Non-contact mountain fan blade state method, system and equipment based on sound characteristics and medium

Abstract

The invention discloses a non-contact mountain fan blade state method, system and equipment based on sound characteristics and a medium, and belongs to the technical field of fan blade state analysis, and the method comprises the steps: collecting a multi-channel sound signal of fan blade operation, and recording an instantaneous rotating speed signal and an instantaneous wind speed signal; periodically segmenting the multi-channel sound signal by taking the blade passing period as a time window to obtain a sound signal segment corresponding to the rotating phase of each blade; performing noise suppression on the sound signal segment by using the filtering parameter to obtain a target sound signal segment; extracting a multi-scale feature set from the target sound signal segment; generating a fusion feature vector; and judging an abnormal blade and a damage type and outputting a state evaluation result. According to the invention, periodic segmentation of phase alignment is carried out according to the period of the blades, and the multi-channel sound signal is segmented into sound signal segments corresponding to the rotation phase of each blade.

Description

Technical Field

[0001] This invention relates to the field of wind turbine blade condition analysis technology, specifically to a non-contact method, system, device, and medium for analyzing the condition of mountain wind turbine blades based on sound characteristics. Background Technology

[0002] Mountain wind turbine blades are exposed to complex climatic environments for extended periods, and are affected by factors such as wind shear, turbulence, dust, and drastic changes in temperature and humidity. As a result, blade surface wear, cracks, and icing are prominent problems, which seriously affect the safe operation and power generation efficiency of the wind turbine.

[0003] Existing blade condition monitoring technologies mainly include manual visual inspection, drone inspection, and vibration monitoring. Manual visual inspection is inefficient and poses safety hazards; drone inspection is limited by weather conditions and cannot achieve real-time monitoring; vibration monitoring requires installing sensors on the blades or hubs, which presents problems such as difficult installation and maintenance, easy sensor damage, unstable data transmission, and difficulty in achieving precise positioning at the single-blade level. Acoustic monitoring, as a non-contact method, has advantages such as simple installation, no downtime required, and continuous monitoring capability. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a non-contact method, system, device and medium for determining the blade status of mountain wind turbines based on sound characteristics.

[0005] Therefore, the technical problem solved by the present invention is that existing wind turbine blade monitoring methods require contact-type sensor installation, are severely affected by strong wind noise in mountainous environments, cannot distinguish the state of individual blades, and are difficult to simultaneously achieve damage location and type identification.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a non-contact method for determining the blade status of a mountain wind turbine based on sound characteristics, comprising, Acoustic sensors arranged in a ring array synchronously collect multi-channel sound signals of the wind turbine blades, and simultaneously record instantaneous rotational speed signals and instantaneous wind speed signals; The blade passage period is calculated based on the instantaneous rotation speed signal. The multi-channel sound signal is periodically segmented using the blade passage period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade. Based on the gradient of the instantaneous wind speed signal, filter parameters are selected from a preset noise parameter library, and the filter parameters are used to suppress noise in the sound signal segment to obtain the target sound signal segment. Extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passing frequency, the energy distribution of each frequency band of the wavelet packet, and the signal envelope kurtosis coefficient. Establish time series of each feature parameter in the multi-scale feature set, assign fusion weights to each feature parameter according to the monotonicity index and volatility index of the time series, and generate a fusion feature vector; The fused feature vector is mapped to a pre-established feature reference space, and the Mahalanobis distance with the feature reference space is calculated. When the Mahalanobis distance exceeds a preset threshold, the abnormal blade and damage type are determined and the status assessment result is output based on the imbalance coefficient of the harmonic amplitude ratio relationship, its phase position, and the growth slope of the high-frequency energy in the energy proportion distribution.

[0007] As a preferred embodiment of the non-contact mountain wind turbine blade state method based on sound characteristics described in this invention, the step of obtaining the sound signal segment corresponding to the rotation phase of each blade includes calculating the time interval between adjacent pulses as the initial blade passage period based on the arrival time of the continuous blade position pulses in the instantaneous rotation speed signal. A time series of the initial blade passage period is established, and the period values ​​in the time series are weighted to obtain the corrected blade passage period. Extract the amplitude feature time of the multi-channel sound signal and the arrival time of the blade position pulse, and calculate the time difference between the two as a time calibration parameter; Based on the modified blade passage period and the time calibration parameters, periodic segmentation boundaries are calibrated on the multi-channel sound signal, and the multi-channel sound signal is divided into multiple time periods according to the periodic segmentation boundaries; Extract the energy distribution curve of the sound signal within each time period, determine the time boundary of each blade according to the extreme value position in the energy distribution curve, and divide each time period into sound signal segments corresponding to each blade according to the time boundary.

[0008] The beneficial effect of this preferred technical solution is that, by correcting the blade passage period and performing time calibration, this technology effectively improves the accuracy of signal segmentation. The corrected period eliminates errors caused by fluctuations in wind speed and rotational speed, ensuring that the sound signal of each blade accurately corresponds to its rotational phase.

[0009] As a preferred embodiment of the non-contact mountain wind turbine blade state method based on sound characteristics described in this invention, the step of obtaining the target sound signal segment includes calculating the ratio of the numerical change of the instantaneous wind speed signal to the time change within a set time window to obtain the wind speed change gradient. The noise frequency band range parameter is determined based on the magnitude of the wind speed change gradient. A reference sensor is set up to collect ambient background sound signals. The ambient background sound signals and the sound signal segments are subjected to time-frequency transformation to obtain the ambient background spectrum and the sound signal segment spectrum. Based on the noise frequency band range parameters, the spectral components of the corresponding frequency bands are extracted from the environmental background spectrum as noise spectrum references; The spectral components in the audio signal segment spectrum that are similar to the noise spectrum reference are marked as noise spectrum components. After subtracting the noise spectrum components from the audio signal segment spectrum, an inverse transformation is performed to obtain the target audio signal segment.

[0010] The beneficial effect of this preferred technical solution is that it solves the interference problem caused by wind speed changes and environmental noise fluctuations in mountainous environments by using a dynamic noise suppression method based on wind speed variation gradients. Real-time calculation of the wind speed variation gradient allows noise suppression to be automatically adjusted according to the actual environment, accurately removing noise components and ensuring signal clarity.

[0011] As a preferred embodiment of the non-contact mountain wind turbine blade status method based on sound features described in this invention, the step of extracting a multi-scale feature set from the target sound signal segment includes: calculating the blade passage frequency based on the instantaneous rotational speed signal and the number of blades; performing spectral analysis on the target sound signal segment to obtain a spectral amplitude sequence; extracting the amplitude corresponding to each harmonic frequency of the blade passage frequency from the spectral amplitude sequence; calculating the ratio of each harmonic amplitude to the first harmonic amplitude to obtain the harmonic amplitude ratio relationship. The target sound signal segment is decomposed into wavelet packet coefficients for each decomposed frequency band. The energy value of the wavelet packet coefficients for each decomposed frequency band is calculated. The ratio of the energy value of each decomposed frequency band to the total energy value of all decomposed frequency bands is used as the energy proportion to obtain the energy proportion distribution of each frequency band. Envelope extraction is performed on the target sound signal segment to obtain a signal envelope sequence. The ratio of the fourth central moment to the square of the second central moment of the signal envelope sequence is calculated to obtain the signal envelope kurtosis coefficient.

