Blade Fault Diagnosis Method and Device Based on Wind Turbine Generator Vibration

Abstract

This invention provides a method and device for blade fault diagnosis based on wind turbine vibration, relating to the field of blade fault diagnosis technology for wind turbine vibration. This invention synchronously collects vibration signals at the nacelle foundation and blade root. After filtering and steady-state condition screening, it accurately identifies the first-order modal frequency and damping ratio of the blade under different operating conditions through phase compensation and frequency response function calculation. The modal parameters of multiple operating points are sequentially connected to construct a dynamic response trajectory. Local mode feature vectors are formed by calculating the relative vectors between each point and other points in the trajectory, and then integrated into a global mode feature matrix. The trajectory mode deviation index is obtained by calculating the Frobenius norm between the monitored matrix and the reference matrix. Structural damage is determined based on the statistical law of this index continuously exceeding a threshold.

Description

Technical Field

[0001] This invention relates to the field of blade fault diagnosis technology for wind turbine generator vibration, specifically to a method and apparatus for blade fault diagnosis based on wind turbine generator vibration. Background Technology

[0002] As the core component for capturing wind energy, wind turbine blades are subjected to complex alternating loads over long periods, making them prone to structural damage such as cracks and delamination. If not detected in time, this can lead to catastrophic fractures, causing huge economic losses and safety accidents. Therefore, conducting vibration-based blade condition monitoring and fault diagnosis is of vital engineering value for ensuring the safe operation of wind turbines and achieving predictive maintenance. Traditional vibration diagnosis methods typically involve placing sensors in the nacelle or tower and inferring the blade's health condition by analyzing changes in the amplitude, frequency, and other characteristics of vibration signals. This type of method is an important component of current wind farm condition monitoring systems.

[0003] In existing technologies, vibration-based blade diagnostic methods mainly rely on identifying blade modal parameters, such as natural frequency and damping ratio, under specific operating conditions, and then identifying faults by monitoring the deviation of these parameters from a healthy baseline. These methods typically assume that changes in modal parameters are isolated, i.e., they only focus on whether the absolute deviation of the frequency or damping ratio exceeds a threshold. However, under different wind speeds, power, and other operating conditions, the modal parameters of the blade itself will exhibit inherent volatility due to load changes. Existing methods treat data points under different operating conditions in isolation or attempt to establish a fixed threshold range that allows for fluctuations. This essentially ignores the continuity and integrity of the blade's dynamic characteristics as they evolve with operating conditions.

[0004] The shortcomings of such existing technologies are that they are difficult to effectively distinguish between the natural fluctuations of modal parameters caused by changes in normal operating loads and the abnormal changes caused by structural damage. For early and minor damage, the parameter change signals are weak and easily drowned out by normal operating condition fluctuations, resulting in low diagnostic sensitivity and inability to achieve early warning. In addition, due to the failure to capture the intrinsic correlation between modal parameters, the existing methods are not very targeted to damage and are easily affected by environmental noise, resulting in a high false alarm rate.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and apparatus for diagnosing blade faults based on the vibration of wind turbine generator sets, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for diagnosing blade faults based on wind turbine vibration, comprising the following steps:

[0009] Step 1: Install sensors at the nacelle foundation and blade root of the wind turbine generator set to synchronously collect the nacelle reference vibration signal and the blade response vibration signal, and filter out the effective steady-state data segment through a preset time window;

[0010] Step 2: Based on the valid data segments, analyze the time delay between the nacelle reference signal and the blade response signal. Through phase compensation and frequency response function calculation, identify the first-order modal parameters of the blade, including modal frequency and modal damping ratio. Repeat the above operation in multiple different valid data segments to obtain a series of modal parameters.

[0011] Step 3: Connect a series of modal parameters in a coordinate system with modal frequency as the abscissa and modal damping ratio as the ordinate to construct the trajectory of the dynamic response. For each modal parameter on the trajectory, calculate its relative vector with all other modal parameters in the trajectory to form a local feature vector. Integrate all local feature vectors to construct a global feature matrix that characterizes the overall trajectory shape.

[0012] Step 4: Calculate the baseline global feature matrix when the blade is healthy and undamaged, following steps 1 to 3. Calculate the Frobenius norm between the global feature matrix and the baseline global feature matrix. Define the norm value as the trajectory deviation. Construct a state transition probability matrix based on the trajectory deviation to determine whether there is structural damage to the wind turbine blade.

[0013] Furthermore, a time window of fixed length T is defined, starting from the current time t and moving backwards. ,in , The time length is 10 complete rotation cycles. Within the time window, synchronous signal acquisition is performed by a reference acceleration sensor located on the main frame of the nacelle and a response acceleration sensor located inside the blade root flange to obtain the nacelle reference vibration signal and the blade response vibration signal.

[0014] A bandpass filter is used to filter the signal, with its passband frequency range covering the first-order flapping mode frequency of the blades. This bandpass filtering preserves the main modal frequency components of the blades. During unit operation, wind speed and output power signals are monitored in real time, and a range of lengths is set. A sliding window, making the sliding window appear within the time window. Within the sliding window, when the fluctuation range of wind speed and output power is less than their respective preset thresholds, the unit is determined to have entered steady-state operation, and the time interval corresponding to this sliding window is taken as a valid data segment.

