Fan blade state monitoring method based on multi-sensor fusion

Abstract

The invention discloses a fan blade state monitoring method based on multi-sensor fusion, relates to the technical field of wind power, and is suitable for wind energy prime mover equipment manufacturing and blade state monitoring technologies of onshore and offshore wind generating sets. The method comprises the following steps: acquiring operation data, a vibration signal, an acoustic signal and a pulse signal of a fan; the current working condition state of the fan is recognized, common-mode fault verification, local damage positioning and transient stress damage analysis are carried out on the vibration signals and the acoustic signals, and a fault analysis result and a first damage analysis result are obtained; performing phase-locked amplification analysis on the vibration signal and the acoustic signal through active excitation to obtain a second damage analysis result; and finally, a comprehensive state monitoring report of the fan blade is generated, so that the problems of difficulty in identification of weak damage and high false alarm rate of blades of land and offshore wind generating sets in wind energy prime mover equipment manufacturing under a non-stable working condition are solved, and the equipment operation and maintenance intelligent level in the wind energy prime mover equipment manufacturing industry is effectively improved.

Description

Technical Field

[0001] This invention relates to the field of wind power technology, and in particular to a method for monitoring the condition of wind turbine blades based on multi-sensor fusion. Background Technology

[0002] As a clean and renewable energy source, wind power is playing an increasingly important role in the global energy structure transformation. As an important component of the national strategic emerging industries, the wind power equipment manufacturing industry is experiencing rapid development. Among them, the installed capacity of onshore and offshore wind turbines continues to grow. As the core equipment for capturing wind energy, onshore and offshore wind turbines are usually installed in harsh environments and are subjected to complex weather conditions year-round. In particular, as a key component in the manufacturing of wind power equipment, the turbine blades are huge and complex in structure, and their health directly determines the power generation efficiency and operational safety of the unit. However, under long-term alternating loads, the blades are prone to failures such as cracks, delamination, breakage, and icing. To address the problem that existing technologies rely solely on internal current and voltage data of wind turbines to monitor blade wear or mechanical structural anomalies, several monitoring schemes based on multi-data feature fusion have emerged. Chinese Patent Application No. 202410841389.6 discloses a wind turbine operation status monitoring method based on data feature extraction. The main idea of ​​this invention is to construct a three-dimensional coordinate system containing three dimensions: noise intensity, frequency, and power. The operating data is mapped into this system and clustered to form several clusters. Finally, the distance between the current data point and the cluster center, as well as the anomaly feature index, are calculated to determine whether the wind turbine is abnormal. However, the aforementioned existing technologies still have significant limitations when diagnosing early, minor damage to wind turbine blades and under complex operating conditions. These existing technologies are essentially passive signal monitoring and statistical analysis methods, relying on signals generated by the wind turbine under natural operating conditions. For minor damage such as early closed-type cracks in the blades, the acoustic or vibration characteristics generated under normal operating loads are extremely weak and easily masked by environmental noise. Clustering analysis based solely on passively received signals can easily lead to missed detections. While existing technologies consider the relationship between power and sound, they primarily rely on time-segmented statistical processing, lacking in-depth processing of the wind turbine's non-stationary operating characteristics. Wind turbine speed fluctuates in real time with wind speed, exhibiting significant non-stationarity and variable speed characteristics. Directly segmenting and statistically analyzing data in the time domain ignores the physical correspondence between rotating machinery fault characteristics and rotation angles, easily causing aliasing or blurring of characteristic frequencies due to speed fluctuations, making it difficult to pinpoint the fault location. This is a key technical bottleneck restricting the improvement of intelligent operation and maintenance levels in wind power equipment manufacturing, especially in scenarios with high maintenance requirements such as offshore wind turbines. Summary of the Invention

[0003] The technical problem solved by this invention is that existing technologies still have significant limitations in diagnosing early, minor damage to wind turbine blades and under complex operating conditions. Existing technologies are essentially passive signal monitoring and statistical analysis methods, relying on signals generated by the wind turbine under natural operating conditions. For minor damage such as early closed-type cracks in the blades, the acoustic or vibration characteristics generated under normal operating loads are extremely weak and easily masked by environmental noise. Clustering analysis based solely on passively received signals easily leads to missed detections. Although existing technologies consider the relationship between power and sound, they mainly rely on time-segmented statistical processing, lacking in-depth processing of the wind turbine's non-stationary operating characteristics. Wind turbine speed fluctuates in real time with wind speed, exhibiting significant non-stationarity and variable speed characteristics. Directly segmenting and statistically analyzing data in the time domain ignores the physical correspondence between rotating machinery fault characteristics and rotation angles, easily causing aliasing or blurring of characteristic frequencies due to speed fluctuations, making it difficult to pinpoint the fault location.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: a wind turbine blade condition monitoring method based on multi-sensor fusion, comprising the following steps: Step S1: Collect the operating data, vibration signals, acoustic signals, and pulse signals of the fan; Step S2: Based on the impeller speed and active power in the operating data, calculate the speed fluctuation rate and power stability, and compare the calculation results with a preset threshold to identify the current operating status of the wind turbine. Step S3: Based on the operating condition, select the corresponding signal processing strategy to extract features from the vibration signal and acoustic signal. The signal processing strategy includes common mode fault verification, local damage location or transient stress damage analysis, and obtain the common mode fault analysis results characterizing the overall machine condition and the first damage analysis results characterizing the suspected features of the blade structure. Step S4: If the first damage analysis result indicates the presence of suspected features, then an active excitation mechanism is triggered to perform phase-locked amplification analysis on the vibration signal and acoustic signal to obtain a second damage analysis result characterizing nonlinear features. Step S5: Based on the common-mode fault analysis results and the second damage analysis results, a comprehensive judgment is made to generate a comprehensive condition monitoring report for the wind turbine blades.

[0005] Preferably, the operating data includes impeller speed, active power, pitch angle, impeller azimuth angle, pitch motor temperature, and wind speed; The vibration signal is a vibration acceleration sequence; The acoustic signal is a time-domain discrete sequence, which is composed of instantaneous response voltage amplitudes. The pulse signal is a key phase pulse sequence, which is composed of digital codes representing triggered and untriggered voltage signals.

[0006] Preferably, in step S2, the process of identifying the current operating status of the wind turbine specifically includes: Obtain the impeller speed and active power within the preset first sliding window from the operating data, and construct the first impeller speed sequence and the first active power sequence respectively; The standard deviation of the first impeller speed sequence is used as the speed fluctuation rate, and the coefficient of variation of the first active power sequence is used as the power stability. The speed fluctuation rate and power stability are compared with the preset speed stability threshold and power stability threshold, respectively. If the speed fluctuation rate is less than the speed stability threshold and the power stability is less than the power stability threshold, then the current wind turbine is determined to be in a quasi-steady-state operating condition. If the speed fluctuation rate is greater than or equal to the speed stability threshold, or the power stability is greater than or equal to the power stability threshold, then the current wind turbine is determined to be in a transient change condition.

[0007] Preferably, in step S3, the common-mode fault verification process specifically includes: The vibration signal is subjected to spectral analysis to extract the first natural frequency of each blade. The mean and coefficient of variation of all first natural frequencies are calculated based on the first natural frequency of each blade. Based on real-time impeller speed and real-time pitch angle, bilinear interpolation is performed in a pre-set health condition association database to obtain the current theoretical natural frequency average. The absolute value of the difference between the mean of the first-order natural frequencies of all blades and the mean of the theoretical natural frequencies is taken as the frequency residual. If the frequency residual exceeds a preset common-mode fault frequency threshold and the coefficient of variation is less than a preset common-mode fault variation threshold, then a common-mode fault is determined to exist. If the frequency residual is less than or equal to the preset common-mode fault frequency threshold, or the coefficient of variation is greater than or equal to the preset common-mode fault variation threshold, then it is determined that there is no common-mode fault. Output the fault analysis results, which indicate whether a common-mode fault exists or not.

