is as follows:

[0060] Collect M groups of data for the same area of the defect-free blade, where M ≥ 50. Calculate the TVR of each group of pixels and construct a normal distribution TVRnormal ∼ N(μ,σ2);

[0061] Let Crmin = μ - k×σ and Crmax = μ + k×σ, where k is the confidence coefficient, usually taking 2 or 3;

[0062] The specific reasons for choosing the total change rate TVR, the mean offset rate MOR, and the period coincidence degree PC as the primary warning conditions include:

[0063] The total change rate TVR. TVR is the sum of the positive / negative change rates at adjacent moments in the temperature sequence, reflecting the severity of temperature changes and the frequency of direction switching. If TVR > Crmax, it may indicate a sudden rise or fall in temperature, such as abnormal local heat dissipation caused by surface cracks; if TVR < Crmin, it may indicate that the temperature change is too slow, such as internal delamination hindering heat conduction and causing the temperature rise to stagnate;

[0064] TVR is the most direct indicator of thermal response anomalies. Defects will significantly change the local thermal resistance, such as cracks and debonding, resulting in the temperature change rate deviating from the normal range, with strong sensitivity. For example, during laser irradiation, the temperature in the normal area should rise steadily at a certain rate, while in the debonded area, due to heat accumulation, the heating rate increases initially, the positive change rate increases, and later may drop suddenly due to thermal saturation, and the overall TVR exceeds the threshold;

[0065] The mean offset rate MOR. MOR measures the deviation of the mean value of the temperature sequence from the defect-free reference area, reflecting the abnormality of the long-term thermal equilibrium state. If MOR > miu1, it may mean that the defect area continuously absorbs heat or has abnormal heat dissipation. For example, coating damage will cause a decrease in local reflectivity and absorb more laser energy, resulting in a mean value higher than the reference area, while damage to the internal honeycomb core may cause the mean value to be lower than the normal area due to the interruption of the heat conduction path;

[0066] The mean offset is the cumulative thermal effect caused by defects, which complements the dynamic change rate of TVR. For example, for some progressive defects, the TVR change is not significant in the initial stage, but the long-term mean value will gradually deviate, such as minor debonding, and MOR can capture such chronic anomalies;

[0067] The PC reflects the synchronization between the temperature change cycle and the laser irradiation cycle. The normal area should strictly follow the laser heating frequency. <miu2表示周期偏移或紊乱,可能因缺陷导致热传导延迟,如脱粘层形成热阻,升温滞后于激光照射,亦或是热扩散异常,如纤维断裂改变热流方向,导致非周期性波动;

[0068] Laser irradiation has a clear periodicity. Normal thermal response will show the same frequency fluctuation. Periodic disorder is a strong feature of defects. For example, when there is stratification inside the blade, the laser energy needs to pass through the air layer. The delay of heat conduction causes the temperature rise peak to lag behind the laser pulse, and PC drops significantly.

[0069] These three indicators cover the core characteristics of thermal response from the three dimensions of TVR, MOR, and PC, and are directly related to the physical mechanism of defects, namely, thermal resistance change, heat accumulation, and abnormal heat conduction path. As the core function of the first-level warning, it is to quickly filter out obviously abnormal pixels, narrow the scope of suspicious areas, and avoid missing serious defects.

[0070] The specific reasons for selecting the standard deviation coefficient CoV and the determination coefficient R2 as secondary confirmation include:

[0071] Standard deviation coefficient CoV, CoV reflects the relative amplitude of temperature fluctuation and measures thermal stability. If CoV>miu3, it means that the temperature fluctuates violently. It may be that defects lead to unstable local thermal state. For example, the edge of the surface crack is intensified by air convection, causing the temperature to fluctuate during laser irradiation. The internal inclusion defects cause periodic thermal stress release due to the difference in thermal expansion coefficient, resulting in an increase in the fluctuation amplitude.

[0072] Some environmental noise may cause instantaneous anomalies in TVR, such as a short-term sudden change in wind speed, but CoV remains stable, that is, the fluctuation amplitude does not continue to increase. However, real defects will destroy the thermal balance mechanism, resulting in continued intensification of fluctuations. CoV can distinguish between accidental noise and structural anomalies.