[0012] As a preferred embodiment of the non-contact mountain wind turbine blade state method based on sound features described in this invention, the generation of the fusion feature vector includes: mapping each harmonic in the harmonic amplitude ratio relationship to the corresponding blade according to the instantaneous speed signal and the number of blades, and extracting the harmonic subset corresponding to each blade as the harmonic feature component of that blade. The target sound signal segment is divided into leaf-shaped time-domain segments, and the wavelet packet energy ratio distribution in each leaf time period is calculated. The high-frequency energy ratio increment is extracted as the energy feature component of each leaf. The signal envelope kurtosis coefficient is calculated for each leaf time period and used as the kurtosis feature component of each leaf. Time series are established for the harmonic characteristic components, energy characteristic components and kurtosis characteristic components of each blade. The monotonically increasing coefficient of each component time series is calculated as a monotonicity index. The stability coefficient of each component time series after excluding the instantaneous wind speed signal change period is calculated as a volatility index. For each leaf, the feature components whose monotonicity index is greater than the threshold and whose volatility index is less than the threshold are selected into the effective feature subset of the leaf. The feature components in the effective feature subset are normalized and then averaged to obtain the fused feature value of the leaf. The fusion feature values ​​of each blade are combined into a fusion feature vector.

[0013] As a preferred embodiment of the non-contact mountain wind turbine blade state method based on sound features described in this invention, the calculation of the Mahalanobis distance with the feature reference space includes: collecting fused feature vectors of different operating points under normal operating conditions as training samples, calculating the mean vector and covariance matrix of the training samples, and establishing a feature reference space with the mean vector as the center and the covariance matrix as the shape parameter. Obtain the fused feature vector and calculate the difference vector between the fused feature vector and the mean vector; Calculate the product of the difference vector and the inverse of the covariance matrix, then multiply it by the transpose of the difference vector, and take the square root of the result to obtain the Mahalanobis distance. The Mahalanobis distance is compared with a preset threshold. If the Mahalanobis distance exceeds the preset threshold, the current blade is determined to be abnormal. If the Mahalanobis distance does not exceed the preset threshold, the current blade is determined to be operating normally.

[0014] As a preferred embodiment of the non-contact mountain wind turbine blade status method based on sound characteristics described in this invention, the step of determining abnormal blades and damage types and outputting status assessment results based on the imbalance coefficient and phase position of the harmonic amplitude ratio relationship and the growth slope of high-frequency energy in the energy proportion distribution includes extracting the harmonic ratio values ​​of each harmonic in the harmonic amplitude ratio relationship of the abnormal blades, calculating the deviation of each harmonic ratio value from the normal reference value, and selecting the harmonic number with the largest deviation. The blade number corresponding to the calculation result is the abnormal blade number, based on the harmonic order with the largest deviation and the number of blades. Extract the proportion values ​​of multiple harmonic orders corresponding to the abnormal blades, calculate the standard deviation of multiple harmonic proportion values, and use it as the unbalance coefficient. Acquire the sound signal of the abnormal blade in the corresponding time period of the target sound signal segment, perform wavelet packet decomposition on the sound signal of the time period to obtain the energy proportion of each decomposition frequency band, extract the time series of the energy proportion of the high frequency decomposition frequency band in multiple monitoring periods, perform linear regression calculation on the time series to obtain the slope, and use it as the energy growth slope of the high frequency band. When the imbalance coefficient is greater than the first judgment threshold and the energy growth slope in the high-frequency band is less than the second judgment threshold, it is judged as an aerodynamic imbalance type. When the imbalance coefficient is less than the first judgment threshold and the energy growth slope in the high-frequency band is greater than the second judgment threshold, it is judged as a surface damage type. When the imbalance coefficient is greater than the first judgment threshold and the energy growth slope in the high-frequency band is greater than the second judgment threshold, it is judged as a composite damage type. The output includes a status assessment result containing the abnormal blade number and damage type.

[0015] This invention provides a non-contact mountain wind turbine blade condition analysis system based on sound characteristics.

[0016] To solve the above technical problems, the present invention provides the following technical solution: a non-contact mountain wind turbine blade state analysis system based on sound characteristics, comprising: a data acquisition module, used to synchronously acquire multi-channel sound signals of wind turbine blade operation through acoustic sensors arranged in a ring array, and synchronously record instantaneous rotational speed signals and instantaneous wind speed signals; The signal segmentation module is used to calculate the blade passing period based on the instantaneous rotation speed signal, and to periodically segment the multi-channel sound signal using the blade passing period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade. The noise suppression module is used to select filtering parameters from a preset noise parameter library based on the change gradient of the instantaneous wind speed signal, and use the filtering parameters to suppress noise in the sound signal segment to obtain the target sound signal segment. The feature extraction module is used to extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passing frequency, the energy distribution of each frequency band of the wavelet packet, and the signal envelope kurtosis coefficient. The feature fusion module is used to establish the time series of each feature parameter in the multi-scale feature set, assign fusion weights to each feature parameter according to the monotonicity index and volatility index of the time series, and generate a fusion feature vector. An anomaly detection module is used to map the fused feature vector to a pre-established feature reference space, calculate the Mahalanobis distance with the feature reference space, and determine that an anomaly exists when the Mahalanobis distance exceeds a preset threshold. The damage diagnosis module is used to determine the abnormal blade and damage type and output the status assessment result based on the imbalance coefficient and phase position of the harmonic amplitude ratio relationship and the growth slope of the high-frequency energy in the energy proportion distribution.

[0017] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the non-contact mountain wind turbine blade state method based on sound characteristics.

[0018] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the aforementioned non-contact mountain wind turbine blade state method based on sound features.

[0019] The beneficial effects of this invention are as follows: By periodically segmenting the multi-channel sound signal according to the phase alignment of the blade's passage period, the multi-channel sound signal is divided into sound signal segments corresponding to the rotation phase of each blade. After an anomaly is detected, the deviation of each harmonic ratio value from the normal reference value is extracted. Based on the harmonic number with the largest deviation and the number of blades, a modulo operation is performed to determine the abnormal blade number, thus achieving precise positioning from frequency domain characteristics to spatial location.

[0020] By calculating the gradient of the instantaneous wind speed signal, the noise frequency band range parameter is dynamically determined based on the magnitude of the gradient. Combined with the background sound signal, a noise spectrum benchmark is established for adaptive noise suppression, so that the noise reduction parameters are dynamically adjusted with the change of mountain wind speed.

[0021] By combining two physically independent indicators—the imbalance coefficient and the high-frequency energy growth slope—three types of damage—aerodynamic imbalance, surface damage, and combined damage—can be distinguished. A multi-scale feature fusion strategy is employed for each blade, generating fused feature values ​​for each blade and then combining them into a multi-dimensional vector. This improves monitoring stability while preserving the spatial distinguishability of each blade. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The above is a flowchart of a non-contact mountain wind turbine blade status method based on sound features, provided as an embodiment of the present invention. Detailed Implementation

[0024] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0025] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a non-contact method for determining the blade status of a mountain wind turbine based on sound characteristics, including: Step 1: Acoustic sensors arranged in a ring array synchronously collect multi-channel sound signals of the wind turbine blades, and synchronously record instantaneous speed signals and instantaneous wind speed signals; Step 2: Calculate the blade passing period based on the instantaneous rotation speed signal, and periodically segment the multi-channel sound signal using the blade passing period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade; Step 3: Select filtering parameters from a preset noise parameter library based on the gradient of the instantaneous wind speed signal, and use the filtering parameters to suppress noise in the sound signal segment to obtain the target sound signal segment; Step 4: Extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passing frequency, the energy distribution of each frequency band of the wavelet packet, and the signal envelope kurtosis coefficient. Step 5: Establish the time series of each feature parameter in the multi-scale feature set, assign fusion weights to each feature parameter according to the monotonicity index and volatility index of the time series, and generate a fusion feature vector; Step 6: Map the fused feature vector to the pre-established feature reference space, calculate the Mahalanobis distance with the feature reference space, and when the Mahalanobis distance exceeds the preset threshold, determine the abnormal blade and damage type and output the status assessment result based on the imbalance coefficient of the harmonic amplitude ratio relationship and its phase position and the growth slope of the high-frequency energy in the energy ratio distribution.