[0015] Furthermore, based on each valid data segment, the first-order modal parameters of the blade are identified, specifically including: calculating the nacelle reference signal under each valid data segment. With blade response signal cross power spectrum and the self-power spectrum of the cabin reference signal ,in and They are respectively and Fourier transform, Where is the expectation operation, For complex conjugate;

[0016] Through the The cross-correlation function is obtained by performing an inverse Fourier transform: ,in This is the inverse Fourier transform;

[0017] Searching for Initial time delay to reach the maximum value ,use , , Parabolic interpolation of the cross-correlation function values ​​at the three points is performed to accurately determine the time delay. The interpolation formula is as follows:

[0018] ;

[0019] in, This represents the sampling time interval.

[0020] Furthermore, in order to eliminate time delay The impact on phase information, in the frequency domain, on the cross-power spectrum of the two. Compensation will be provided, and the compensation formula is as follows:

[0021] ;

[0022] in, The imaginary unit, The positive sign of the compensation, representing the frequency, is used to offset the time delay in the blade response signal relative to the nacelle reference signal. This corrects the phase of the cross power spectrum;

[0023] And using the compensated cross power spectrum Self-power spectrum of the cabin reference signal Calculate the frequency response function from the nacelle to the blades: ,exist In the amplitude spectrum, the highest peak value located within the theoretical frequency range of the first-order flapping mode of the blade is identified, the frequency corresponding to the peak value is determined as the modal frequency, and the modal damping ratio of the first-order mode is calculated using the half-power bandwidth method.

[0024] Furthermore, when constructing the trajectory of the dynamic response, the modal frequency and modal damping ratio corresponding to each effective data segment are treated as a single data point. A total of N data points were obtained, and the data points were connected sequentially according to the starting time of the valid data segments to form a trajectory. Let i be the modal frequency of the i-th data point. Let i be the modal damping ratio of the i-th data point;

[0025] For the i-th data point on the trajectory Its local feature vector It is A column vector of dimension 1 has the following mathematical expression:

[0026] ;

[0027] in, `i` is the transpose operator, where `i` is the index of the data point, and .

[0028] Furthermore, Z-score normalization is performed on the modal frequencies and modal damping ratios within all valid data segments in the time window. Then, all N normalized local feature vectors are analyzed according to their corresponding data points. The elements are stacked sequentially as row vectors to form the global feature matrix. Signals from the same unit's blades under healthy and undamaged conditions are processed using the same method to obtain a baseline global feature matrix. ;

[0029] Define trajectory deviation index The Frobenius norm of the difference between two matrices is calculated as follows:

[0030] ;

[0031] in, As the baseline global feature matrix, The global feature matrix, This represents the Frobenius norm of the matrix.

[0032] Furthermore, under the condition that the blades of the same unit are healthy and undamaged, historical data on the trajectory deviation index are obtained through multiple historical time windows, and their average value is calculated. and standard deviation The trajectory deviation index obtained through real-time monitoring abnormal probability Calculated by the following formula:

[0033] ;

[0034] in, The cumulative distribution function of the standard normal distribution;

[0035] When the abnormal probability is continuously higher than a set threshold for K consecutive time windows, it is determined that the blade has structural damage, where K is greater than or equal to 3.

[0036] The present invention also provides a blade fault diagnosis device based on wind turbine generator vibration, the device being used to perform the above-described blade fault diagnosis method based on wind turbine generator vibration, comprising:

[0037] The signal acquisition module is used to deploy sensors at the nacelle foundation and blade root of the wind turbine generator set to simultaneously acquire the nacelle reference vibration signal and the blade response vibration signal, and to filter out the effective steady-state data segments through a preset time window.

[0038] The modal calculation module is used to analyze the time delay between the nacelle reference signal and the blade response signal based on the effective data segment. Through phase compensation and frequency response function calculation, it identifies the first-order modal parameters of the blade, including the modal frequency and modal damping ratio. The above operation is repeated in multiple different effective data segments to obtain a series of modal parameters.

[0039] The feature construction module is used to connect a series of modal parameters in a coordinate system with modal frequency as the abscissa and modal damping ratio as the ordinate to construct the trajectory of the dynamic response. For each modal parameter on the trajectory, its relative vector with all other modal parameters in the trajectory is calculated to form a local feature vector. All local feature vectors are integrated to construct a global feature matrix that characterizes the overall trajectory shape.

[0040] The result judgment module is used to calculate the baseline global feature matrix when the blade is healthy and undamaged according to steps 1 to 3, calculate the Frobenius norm between the global feature matrix and the baseline global feature matrix, define the norm value as the trajectory deviation, and construct a state transition probability matrix based on the trajectory deviation to determine whether there is structural damage to the wind turbine blade.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This invention ensures the accuracy of the basic data points for constructing the trajectory through precise phase compensation and modal parameter identification. The defined local mode feature vector and global mode feature matrix transform the geometric shape of the trajectory, i.e., the synergistic relationship between modal frequency and damping ratio as load changes, into a quantifiable mathematical object. This technical feature enables the method to keenly capture the trajectory shape distortion caused by structural damage. Such changes in overall shape are incomparable to the slight offset of a single parameter point, thus achieving high-sensitivity detection of early and minor damage.