[0073] The coefficient of determination R2 reflects the quality of the linear fit of the temperature series and measures the consistency of the trend. <miu4,则表示温度变化趋势紊乱,无明确规律性,正常区域在稳定热输入下应呈现近似线性升温或降温,如激光持续照射时温度线性上升,而缺陷区域可能因热阻非线性变化,如脱粘面积随温度扩大,导致趋势曲线弯曲;

[0074] CoV and R2 focus on fluctuation stability and trend rationality to solve the problem of false alarms in the first-level warning. The first-level condition may trigger an alarm due to instantaneous interference, such as a short-term occlusion of the light spot, while the second-level condition requires that the fluctuation amplitude is abnormal and the trend is disordered, ensuring that only anomalies with physical consistency are judged as suspicious;

[0075] Combine all the pixel points of each thermal imaging picture to make a judgment on suspicion and normality, obtaining the total discrimination sequence of the thermal imaging image. Extract the pixel points marked as suspicious in the total discrimination sequence of the thermal imaging image, and locate them on the actual three-dimensional coordinates of the wind turbine blade, and mark this three-dimensional coordinate with 1.

[0076] S103: Fit the three-dimensional coordinates marked as 0 on the wind turbine blade through an oriented bounding box to obtain the optical reflection defect area, fit the three-dimensional coordinates marked as 1 through an oriented bounding box to obtain the thermal distribution defect area, and find the intersection of the optical reflection area and the thermal distribution fault area to obtain the specific area of the wind turbine blade defect;

[0077] Use the acoustic emission sensor on the unmanned aerial vehicle to collect acoustic signals in the specific area of the wind turbine blade defect, obtaining multiple groups of acoustic signal data to be detected. After denoising the acoustic signal data to be detected through local mean decomposition LMD combined with multi-scale fuzzy slope entropy MFSE, extract the feature vectors, and substitute the feature vectors into the trained GRU wind turbine blade defect judgment model to obtain the specific defect types inside the wind turbine blade;

[0078] The specific steps of using LMD combined with MFSE to denoise the acoustic signal are as follows:

[0079] (1) Decompose the original acoustic signal by LMD into multiple product function components PF and a residual term. Each PF corresponds to the signal components of different frequency scales;

[0080] The original acoustic signals collected are mostly non-linear and non-stationary signals. LMD can adaptively decompose complex signals into several PFs with clear physical meanings. Each PF corresponds to a specific frequency modulation feature, facilitating the subsequent separation of noise and effective signals;

[0081] Through decomposition, signals of different time scales and frequency ranges can be separated into different PFs, such as the low-frequency signals generated by defects and the high-frequency interference of environmental noise, providing a basis for noise identification;

[0082] (2) Calculate the MFSE mean Argmfse of each PF to measure its complexity at different time scales. The complexity of the noise component is higher. If Argmfse >= Argth, then classify this PF component as a noise component; otherwise, classify it as a defect signal component;

[0083] MFSE measures the complexity of the signal at different time scales. Calculating the mean can integrate the information of multiple scales, avoid the one-sidedness of single-scale analysis, and more comprehensively reflect the signal characteristics of the PF component;

[0084] The defect signals are regular at a specific scale and have low complexity; the noise shows random fluctuations at all scales and has high complexity. Calculating the mean value can highlight this overall difference and facilitate threshold setting;

[0085] The MFSE at a single scale may be affected by local noise interference. The mean value calculation can smooth out random fluctuations, improve the feature stability, and ensure a reliable distinction between the noise and signal components;

[0086] (3) Add the defect signal component and the residual term to obtain the denoised acoustic signal;

[0087] The residual term usually represents the low-frequency trend of the signal, does not contain periodic or high-frequency oscillation features, has extremely low complexity, and is essentially different from the high-complexity characteristics of the noise. It directly reflects the overall trend of the signal and should be directly retained to avoid information loss;

[0088] The specific method for constructing the GRU wind turbine blade defect judgment model is as follows:

[0089] Obtain several groups of normal and defective acoustic signals collected historically. After denoising these acoustic signals using LMD combined with MFSE, obtain pure normal signals and defective signals;

[0090] Extract time-domain features, frequency-domain features, time-frequency domain features, and non-linear features from the normal signals and defective signals, and combine the feature vectors of each acoustic signal;

[0091] The time-domain features include peak value, root mean square value, rise time, pulse width, and zero-crossing rate;

[0092] The peak value represents the maximum absolute value in the time series of the signal and can reflect the degree of stress concentration caused by defects;