[0008] Preferably, in step S3, the process of localizing the damage specifically includes: The vibration signal is resampled in the angular domain based on the pulse signal to obtain the first average angular domain waveform. The local energy density of the first average angular domain waveform is calculated by sliding window method. The process of calculating the local energy density specifically includes: calculating the mean of the sum of squares of all vibration accelerations within a preset second sliding window; The maximum value of all local energy densities is compared with a preset local damage threshold; If the maximum value exceeds the preset local damage threshold, the first damage analysis result is determined to be a detected static suspected feature, and the blade number and impeller azimuth angle corresponding to the maximum value are recorded.

[0009] Preferably, in step S3, the transient stress damage analysis process specifically includes: The impeller speed and active power within the preset third sliding window are obtained from the wind turbine's operating data. The statistical variance of the impeller speed is used as the speed fluctuation index, and the maximum absolute value of the difference between adjacent sampling points of active power is used as the power change rate index. When the speed fluctuation index is greater than the preset speed variance threshold or the power change rate index is greater than the preset power change rate threshold, it is determined that the current moment is in an unsteady operating condition and the current sliding window is marked as a transient analysis window. Obtain the transient vibration acceleration sequence and the corresponding transient impeller azimuth sequence within the transient analysis window. The transient vibration acceleration sequence is composed of the original vibration acceleration within the transient analysis window, and the transient impeller azimuth sequence is composed of the impeller azimuth within the transient analysis window.

[0010] Preferably, in step S3, the transient stress damage analysis process further includes: Within the transient analysis window, a fixed angular domain resampling resolution is set, and the vibration signal is resampled in the angular domain to generate a second average angular domain waveform. The second average angular domain waveform includes the vibration acceleration at equal impeller azimuth intervals for the entire rotation cycle. Based on the current impeller azimuth phase, the second average angular domain waveform is synchronized with the pre-stored steady-state tower shadow reference template by performing phase alignment and synchronous differential operation to obtain the vibration residual signal vector. The acoustic signal is acquired, and the acoustic signal is decomposed into an acoustic time-frequency complex matrix using short-time Fourier transform. The spectral kurtosis vector of each frequency slice in the acoustic time-frequency complex matrix is ​​calculated, and the maximum acoustic kurtosis value is searched. Calculate the energy signal-to-noise ratio of the vibration residual signal vector within a local sliding window of the preset transient analysis window; If the maximum acoustic kurtosis value is greater than the preset acoustic kurtosis threshold, and the energy signal-to-noise ratio of the vibration residual signal vector is greater than the preset energy mutation threshold, then the first damage analysis result is determined to be a detected dynamic suspected feature.

[0011] Preferably, step S4 includes the following sub-steps: Step S41: If a static suspected feature or a dynamic suspected feature is detected, the real-time wind speed and pitch motor temperature are obtained, and a safety assessment is performed. The safety assessment is whether the real-time wind speed is lower than the preset safety upper limit and whether the pitch motor temperature is lower than the preset overheat alarm threshold. If the real-time wind speed is lower than the preset safety upper limit and the pitch motor temperature is lower than the preset overheat alarm threshold, the safety assessment is passed. Step S42: After passing the safety assessment, a micro-disturbance command is injected into the pitch system. The micro-disturbance command is a preset sine wave signal, which includes a preset specific excitation frequency and angle.

[0012] Preferably, step S4 further includes the following sub-steps: Step S43: Using the micro-perturbation command as a reference signal, the vibration signal and acoustic signal are synchronously subjected to phase-locked amplification analysis. For the vibration signal, quadrature demodulation is performed using the specific excitation frequency of the preset sine wave signal and the second harmonic of the specific excitation frequency as reference frequencies to extract the vibration fundamental frequency response amplitude and the vibration second harmonic response amplitude. For the acoustic signal, orthogonal demodulation is performed with the specific excitation frequency as the reference frequency. The acoustic modulation response amplitude synchronized with the excitation frequency is extracted, and the ratio of the acoustic modulation response amplitude to the background acoustic noise amplitude is calculated as the acoustic synchronization signal-to-noise ratio. Step S44: The ratio of the vibration second harmonic response amplitude to the vibration fundamental frequency response amplitude is taken as the proportion of vibration nonlinear harmonics. The proportion of vibration nonlinear harmonics is compared with a preset nonlinear damage threshold, and the acoustic synchronization signal-to-noise ratio is compared with a preset acoustic response threshold. If the proportion of nonlinear harmonic vibrations is greater than the nonlinear damage threshold and the acoustic synchronization signal-to-noise ratio is greater than the acoustic response threshold, then nonlinear damage is determined to exist. If the proportion of vibration nonlinear harmonics is less than or equal to the nonlinear damage threshold, or the acoustic synchronization signal-to-noise ratio is less than or equal to the acoustic response threshold, then it is determined that there is no nonlinear damage. The second damage analysis result includes the presence of nonlinear damage and the absence of nonlinear damage.

[0013] Preferably, in step S5, the logic for generating the final comprehensive status monitoring report of the wind turbine blades specifically includes: If no common-mode fault characteristics, no static suspected characteristics, and no dynamic suspected characteristics are detected, the final comprehensive status monitoring report will be output as "wind turbine blade health". If common-mode fault characteristics are detected, the presence of common-mode faults will be marked in the final integrated condition monitoring report. Common-mode faults include blade icing or overall aging. If nonlinear damage is determined to exist, the confirmed blade crack damage is marked in the final integrated condition monitoring report, and the blade number and impeller azimuth angle recorded in step S3 are associated with the output. If a static or dynamic suspected feature is detected, and there is no nonlinear damage, then the static or dynamic suspected feature is determined to be a false signal caused by environmental interference, and the final comprehensive status monitoring report is output as the health of the wind turbine blades.

[0014] The beneficial effects of this invention are as follows: This invention proposes an active diagnostic mechanism. Unlike traditional methods that rely solely on passive monitoring, this invention, upon detecting suspected fault characteristics, injects a micro-perturbation sinusoidal wave command into the pitch system as an active excitation source. Lock-in amplification technology is then used to synchronously demodulate and analyze the vibration and acoustic signals. This active excitation mechanism can induce the nonlinear breathing effect of closed cracks at specific frequencies. By extracting the second harmonic response from the vibration signal and the modulation response of the acoustic signal, the signal-to-noise ratio of the fault characteristics is greatly enhanced. This not only detects faults but also diagnoses them using nonlinear characteristics, effectively avoiding false alarms caused by environmental noise or random interference. For real-time changes in rotational speed caused by wind speed fluctuations, this invention utilizes key-phase pulse signals to construct a continuous time-azimuth mapping relationship, transforming the time domain... Non-stationary vibration signals are resampled into synchronous signals uniformly distributed in the angular domain. This angular domain analysis method eliminates spectral aliasing and energy dispersion caused by speed fluctuations, and achieves the locking of the blade's physical position. Combined with synchronous averaging technology and local energy density analysis, the method of this invention can extract weak periodic fault features that are strictly locked to the rotation period from strong background noise. Even in complex environments with fluctuating wind speeds, it can accurately capture local damage at specific angles of the blade. The mapping relationship between vibration acceleration and impeller azimuth angle within the transient analysis window, and the differential operation after phase alignment with the historical steady-state reference template, can effectively filter out the inherent tower shadow interference during wind turbine rotation. Combined with the energy signal-to-noise ratio calculation of the local sliding window, it can sensitively capture transient crack impact signals submerged in conventional vibrations. Attached Figure Description

[0015] Figure 1 A flowchart illustrating the steps of a wind turbine blade condition monitoring method based on multi-sensor fusion, as provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of 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 some embodiments of the present invention, and not all embodiments.