[0043] The diagnostic basis of this invention is the morphological similarity of the entire trajectory, which is achieved by calculating the Frobenius norm of two global pattern feature matrices. Therefore, it is not sensitive to fluctuations caused by random noise or instantaneous disturbances in a single data point. Even under different normal operating loads, the trajectory morphology of a healthy blade maintains inherent consistency. However, once damage occurs, this consistency will be disrupted, leading to a significant increase in the deviation index. This judgment logic based on pattern similarity rather than absolute parameter values ​​makes the diagnostic results less susceptible to fluctuations in normal operating conditions.

[0044] This invention, by setting statistically based continuous monitoring and judgment conditions, can accurately capture real fault trends while effectively filtering out accidental interference signals, providing wind farm operation and maintenance personnel with highly reliable early warning information, thereby achieving truly effective predictive maintenance. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0046] Figure 2 This is a histogram of the trajectory deviation index and anomaly probability of the present invention.

[0047] Figure 3 This is a spline plot of the modal frequency-modal damping ratio of the present invention;

[0048] Figure 4 This is a curve showing the trajectory deviation index and anomaly probability fitting of the present invention.

[0049] Figure 5 This is a normal probability diagram of the abnormal probability of the present invention;

[0050] Figure 6 This is a flowchart of the overall device structure of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0052] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0053] Example:

[0054] Please see Figures 1-5 The present invention provides a technical solution:

[0055] A method for diagnosing blade faults based on wind turbine vibration, comprising the following steps:

[0056] Step 1: Install sensors at the nacelle foundation and blade root of the wind turbine generator set to synchronously collect the nacelle reference vibration signal and the blade response vibration signal, and filter out the effective steady-state data segment through a preset time window;

[0057] For the nacelle foundation, this location senses the combined vibration transmitted from the entire drivetrain (including the rotor, main shaft, gearbox, and generator) to the nacelle. It serves as a reference signal, representing the combined effect of the excitation source and the transmission path. For the blade root, this location directly senses the local dynamic response of the blade. Since the blade is a flexible body, its vibration contains rich structural dynamic information. Combining the root response with the nacelle reference signal allows for the isolation of the blade's own dynamic characteristics. Synchronous acquisition ensures strict time alignment between the two signals, which is fundamental for subsequent time delay estimation, cross-correlation analysis, and frequency response function calculation. Without synchronization, all subsequent analyses will contain systematic errors.

[0058] The time window refers to the total duration of data acquisition, while the effective data segment refers to the high-quality short data fragments selected from the total acquired data that are suitable for subsequent precise modal analysis. The selection criterion is that the unit is in steady-state operating condition.

[0059] Define a time window of fixed length T that ends at the current time t and traces backwards. ,in , The time window is defined as 10 complete rotation cycles. Within this time window, synchronous signal acquisition is performed using a reference accelerometer located on the main frame of the nacelle and a response accelerometer located inside the blade root flange to obtain the nacelle reference vibration signal and the blade response vibration signal. The sampling frequency of the reference accelerometer and the response accelerometer is set to 100Hz, i.e., the sampling time interval is... This setting, within seconds, satisfies the Nyquist sampling theorem, effectively covering the main frequency components of blade vibration (below 50Hz) and providing sufficient resolution for subsequent cross-correlation calculations and time delay interpolation; within the time window Internally, synchronous signal acquisition is used for To ensure that the time series of the nacelle reference vibration signal and the blade response vibration signal have a consistent time reference, the two tests are performed at intervals.

[0060] ,in This represents the current rotational speed of the wind turbine blades. Ensure the data cache is large enough, for example if If the rotational speed is approximately 0.33 Hz (corresponding to a typical fan speed of about 0.33 Hz, 10 revolutions is 30 seconds), then T must be at least greater than 90 seconds; Ten rotation cycles were chosen to be statistically representative—too short a time would fail to capture a stable vibration cycle and be susceptible to random fluctuations; too long a time would reduce the system's response speed to changes in operating conditions. Ten cycles is an empirical value that strikes a balance between stability and timeliness.

[0061] A bandpass filter is used to filter the signal, with its passband frequency range covering the flapping mode frequencies of the blades. This bandpass filtering preserves the main modal frequency components of the blades. During unit operation, wind speed and output power signals are monitored in real time, and a range of lengths is set... A sliding window, making the sliding window appear within the time window. Within the sliding window, when the fluctuation range of wind speed and output power is consistently less than their respective preset thresholds, the unit is determined to have entered steady-state operation, and the time interval corresponding to this sliding window is taken as a valid data segment.

[0062] The filter uses a zero-phase digital filter to avoid phase distortion. The filter removes high-frequency noise, such as gearbox and generator vibration, and low-frequency interference, such as tower sway, highlighting vibration components related to blade flapping modes and improving the signal-to-noise ratio.