[0026] Environmental noise in mountain wind farms varies significantly with instantaneous wind speed, and existing noise suppression methods with fixed parameters are ill-suited to these fluctuations. This embodiment synchronously acquires instantaneous wind speed signals in step 1, and selects filtering parameters from a pre-set noise parameter library in step 3 based on the gradient of the instantaneous wind speed signal. When the wind speed gradient is large, the frequency band of wind noise expands, allowing for the selection of filtering parameters covering a wider frequency band; when the wind speed gradient is small, wind noise is concentrated in a specific frequency band, requiring the selection of more targeted filtering parameters. By using the gradient of the instantaneous wind speed signal as an indicator variable of noise characteristics, the noise suppression process adapts to fluctuations in mountainous conditions, solving the problem of noise reduction failure caused by dynamic changes in the noise spectrum in mountainous environments. After noise suppression of the sound signal segment using the aforementioned filtering parameters, the resulting target sound signal segment retains the blade passage frequency and its harmonic components, providing a reliable signal basis for subsequent extraction of harmonic amplitude ratios and energy distribution.

[0027] Traditional acoustic monitoring assesses the wind turbine as a whole, failing to pinpoint the exact location of abnormal blades. This embodiment calculates the blade passage period based on the instantaneous rotational speed signal in step 2, periodically segments the multi-channel sound signal according to the blade passage period, obtaining sound signal segments corresponding to the rotation phase of each blade, thus establishing a correspondence between each blade and the sound signal segment. In step 5, time series of each characteristic parameter are established, and a fused feature value is generated for each blade. The fused feature values ​​of each blade are combined into a fused feature vector. In step 6, when the Mahalanobis distance exceeds a preset threshold, the harmonic order with the largest deviation in the harmonic amplitude ratio is extracted. Modulus operation is performed based on this harmonic order and the number of blades to determine the abnormal blade number. The mapping from frequency domain features to spatial location is achieved through the modulus operation relationship between the phase position of the harmonic order and the blade number. Furthermore, the damage type is determined based on the imbalance coefficient and high-frequency energy growth slope of the abnormal blade. This method refines the monitoring object to a single blade and completes the entire analysis process of anomaly detection, spatial location, and damage classification.

[0028] Example 2, an embodiment of the present invention, provides a non-contact mountain wind turbine blade status method based on sound characteristics, based on the previous embodiment, including: In this embodiment, step 1 involves synchronously acquiring multi-channel sound signals from the wind turbine blades using acoustic sensors arranged in a ring array, and simultaneously recording instantaneous rotational speed and instantaneous wind speed signals. The ring array of acoustic sensors is achieved by: uniformly arranging multiple acoustic sensors along the horizontal circumference outside the wind turbine tower to form a ring array; the installation distance of the acoustic sensors is determined based on the diameter of the blade sweep plane; and the axial direction of each acoustic sensor is adjusted to point towards the geometric center of the blade sweep plane. A rotational speed sensor is installed at the wind turbine hub, outputting a position pulse when each blade passes a fixed reference position. A wind speed sensor is installed on the top of the wind turbine nacelle to measure the instantaneous wind speed. Each sensor is connected to a data acquisition device, which receives the position pulse from the rotational speed sensor as a trigger signal and synchronously starts data acquisition for each channel upon receiving the trigger signal. In the acquired multi-channel sound signals, each channel corresponds to the output of an acoustic sensor at the same moment. The instantaneous rotational speed signal records the arrival time of each blade position pulse. The instantaneous wind speed signal records the wind speed value within the same time period as the multi-channel sound signals.

[0029] In an optional implementation, step 1 involves synchronously acquiring multi-channel sound signals from the wind turbine blades using acoustic sensors arranged in a ring array, and simultaneously recording instantaneous rotational speed and instantaneous wind speed signals. The ring array of acoustic sensors can be achieved by uniformly arranging multiple sensors along a vertical circumference on the outside of the wind turbine tower, forming a ring array. The plane of the vertical circumference coincides with or nearly coincides with the blade sweeping plane. Each acoustic sensor is distributed along the vertical circumference, covering the upper and lower regions of the blade sweeping plane. Rotational speed sensors are installed on the generator shaft or main shaft, outputting pulse signals synchronized with the rotational speed. The position pulses of each blade are obtained by frequency division or multiplication of the pulse signals. Wind speed sensors are installed on the side of the nacelle, measuring the wind speed in the direction of the incoming flow. The data acquisition device achieves synchronous acquisition of signals from each channel through hardware or software triggering, maintaining a time correspondence between the acquisition time and the position pulses of the rotational speed sensors. The vertical circumferential arrangement allows the ring array to acquire sound signals as the blades pass through both the top dead center and bottom dead center positions.

[0030] In another optional implementation, step 1 involves synchronously acquiring multi-channel sound signals from the wind turbine blades using acoustic sensors arranged in a ring array, and simultaneously recording instantaneous rotational speed and instantaneous wind speed signals. The ring array of acoustic sensors can also be achieved by uniformly arranging multiple acoustic sensors along an oblique circumference at an angle to the blade sweep plane on the outer side of the wind turbine tower, forming a ring array. The oblique direction of the circumference is at a certain angle to the ground normal. The axial direction of each acoustic sensor is adjusted according to its position on the circumference, so that the acoustic axis of each sensor points to the geometric center or a region near the center of the blade sweep plane. The rotational speed sensor acquires position pulses by detecting marks or magnets on the rotating shaft; the circumferential position of the marks or magnets corresponds to the installation angle of each blade. The wind speed sensor measures the horizontal and vertical components of the wind speed and records wind direction information. The sensor signals are transmitted to a data processing device via wired or wireless connection. The data processing device adds time stamps to each signal, achieving time alignment of the signals from each channel. The oblique circumferential arrangement avoids interference from ground-reflected sound waves on the sound signals.

[0031] Step 2: Calculate the blade passage period based on the instantaneous rotational speed signal, and periodically segment the multi-channel sound signal using the blade passage period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade, including the following steps A1-A5: A1: Based on the arrival time of the continuous blade position pulses in the instantaneous rotational speed signal, the time interval between adjacent pulses is calculated as the initial blade passing period; A2: Establish a time series of the initial blade passage period, and perform weighted processing on the period values ​​in the time series to obtain the corrected blade passage period; A3: Extract the amplitude feature time of the multi-channel sound signal and the arrival time of the blade position pulse, and calculate the time difference between the two as a time calibration parameter; A4: Based on the corrected blade passage period and the time calibration parameters, periodic segmentation boundaries are calibrated on the multi-channel sound signal, and the multi-channel sound signal is divided into multiple time periods according to the periodic segmentation boundaries; A5: Extract the energy distribution curve of the sound signal within each time period, determine the time boundary of each blade according to the extreme value position in the energy distribution curve, and divide each time period into sound signal segments corresponding to each blade according to the time boundary.

[0032] Step 3: Select filtering parameters from a preset noise parameter library based on the gradient of the instantaneous wind speed signal, and use the filtering parameters to suppress noise in the sound signal segment to obtain the target sound signal segment, including the following steps B1-B5: B1: Calculate the ratio of the numerical change to the time change of the instantaneous wind speed signal within a set time window to obtain the wind speed change gradient; B2: Determine the noise frequency band range parameters based on the magnitude of the wind speed change gradient; B3: Set a reference sensor to collect ambient background sound signals, and perform time-frequency transformation on the ambient background sound signals and the sound signal segments respectively to obtain the ambient background spectrum and the sound signal segment spectrum; B4: Extract the spectral components of the corresponding frequency band from the environmental background spectrum based on the noise frequency band range parameters as the noise spectrum reference; B5: Mark the spectral components in the sound signal segment spectrum that are similar to the noise spectrum reference as noise spectral components, subtract the noise spectral components from the sound signal segment spectrum and then perform an inverse transformation to obtain the target sound signal segment.

