An intelligent analysis system for wind power blade vibration signals

Abstract

The application discloses a kind of wind power generation blade vibration signal intelligent analysis system, the system includes signal acquisition module, matrix generation module, space-time joint module, damage locking module, damage evolution module, working condition adaptive module and damage classification determination module.Signal acquisition module acquires blade multi-measuring point vibration time sequence and identifies rotation period, generates original data set;Matrix generation module divides original data set into single-cycle signal segment, generates period angle two-dimensional signal matrix;Space-time joint module executes coherent accumulation on period angle two-dimensional signal matrix, inverts damage space distribution, generates two-dimensional damage space distribution diagram;Damage locking module executes peak detection and centroid fitting on two-dimensional damage space distribution diagram, generates damage space trajectory sequence;Damage evolution module executes curve fitting of fracture mechanics model constraint on damage equivalent area, outputs damage development prediction result.The application realizes the accurate positioning and evolution prediction of blade damage.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment condition monitoring technology, and in particular to an intelligent analysis system for vibration signals of wind turbine blades. Background Technology

[0002] Wind turbine blades operate in complex and harsh environments for extended periods, and are susceptible to structural damage such as cracks, delamination, and leading-edge corrosion due to factors such as wind speed variations, turbulence disturbances, and tower shadow effects. As a key component for capturing wind energy in wind turbine units, the structural integrity of the blades directly affects the safe operation of the unit.

[0003] Currently, the monitoring of blade damage in wind turbines mainly relies on manual inspections, drone image recognition, and indirect diagnosis based on operational data. However, these methods have limitations in detecting early damage and in locating the specific spatial position of the damage.

[0004] Therefore, there is an urgent need for an intelligent analysis system for wind turbine blade vibration signals that can spatially locate and predict the evolution of wind turbine blade damage. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent analysis system for vibration signals of wind turbine blades, which solves the problems of existing technologies that make it difficult to detect early damage to blades and locate the specific spatial position of the damage.

[0006] This invention provides an intelligent analysis system for vibration signals of wind turbine blades, employing the following technologies: An intelligent analysis system for vibration signals of wind turbine blades includes: The signal acquisition module is used to collect the vibration time sequence of multiple measuring points on the blade and identify the rotation period to generate the raw dataset; The matrix generation module is used to divide the original dataset into single-period signal segments, and resample and stack them at equal angular intervals to generate a periodic angular two-dimensional signal matrix. The space-time joint module is used to perform coherent accumulation on the periodic angle two-dimensional signal matrix and invert the damage spatial distribution to generate a two-dimensional damage spatial distribution map of the blade surface. The damage locking module is used to perform peak detection and centroid fitting on the two-dimensional damage spatial distribution map to obtain damage spatial coordinates and generate a damage spatial trajectory sequence. The damage evolution module is used to perform curve fitting of the equivalent area of ​​damage in the damage spatial trajectory sequence under the constraints of the fracture mechanics model, and output the damage development prediction result. The adaptive operating condition module is used to calculate the aerodynamic load influence coefficient based on real-time operating condition data in order to eliminate the pseudo-damage area caused by changes in operating conditions. The damage grading module is used to compare the adaptive damage equivalent area with a preset threshold range and output a damage level label.

[0007] Optionally, the signal acquisition module includes a multi-point synchronous acquisition unit, a rotation period identification unit, and a data aggregation unit, wherein: The multi-point synchronous acquisition unit is used to synchronously read the original vibration time sequence of each blade's spanwise and chordwise acquisition points. The rotation cycle identification unit is used to extract the rotation cycle from the original vibration time sequence and identify the start and end times of each rotation cycle to output a vibration time sequence segment with a cycle number. The data aggregation unit is used to bind the original vibration time series with the period number, blade number and measuring point coordinates in the vibration time series segment with period number to generate the original dataset.