[0063] The most easily excited and most sensitive to damage modes of blades are usually low-order modes, especially the first-order flapping mode. First, it is necessary to estimate the frequency range of the first-order flapping mode of the blade through simulation calculations or unit design data. For example, for large wind turbine blades, this range is between 0.8Hz and 1.5Hz. Then the passband of the bandpass filter should be set within this range, such as 0.5Hz-2.0Hz. Doing so can significantly improve the signal-to-noise ratio and make subsequent mode identification more accurate.

[0064] Length is The sliding window is the execution unit for steady-state judgment, with... For step size, in The window is slid inwards, and the data within each window is examined to assess steady-state characteristics. The quantification standard for steady-state is based on the statistical volatility of the data within the window, requiring the volatility range to be less than a preset threshold. Specifically: for wind speed data, the coefficient of variation (the ratio of standard deviation to mean) of the wind speed values ​​within the window is calculated, and this coefficient of variation is required to be less than a preset threshold (e.g., not exceeding 0.1). For power data, the coefficient of variation (the ratio of standard deviation to mean) of the power values ​​within the window is calculated, and this coefficient of variation is required to be less than a preset threshold (e.g., not exceeding 0.03). The coefficient of variation is defined as the ratio of standard deviation to mean. Only when the volatility statistics of all monitored parameters within the window meet their respective preset threshold requirements is the time window determined to be in steady-state condition.

[0065] Step 2: Based on the valid data segments, analyze the time delay between the nacelle reference signal and the blade response signal. Through phase compensation and frequency response function calculation, identify the first-order modal parameters of the blade, including modal frequency and modal damping ratio. Repeat the above operation in multiple different valid data segments to obtain a series of modal parameters.

[0066] In physics, time delay analysis involves the time required for vibration to travel from the nacelle to the blade root. This delay includes information about the signal's propagation along its structural path. Accurately estimating this delay is the first step in ensuring the accuracy of subsequent analyses. Due to this time delay, the blade response signal... It will lag behind the cabin reference signal in phase. This phase lag distorts the calculated frequency response function. Phase compensation is an alignment operation designed to eliminate this propagation effect, allowing us to observe the characteristics of the pure blade dynamics subsystem.

[0067] The frequency response function describes the relationship between the output (response) and input (excitation) of the turbine blades in the frequency domain. At the modal frequency, the turbine blades will resonate, and the response will be significantly amplified. This will be manifested as a prominent peak in the frequency response function. By analyzing this peak, the key modal frequencies and damping ratios can be extracted.

[0068] Identifying the first-order modal parameters of the blade based on each valid data segment specifically includes: calculating the nacelle reference signal under each valid data segment. With blade response signal cross power spectrum and the self-power spectrum of the cabin reference signal ,in and They are respectively and Fourier transform, Where is the expectation operation, For complex conjugate;

[0069] It is a complex number, representing the frequency component in two signals. The common power and phase relationship on the frequency reflects the frequency. At this point, the degree of linear correlation between the blade response signal and the nacelle reference signal is considered; the larger the amplitude, the stronger the correlation for that frequency component. from Passed to The stronger the energy, the more likely this frequency is to be the system's natural frequency. The size directly depends on and The amplitude, if input If there is energy at this frequency, and the system is easily excited at this frequency (i.e., close to the modal frequency), then the output... The amplitude will be very large, thus leading to The amplitude is also very large;

[0070] It is a real number representing the blade reference signal. The distribution of its own power in the frequency domain reflects the frequency components of the excitation signal. The amount of energy contained within indicates at which frequencies the system is excited. Directly by The square of the amplitude determines, The larger the amplitude, the better. The larger the value at that frequency;

[0071] Through the The cross-correlation function is obtained by performing an inverse Fourier transform: ,in This is the inverse Fourier transform;

[0072] This indicates that two signals are delayed at different times. The similarity under the signal reflects the signal With time delay The degree of similarity, when When equal to the actual time delay between two signals, It will reach its maximum value, therefore it is a key function indicating the time delay between signals; The larger the value, the more time is delayed. Below, the more similar the waveforms of two signals, the closer the global maximum value corresponds to... The main time delay to be found;

[0073] Searching for Initial time delay to reach the maximum value ,use , , Parabolic interpolation of the cross-correlation function values ​​at the three points is performed to accurately determine the time delay. The interpolation formula is as follows:

[0074] ;

[0075] in, The sampling time interval;

[0076] Since the signal is discretely sampled, the initial maximum point is found. Only accurate to the sampling interval However, the true maximum point lies between the two sampling points. Parabolic interpolation uses the maximum point and its left and right adjacent points to fit a parabola and find the vertex of this parabola, thereby reducing the time delay. The accuracy is improved to the subsampling level;

[0077] The above formula uses these three function values , , Calculate the x-coordinate of the vertex of the parabola. The larger the values ​​at the left and right points (i.e., the sharper the peak), the better the calculated value. The closer If the peak values ​​are asymmetrical, It will shift towards the side with a higher function value;

[0078] To eliminate time delay The impact on phase information, in the frequency domain, on the cross-power spectrum of the two. Compensation will be provided, and the compensation formula is as follows:

[0079] ;

[0080] in, The imaginary unit, The positive sign of the compensation, representing the frequency, is used to offset the time delay in the blade response signal relative to the nacelle reference signal. This corrects the phase of the cross-power spectrum; the formula is multiplied by To compensate for (offset) the time delay in the response signal The resulting phase lag The positive sign compensates for the delay itself, which has the effect of advancing the phase of the response signal, thereby aligning it with the reference signal.