[0033] In this embodiment of the application, in step B5, the target sound signal segment is determined by: extracting the value of the instantaneous wind speed signal at the start time within a set time window. and the value at the termination time Set the duration of the time window to be... Calculate the numerical change of the instantaneous wind speed signal. The ratio of the numerical change to the time change is the wind speed gradient. When wind speed changes drastically in mountainous wind farms, turbulence intensity increases, and the wind noise frequency band expands to higher frequencies. The method for establishing a preset noise parameter database is as follows: before wind turbine operation, measure the environmental noise spectrum under different wind speed variation gradients, statistically analyze the main noise frequency band range corresponding to each wind speed variation gradient, and record this as the noise frequency band range parameter. Based on the absolute value of the wind speed variation gradient... Search the preset noise parameter library, when The noise frequency band range parameter is ,when At that time ,when At that time A reference sensor is placed outside the blade sweep area to collect ambient background sound signals. For ambient background sound signals Perform a short-time Fourier transform:

[0034] in For frequency index, For frame index, For the length of the window, For frame shift, Using a window function, the ambient background spectrum is obtained. For audio signal segments The spectrum of the sound signal segment is obtained by performing a short-time Fourier transform with the same parameters. Extracting the frequency index corresponding to the frequency value from the environmental background spectrum. The spectral components within the specified range are denoted as the noise spectrum reference. Calculate the amplitude ratio of the audio signal spectrum segment to the noise spectrum reference. When the amplitude ratio is less than the similarity threshold At that time, the spectral components corresponding to the frequency index and frame index are labeled as noise spectral components. Similarity threshold. A value of 1.5 indicates that spectral components whose amplitude does not exceed 1.5 times the noise spectrum reference amplitude are considered noise. For the frequency index and frame index marked as noise spectral components, the corresponding noise spectrum reference is subtracted from the audio signal segment spectrum.

[0035] in This is the noise suppression coefficient in the filtering parameters, with a value of 0.9. Unlabeled spectral components remain unchanged. The processed spectrum... The target sound signal segment is obtained by performing a short-time inverse Fourier transform.

[0036] In an optional implementation, in step B5, the target sound signal segment can be determined by: calculating the ratio of the instantaneous wind speed signal's numerical change to its temporal change within a set time window to obtain the wind speed change gradient; and determining the noise frequency band range parameter from a preset noise parameter library based on the magnitude of the wind speed change gradient. A reference sensor is set upwind of the blade sweep area. Short-time Fourier transforms are performed on the ambient background sound signal and the sound signal segment's spectrum to obtain the ambient background spectrum and the sound signal segment's spectrum, respectively. The spectral components corresponding to the noise frequency band range parameter are extracted from the ambient background spectrum, and the median amplitude of all frames for each frequency index is calculated as the noise spectrum reference. The difference between the amplitude of each frame at each frequency index of the sound signal segment's spectrum and the corresponding frequency index noise spectrum reference is calculated; differences less than half of the noise spectrum reference are marked as noise spectrum components. The noise spectrum reference amplitude corresponding to the marked spectral components is subtracted from the sound signal segment's spectrum; amplitudes less than zero after subtraction are set to zero. A short-time Fourier inverse transform is performed on the processed spectrum to obtain the target sound signal segment.

[0037] In another optional implementation, in step B5, the target sound signal segment can also be determined by: calculating the ratio of the instantaneous wind speed signal's numerical change to its temporal change within a set time window to obtain the wind speed change gradient; and determining the noise frequency band range parameter from a preset noise parameter library based on the magnitude and sign of the wind speed change gradient. A reference sensor is installed at the bottom of the wind turbine tower. Short-time Fourier transforms are performed on the ambient background sound signal and the sound signal segment to obtain the ambient background spectrum and the sound signal segment spectrum, respectively. The spectral components corresponding to the noise frequency band range parameter are extracted from the ambient background spectrum as the noise spectrum reference. The amplitude difference between each frame of each frequency index in the sound signal segment spectrum and the amplitude of the corresponding frame of the noise spectrum reference frequency index is calculated, and the difference is divided by the noise spectrum reference amplitude to obtain a normalized difference. Normalized differences less than 0.3 are marked as noise spectrum components. The target sound signal segment is obtained by subtracting the noise spectrum reference amplitude corresponding to the marked spectral components multiplied by the reciprocal of the normalized difference and then multiplied by 0.9 from the sound signal segment spectrum, and then performing a short-time Fourier inverse transform on the processed spectrum.

[0038] Step 4: Extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passage frequency, the energy distribution of each frequency band of the wavelet packet, and the kurtosis coefficient of the signal envelope, including the following steps C1-C3: C1: Calculate the blade passing frequency based on the instantaneous rotation speed signal and the number of blades, perform spectral analysis on the target sound signal segment to obtain a spectral amplitude sequence, extract the amplitude corresponding to each harmonic frequency of the blade passing frequency from the spectral amplitude sequence, calculate the ratio of each harmonic amplitude to the first harmonic amplitude, and obtain the harmonic amplitude ratio relationship. C2: Perform wavelet packet decomposition on the target sound signal segment to obtain the wavelet packet coefficients of each decomposed frequency band, calculate the energy value of the wavelet packet coefficients of each decomposed frequency band, and take the ratio of the energy value of each decomposed frequency band to the total energy value of all decomposed frequency bands as the energy proportion to obtain the energy proportion distribution of each frequency band. C3: Extract the envelope of the target sound signal segment to obtain the signal envelope sequence, calculate the ratio of the fourth central moment to the square of the second central moment of the signal envelope sequence, and obtain the signal envelope kurtosis coefficient.

[0039] In this embodiment of the application, in step C3, the signal envelope kurtosis coefficient is determined by: The instantaneous rotational speed is calculated based on the time interval between consecutive blade position pulses in the instantaneous rotational speed signal. Multiplying the instantaneous rotational speed by the number of blades yields the blade passing frequency. For the target sound signal segment The frequency domain representation is obtained by performing a fast Fourier transform. The spectral amplitude sequence is Extract the amplitude corresponding to each harmonic frequency of the blade passage frequency from the spectral amplitude sequence. The frequency of the second harmonic is The corresponding amplitude is denoted as Calculate the ratio of the amplitude of each harmonic to the amplitude of the first harmonic. The harmonic amplitude ratio relationship is obtained. ,in To extract the highest harmonic order, wavelet packet decomposition was performed on the target audio signal segment using the db4 wavelet basis function. The decomposition consisted of three levels, yielding wavelet packet coefficients for eight frequency bands. The wavelet packet coefficients of each decomposed frequency band are denoted as follows: Using coefficient indexes, the energy value of this frequency band is calculated as follows:

[0040] The total energy value of all decomposed frequency bands is , No. The energy percentage of each decomposed frequency band is The energy distribution of each frequency band was obtained. The envelope of the target sound signal segment is extracted, and the analytic signal of the target sound signal segment is calculated using the Hilbert transform method.

[0041] in For Hilbert transform operators, This is an analytic signal. The signal envelope sequence is the amplitude of the analytic signal:

[0042] Calculate the mean of the signal envelope sequence. ,in Let be the sequence length. Calculate the second-order central moments of the signal envelope sequence:

[0043] Calculate the fourth central moments of the signal envelope sequence:

[0044] The signal envelope kurtosis coefficient is the ratio of the fourth central moment to the square of the second central moment:

[0045] The kurtosis coefficient of a signal envelope reflects the peak characteristics of the signal envelope sequence. The impact caused by damage to the blade surface leads to peaks in the envelope sequence, increasing the kurtosis coefficient.