[0017] Example, refer to Figure 1A method for monitoring the condition of wind turbine blades based on multi-sensor fusion is provided, including the following steps: The method for monitoring the condition of wind turbine blades based on multi-sensor fusion includes the following steps: Step S1: Collect the operating data, vibration signals, acoustic signals, and pulse signals of the fan; Step S2: Based on the impeller speed and active power in the operating data, calculate the speed fluctuation rate and power stability, and compare the calculation results with the preset threshold to identify the current operating status of the wind turbine. Step S3: Based on the operating conditions, select the corresponding signal processing strategy to extract features from the vibration and acoustic signals. The signal processing strategies include common-mode fault verification, local damage location, or transient stress damage analysis. Obtain the common-mode fault analysis results that characterize the overall machine condition and the first damage analysis results that characterize the suspected features of the blade structure. Step S4: If the first damage analysis result indicates the presence of suspected features, then the active excitation mechanism is triggered to perform phase-locked amplification analysis on the vibration signal and acoustic signal to obtain the second damage analysis result characterizing the nonlinear features. Step S5: Based on the common-mode fault analysis results and the second damage analysis results, a comprehensive judgment is made to generate a comprehensive condition monitoring report for the wind turbine blades.

[0018] This invention deeply integrates traditional passive monitoring with an innovative active incentive confirmation mechanism. By collecting multi-source heterogeneous signals in real time and first identifying the operating conditions of the wind turbine, it can adaptively switch analysis strategies between two distinct modes: quasi-steady state and transient change. This avoids the failure problem of a single algorithm under non-steady wind conditions. After detecting suspected damage, an active incentive stage is introduced, and phase-locked amplification technology is used for secondary confirmation. This effectively solves the pain point of high false alarm rate caused by environmental noise or random interference in the field of wind power equipment manufacturing. It significantly improves the accuracy and intelligence level of blade condition monitoring in the manufacturing and life-cycle operation and maintenance of onshore and offshore wind turbine generators and other generators.

[0019] Operating data includes impeller speed, active power, pitch angle, impeller azimuth angle, pitch motor temperature, and wind speed; The vibration signal is a sequence of vibration accelerations; Acoustic signals are discrete-time sequences, which are composed of instantaneous response voltage amplitudes. The pulse signal is a key phase pulse sequence, which is composed of digital codes representing the triggered and untriggered voltage signals; In a specific embodiment of the present invention, in order to meet the monitoring needs in the manufacturing scenario of wind power equipment, the vibration signal comes from the piezoelectric vibration acceleration sensor installed at the blade root and tower top of the onshore or offshore wind turbine generator set, and the vibration signal is an acceleration sequence composed of vibration acceleration. The acoustic emission sensor installed on the inner wall of the blade root is a time-domain discrete sequence consisting of instantaneous response voltage amplitudes; The elements in the key-phase pulse sequence of the pulse signal are voltage signals representing triggered (1 in this embodiment) and untriggered (0 in this embodiment), where 1 represents zero phase and represents the location of the pulse sensor; The impeller azimuth angle is obtained based on pulse signals and time interpolation algorithms. Based on the key-phase pulse sequence, the zero-phase reference time and real-time rotation period of the impeller rotation are determined. The current sampling time is obtained, and the time interval between the current sampling time and the most recent zero-phase reference time is calculated. Based on the ratio of the time interval to the real-time rotation period, the mathematical expression for the impeller azimuth angle is: ; in, This is the real-time azimuth angle. At the current sampling time, The sampling time (i.e., the 0-phase reference time) is the time when the most recent rising edge of the bond phase pulse sequence changes from 0 to 1 is detected. The real-time rotation period refers to the time difference between the trigger times of the two most recent adjacent pulse signals in the bond phase pulse sequence.

[0020] Step S2, the process of identifying the current operating status of the wind turbine, specifically includes: Obtain the impeller speed and active power within the preset first sliding window from the operating data, and construct the first impeller speed sequence and the first active power sequence respectively; The standard deviation of the first impeller speed sequence is used as the speed fluctuation rate, and the coefficient of variation of the first active power sequence is used as the power stability. The speed fluctuation rate and power stability are compared with the preset speed stability threshold and power stability threshold, respectively. If the speed fluctuation rate is less than the speed stability threshold and the power stability is less than the power stability threshold, then the current wind turbine is determined to be in a quasi-steady-state operating condition. If the speed fluctuation rate is greater than or equal to the speed stability threshold, or the power stability is greater than or equal to the power stability threshold, then the current wind turbine is determined to be in a transient change condition.

[0021] In one specific embodiment of the present invention, the impeller speed and active power within a first sliding window are obtained from the wind turbine's operating data. In this embodiment, the length of the first sliding window is 30 seconds, the acquisition frequency is 1 Hz, and the first sliding window contains 30 impeller speed and 30 active power values ​​for spectrum monitoring of the raw time-domain data, capturing structural resonance, and sorting the acquired impeller speed and active power in chronological order to obtain the first impeller speed sequence. and the first active power sequence ; Calculate the speed fluctuation rate of the first impeller speed sequence using the standard deviation formula. The mathematical expression for the rotational speed fluctuation rate is: ; in, This represents the speed fluctuation rate, used to quantify the degree of dispersion of the impeller speed within the current time window. This represents the total number of sampling points within the sliding time window. The index of the sampling point is an integer variable. The value range increases from 1 to (In this embodiment) (30) Indicates the first The instantaneous value of the impeller speed corresponding to the sampling point, that is, the first sampling point in the first impeller speed sequence. One data point, This represents the arithmetic mean of the impeller speeds within the sliding time window, calculated by summing all speed data within the window and dividing by the total number of sampling points. ; Calculate the power stability of the first active power sequence using the coefficient of variation formula. The mathematical expression for power stability is: ; in, This represents the arithmetic mean of the active power within the sliding time window. The coefficient of variation (COV) represents power stability and is a dimensionless physical quantity used to eliminate the influence of the magnitude of the power itself, quantifying the relative severity of power fluctuations. This represents the total number of sampling points within the sliding time window. This represents the index of the sampling point, with values ​​increasing from 1 to... , Indicates the first The instantaneous value of active power corresponding to the sampling point, that is, the first active power in the first active power sequence. One data point, It represents the sum of squares of power deviations, which is the sum of the squares of the differences between the instantaneous power values ​​and the power mean values ​​at all sampling points; The standard deviation of the first active power series reflects the absolute fluctuation range of the power data. In this embodiment, the speed stability threshold is set to 0.2 rpm, meaning that if the speed standard deviation is less than 0.2 rpm within 30 seconds, the speed is considered to be in a stable state. The power stability threshold is set to 0.05 (i.e. 5%), meaning that if the power fluctuation amplitude does not exceed 5% relative to the mean, the load is considered to be stable.

[0022] By constructing a sliding window and calculating the standard deviation of impeller speed and the coefficient of variation of active power respectively, the system can quantify the dispersion of speed and the severity of power fluctuations. By using a dual judgment based on speed stability threshold and power stability threshold, the system can accurately classify the wind turbine operating state into quasi-steady-state and transient abrupt change conditions. This classification processing mechanism ensures the targeting of subsequent algorithms: focusing on cumulative damage analysis in steady state and impact response capture in transient state, avoiding feature overload or misjudgment caused by blindly using the same algorithm to process non-stationary data.