[0081] And using the compensated cross power spectrum Self-power spectrum of the cabin reference signal Calculate the frequency response function from the nacelle to the blades: ,exist In the amplitude spectrum, the highest peak value located within the theoretical frequency range of the first-order flapping mode of the blade is identified, the frequency corresponding to the peak value is determined as the modal frequency, and the modal damping ratio of the first-order mode is calculated using the half-power bandwidth method.

[0082] It is a complex number containing amplitude and phase information. It describes the frequency domain relationship between the vibration response of the blade as a system and the excitation it receives, at the blade's modal frequencies. The amplitude will show a significant peak because the system resonates at this frequency, and a small excitation can produce a huge response. The larger the amplitude, the stronger the system performance at that frequency. The stronger the vibration response amplification capability at a given point, the stronger the implication that the frequency is a natural frequency of the system.

[0083] Within the estimated theoretical frequency range, for example, obtained through design drawings or simulations, search The highest peak in the amplitude spectrum, the frequency corresponding to this peak. This refers to the identified first-order flapping mode frequency, which is a fundamental inherent property of the blade.

[0084] In the identified modal frequencies Find the amplitude at the peak. Calculate the -3d point, i.e., the amplitude drops to The two frequencies corresponding to time and , Calculate half-power bandwidth Modal damping ratio The calculation formula is: Modal damping ratio reflects the ability of a structural system to dissipate vibrational energy. The larger the damping ratio, the stronger the system's ability to dissipate energy and the faster the vibration decays. For a healthy blade, the modal damping ratio is usually within a stable and small range. If the modal damping ratio increases significantly, it means that there is frictional damage in the structure, such as energy dissipation due to friction between crack surfaces. If it decreases significantly, it means that the structural connections are loose and the energy dissipation mechanism is weakened. Therefore, the modal damping ratio, like the modal frequency, is a key indicator for diagnosing the health status of a structure.

[0085] Step 3: Connect a series of modal parameters in a coordinate system with modal frequency as the abscissa and modal damping ratio as the ordinate to construct the trajectory of the dynamic response. For each modal parameter on the trajectory, calculate its relative vector with all other modal parameters in the trajectory to form a local feature vector. Integrate all local feature vectors to construct a global feature matrix that characterizes the overall trajectory shape.

[0086] A single data point can only reflect the static characteristics of a blade under a fixed operating condition, while the trajectory describes the dynamic process of how the blade's dynamic characteristics change continuously with the operating conditions (load). Each data point represents the "dynamic fingerprint" of the blade under a specific steady-state operating condition. The order of the starting time is essentially the order of the operating conditions. Since the data is collected under steady-state operating conditions at different wind speeds (i.e., different loads), this order implies the change of load from small to large. Therefore, this trajectory reveals the inherent law of the blade's modal parameters (frequency and damping) changing with wind load. A healthy blade has a stable and repeatable trajectory shape.

[0087] When constructing the trajectory of the dynamic response, the modal frequency and modal damping ratio corresponding to each effective data segment are treated as a data point. A total of N data points were obtained, and the data points were connected sequentially according to the starting time of the valid data segments to form a trajectory. Let i be the modal frequency of the i-th data point. Let i be the modal damping ratio of the i-th data point;

[0088] For the i-th data point on the trajectory Its local feature vector It is A column vector of dimension 1 has the following mathematical expression:

[0089] ;

[0090] in, `i` is the transpose operator, where `i` is the index of the data point, and ;

[0091] To quantify the geometry of the entire trajectory, a mathematical tool is needed to describe the relationship between each data point and the global context: local feature vectors. This is to create a "global context description" for each data point on the trajectory;

[0092] It is itself a high-dimensional vector, and its elements are the current data points. The difference between the coordinates of the data points and all other data points on the trajectory. Reflects data points Its unique position and role in the entire trajectory pattern: it is not merely information from a single data point, but rather encodes... The relative relationship with all other data points on the trajectory, if If it is an anomaly (e.g., caused by injury), then its relative relationship with most points on the healthy trajectory will change drastically, thus causing its... Significant changes occurred;

[0093] The size of each element in the representation The components of the Euclidean distance to other data points in the modal parameter space on each coordinate axis (frequency, damping), if Because of the damage, it has deviated significantly from its proper healthy trajectory, so these differences... The absolute value will generally increase;

[0094] Each element directly depends on these coordinate values. If the turbine blades are healthy, the trajectory formed by all data points is stable, then each... It is also stable; however, if damage occurs, the coordinates of some data points will change, which will cause that point to... Almost all elements in the data change, which also slightly affects other data points. This characteristic, which has a ripple effect on the whole system, makes this feature extremely sensitive to local damage;

[0095] Z-score normalization is performed on the modal frequencies and modal damping ratios within all valid data segments in the time window. Then, all N normalized local feature vectors are analyzed according to their corresponding data points. The elements are stacked sequentially as row vectors to form the global feature matrix. Signals from the same unit's blades under healthy and undamaged conditions are processed using the same method to obtain a baseline global feature matrix. ;

[0096] The global feature matrix combines N vectors that describe the local feature vectors of each point into a mathematical object that can describe the complete shape of the entire trajectory.