[0093] The root mean square value characterizes the signal energy intensity and is sensitive to material damage, such as fiber fracture;

[0094] The rise time represents the time for the signal to rise from 10% to 90% of the peak value. The rise time may be extended in the defect area due to damping changes;

[0095] The pulse width represents the duration for the signal to be above the threshold, such as when the threshold is taken as half of the peak value;

[0096] The zero-crossing rate represents the number of times the signal crosses the zero point per unit time and reflects the oscillation frequency of the signal. Defects will change the material stiffness and cause changes in the zero-crossing rate;

[0097] The frequency-domain features include the main frequency of the fast Fourier transform FFT, the frequency band energy, and the skewness and kurtosis of the power spectral density PSD;

[0098] The main frequency of the FFT represents the frequency component with the highest energy, reflecting the defect size. Small defects correspond to high frequencies, and large defects correspond to low frequencies.

[0099] Band energy: Calculate the proportion of energy in each frequency band. Different defect types will lead to differences in the energy of characteristic frequency bands.

[0100] Skewness and kurtosis of the power spectral density PSD. Skewness reflects the symmetry of the frequency spectrum, and kurtosis reflects the sharpness of the frequency spectrum. Defects will cause abnormal frequency spectrum distribution.

[0101] Time-frequency domain features include wavelet energy entropy and the variance of the short-time Fourier transform STFT.

[0102] The entropy value of wavelet energy entropy reflects the uniformity of energy distribution in different frequency bands. Defects will cause energy concentration in specific frequency bands and a decrease in the entropy value.

[0103] The variance of STFT reflects the distribution characteristics of the signal in the time-frequency plane. The time-frequency distribution in the defect area is more disordered, and the variance increases.

[0104] Nonlinear features include permutation entropy PE and fractal dimension FD.

[0105] Permutation entropy calculates the probability distribution entropy value of different permutation patterns. Defects will increase the complexity of the signal.

[0106] The fractal dimension reflects the self-similarity of the signal. Material damage will cause changes in the surface or internal structure, and the FD value will change.

[0107] Divide the acoustic signal into a training set and a test set according to a 7:3 ratio to ensure the balance of each sample.

[0108] The initial parameters of the GRU model are as follows:

[0109] The input dimension is 13, the number of GRU layers is 2, the number of neurons in each layer is 64, the activation function is selected as tanh, the dropout rate is 0.2, the output dimension is q + 1, the learning rate is 0.001, the number of training epochs is 50, and the batch size is 32. Among them, q is the number of types of internal defects in the wind turbine blade.

[0110] Substitute the training set into the GRU model for training, select the categorical cross-entropy as the loss function, substitute the test set to adjust the internal parameters, and obtain the final GRU wind turbine blade defect judgment model.

[0111] Some of the data in the above formulas are numerically calculated after removing their dimensions. At the same time, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0112] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An acoustic emission-based defect detection method for wind turbine blades, characterized in that, Including: S101: Based on the collected multiple multispectral images, calculate the four-dimensional spectral vector corresponding to each pixel, calculate the angle between it and the reference spectral vector of the normal area. If it is greater than the angle threshold, mark this pixel as a suspicious defect pixel point to obtain a single multispectral image discrimination sequence. Process multiple discrimination sequences to obtain a total discrimination sequence, map the suspicious pixel points to the actual three-dimensional coordinates of the wind turbine blade, and mark them with 0; S102: Calculate the temperature value of each pixel point in each thermal imaging picture to obtain the temperature value sequence during this period. Calculate the total change rate, mean deviation rate, standard deviation coefficient, period coincidence degree, and determination coefficient. If the total change rate is not within the specified interval, or the mean deviation rate is greater than the first threshold, or the period coincidence degree is less than the second threshold, start the first-level warning and enter the second-level confirmation. If the standard deviation coefficient is greater than the third threshold and the determination coefficient is less than the fourth threshold, determine that this pixel point is a suspicious pixel point to obtain the total discrimination sequence of the thermal imaging image, map the suspicious pixel points to the actual three-dimensional coordinates, and mark them with 1; S103: Fit the three-dimensional coordinates marked as 0 and 1 on the wind turbine blade with an oriented bounding box to obtain the optical reflection defect and thermal distribution defect areas respectively. Take the intersection to determine the specific area of the surface defect of the wind turbine blade. Use the unmanned aerial vehicle acoustic emission sensor to locate this area and obtain the acoustic signal to be detected. After denoising by LMD combined with MFSE, extract the feature vector and substitute it into the GRU wind turbine blade defect judgment model to further determine whether there is an internal fault in this area.