[0023] In step S3, the common-mode fault verification process specifically includes: Spectral analysis was performed on the vibration signal to extract the first natural frequency of each blade. The mean and coefficient of variation of all first natural frequencies were calculated based on the first natural frequency of each blade. Based on real-time impeller speed and real-time pitch angle, bilinear interpolation is performed in a pre-set health condition association database to obtain the current theoretical natural frequency average. The absolute value of the difference between the mean of the first-order natural frequencies of all blades and the mean of the theoretical natural frequencies is taken as the frequency residual. If the frequency residual exceeds the preset common-mode fault frequency threshold and the coefficient of variation is less than the preset common-mode fault variation threshold, then a common-mode fault is determined to exist. If the frequency residual is less than or equal to the preset common-mode fault frequency threshold, or the coefficient of variation is greater than or equal to the preset common-mode fault variation threshold, then it is determined that there is no common-mode fault. Output the fault analysis results, which indicate whether a common-mode fault exists or not.

[0024] In one specific embodiment of the present invention, the health condition association database is used to store the baseline characteristics of the wind turbine in a healthy state, and the specific process of constructing the health condition association database includes: The first month after the wind turbine is put into operation is taken as the health baseline period. The impeller speed and pitch angle collected during the health baseline period are taken as the historical impeller speed and historical pitch angle. The historical vibration acceleration and the historical first-order natural frequency of the blade are obtained by vibration sensors installed on the tower and obtained by fast Fourier transform. The historical impeller speed and historical pitch angle are discretized using grid binning technology. The specific discretization process includes: Obtain the range of historical impeller speeds during the health baseline period, divide the speed range into several intervals with a step size of 0.5 rpm. If the maximum value of the historical impeller speed range is not an integer multiple of 0.5 rpm, expand it to the next value that is divisible by 0.5. Use each of the divided historical impeller speed range intervals as the row index of the health condition association database. Obtain the range of historical pitch angles during the health baseline period, divide the historical pitch angles into several intervals with a step size of 0.5°. If the maximum value of the range of historical pitch angles is not an integer multiple of 0.5°, expand it to the next value that is divisible by 0.5. Use each of the divided historical pitch angle intervals as a column index of the health condition association database. The health condition association database is represented as a two-dimensional matrix, and the elements in the two-dimensional matrix are... What is stored is the number that falls into the first The first speed range and the first The arithmetic mean of all historical first-order natural frequency data within each propeller angle interval; During real-time monitoring, the mean theoretical natural frequency representing the real-time operating condition of the wind turbine often lies between the grids in the health condition correlation database. The mean theoretical natural frequency under real-time operating conditions is obtained using bilinear interpolation. The bilinear interpolation process specifically includes: If the real-time rotational speed of the operating point collected at the current moment is The pitch angle is In the health condition association database ( In a two-dimensional matrix, the four nearest neighbor grid points (two-dimensional matrix elements) surrounding the working point are retrieved as reference points. These four points are: ,in, Corresponding rotational speed coordinates Corresponding strain gauge angle coordinates; The current mathematical expression for the theoretical natural frequency mean is: ; in, The desired theoretical natural frequency is the frequency value that a healthy fan should exhibit under the current specific non-grid point operating conditions. Let be the distance weight vector in the direction of rotational speed. This represents the distance between the current rotational speed and the upper limit rotational speed. The smaller the distance (i.e., the closer to the upper limit), the lower limit point (...). The smaller the weight of the upper bound, the larger the weight of the upper bound (inverse relationship). The vector is used for linear weighting in the row direction (rotation dimension). , , and These correspond to the historical first-order natural frequencies of the four grid points, respectively. This is the distance weight vector in the direction of the pitch angle, used for linear weighting in the column direction; The normalized area factor is used to normalize the area of ​​the rectangular region enclosed by the four grid points. The normalized area factor is used to standardize the subsequent weighting and to ensure that the interpolation results are dimensionally correct. Calculate the arithmetic mean of the first natural frequencies of the blade and the arithmetic mean of the theoretical natural frequencies. Take the absolute value of the difference between the arithmetic mean of the first natural frequencies of the blade and the arithmetic mean of the theoretical natural frequencies as the frequency residual. Take the ratio of the standard deviation of the first natural frequencies of the blade to the arithmetic mean of the first natural frequencies of the blade as the coefficient of variation. In this embodiment, the common-mode fault frequency threshold is set to The frequency resolution of conventional accelerometers is usually 0.01Hz. Considering the frequency drift caused by changes in ambient temperature, setting 0.03Hz as the threshold can effectively filter out false alarms caused by measurement noise and environmental thermal effects, ensuring the robustness of fault identification; the common-mode fault variation threshold is set to 0.01.

[0025] By performing grid binning and bilinear interpolation on historical health data, the theoretical mean natural frequency under the current operating conditions can be dynamically calculated based on the current real-time speed and pitch angle. By calculating the residual between the measured mean frequency and the theoretical value, and combining it with coefficient of variation analysis, it is possible to effectively distinguish whether the overall frequency shift is caused by environmental factors (such as temperature drift) or the local frequency anomaly is caused by damage to a single blade. Thus, without relying on external meteorological data, it is possible to accurately identify blade icing or common mode faults caused by overall aging.

[0026] Step S3, the process of local damage localization specifically includes: The vibration signal is resampled in the angular domain based on the pulse signal to obtain the first average angular domain waveform. The local energy density of the first average angular domain waveform is then calculated using the sliding window method. The process of calculating local energy density specifically includes: calculating the mean of the sum of squares of all vibration accelerations within a preset second sliding window; The maximum value of all local energy densities is compared with a preset local damage threshold; If the maximum value exceeds the preset local damage threshold, the first damage analysis result is determined to be a detected static suspected feature, and the blade number and impeller azimuth angle corresponding to the maximum value are recorded.

[0027] In one specific embodiment of the present invention, the peak value of the pulse signal is calculated, the trigger level is set to 50% of the signal amplitude (i.e., the middle position of the waveform amplitude), and the pulse signal sequence is scanned point by point to find the voltage value that is lower than... Towards higher For adjacent sampling points of the jump, the pulse signal crossing is calculated using a linear interpolation algorithm. The subsampling time point is marked as the th subsampling time point. The start time of each rotation cycle The next subsampling time point is the [number]th [time point]. The end time of each rotation cycle , It is also the first The start time of each rotation cycle; No. The time length of each rotation cycle is Based on the start time and termination time A mapping function between continuous time and impeller azimuth angle is constructed using cubic spline interpolation. Through mapping function It is possible to calculate the time of each vibration sampling point. The corresponding precise impeller angle , to each rotation Divided into A fixed constant impeller azimuth increment (in this embodiment, it is every...) (a point), for each impeller azimuth point, based on The inverse function relationship is used to calculate this. The time coordinates corresponding to each impeller azimuth angle point are used to interpolate and resample the original non-equally spaced vibration signal to obtain the corresponding vibration acceleration. The vibration acceleration values ​​constitute the angular domain sequence of the rotation period; Using the same time-impeller azimuth mapping relationship, an angular domain sequence of C consecutive rotation cycles is obtained, where C is the number of complete rotation cycles contained within the first window, and this sequence is constructed as follows: angular domain matrix angular matrix The mathematical expression is: ; Each row Representing the vibration acceleration within a complete rotation cycle, for each column Representing the same physical impeller azimuth position (the first) column corresponding impeller azimuth angle ), Representative at the Circle, go to the first Vibration acceleration amplitude at each impeller azimuth angle division, diagonal domain matrix By averaging along the columns, a first average angular domain waveform vector with a length of 3600 is obtained. , No. Vibration acceleration amplitude at each impeller azimuth angle point The mathematical expression is: ; in, For the first The vibration acceleration amplitude at each impeller azimuth angle point , Representative at the Circle, go to the first Vibration acceleration amplitude at each impeller azimuth angle division; In order to obtain the waveform vector from the first average angular domain To identify minute crack impacts, it is necessary to calculate the local energy density. The length of the second sliding window is set as follows: Each data collection point, in this embodiment (correspond The impeller azimuth angle range), the second sliding window is used to perform sliding average processing on the pulse interval to suppress speed calculation noise. When the structural crack is excited by gravity or centrifugal force, the transient response of the structural crack is a damped oscillation, the duration of which corresponds to 3°-5° of the rotation angle. If the window is too narrow, it will cut off the signal; if it is too wide, it will dilute the energy density. Along the first average angular domain waveform vector Slide point by point, for the first The local energy density at each impeller azimuth position The calculation is as follows: ; For local energy density, used to quantize the first The degree of vibration energy concentration near the azimuth angle of each impeller The window length (50 points) is used as the denominator for normalization. Represents the first average angular domain waveform vector. The vibration amplitude at each location, considering the circumferential closure of the blade structure, needs to be processed using a cyclic index, i.e., under the index... Greater than or equal to the vector length (In this embodiment, 3600) When taking As the actual index value, i.e., reading the data from the beginning segment of the vector, ensures that the sliding window can span... The boundary is continuously calculated, and mod is the remainder in mathematical operations; In this embodiment, the local damage threshold is: Based on the wind turbine geometry, blade intervals are defined. In this embodiment, the wind turbine has 3 blades, therefore blade interval 1 is defined as... blade 2 interval , blade 3-section ; All impeller azimuth angles Arranged in ascending order of impeller azimuth angle, forming a local energy density sequence. The local energy density sequence visually reflects the energy distribution at various locations during one rotation of the blade. Traversing the entire local energy density sequence... Find The maximum value in is denoted as And record the impeller azimuth index corresponding to the maximum value. In the impeller azimuth index At that location, calculations yielded ,like If so, it is determined that there is local structural damage to the blade, and the data is obtained. The corresponding impeller azimuth angle is used to determine the blade number based on the blade interval entered by the impeller azimuth angle, and the blade number and impeller azimuth angle are recorded. The mathematical expression for the corresponding physical impeller azimuth angle is: ; in, This is the impeller azimuth index corresponding to the maximum value in the local energy density sequence.