[0097] Z-score normalization is used to eliminate the bias caused by the different dimensions and orders of magnitude of the two parameters, modal frequency and damping ratio. The frequency value is usually around 1 Hz, while the damping ratio is only 0.01. If not processed, the small fluctuations in frequency will dominate in the feature vector and mask the changes in damping ratio. After normalization, the modal frequency and damping ratio are placed on a comparable scale, ensuring that the changes of both can be fairly considered when constructing the feature vector.

[0098] The mean and standard deviation used for normalization must come from long-term monitoring data of leaf health. For the monitoring data, this set of fixed health baseline statistics should be used for normalization, rather than using its own data to calculate new mean and standard deviation. This ensures that all data are compared under the same baseline.

[0099] It is no longer merely a "digital template" of health trajectory patterns; more accurately, it is a health trajectory behavior code embedded with temporal sequence information and based on relative spatial relationships. It contains the relative vector relationships (spatial information) between all data points, and through the strict order of rows, The timestamp sequence (time information) of these data points when they were confirmed is strongly preserved; The specific spatial configuration formed by the dynamic parameters of a healthy blade under the stated time series is defined. Any factor that disrupts this spatiotemporal configuration (such as structural damage) will affect the currently constructed global feature matrix. and This results in significant differences.

[0100] Step 4: Calculate the baseline global feature matrix when the blade is healthy and undamaged according to steps 1 to 3, calculate the Frobenius norm between the global feature matrix and the baseline global feature matrix, define the norm value as the trajectory deviation, and construct the state transition probability matrix based on the trajectory deviation to determine whether there is structural damage to the wind turbine blade.

[0101] Define trajectory deviation index The Frobenius norm of the difference between two matrices is calculated as follows:

[0102] ;

[0103] in, As the baseline global feature matrix, The global feature matrix, Denotes the Frobenius norm of a matrix;

[0104] The Frobenius norm is the mathematical tool that compresses the overall difference between these two complex matrices into a single, comparable scalar value. It quantitatively reflects the overall deviation between the overall dynamic behavior pattern of the monitored blade and its health baseline pattern. It calculates the overall level of the difference between all corresponding elements in the two global feature matrices. The larger the value, the greater the difference in overall morphology between the current dynamic response trajectory of the monitored blade and its healthy trajectory. This difference is caused by changes in the structural characteristics of the blade (i.e., damage). A value close to 0... The value indicates that the current behavior is almost identical to healthy behavior;

[0105] It is the square root of the sum of the squares of these differences, therefore, and An increase in the difference between any corresponding element will lead to The value increases, It is sensitive to any tiny changes in the matrix and can comprehensively and thoroughly capture any local or global morphological distortions caused by damage.

[0106] Under the condition that the blades of the same unit are healthy and undamaged, historical data of the trajectory deviation index are obtained through multiple historical time windows, and its average value is calculated. and standard deviation The trajectory deviation index obtained through real-time monitoring abnormal probability Calculated by the following formula:

[0107] ;

[0108] in, The cumulative distribution function of the standard normal distribution;

[0109] Specific data on some time windows and anomaly probabilities are shown in Table 1.

[0110] Table 1. Statistical Chart of Anomalies

[0111]

[0112] Analysis of the data revealed a certain correlation between different characteristic parameters. For example, the data showed a negative correlation between modal frequency and trajectory deviation index. As the time window increased, the modal frequency gradually decreased (from 0.5995 to 0.5858), while the trajectory deviation index increased (from 0.027 to 0.064). This indicates that under conditions of lower modal frequency, the vibration characteristics of the blade change, leading to an increase in trajectory deviation and reflecting a deterioration in the structural condition.

[0113] When analyzing the relationship between modal frequency and anomaly probability, the anomaly probability gradually increases as the modal frequency decreases. This is because the decrease in modal frequency is related to blade structural damage, thereby improving the sensitivity of anomaly detection. For example, the modal frequency in time window 1 is 0.5995, and the corresponding anomaly probability is 0.556, while the modal frequency in time window 15 is 0.5858, and the anomaly probability rises to 0.811. The combined influence of modal frequency and trajectory deviation should be considered to optimize the accuracy and reliability of blade health monitoring.

[0114] When the anomaly probability remains above a set threshold for K consecutive time windows, structural damage to the blade is determined, where K is greater than or equal to 3. This determination strategy perfectly balances sensitivity and reliability, providing sufficient sensitivity to early damage (through...). It can capture signals and, through continuous requirements, has strong anti-interference capabilities, effectively avoiding false alarms, making the final diagnostic conclusion very robust and sufficient as the basis for shutdown and maintenance decisions.