[0046] In an optional implementation, in step C3, the signal envelope kurtosis coefficient can be obtained by: calculating the blade passage frequency based on the instantaneous rotation speed signal and the number of blades; performing a fast Fourier transform on the target sound signal segment to obtain a spectral amplitude sequence; extracting the harmonic amplitudes of each harmonic passage frequency; and calculating the ratio of each harmonic amplitude to the first harmonic amplitude to obtain the harmonic amplitude ratio. Wavelet packet decomposition is performed on the target sound signal segment using the sym8 wavelet basis function, with a decomposition level of 4 layers, resulting in wavelet packet coefficients for 16 decomposed frequency bands. The sum of the squares of the wavelet packet coefficients for each decomposed frequency band is calculated as the energy value. The energy value is divided by the total energy value of all frequency bands to obtain the energy proportion, thus obtaining the energy proportion distribution of each frequency band. Envelope extraction is performed on the target sound signal segment using the square envelope detection method. The squared values ​​of each sampling point of the target sound signal segment are passed through a low-pass filter to obtain the squared envelope. The square root of the squared envelope is then used to obtain the signal envelope sequence. Calculate the mean and standard deviation of the signal envelope sequence. Subtract the mean from the signal envelope sequence and divide by the standard deviation to obtain the standardized envelope sequence. Calculate the fourth moment of the standardized envelope sequence as the signal envelope kurtosis coefficient.

[0047] In another optional implementation, in step C3, the signal envelope kurtosis coefficient can also be obtained by: calculating the blade passage frequency based on the instantaneous rotation speed signal and the number of blades; performing a fast Fourier transform on the target sound signal segment to obtain a spectral amplitude sequence; extracting the harmonic amplitudes of each harmonic at the blade passage frequency; and calculating the ratio of each harmonic amplitude to the first harmonic amplitude to obtain the harmonic amplitude ratio. Wavelet packet decomposition is performed on the target sound signal segment using the coif5 wavelet basis function, with a decomposition level of 3 layers, resulting in wavelet packet coefficients for 8 decomposition frequency bands. The energy value of the wavelet packet coefficients in each decomposition frequency band is calculated, and the energy value is divided by the total energy value of all frequency bands to obtain the energy proportion, thus obtaining the energy proportion distribution of each frequency band. Envelope extraction is performed on the target sound signal segment. First, bandpass filtering is applied to the target sound signal segment, with the filtering frequency band ranging from 0.5 to 10 times the blade passage frequency. The filtered signal is then subjected to full-wave rectification, and the rectified signal is passed through a low-pass filter with a cutoff frequency of 0.2 times the blade passage frequency to obtain the signal envelope sequence. Calculate the second and fourth central moments of the signal envelope sequence. Divide the fourth central moment by the square of the second central moment and subtract 3 to obtain the excess kurtosis. Add 3 to the excess kurtosis to obtain the signal envelope kurtosis coefficient.

[0048] Step 5: Establishing the time series of each feature parameter in the multi-scale feature set, assigning fusion weights to each feature parameter according to the monotonicity and volatility indices of the time series, and generating the fused feature vector includes the following steps D1-D6: D1: Based on the instantaneous rotational speed signal and the number of blades, map each harmonic in the harmonic amplitude ratio relationship to the corresponding blade, and extract the harmonic subset corresponding to each blade as the harmonic feature component of that blade. D2: Divide the target sound signal segment into leaf-shaped time-domain segments, calculate the wavelet packet energy ratio distribution in each leaf time period, and extract the high-frequency energy ratio increment as the energy feature component of each leaf. D3: Calculate the signal envelope kurtosis coefficient for each leaf time period, and use it as the kurtosis feature component of each leaf; D4: Establish time series for the harmonic characteristic component, energy characteristic component and kurtosis characteristic component of each blade, calculate the monotonically increasing coefficient of each component time series as a monotonicity index, and calculate the stability coefficient of each component time series after excluding the instantaneous wind speed signal change period as a volatility index. D5: For each leaf, select the feature components whose monotonicity index is greater than the threshold and whose volatility index is less than the threshold into the effective feature subset of the leaf. Normalize each feature component in the effective feature subset and then calculate the average to obtain the fused feature value of the leaf. D6: Combine the fusion feature values ​​of each blade into a fusion feature vector.

[0049] It should be noted that for three-bladed fans, the harmonics in the harmonic amplitude ratio relationship satisfy the modulo operation relationship with the blade number, and the harmonic order... The corresponding blade is determined by taking the modulus of the number 3 blades. Corresponding to the first blade, Corresponding to the second blade, Corresponding to the 3rd blade. Extract the harmonic amplitude ratio. The harmonic order corresponding to the first blade is The corresponding harmonic amplitude ratio is The arithmetic mean of this harmonic subset is calculated as the harmonic characteristic component of the first blade. Repeat the above calculations for the second and third blades to obtain the harmonic characteristic components of each blade. Using the sound signal segments corresponding to the rotation phase of each blade obtained in step 2, segment the target sound signal segments into blade time-domain segments according to the time boundaries corresponding to each blade. For the first... Wavelet packet decomposition was performed on the target sound signal during each blade passage time period to obtain the energy distribution of each frequency band. The frequency band with a center frequency greater than 5 times the blade passage frequency was defined as the high-frequency band. The energy distribution of all high-frequency bands was summed to obtain the high-frequency energy proportion. Record the first [item] in multiple consecutive monitoring cycles. High-frequency energy percentage sequence of each blade Calculate the difference in high-frequency energy proportion between adjacent monitoring periods, and take the difference of the most recent monitoring period as the energy characteristic component of the leaf. For the first Envelope extraction is performed on the target sound signal for each leaf segment to obtain the signal envelope sequence. The kurtosis coefficient of the signal envelope is then calculated as the kurtosis feature component of that leaf. .

[0050] In the recent In the monitoring cycle, the first The harmonic characteristic components of each blade form a time series. Energy characteristic components form time series kurtosis feature components form time series - For the time series of harmonic characteristic components, calculate the difference between characteristic components at adjacent time points and count the number of positive differences. Total number of differences The ratio of is used as a monotonically increasing coefficient. A monotonically increasing coefficient close to 1 indicates a monotonically increasing time series, close to 0 indicates a monotonically decreasing time series, and close to 0.5 indicates no significant monotonicity. The above calculation is repeated for the time series of energy and kurtosis feature components to obtain the monotonically increasing coefficients. and This serves as a monotonicity index for each characteristic component. The gradient of the instantaneous wind speed signal over each monitoring period is calculated, excluding gradients with absolute values ​​greater than a certain threshold. The standard deviation of the harmonic characteristic component time series is calculated for the monitoring period corresponding to the abrupt change period and for the remaining monitoring periods. The coefficient of variation is obtained by dividing the standard deviation by the mean. The reciprocal of the coefficient of variation is used as the stability coefficient. A higher stability coefficient indicates a more stable characteristic component. The stability coefficients are obtained by repeating the above calculations for the energy and kurtosis characteristic components. and , serving as a volatility indicator for each characteristic component.

[0051] For the first Each blade is used to determine the monotonicity index of harmonic characteristic components. Is it greater than the monotonicity interval value of 0.6 and the volatility index? If the harmonic characteristic component exceeds the volatility threshold of 5, it is selected into the effective feature subset. The same criteria are applied to the energy characteristic component and the kurtosis characteristic component to obtain the result. The effective feature subset of each leaf. Each feature component in the effective feature subset is normalized by subtracting the minimum value of that component from all monitoring periods and then dividing by the range of that component.

[0052] Calculate the arithmetic mean of all normalized feature components in the effective feature subset to obtain the first... Fusion characteristic value of individual blades Calculate the fusion eigenvalues ​​for each of the three blades. Combined into a fused feature vector .

[0053] Step 6: Map the fused feature vector to a pre-established feature reference space, and calculate the Mahalanobis distance with the feature reference space, including the following steps E1-E5: E1: Collect the fused feature vectors of different working conditions under normal operating conditions as training samples, calculate the mean vector and covariance matrix of the training samples, and establish a feature reference space with the mean vector as the center and the covariance matrix as the shape parameter. E2: Obtain the fused feature vector and calculate the difference vector between the fused feature vector and the mean vector; E3: Calculate the product of the difference vector and the inverse of the covariance matrix, then multiply it by the transpose of the difference vector, and take the square root of the result to obtain the Mahalanobis distance. E4: Compare the Mahalanobis distance with a preset threshold. If the Mahalanobis distance exceeds the preset threshold, the current blade is determined to be abnormal. If the Mahalanobis distance does not exceed the preset threshold, the current blade is determined to be operating normally.