[0028] By introducing angular domain resampling and local energy density analysis techniques, and using the fitted continuous time and azimuth mapping function, the original non-equal interval time domain vibration signal is resampled into an angular domain sequence with equal angular intervals. This eliminates the spectral ambiguity and energy leakage caused by speed fluctuations. By using synchronous averaging and local sliding windows to calculate energy density, it is possible to keenly capture the damped oscillations caused by structural cracks from strong background noise. This method not only significantly improves the signal-to-noise ratio, but also pinpoints the physical angle at which the fault occurs.

[0029] Step S3, the transient stress damage analysis process specifically includes: The impeller speed and active power within the preset third sliding window are obtained from the wind turbine's operating data. The statistical variance of the impeller speed is used as the speed fluctuation index, and the maximum absolute value of the difference between adjacent sampling points of active power is used as the power change rate index. When the speed fluctuation index is greater than the preset speed variance threshold or the power change rate index is greater than the preset power change rate threshold, it is determined that the current moment is in an unsteady operating condition and the current sliding window is marked as a transient analysis window. Obtain the transient vibration acceleration sequence and the corresponding transient impeller azimuth sequence within the transient analysis window. The transient vibration acceleration sequence is composed of the original vibration acceleration within the transient analysis window, and the transient impeller azimuth sequence is composed of the impeller azimuth within the transient analysis window.

[0030] In a specific embodiment of the present invention, the time length of the third sliding window is set to 10s and the sliding step size is 1s. The third sliding window is used to prevent high-frequency aliasing during the mapping interpolation process. Within the third sliding window, the second impeller speed sequence is simultaneously acquired over time. Second active power sequence ,in The number of sampling points within the third sliding window; in this embodiment, the sampling frequency for rotational speed and power is set to 50Hz, i.e. ; For the second impeller speed sequence ,calculate The variance value is used as an indicator of speed fluctuation. Variance can sensitively reflect the degree of dispersion of the rotational speed from the mean, and its calculation formula is as follows: ; in, This represents the arithmetic mean of the impeller speeds within the third sliding window. The impeller speed; For the second active power sequence Calculate the absolute value of the difference between adjacent sampling points, and take the maximum value as the power change rate index. The calculation formula is: ; The preset speed variance threshold is 0.05 r / min. 2 The power change rate threshold is 50kW.

[0031] By monitoring the speed variance and power change rate within a shorter time window, the system can quickly pinpoint the moment of drastic speed fluctuations or power abrupt changes. This prevents the averaging and dilution of short-term transient impact energy by long-term steady-state analysis, ensuring that even in harsh operating environments and when blades are subjected to complex stresses on onshore and offshore wind turbines, the monitoring system can still retain the most diagnostically valuable dynamic stress damage characteristics through high-frequency sampling and transient sequence extraction.

[0032] Step S3, the transient stress damage analysis process specifically includes: Within the transient analysis window, a fixed angular domain resampling resolution is set, and the vibration signal is resampled in the angular domain to generate a second average angular domain waveform. The second average angular domain waveform includes the vibration acceleration at equal impeller azimuth intervals for the entire rotation cycle. Based on the current impeller azimuth phase, the second average angular domain waveform is phase-aligned with the pre-stored steady-state tower shadow reference template and then synchronous differential operation is performed to obtain the vibration residual signal vector. Acquire acoustic signals, decompose them into acoustic time-frequency complex matrices using short-time Fourier transform, calculate the spectral kurtosis vector of each frequency slice in the acoustic time-frequency complex matrix, and search for the maximum acoustic kurtosis value. Calculate the energy signal-to-noise ratio of the vibration residual signal vector within a local sliding window of the preset transient analysis window; If the maximum acoustic kurtosis value is greater than the preset acoustic kurtosis threshold, and the energy signal-to-noise ratio of the vibration residual signal vector is greater than the preset energy mutation threshold, then the first damage analysis result is determined to be a detected dynamic suspected feature.

[0033] In a specific embodiment of the present invention, the impeller azimuth angle within the transient analysis window is obtained. The transient impeller azimuth angle sequence and the transient vibration acceleration sequence are formed by combining the transient impeller azimuth angle sequence and the transient vibration acceleration sequence. The transient impeller azimuth angle sequence and the transient vibration acceleration sequence are aligned in time, that is, each impeller azimuth angle corresponds to a vibration acceleration. In this embodiment, the preset angular resolution is... When the number of vibration accelerations generated by interpolation reaches 3600 according to the angular resolution, the interpolation calculation is stopped. All the vibration accelerations constitute the second average angular domain waveform, that is, the second average angular domain waveform contains a complete impeller rotation cycle. The first month after the wind turbine is put into operation is taken as the health baseline period. The impeller azimuth angle collected during the health baseline period is the historical impeller azimuth angle, and the corresponding historical vibration acceleration and the historical first-order natural frequency of the blade obtained by fast Fourier transform are also obtained. Based on the historical impeller azimuth angle and historical vibration acceleration, the historical second average angular domain waveform is obtained using the same method. Multiple historical second average angular domain waveforms (500 in this embodiment) from different time periods within the health reference period are selected, aligned according to the impeller azimuth angle, and the average value is calculated to obtain the steady-state tower shadow reference template. The second average angular domain waveform of the transient analysis window and the steady-state tower shadow reference template are synchronously differentially calculated. Since the second average angular domain waveform of the transient analysis window and the steady-state tower shadow reference template are aligned in the impeller azimuth angle, the regular tower interference can be removed by direct subtraction, and the vibration residual signal vector is obtained. The acoustic signal is decomposed into an acoustic time-frequency complex matrix using the short-time Fourier transform. The short-time Fourier transform window length is set to 1024 points, the overlap rate is 50%, and the window function is the Hanning window. For each frequency slice of the time-frequency matrix, the spectral kurtosis of the frequency slice is calculated. Spectral kurtosis is a statistical index that measures the non-Gaussianity of the signal. The maximum acoustic kurtosis value is searched among all frequency slices. The energy signal-to-noise ratio is calculated as follows: Set length as A local sliding window (50 in this embodiment) with a step size (For example (point) at a length of ( The signal is slid point by point within the transient analysis window. At each sliding position, the root mean square value of the vibration residual signal within the local sliding window is calculated as the local signal energy, and the root mean square value of the vibration residual signal within the entire transient analysis window is calculated as the background noise reference energy. The ratio of the local signal energy to the background noise reference energy is calculated, which is the energy signal-to-noise ratio. Since the vibration impact caused by dynamic cracks in blades is usually short in duration, if the average energy of the entire transient analysis window is directly calculated, the weak high-frequency energy generated by the fault will be diluted by the background noise over a long period of time, resulting in a signal-to-noise ratio that is too low to be identified. By setting a short local sliding window and scanning point by point, when the sliding window moves to the time period in which the fault impact occurs, the local energy density in the window will be significantly higher than the background noise reference energy, thereby greatly amplifying the signal-to-noise ratio of the fault characteristics and ensuring that the weak damage signal submerged in the noise can be captured sensitively. Traverse each local sliding window and take the maximum value of the energy signal-to-noise ratio as the vibration residual energy signal-to-noise ratio of the transient analysis window; Set the acoustic kurtosis threshold to 4.5 and the preset energy mutation threshold to 6dB.