[0115] When the leaves are healthy, steps 1-4 are performed repeatedly over a long period of time to calculate a large number of data. Values, these values ​​constitute the health status Sample distribution, using its mean and standard deviation To fully describe; mean Represents healthy leaves The normal average value should ideally be close to 0, but due to various random factors, it is usually a value slightly greater than 0. Represents healthy leaves The normal range of fluctuations in values The larger the value, the better the health status. It is inherently unstable and fluctuates greatly; and They jointly defined a reference benchmark for health status, for example This is the limit of normal fluctuations;

[0116] The current The value is transformed into a more intuitive and interpretable metric—it represents the probability of outliers, which is more accurate than using it directly. Using the absolute value of the threshold for judgment is more precise, and the anomaly probability... It is a value between 0 and 1. Indicator of current observed trajectory deviation The probability of belonging to a healthy distribution; The larger the value, the more it indicates the current... The value originates from an abnormal state, therefore the possibility of abnormal leaf condition (damage) is extremely high, for example... This means that in a healthy state, there is only a 5% probability of observing the current state. The value is either greater than the value, therefore there is a high degree of confidence that the unit's blades are abnormal;

[0117] It is standardized Value, that is It indicates the current How many standard deviations from the healthy mean? The larger the value, the better. The closer the cumulative probability of this value in the standard normal distribution is to 1, the better. The closer it gets to 1, the more monotonically increasing it becomes: standardized. The larger the value, the higher the probability of an anomaly. The larger it is;

[0118] when The value remains below the set threshold for K consecutive time windows. ,For example If the abnormal signal is detected when K is greater than or equal to 3, then structural damage to the blade is determined. This determination strategy balances sensitivity and specificity. It requires the abnormal signal to be continuous to avoid false alarms caused by transient interference, such as when there are abnormal signals in three consecutive time windows. Then there is a high degree of confidence in determining the damage.

[0119] Please see Figure 6 The present invention also provides a blade fault diagnosis device based on wind turbine generator vibration, the device being used to perform the above-described blade fault diagnosis method based on wind turbine generator vibration, comprising:

[0120] The signal acquisition module is used to deploy sensors at the nacelle foundation and blade root of the wind turbine generator set to simultaneously acquire the nacelle reference vibration signal and the blade response vibration signal, and to filter out the effective steady-state data segments through a preset time window.

[0121] The modal calculation module is used to analyze the time delay between the nacelle reference signal and the blade response signal based on the effective data segment. Through phase compensation and frequency response function calculation, it identifies the first-order modal parameters of the blade, including the modal frequency and modal damping ratio. The above operation is repeated in multiple different effective data segments to obtain a series of modal parameters.

[0122] The feature construction module is used to connect a series of modal parameters in a coordinate system with modal frequency as the abscissa and modal damping ratio as the ordinate to construct the trajectory of the dynamic response. For each modal parameter on the trajectory, its relative vector with all other modal parameters in the trajectory is calculated to form a local feature vector. All local feature vectors are integrated to construct a global feature matrix that characterizes the overall trajectory shape.

[0123] The result judgment module is used to calculate the baseline global feature matrix when the blade is healthy and undamaged according to steps 1 to 3, calculate the Frobenius norm between the global feature matrix and the baseline global feature matrix, define the norm value as the trajectory deviation, and construct a state transition probability matrix based on the trajectory deviation to determine whether there is structural damage to the wind turbine blade.

[0124] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0125] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0126] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A blade fault diagnosis method based on wind turbine vibration, characterized by, The specific steps include: Step 1: arranging sensors at the nacelle base position and the blade root of the wind turbine generator set, synchronously collecting the nacelle reference vibration signal and the blade response vibration signal, and screening out the steady-state effective data segment through a preset time window; Step 2: according to the effective data segment, analyzing the time delay between the nacelle reference signal and the blade response signal, identifying the modal parameters of the first order of the blade, including the modal frequency and the modal damping ratio, repeating the above operation on multiple different effective data segments to obtain a series of modal parameters; Step 3: connecting the series of modal parameters in the coordinate system with the modal frequency as the horizontal coordinate and the modal damping ratio as the vertical coordinate in turn to construct the trajectory of the dynamic response, calculating the relative vector of each modal parameter on the trajectory with all other modal parameters in the trajectory to form a local feature vector, integrating all local feature vectors to construct a global feature matrix representing the overall trajectory form; Step 4: calculating the baseline global feature matrix of the blade when it is healthy and undamaged according to steps 1 to 3, calculating the Frobenius norm between the global feature matrix and the baseline global feature matrix, defining the norm value as the trajectory deviation degree, and judging whether the wind turbine generator set blade has structural damage according to the state transition probability matrix constructed according to the trajectory deviation.

2. The blade fault diagnosis method based on vibration of a wind turbine generator set according to claim 1, characterized in that: Set a time window with a fixed length of T backward from the current time t as the end point wherein , is the time length of 10 complete rotation periods, and synchronous signal acquisition is performed in the time window by the reference acceleration sensor located on the main frame of the cabin and the response acceleration sensor located on the inside of the blade root flange to obtain the cabin reference vibration signal and the blade response vibration signal; The signal is filtered by a band-pass filter, and the pass-band frequency range covers the first flapping mode frequency of the blade, that is, the vibration signal is band-pass filtered to retain the main mode frequency component of the blade; during the operation of the unit, the wind speed and output power signals are monitored in real time, a sliding window with a length of is set, and the sliding window is slid within a time window , and when the fluctuation ranges of the wind speed and the output power are both less than the preset thresholds of the wind speed and the output power respectively within the sliding window, it is determined that the unit enters a steady state condition, and the time interval corresponding to the sliding window is regarded as an effective data segment.