2. The method for detecting defects in wind turbine blades based on acoustic emission according to claim 1, wherein, It also includes S100. The S100 irradiates the wind turbine blade with multiple lasers on the unmanned aerial vehicle, and uses a multispectral camera and a thermal imaging sensor to take a number of multispectral images and thermal imaging images within a fixed time period by carrying multiple lasers, a multispectral camera, a thermal imaging sensor and several acoustic emission sensors on the unmanned aerial vehicle.

3. The method for detecting defects of a wind turbine blade based on acoustic emission according to claim 2, wherein The bands selected by the multiple lasers include the visible band, the near-infrared band, and the short-wave infrared band.

4. The method for detecting defects of wind turbine blades based on acoustic emission according to claim 2, characterized in that, When collecting information about the wind turbine blade, the specific requirements for the surrounding environment include that the ambient temperature needs to be stable at 10 - 30 °C, the humidity needs to be <= 80% RH, the wind speed needs to be <= 5 m / s, the multispectral imaging needs uniform natural light, the thermal imaging needs to avoid direct sunlight, be far away from strong electromagnetic sources, and the temperature difference between the background temperature of the shooting area and the blade needs to be < 5 °C.

5. The method for detecting defects of a wind turbine blade based on acoustic emission according to claim 1, wherein Process the multiple multispectral image discrimination sequences, calculate the proportion PerDou and PerNor of the suspicious quantity and the normal quantity in each pixel point to the total quantity, and select max{PerDou, PerNor} as the overall judgment situation of this pixel point to obtain the total discrimination sequence of the multispectral image.

6. The method for detecting defects of a wind turbine blade based on acoustic emission according to claim 5, wherein, If the number of multiple multispectral images is even, calculate the proportion PerNei of the suspicious pixel points in the 3×3 neighborhood around this pixel point. If PerNei >= 0.5, then the current pixel point is determined to be suspicious, otherwise it is determined to be normal. If this pixel point is a corner pixel, the effective neighborhood is three pixels. If this pixel point is an edge pixel, the effective neighborhood is five pixels. The corner pixels are the upper left, lower left, upper right, and lower right, and the edge pixels are the left middle, right middle, upper middle, and lower middle.

7. The method for detecting defects of wind turbine blades based on acoustic emission according to claim 1, characterized in that The specific dynamic threshold construction method for each group of TVR is as follows: Collect M groups of data for the same area of defect-free blades, calculate the TVR of each group of pixels, and construct a normal distribution TVRnormal~N(μ,σ 2 ); Let Crmin = μ - k×σ, Crmax = μ + k×σ, where k is the quality confidence coefficient and can take 2 or 3.

8. The method for detecting defects of a wind turbine blade based on acoustic emission according to claim 1, characterized in that The specific steps for denoising the acoustic signal using LMD combined with MFSE are as follows: Decompose the original acoustic signal into multiple product function components PF and a residual term by LMD; Calculate the MFSE mean Argmfse of each PF. If Argmfse >= Argth, classify this PF component as a noise component; otherwise, classify it as a defect signal component; Add the defect signal components and the residual term to obtain the denoised acoustic signal.

9. The method for detecting defects of wind turbine blades based on acoustic emission according to claim 1, wherein, Extract time-domain features, frequency-domain features, time-frequency domain features, and non-linear features from the denoised acoustic signal. The time-domain features include peak value, root mean square value, rise time, pulse width, and zero-crossing rate. The frequency-domain features include the main frequency of the fast Fourier transform FFT, band energy, skewness and kurtosis of the power spectral density PSD. The time-frequency domain features include wavelet energy entropy and variance of the short-time Fourier transform STFT. The non-linear features include permutation entropy PE and fractal dimension FD.

10. The method for detecting defects of a wind turbine blade based on acoustic emission according to claim 1, characterized in that, The initial parameters of the GRU model specifically include an input dimension of 13, 2 GRU layers, 64 neurons in each layer, the activation function is selected as tanh, the dropout rate is 0.2, the output dimension is q + 1, the learning rate is 0.001, the number of training epochs is 50, and the batch size is 32, where q is the number of types of internal defects of the wind turbine blade.

Images