[0034] By performing phase alignment and differential calculations between the transient angular domain waveform and the historical steady-state tower shadow reference template, the inherent periodic tower interference during wind turbine rotation is effectively filtered out, allowing the masked vibration residual signal to be revealed. Combining the spectral kurtosis analysis of the acoustic signal with the energy signal-to-noise ratio calculation of the vibration residual, the method of this invention achieves cross-validation of acoustic and vibration characteristics. Anomalies are only judged when both acoustic non-Gaussianity (kurtosis) and vibration energy mutation exceed the standard. This multi-dimensional fusion judgment logic greatly reduces the false alarm rate of a single signal source under transient conditions.

[0035] Step S4 includes the following sub-steps: Step S41: If a static suspected feature or a dynamic suspected feature is detected, obtain the real-time wind speed and pitch motor temperature, and perform a safety assessment. The safety assessment is whether the real-time wind speed is lower than the preset safety upper limit and whether the pitch motor temperature is lower than the preset overheat alarm threshold. If the real-time wind speed is lower than the preset safety upper limit and the pitch motor temperature is lower than the preset overheat alarm threshold, then the safety assessment is passed. Step S42: After passing the safety assessment, a micro-disturbance command is injected into the pitch system. The micro-disturbance command is a preset sine wave signal, which includes a preset specific excitation frequency and angle.

[0036] In one specific embodiment of the present invention, the preset safety upper limit of wind speed is 12m / s, which is set based on the rated wind speed of the wind turbine, and the overheat alarm threshold of the pitch motor temperature is 65°C. After a safety assessment, superimposed micro-disturbance commands are sent to the control loop of the pitch system, and vibration and acoustic signals under the excitation response are collected simultaneously. The vibration and acoustic signals under the excitation response are then subjected to phase-locked amplification analysis. Angle commands sent to the control loop of the pitch system Defined as: ; in, This refers to the micro-perturbation command in this embodiment. The micro-perturbation command is a discretized sine wave sequence, and its mathematical expression is: ; This is the original pitch control command (reference value) calculated by the wind turbine's main controller based on the current wind speed. During active excitation, the reference value remains unchanged to ensure that the wind turbine maintains normal power generation. Index representing the time step of a discrete control system ( The value can be 1, 2, 3, ..., n). Discrete control system refers to the operating environment of the wind turbine main control program. The wind turbine main control program is a digital system based on time steps, which controls the wind turbine's actions by sending instructions frame by frame. This represents the sampling period of the control system, which is 0.01 seconds (i.e., 10ms) in this embodiment. This indicates that the initial phase is set to 0 to ensure that the excitation signal starts smoothly from zero and avoids step shock. The vibration frequency representing the active excitation signal is set to 5Hz; The preset angle of the sinusoidal signal is 0.3°, and the preset specific excitation frequency of the sinusoidal signal is 5Hz. The impeller rotation frequency of the wind turbine is usually between 0.2-0.3Hz, and the first natural frequency of the tower is usually between 0.3-0.5Hz. Selecting 5Hz can avoid these low-frequency interference areas and prevent the whole machine from resonating. To facilitate computer processing and subsequent phase-locked loop analysis, an excitation control matrix is ​​constructed during step S52. ; ; Each row of the excitation control matrix represents a discrete time control point. If the excitation duration is 5 seconds and the sampling frequency is 100Hz, then the excitation control matrix has 500 rows. The first column of the excitation control matrix represents a relative timestamp, used to ensure strict timing synchronization of instructions; The second column of the excitation control matrix represents the reference pitch command of the master control system at the corresponding moment; The third column of the excitation control matrix represents the sinusoidal micro-perturbation value calculated at the corresponding time moment; The specific values ​​in the element matrix of the excitation control matrix are the underlying instruction codes sent to the pitch actuator. During execution, the excitation control matrix is ​​read row by row, the elements of the second and third columns are added together, and the result is output to the pitch control loop in real time. The main purpose of constructing the excitation control matrix is ​​to ensure that the excitation signal applied to the pitch system is strictly aligned with the reference signal used in subsequent signal processing steps on the time axis. This is a prerequisite for the lock-in amplification algorithm to effectively extract weak signals. The excitation control matrix stores the reference pitch command and the micro-disturbance command separately, which can independently superimpose and record high-frequency test signals (micro-disturbance commands) without interfering with the normal aerodynamic control of the wind turbine (reference command). By pre-generating a discretized instruction sequence, the computational delay caused by real-time calculation of the sine function is avoided, ensuring the real-time performance and smoothness of high-frequency jitter control.

[0037] The wind speed and temperature safety assessment conducted before the excitation was performed ensured that the testing process would not impose an additional burden on the unit. By constructing an excitation control matrix that includes a reference command and a micro-disturbance sine wave, extremely small and precisely synchronized jitter signals can be injected into the pitch system without interfering with the normal power generation of onshore and offshore wind turbines and other generators. The excitation control matrix ensures that the excitation signal and the reference signal for subsequent analysis are strictly aligned on the time axis, providing the necessary time reference for extracting deep nonlinear features using the lock-in amplification algorithm. This solves the technical problem of synchronizing excitation and response in wind power equipment manufacturing technology.

[0038] Step S4 also includes the following sub-steps: Step S43: Using the micro-perturbation command as a reference signal, the vibration signal and acoustic signal are synchronously subjected to phase-locked amplification analysis. For the vibration signal, quadrature demodulation is performed using a specific excitation frequency of a preset sine wave signal and a second harmonic of the specific excitation frequency as reference frequencies to extract the fundamental frequency response amplitude and the second harmonic response amplitude of the vibration. For acoustic signals, quadrature demodulation is performed with a specific excitation frequency as the reference frequency. The acoustic modulation response amplitude synchronized with the excitation frequency is extracted, and the ratio of the acoustic modulation response amplitude to the background acoustic noise amplitude is calculated as the acoustic synchronization signal-to-noise ratio. Step S44: The ratio of the vibration second harmonic response amplitude to the vibration fundamental frequency response amplitude is taken as the proportion of vibration nonlinear harmonics. The proportion of vibration nonlinear harmonics is compared with a preset nonlinear damage threshold, and the acoustic synchronization signal-to-noise ratio is compared with a preset acoustic response threshold. If the proportion of nonlinear harmonic vibrations is greater than the nonlinear damage threshold and the acoustic synchronization signal-to-noise ratio is greater than the acoustic response threshold, then nonlinear damage is determined to exist. If the proportion of vibration nonlinear harmonics is less than or equal to the nonlinear damage threshold, or the acoustic synchronization signal-to-noise ratio is less than or equal to the acoustic response threshold, then it is determined that there is no nonlinear damage. The second damage analysis result includes the presence of nonlinear damage and the absence of nonlinear damage.