[0054] In this embodiment, step 6, calculating the Mahalanobis distance to the feature reference space involves: selecting a time period during which the wind turbine is in normal operation; and collecting corresponding fused feature vectors as training samples under different operating condition combinations, including multiple speed levels from cut-in speed to rated speed and multiple wind speed levels from cut-in wind speed to rated wind speed. The operating condition coverage of the mountain wind farm needs to include typical operating states such as low wind speed and low speed, medium wind speed and medium speed, and high wind speed and high speed. The training samples reflect the normal state characteristic distribution of the three blades under complex mountain conditions. For a three-bladed wind turbine, the fused feature vector contains three components, denoted as... ,in This is the fusion feature value of the first leaf. This is the fusion feature value of the second leaf. Let be the fusion feature value of the third leaf, with the superscript T denoteing vector transpose. N training samples are collected, denoted as . , where the superscript (k) denotes the k-th training sample. Calculate the mean vector. :

[0055] in The first leaf fusion feature value is the arithmetic mean of all training samples. The arithmetic mean of the second leaf fusion characteristic values. The arithmetic mean of the third leaf fusion characteristic values. This represents the summation of k from 1 to N, where N is the total number of training samples. Calculate the covariance matrix. :

[0056] in For the first The difference between each training sample and the mean vector. The product of the two is the transpose of the difference, and the product of the two is the result of a 3x3 matrix. The sum of these matrices over all training samples is then divided by the product of the two matrices. The covariance matrix is ​​obtained. In the covariance matrix... Let be the covariance between the fused eigenvalues ​​of the i-th leaf and the j-th leaf. Let be the variance of the fused eigenvalues ​​of the i-th blade. The covariance matrix of the three blades reflects the correlation between the aerodynamic performance and surface condition of the three blades of the mountain wind turbine. The covariance is larger when the three blades are well-coordinated, and decreases when one blade deviates from the others. For the covariance matrix... Finding the inverse matrix using LU decomposition ,Will Decompose into the product of a lower triangular matrix L and an upper triangular matrix U. Given a system of linear equations, where the diagonal elements of L are all 1s, solve the system of equations. Obtain the intermediate matrix ,in Given a 3x3 identity matrix, solve the system of linear equations. Obtain the inverse matrix Obtain the fused feature vector x at the current monitoring time, and calculate the difference vector. The three components of the difference vector represent the deviations between the fused feature values ​​of the current three blades and the mean value in the normal state. Calculate the Mahalanobis distance:

[0057] in It is the transpose of the difference vector, i.e., a row vector. It is the inverse of the covariance matrix. The row vector is obtained by multiplying the row vector by the matrix, and then multiplied by the column vector d to obtain a scalar. The square root of the scalar is taken to obtain the Mahalanobis distance D. The Mahalanobis distance uses the inverse of the covariance matrix to weight the deviations of the three blades. When the deviation direction of a blade is consistent with the fluctuation direction of that blade in the training sample, the weight is smaller; when the deviation direction is orthogonal to the fluctuation direction of the training sample, the weight is larger, thus identifying abnormal states that disrupt the normal cooperative relationship between the blades. The Mahalanobis distances of all training samples are calculated, sorted from smallest to largest, and the Mahalanobis distance value corresponding to the 95th percentile is selected as the preset inter-value. The 95th percentile indicates The Mahalanobis distance of the training samples is less than this value. If If the current blade is abnormal, it is determined that the current blade is operating normally; otherwise, it is determined that the current blade is operating normally.

[0058] In an optional implementation, step 6, calculating the Mahalanobis distance with the feature reference space, can be achieved by: collecting fused feature vectors from the wind turbine under normal operating conditions as training samples, and dividing the training samples into multiple speed intervals based on the speed values ​​corresponding to the instantaneous speed signals. For each speed interval, the mean vector and covariance matrix of the training samples within that interval are calculated, establishing feature reference spaces corresponding to multiple speed intervals. When acquiring the fused feature vector at the current monitoring moment, the corresponding instantaneous speed signal is also acquired, and the speed interval to which it belongs is determined based on the value of the instantaneous speed signal. The mean vector and covariance matrix corresponding to the speed interval are selected, and the difference vector between the current fused feature vector and the mean vector of the speed interval is calculated. The inverse matrix of the covariance matrix of the speed interval is obtained, and the difference vector is written as a row vector and multiplied by the inverse matrix of the covariance matrix to obtain the intermediate row vector. Then, the difference vector is written as a column vector and multiplied by the intermediate row vector to obtain a scalar. The square root of the scalar is taken to obtain the Mahalanobis distance. Each speed range has a preset threshold set according to the statistical distribution of Mahalanobis distance of the training samples in that range. The Mahalanobis distance of the current fused feature vector is compared with the preset threshold of the speed range to determine whether there is an anomaly.

[0059] In another optional implementation, in step 6, the Mahalanobis distance with the feature reference space can also be calculated by: when collecting fused feature vectors under normal wind turbine operation as training samples, calculating the gradient of instantaneous wind speed signal changes within a time window, filtering time periods where the gradient of instantaneous wind speed signal changes is less than a gradient threshold, and collecting only the fused feature vectors corresponding to the filtered time periods as training samples. After calculating the mean vector and covariance matrix of the training samples, regularization terms are added to the diagonal elements of the covariance matrix, and the value of the regularization terms is determined according to the number of training samples and the dimension of the fused feature vectors. The inverse matrix of the covariance matrix after adding the regularization terms is obtained to acquire the fused feature vector at the current monitoring time, and the difference vector between the current fused feature vector and the mean vector is calculated. The difference vector is multiplied by the inverse matrix of the regularized covariance matrix and then multiplied by the transpose of the difference vector. The square root of the result is taken to obtain the Mahalanobis distance. Calculate the mean and standard deviation of the Mahalanobis distance of the training samples. A preset threshold is set as the mean of the Mahalanobis distance of the training samples plus a certain multiple of the standard deviation; the specific multiple is determined based on the desired anomaly detection rate. An anomaly is determined to exist when the Mahalanobis distance of the current fused feature vector exceeds the preset threshold.

[0060] Furthermore, when the Mahalanobis distance exceeds a preset threshold, based on the imbalance coefficient of the harmonic amplitude ratio and its phase position, and the growth slope of the high-frequency energy in the energy proportion distribution, the abnormal blade and damage type are determined, and the status assessment result is output, including the following steps F1-F2: F1: Extract the harmonic ratio values ​​of each harmonic from the harmonic amplitude ratio relationship of the abnormal blades, calculate the deviation of each harmonic ratio value from the normal reference value, and select the harmonic number with the largest deviation. F2: Perform a modulo operation based on the harmonic order with the largest deviation and the number of blades. The blade number corresponding to the calculation result is the abnormal blade number. F3: Extract the proportion of multiple harmonic orders corresponding to the abnormal blades, calculate the standard deviation of multiple harmonic proportions, and use it as the unbalance coefficient. F4: Obtain the sound signal of the abnormal blade in the corresponding time period in the target sound signal segment, perform wavelet packet decomposition on the sound signal of the time period to obtain the energy proportion of each decomposition frequency band, extract the energy proportion of the high frequency decomposition frequency band in the time series of multiple monitoring periods, perform linear regression calculation on the time series to obtain the slope, and use it as the energy growth slope of the high frequency band. F5: When the imbalance coefficient is greater than the first judgment threshold and the energy growth slope in the high-frequency band is less than the second judgment threshold, it is judged as aerodynamic imbalance type. F6: When the imbalance coefficient is less than the first judgment threshold and the energy growth slope in the high-frequency band is greater than the second judgment threshold, it is judged as a surface damage type. F7: When the imbalance coefficient is greater than the first judgment threshold and the energy growth slope in the high-frequency band is greater than the second judgment threshold, it is judged as a composite damage type. F8: Outputs status assessment results including abnormal blade numbers and damage types.