3. The blade fault diagnosis method based on vibration of wind turbine generator set according to claim 2, characterized in that: Based on each valid data segment, modal parameters of the blade first order are identified, specifically comprising: calculating a machine cabin reference signal under each valid data segment and a cross-power spectrum of the blade response signal and the machine cabin reference signal , wherein and are Fourier transforms of and , wherein is an expectation operation, is a complex conjugate; The cross-correlation function is obtained by inverse Fourier transformation of where is the inverse Fourier transformation.​ Finding the initial time delay that maximizes the value of the cross-correlation function , , Parabolic interpolation of the cross-correlation function values at three points to determine the time delay accurately The interpolation formula is as follows: ； wherein is the sampling time interval.

4. The blade fault diagnosis method based on vibration of wind turbine generator set according to claim 3, characterized in that: To eliminate time delay The impact on the phase information, in the frequency domain to both the cross power spectrum Compensation, compensation formula is: ； wherein, is the imaginary unit, is the frequency, the positive sign of the compensation is used to counteract the time delay in the blade response signal relative to the nacelle reference signal so as to correct the phase of the cross-power spectrum; and using the compensated cross-power spectrum and the auto-power spectrum of the cabin reference signal , the frequency response function from the cabin to the blade is calculated: , in the amplitude spectrum of the highest peak value in the range of the first flapping mode theoretical frequency of the blade is identified, the frequency corresponding to the peak value is determined as the modal frequency, and the modal damping ratio of the first mode is calculated using the half-power bandwidth method.

5. The method for diagnosing blade failure based on vibration of wind turbine generator set according to claim 4, characterized in that: When constructing the trajectory of the dynamic response, the modal frequency and the modal damping ratio corresponding to each effective data segment are taken as a data point , a total of N data points are obtained, and each data point is sequentially connected to form a trajectory according to the starting time of the effective data segment in chronological order, wherein is the modal frequency of the i th data point, is the modal damping ratio of the i th data point; for the i-th data point on the trajectory whose local feature vector is a column vector of dimension d, mathematically expressed as follows: ； wherein is a transpose symbol, i is an index of a data point, and .

6. The blade fault diagnostic method based on vibration of a wind turbine generator set according to claim 5, characterized in that: Z-score normalization is performed on the modal frequencies and modal damping ratios within all valid data segments in the time window. Then, all N normalized local feature vectors are analyzed according to their corresponding data points. The elements are stacked sequentially as row vectors to form the global feature matrix. Signals from the same unit's blades under healthy and undamaged conditions are processed using the same method to obtain a baseline global feature matrix. ; Defining a trajectory deviation index The Frobenius norm of the difference of two matrices is calculated as follows: ； wherein, is the reference global feature matrix, is the global feature matrix, denotes the Frobenius norm of a matrix.

7. The method of diagnosing blade failure based on vibration of wind turbine generator set according to claim 6, characterized in that: Under the condition that the blades of the same unit are healthy and undamaged, historical data of the trajectory deviation index are obtained through multiple historical time windows, and its average value is calculated. and standard deviation The trajectory deviation index obtained through real-time monitoring abnormal probability Calculated by the following formula: ； wherein is the cumulative distribution function of the standard normal distribution; When the abnormal probability is continuously higher than the set threshold in the continuous K time windows, it is determined that the blade has structural damage, where K is greater than or equal to 3.

8. A blade fault diagnosis device based on vibration of a wind turbine generator unit, characterized by: The device is used to execute the wind turbine generator set vibration-based blade fault diagnosis method of any one of claims 1-7, comprising: A signal acquisition module is arranged for arranging sensors at the nacelle base position and the blade root of the wind turbine generator set, synchronously collecting the nacelle reference vibration signal and the blade response vibration signal, and screening out the steady-state effective data segment through a preset time window; A modal calculation module is arranged for analyzing the time delay between the nacelle reference signal and the blade response signal according to the effective data segment, identifying the modal parameters of the first order of the blade, including the modal frequency and the modal damping ratio, repeating the above operation on multiple different effective data segments to obtain a series of modal parameters; A feature construction module is arranged for connecting the series of modal parameters in the coordinate system with the modal frequency as the horizontal coordinate and the modal damping ratio as the vertical coordinate in turn to construct the trajectory of the dynamic response, calculating the relative vector of each modal parameter on the trajectory with all other modal parameters in the trajectory to form a local feature vector, integrating all local feature vectors to construct a global feature matrix representing the overall trajectory form; A result judgment module is arranged for calculating the baseline global feature matrix of the blade when it is healthy and undamaged according to steps 1 to 3, calculating the Frobenius norm between the global feature matrix and the baseline global feature matrix, defining the norm value as the trajectory deviation degree, and judging whether the wind turbine generator set blade has structural damage according to the state transition probability matrix constructed according to the trajectory deviation.

Images