[0039] In a specific embodiment of the present invention, when performing active excitation, a lock-in amplification algorithm is used to extract features from the vibration signal and the acoustic signal; In the context of continuous proactive incentives Within a given time period, two sets of discrete time-series data were simultaneously acquired. The discrete time-series data included vibration signal sequences. Harmony and acoustic signal sequence The micro-perturbation command is upsampled, and the upsampling specifically includes: Obtain the sampling frequency of the vibration signal Total number of sampling points Construct a length of discrete-time vector , of which element , time vector Substituting each time point into the mathematical expression of the micro-perturbation command The calculation is performed to obtain the length as... The first standard sine reference sequence; Obtain the sampling frequency and total number of sampling points of the acoustic signal, and use the same upsampling operation to obtain a second standard sinusoidal reference sequence with a length equal to the total number of sampling points of the acoustic signal; The lock-in amplification algorithm uses quadrature demodulation to filter out only the signal that is completely consistent with the specific excitation frequency of the micro-perturbation command. The input to the lock-in amplification algorithm is a vibration signal sequence, an acoustic signal sequence, a first standard sinusoidal reference sequence, and a second standard sinusoidal reference sequence. After upsampling, the vibration signal sequence and the standard sinusoidal reference sequence are consistent in dimension, and the acoustic signal sequence and the second standard sinusoidal reference sequence are consistent in dimension, allowing the lock-in amplification algorithm to be executed. Based on the first standard sinusoidal reference sequence, the vibration signal sequence is processed by orthogonal demodulation (multiplication) and low-pass filtering (integration and summation) to output the fundamental frequency response amplitude, the second harmonic response amplitude, and the acoustic modulation response amplitude. Based on the second standard sinusoidal reference sequence, the acoustic signal sequence is quadrature demodulated, and the output is the acoustic modulation response amplitude. In this embodiment, the nonlinear damage threshold is 0.05, and the acoustic response threshold is 2.0; Only when macroscopic motion distortion (vibration) and microscopic frictional release (acoustics) occur simultaneously under the same frequency excitation do they conform to the complete physical characteristics of a closed crack.

[0040] Lock-in amplification technology was used to deeply extract the nonlinear characteristics of closed cracks. By upsampling the micro-perturbation command and using it as a reference signal, the vibration and acoustic signals could be orthogonally demodulated to extract the fundamental frequency response, second harmonic response, and acoustic modulation component that are strictly synchronized with the excitation frequency. Using the proportion of vibration nonlinear harmonics and the acoustic synchronization signal-to-noise ratio as dual criteria, the method of this invention can capture the breathing effect (i.e., nonlinear stiffness change) generated by closed cracks under forced vibration. This nonlinear diagnostic method based on physical mechanism can effectively eliminate the interference of linear environmental noise.

[0041] In step S5, the logic for generating the final comprehensive condition monitoring report for the wind turbine blades specifically includes: If no common-mode fault characteristics, no static suspected characteristics, and no dynamic suspected characteristics are detected, the final comprehensive status monitoring report will be output as "wind turbine blade health". If common mode fault characteristics are detected, the presence of common mode faults will be marked in the final integrated condition monitoring report. Common mode faults include blade icing or overall aging. If nonlinear damage is determined to exist, the confirmed blade crack damage is marked in the final integrated condition monitoring report, and the blade number and impeller azimuth angle recorded in step S3 are associated with the output. If a static or dynamic suspected feature is detected, and there is no nonlinear damage, then the static or dynamic suspected feature is determined to be a false signal caused by environmental interference, and the final comprehensive status monitoring report is output as the health of the wind turbine blades.

[0042] This invention uses a hierarchical judgment logic to clearly distinguish between common-mode faults, confirmed crack damage, and false signals caused by environmental interference. In particular, it can automatically identify and exclude cases with suspected characteristics that have not passed active excitation verification, thus avoiding unnecessary shutdown checks by maintenance personnel due to invalid alarms.

[0043] The beneficial effects of this invention are as follows: This invention proposes an active diagnostic mechanism for the hidden faults of closed cracks. Unlike traditional methods that rely solely on passive monitoring, this invention, after detecting suspected fault characteristics, injects a micro-perturbation sinusoidal wave command into the pitch system as an active excitation source and utilizes lock-in amplification technology to synchronously demodulate and analyze vibration and acoustic signals. This active excitation mechanism can induce the nonlinear breathing effect of closed cracks at specific frequencies. By extracting the second harmonic response from the vibration signal and the modulation response of the acoustic signal, the signal-to-noise ratio of the fault characteristics is greatly enhanced. This not only detects faults but also diagnoses them using nonlinear characteristics, effectively avoiding false alarms caused by environmental noise or random interference. For real-time changes in rotational speed caused by wind speed fluctuations, this invention utilizes key-phase pulse signals to construct continuous time and azimuth mappings. By resampling the non-stationary vibration signal in the time domain into a synchronous signal uniformly distributed in the angular domain, this angular domain analysis method eliminates spectral aliasing and energy dispersion caused by speed fluctuations, and achieves the locking of the physical position of the blade. Combined with synchronous averaging technology and local energy density analysis, the method of this invention can extract weak periodic fault features that are strictly locked to the rotation period from strong background noise. Even in complex environments with fluctuating wind speeds, it can accurately capture local damage at specific angles of the blade. The mapping relationship between vibration acceleration and impeller azimuth angle in the transient analysis window, and the differential operation after phase alignment with the historical steady-state reference template, can effectively filter out the inherent tower shadow interference during wind turbine rotation. Combined with the energy signal-to-noise ratio calculation of the local sliding window, it can sensitively capture transient crack impact signals submerged in conventional vibrations.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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 protection of the present invention.

Claims

1. A method for monitoring the condition of wind turbine blades based on multi-sensor fusion, characterized in that, Includes the following steps: Step S1: Collect the operating data, vibration signals, acoustic signals, and pulse signals of the fan; Step S2: Based on the impeller speed and active power in the operating data, calculate the speed fluctuation rate and power stability, and compare the calculation results with a preset threshold to identify the current operating status of the wind turbine. Step S3: Based on the operating condition, select the corresponding signal processing strategy to extract features from the vibration signal and acoustic signal. The signal processing strategy includes common mode fault verification, local damage location or transient stress damage analysis, and obtain the common mode fault analysis results characterizing the overall machine condition and the first damage analysis results characterizing the suspected features of the blade structure. Step S4: If the first damage analysis result indicates the presence of suspected features, then an active excitation mechanism is triggered to perform phase-locked amplification analysis on the vibration signal and acoustic signal to obtain a second damage analysis result characterizing nonlinear features. Step S5: Based on the common-mode fault analysis results and the second damage analysis results, a comprehensive judgment is made to generate a comprehensive condition monitoring report for the wind turbine blades.

2. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 1, characterized in that, The operating data includes impeller speed, active power, pitch angle, impeller azimuth angle, pitch motor temperature, and wind speed. The vibration signal is a vibration acceleration sequence; The acoustic signal is a time-domain discrete sequence, which is composed of instantaneous response voltage amplitudes. The pulse signal is a key phase pulse sequence, which is composed of digital codes representing triggered and untriggered voltage signals.

3. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 2, characterized in that, In step S2, the process of identifying the current operating status of the wind turbine specifically includes: Obtain the impeller speed and active power within the preset first sliding window from the operating data, and construct the first impeller speed sequence and the first active power sequence respectively; The standard deviation of the first impeller speed sequence is used as the speed fluctuation rate, and the coefficient of variation of the first active power sequence is used as the power stability. The speed fluctuation rate and power stability are compared with the preset speed stability threshold and power stability threshold, respectively. If the speed fluctuation rate is less than the speed stability threshold and the power stability is less than the power stability threshold, then the current wind turbine is determined to be in a quasi-steady-state operating condition. If the speed fluctuation rate is greater than or equal to the speed stability threshold, or the power stability is greater than or equal to the power stability threshold, then the current wind turbine is determined to be in a transient change condition.

4. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 3, characterized in that, In step S3, the common-mode fault verification process specifically includes: The vibration signal is subjected to spectral analysis to extract the first natural frequency of each blade. The mean and coefficient of variation of all first natural frequencies are calculated based on the first natural frequency of each blade. Based on real-time impeller speed and real-time pitch angle, bilinear interpolation is performed in a pre-set health condition association database to obtain the current theoretical natural frequency average. The absolute value of the difference between the mean of the first-order natural frequencies of all blades and the mean of the theoretical natural frequencies is taken as the frequency residual. If the frequency residual exceeds a preset common-mode fault frequency threshold and the coefficient of variation is less than a preset common-mode fault variation threshold, then a common-mode fault is determined to exist. If the frequency residual is less than or equal to the preset common-mode fault frequency threshold, or the coefficient of variation is greater than or equal to the preset common-mode fault variation threshold, then it is determined that there is no common-mode fault. Output the fault analysis results, which indicate whether a common-mode fault exists or not.

5. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 4, characterized in that, In step S3, the process of local damage localization specifically includes: The vibration signal is resampled in the angular domain based on the pulse signal to obtain the first average angular domain waveform. The local energy density of the first average angular domain waveform is calculated by sliding window method. The process of calculating the local energy density specifically includes: calculating the mean of the sum of squares of all vibration accelerations within a preset second sliding window; The maximum value of all local energy densities is compared with a preset local damage threshold; If the maximum value exceeds the preset local damage threshold, the first damage analysis result is determined to be a detected static suspected feature, and the blade number and impeller azimuth angle corresponding to the maximum value are recorded.

6. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 5, characterized in that, In step S3, the transient stress damage analysis process specifically includes: The impeller speed and active power within the preset third sliding window are obtained from the wind turbine's operating data. The statistical variance of the impeller speed is used as the speed fluctuation index, and the maximum absolute value of the difference between adjacent sampling points of active power is used as the power change rate index. When the speed fluctuation index is greater than the preset speed variance threshold or the power change rate index is greater than the preset power change rate threshold, it is determined that the current moment is in an unsteady operating condition and the current sliding window is marked as a transient analysis window. Obtain the transient vibration acceleration sequence and the corresponding transient impeller azimuth sequence within the transient analysis window. The transient vibration acceleration sequence is composed of the original vibration acceleration within the transient analysis window, and the transient impeller azimuth sequence is composed of the impeller azimuth within the transient analysis window.

7. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 6, characterized in that, In step S3, the transient stress damage analysis process specifically includes: Within the transient analysis window, a fixed angular domain resampling resolution is set, and the vibration signal is resampled in the angular domain to generate a second average angular domain waveform. The second average angular domain waveform includes the vibration acceleration at equal impeller azimuth intervals for the entire rotation cycle. Based on the current impeller azimuth phase, the second average angular domain waveform is synchronized with the pre-stored steady-state tower shadow reference template by performing phase alignment and synchronous differential operation to obtain the vibration residual signal vector; The acoustic signal is acquired, and the acoustic signal is decomposed into an acoustic time-frequency complex matrix using short-time Fourier transform. The spectral kurtosis vector of each frequency slice in the acoustic time-frequency complex matrix is ​​calculated, and the maximum acoustic kurtosis value is searched. Calculate the energy signal-to-noise ratio of the vibration residual signal vector within a local sliding window of the preset transient analysis window; If the maximum acoustic kurtosis value is greater than the preset acoustic kurtosis threshold, and the energy signal-to-noise ratio of the vibration residual signal vector is greater than the preset energy mutation threshold, then the first damage analysis result is determined to be a detected dynamic suspected feature.

8. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 7, characterized in that, Step S4 includes the following sub-steps: Step S41: If a static suspected feature or a dynamic suspected feature is detected, the real-time wind speed and pitch motor temperature are obtained, and a safety assessment is performed. The safety assessment is whether the real-time wind speed is lower than the preset safety upper limit and whether the pitch motor temperature is lower than the preset overheat alarm threshold. If the real-time wind speed is lower than the preset safety upper limit and the pitch motor temperature is lower than the preset overheat alarm threshold, the safety assessment is passed. Step S42: After passing the safety assessment, a micro-disturbance command is injected into the pitch system. The micro-disturbance command is a preset sine wave signal, which includes a preset specific excitation frequency and angle.

9. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 8, characterized in that, Step S4 further includes the following sub-steps: Step S43: Using the micro-perturbation command as a reference signal, the vibration signal and acoustic signal are synchronously subjected to phase-locked amplification analysis. For the vibration signal, quadrature demodulation is performed using the specific excitation frequency of the preset sine wave signal and the second harmonic of the specific excitation frequency as reference frequencies to extract the vibration fundamental frequency response amplitude and the vibration second harmonic response amplitude. For the acoustic signal, orthogonal demodulation is performed with the specific excitation frequency as the reference frequency. The acoustic modulation response amplitude synchronized with the excitation frequency is extracted, and the ratio of the acoustic modulation response amplitude to the background acoustic noise amplitude is calculated as the acoustic synchronization signal-to-noise ratio. Step S44: The ratio of the vibration second harmonic response amplitude to the vibration fundamental frequency response amplitude is taken as the proportion of vibration nonlinear harmonics. The proportion of vibration nonlinear harmonics is compared with a preset nonlinear damage threshold, and the acoustic synchronization signal-to-noise ratio is compared with a preset acoustic response threshold. If the proportion of nonlinear harmonic vibrations is greater than the nonlinear damage threshold and the acoustic synchronization signal-to-noise ratio is greater than the acoustic response threshold, then nonlinear damage is determined to exist. If the proportion of vibration nonlinear harmonics is less than or equal to the nonlinear damage threshold, or the acoustic synchronization signal-to-noise ratio is less than or equal to the acoustic response threshold, then it is determined that there is no nonlinear damage. The second damage analysis result includes the presence of nonlinear damage and the absence of nonlinear damage.

10. The wind turbine blade condition monitoring method based on multi-sensor fusion as described in claim 9, characterized in that, In step S5, the logic for generating the final comprehensive status monitoring report of the wind turbine blades specifically includes: If no common-mode fault characteristics, no static suspected characteristics, and no dynamic suspected characteristics are detected, the final comprehensive status monitoring report will be output as "wind turbine blade health". If common mode fault characteristics are detected, the presence of common mode faults will be marked in the final integrated status monitoring report. Common mode faults include blade icing or overall aging. If nonlinear damage is determined to exist, the confirmed blade crack damage is marked in the final integrated condition monitoring report, and the blade number and impeller azimuth angle recorded in step S3 are associated with the output. If a static or dynamic suspected feature is detected, and there is no nonlinear damage, then the static or dynamic suspected feature is determined to be a false signal caused by environmental interference, and the final comprehensive status monitoring report is output as the health of the wind turbine blades.

Images