[0061] It should be noted that after step E4 determines that an anomaly exists, the harmonic amplitude ratio at the current moment is extracted. This harmonic amplitude ratio includes the proportions of the 2nd to 10th harmonics of the blade's passing frequency, denoted as... The mean value of each harmonic proportion recorded during the training sample phase under normal conditions is extracted as the normal baseline value, denoted as . Calculate the deviation of the proportion values ​​for each harmonic, the first... The deviation of the subharmonic is Select the harmonic order with the largest deviation from all harmonic orders. The blades of a three-bladed fan pass through a frequency three times the rotational speed. The phase position of the nth harmonic and the blade number satisfy a modulo operation relationship. The blade number... The modulo operation result is 0 for the 3rd blade, 1 for the 1st blade, and 2 for the 2nd blade. Based on the blade number 'b', all harmonic orders corresponding to that blade are determined. For a three-bladed wind turbine, the harmonic order corresponding to the 1st blade is... Wait until satisfied The harmonic order, the second blade corresponds to Wait until satisfied The harmonic order, the third blade corresponds to Wait until satisfied The harmonic orders are determined. The proportions of each harmonic order corresponding to the abnormal blades are extracted, and the standard deviation of these proportions is calculated. As the unbalance coefficient, the standard deviation is calculated using the following formula:

[0062] Where m is the number of harmonic orders corresponding to that blade. Let R be the proportion of the i-th corresponding harmonic, and R be the mean of these proportions. The imbalance coefficient reflects the consistency of the amplitude of each harmonic corresponding to the same blade. When the blade's aerodynamic performance is balanced, the standard deviation of the coordinated change of each harmonic is small. When there are deviations in the blade's shape, uneven mass distribution, or installation angle, the standard deviation of the dispersion of each harmonic amplitude is large. Retracing back to the sound signal segments corresponding to the rotation phase of each blade obtained in step 2, the sound signal of the blade corresponding to the most recent N consecutive monitoring periods is extracted based on the abnormal blade number. The value of N is not less than 10. The sound signal of each monitoring period is decomposed into 3-level wavelet packets to obtain 8 decomposed frequency bands. The center frequencies of each frequency band are as follows: , Where fs is the sampling frequency. The frequency band with a center frequency greater than 5 times the blade passage frequency is defined as the high-frequency decomposition band. The energy proportion of each high-frequency decomposition band is calculated, and the energy proportions of all high-frequency bands are summed to obtain the high-frequency energy proportion E. The high-frequency energy proportion sequence for N monitoring periods is extracted. Perform linear regression on the sequence to fit a straight line. Where t is the monitoring period number, k is the slope, and c is the intercept, k is the high-frequency energy growth slope. A positive energy growth slope indicates that the high-frequency energy increases with the monitoring period, and that turbulence noise caused by increased blade surface roughness or crack propagation leads to continuous high-frequency energy growth. A negative energy growth slope or one close to zero indicates that the high-frequency energy is stable and the surface condition has not degraded. The first judgment threshold is determined based on the 90th percentile of the imbalance coefficient of each blade in the training sample, and the second judgment threshold is determined based on the 90th percentile of the high-frequency energy growth slope of each blade in the training sample. First judgment interval value and When the second threshold is reached, it is determined to be an aerodynamic imbalance type, where the aerodynamic performance of this blade deviates from that of other blades, but the surface condition is intact. First determination threshold and When the second threshold is reached, it is determined to be surface damage type; the blade is aerodynamically balanced but its surface condition is degraded. First determination threshold and At the second threshold, the damage is classified as a combined damage type, meaning the blade exhibits both aerodynamic imbalance and surface damage. The output status assessment results include the abnormal blade number, damage type, imbalance coefficient value, and energy growth slope value.

[0063] Example 3 is an embodiment of the present invention, which provides a non-contact mountain wind turbine blade condition analysis system based on sound characteristics, including: The data acquisition module is used to synchronously acquire multi-channel sound signals of the wind turbine blades through acoustic sensors arranged in a ring array, and to synchronously record instantaneous speed signals and instantaneous wind speed signals. The signal segmentation module is used to calculate the blade passing period based on the instantaneous rotation speed signal, and to periodically segment the multi-channel sound signal using the blade passing period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade. The noise suppression module is used to select filtering parameters from a preset noise parameter library based on the change gradient of the instantaneous wind speed signal, and use the filtering parameters to suppress noise in the sound signal segment to obtain the target sound signal segment. The feature extraction module is used to extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passing frequency, the energy distribution of each frequency band of the wavelet packet, and the signal envelope kurtosis coefficient. The feature fusion module is used to establish the time series of each feature parameter in the multi-scale feature set, assign fusion weights to each feature parameter according to the monotonicity index and volatility index of the time series, and generate a fusion feature vector. An anomaly detection module is used to map the fused feature vector to a pre-established feature reference space, calculate the Mahalanobis distance with the feature reference space, and determine that an anomaly exists when the Mahalanobis distance exceeds a preset threshold. The damage diagnosis module is used to determine the abnormal blade and damage type and output the status assessment result based on the imbalance coefficient and phase position of the harmonic amplitude ratio relationship and the growth slope of the high-frequency energy in the energy proportion distribution.

[0064] This embodiment also provides an electronic device applicable to a non-contact mountain wind turbine blade status method based on sound features, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the non-contact mountain wind turbine blade status method based on sound features as proposed in the above embodiment.

[0065] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a non-contact mountain wind turbine blade status method based on sound characteristics as proposed in the above embodiment.

[0066] The storage medium proposed in this embodiment and the method for realizing a non-contact mountain wind turbine blade status based on sound characteristics proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0067] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0068] 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A non-contact method for determining the blade status of a mountain wind turbine based on sound characteristics, characterized in that: include, Acoustic sensors arranged in a ring array synchronously collect multi-channel sound signals of the wind turbine blades, and simultaneously record instantaneous rotational speed signals and instantaneous wind speed signals; The blade passage period is calculated based on the instantaneous rotation speed signal. The multi-channel sound signal is periodically segmented using the blade passage period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade. Based on the gradient of the instantaneous wind speed signal, filter parameters are selected from a preset noise parameter library, and the filter parameters are used to suppress noise in the sound signal segment to obtain the target sound signal segment. Extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passing frequency, the energy distribution of each frequency band of the wavelet packet, and the signal envelope kurtosis coefficient. Establish time series of each feature parameter in the multi-scale feature set, assign fusion weights to each feature parameter according to the monotonicity index and volatility index of the time series, and generate a fusion feature vector; The fused feature vector is mapped to a pre-established feature reference space, and the Mahalanobis distance with the feature reference space is calculated. When the Mahalanobis distance exceeds a preset threshold, the abnormal blade and damage type are determined and the status assessment result is output based on the imbalance coefficient of the harmonic amplitude ratio relationship, its phase position, and the growth slope of the high-frequency energy in the energy proportion distribution.

2. The non-contact mountain wind turbine blade status method based on sound characteristics as described in claim 1, characterized in that: The process of obtaining the sound signal segment corresponding to the rotation phase of each blade includes calculating the time interval between adjacent pulses as the initial blade passage period based on the arrival time of the continuous blade position pulses in the instantaneous rotation speed signal. A time series of the initial blade passage period is established, and the period values ​​in the time series are weighted to obtain the corrected blade passage period. Extract the amplitude feature time of the multi-channel sound signal and the arrival time of the blade position pulse, and calculate the time difference between the two as a time calibration parameter; Based on the modified blade passage period and the time calibration parameters, periodic segmentation boundaries are calibrated on the multi-channel sound signal, and the multi-channel sound signal is divided into multiple time periods according to the periodic segmentation boundaries; Extract the energy distribution curve of the sound signal within each time period, determine the time boundary of each blade according to the extreme value position in the energy distribution curve, and divide each time period into sound signal segments corresponding to each blade according to the time boundary.

3. The non-contact mountain wind turbine blade status method based on sound characteristics as described in claim 2, characterized in that: The process of obtaining the target sound signal segment includes calculating the ratio of the numerical change to the time change of the instantaneous wind speed signal within a set time window to obtain the wind speed change gradient. The noise frequency band range parameter is determined based on the magnitude of the wind speed change gradient. A reference sensor is set up to collect ambient background sound signals. The ambient background sound signals and the sound signal segments are subjected to time-frequency transformation to obtain the ambient background spectrum and the sound signal segment spectrum. Based on the noise frequency band range parameters, the spectral components of the corresponding frequency bands are extracted from the environmental background spectrum as noise spectrum references; The spectral components in the audio signal segment spectrum that are similar to the noise spectrum reference are marked as noise spectrum components. After subtracting the noise spectrum components from the audio signal segment spectrum, an inverse transformation is performed to obtain the target audio signal segment.

4. The non-contact mountain wind turbine blade status method based on sound characteristics as described in claim 3, characterized in that: The step of extracting a multi-scale feature set from the target sound signal segment includes: calculating the blade passing frequency based on the instantaneous rotation speed signal and the number of blades; performing spectral analysis on the target sound signal segment to obtain a spectral amplitude sequence; extracting the amplitude corresponding to each harmonic frequency of the blade passing frequency from the spectral amplitude sequence; calculating the ratio of each harmonic amplitude to the first harmonic amplitude; and obtaining the harmonic amplitude ratio. The target sound signal segment is decomposed into wavelet packet coefficients for each decomposed frequency band. The energy value of the wavelet packet coefficients for each decomposed frequency band is calculated. The ratio of the energy value of each decomposed frequency band to the total energy value of all decomposed frequency bands is used as the energy proportion to obtain the energy proportion distribution of each frequency band. Envelope extraction is performed on the target sound signal segment to obtain a signal envelope sequence. The ratio of the fourth central moment to the square of the second central moment of the signal envelope sequence is calculated to obtain the signal envelope kurtosis coefficient.

5. The non-contact mountain wind turbine blade status method based on sound characteristics as described in claim 4, characterized in that: The generation of the fusion feature vector includes mapping each harmonic in the harmonic amplitude ratio relationship to the corresponding blade according to the instantaneous rotation speed signal and the number of blades, and extracting the harmonic subset corresponding to each blade as the harmonic feature component of that blade. The target sound signal segment is divided into leaf-shaped time-domain segments, and the wavelet packet energy ratio distribution in each leaf time period is calculated. The high-frequency energy ratio increment is extracted as the energy feature component of each leaf. The signal envelope kurtosis coefficient is calculated for each leaf time period and used as the kurtosis feature component of each leaf. Time series are established for the harmonic characteristic components, energy characteristic components and kurtosis characteristic components of each blade. The monotonically increasing coefficient of each component time series is calculated as a monotonicity index. The stability coefficient of each component time series after excluding the instantaneous wind speed signal change period is calculated as a volatility index. For each leaf, the feature components whose monotonicity index is greater than the threshold and whose volatility index is less than the threshold are selected into the effective feature subset of the leaf. The feature components in the effective feature subset are normalized and then averaged to obtain the fused feature value of the leaf. The fusion feature values ​​of each blade are combined into a fusion feature vector.

6. The non-contact mountain wind turbine blade status method based on sound characteristics as described in claim 5, characterized in that: The calculation of the Mahalanobis distance with the feature reference space includes collecting fused feature vectors of different working conditions under normal operating conditions as training samples, calculating the mean vector and covariance matrix of the training samples, and establishing a feature reference space with the mean vector as the center and the covariance matrix as the shape parameter. Obtain the fused feature vector and calculate the difference vector between the fused feature vector and the mean vector; Calculate the product of the difference vector and the inverse of the covariance matrix, then multiply it by the transpose of the difference vector, and take the square root of the result to obtain the Mahalanobis distance. The Mahalanobis distance is compared with a preset threshold. If the Mahalanobis distance exceeds the preset threshold, the current blade is determined to be abnormal. If the Mahalanobis distance does not exceed the preset threshold, the current blade is determined to be operating normally.

7. The non-contact mountain wind turbine blade status method based on sound characteristics as described in claim 6, characterized in that: The step of determining the abnormal blade and damage type and outputting the status assessment result based on the imbalance coefficient and phase position of the harmonic amplitude ratio relationship and the growth slope of the high-frequency energy in the energy proportion distribution includes extracting the harmonic ratio values ​​of each harmonic in the harmonic amplitude ratio relationship of the abnormal blade, calculating the deviation of each harmonic ratio value from the normal reference value, and selecting the harmonic number with the largest deviation. The blade number corresponding to the calculation result is the abnormal blade number, based on the harmonic order with the largest deviation and the number of blades. Extract the proportion values ​​of multiple harmonic orders corresponding to the abnormal blades, calculate the standard deviation of multiple harmonic proportion values, and use it as the unbalance coefficient. Acquire the sound signal of the abnormal blade in the corresponding time period of the target sound signal segment, perform wavelet packet decomposition on the sound signal of the time period to obtain the energy proportion of each decomposition frequency band, extract the time series of the energy proportion of the high frequency decomposition frequency band in multiple monitoring periods, perform linear regression calculation on the time series to obtain the slope, and use it as the energy growth slope of the high frequency band. When the imbalance coefficient is greater than the first judgment threshold and the energy growth slope in the high-frequency band is less than the second judgment threshold, it is judged as an aerodynamic imbalance type. When the imbalance coefficient is less than the first judgment threshold and the energy growth slope in the high-frequency band is greater than the second judgment threshold, it is judged as a surface damage type. When the imbalance coefficient is greater than the first judgment threshold and the energy growth slope in the high-frequency band is greater than the second judgment threshold, it is judged as a composite damage type. The output includes a status assessment result containing the abnormal blade number and damage type.

8. A non-contact mountain wind turbine blade state analysis system based on sound characteristics, employing the non-contact mountain wind turbine blade state analysis method based on sound characteristics as described in any one of claims 1 to 7, characterized in that, Includes: a data acquisition module, used to synchronously acquire multi-channel sound signals of wind turbine blade operation through acoustic sensors arranged in a ring array, and synchronously record instantaneous speed signals and instantaneous wind speed signals; The signal segmentation module is used to calculate the blade passing period based on the instantaneous rotation speed signal, and to periodically segment the multi-channel sound signal using the blade passing period as a time window to obtain sound signal segments corresponding to the rotation phase of each blade. The noise suppression module is used to select filtering parameters from a preset noise parameter library based on the change gradient of the instantaneous wind speed signal, and use the filtering parameters to suppress noise in the sound signal segment to obtain the target sound signal segment. The feature extraction module is used to extract a multi-scale feature set from the target sound signal segment. The multi-scale feature set includes the harmonic amplitude ratio based on the blade passing frequency, the energy distribution of each frequency band of the wavelet packet, and the signal envelope kurtosis coefficient. The feature fusion module is used to establish the time series of each feature parameter in the multi-scale feature set, assign fusion weights to each feature parameter according to the monotonicity index and volatility index of the time series, and generate a fusion feature vector. An anomaly detection module is used to map the fused feature vector to a pre-established feature reference space, calculate the Mahalanobis distance with the feature reference space, and determine that an anomaly exists when the Mahalanobis distance exceeds a preset threshold. The damage diagnosis module is used to determine the abnormal blade and damage type and output the status assessment result based on the imbalance coefficient and phase position of the harmonic amplitude ratio relationship and the growth slope of the high-frequency energy in the energy proportion distribution.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the non-contact mountain wind turbine blade status method based on sound features as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the non-contact mountain wind turbine blade status method based on sound features as described in any one of claims 1 to 7.

Images