[0008] Optionally, the rotation period identification unit selects the signal of the leaf root measuring point in the original vibration time sequence as the reference measuring point signal, performs autocorrelation analysis on the reference measuring point signal, and detects the time delay corresponding to the first peak position of the autocorrelation function as the rotation period; Using the first peak position of the autocorrelation function as the period segmentation point, the original vibration time sequence is divided into single-period signal segments, and the period number is incremented in time sequence. By comparing the start time of each cycle with the timestamp of the original vibration time sequence, the vibration data segment within each rotation cycle is extracted, and the vibration time sequence segment with cycle number is output.

[0009] Optionally, the matrix generation module includes a periodic slicing unit, an angle resampling unit, and a periodic stacking unit, wherein: The periodic slicing unit is used to extract the corresponding single-cycle signal segment from the original dataset using the period number output by the rotating period identifier unit as an index. The angle resampling unit is used to resample the single-cycle signal segment at equal angular intervals to generate an angle-aligned single-cycle signal. The periodic stacking unit is used to stack multiple angle-aligned single-cycle signals according to their cycle numbers to generate a two-dimensional periodic angle signal matrix.

[0010] Optionally, the angle resampling unit performs cross-correlation analysis on the signals of each measuring point and the reference measuring point in the single-cycle signal segment to obtain the phase shift difference between each measuring point and the reference measuring point. Based on the phase offset difference, the timing waveforms of each measurement point are shifted, and the timing zeros of all measurement points are unified to obtain a phase-aligned single-cycle signal. The duration of the phase-aligned single-cycle signal is divided into a preset number of angular intervals, and linear interpolation is performed in each angular interval to generate equiangular interval sampling values. Arrange the angularly spaced sampled values ​​of each angular interval in angular order to generate an angularly aligned single-cycle signal.

[0011] Optionally, the space-time joint module includes a transfer function estimation unit, a coherent accumulation unit, and a spatial inversion unit, wherein: The transfer function estimation unit is used to calculate the transfer function matrix of the vibration response from each spatial position on the blade surface to each measuring point based on the blade structure dynamics model. The coherent accumulation unit is used to perform phase coherent accumulation on a two-dimensional signal matrix of periodic angles for multiple periods; The spatial inversion unit is used to invert the spatial distribution of blade surface damage from the coherently accumulated measurement point response to generate a two-dimensional damage spatial distribution map.

[0012] Optionally, the spatial inversion unit separates the periodic angle two-dimensional signal matrix according to the measurement points to obtain a two-dimensional array of angle responses for each measurement point; Perform a discrete Fourier transform on the two-dimensional array of angle responses at each measurement point to extract the frequency domain components corresponding to the fundamental frequency and harmonics of the rotation. The frequency domain components extracted from each measurement point are arranged according to the frequency measurement points to construct the frequency domain measurement point response matrix; The least squares inversion is performed on the frequency domain measurement point response matrix to obtain the damage response amplitude at each spatial location on the blade surface; The damage response amplitudes are arranged according to spanwise and chordwise coordinates to generate a two-dimensional damage spatial distribution map of the blade surface.

[0013] Optionally, the damage locking module includes a damage area identification unit, a positioning unit, and a periodic locking unit, wherein: The damage area identification unit is used to perform peak detection on the two-dimensional damage spatial distribution map and identify the location of the damage center and the boundary of the damage area. The positioning unit is used to perform centroid fitting on the location of the damage center and calculate the damage spatial coordinates. The periodic locking unit is used to utilize the periodicity of blade rotation to align the damage space coordinates of different rotation periods to the same blade reference frame, thereby generating a damage space trajectory sequence.

[0014] Optionally, the positioning unit takes the damage center pixel output by the damage area identification unit as the center and takes the pixel value within the neighborhood window; Calculate the first and second moments of the pixel values ​​within the neighborhood window, and calculate the coordinate offset of the centroid of the damage response. Add the coordinate offset to the pixel coordinates of the damage center to obtain the damage space coordinates; For the damage spatial coordinates obtained from different rotation cycles, the difference between the coordinates of adjacent cycles is calculated. When the difference exceeds the preset drift threshold, it is determined to be a false peak and is removed.

[0015] Optionally, the damage evolution module includes a damage area calculation unit, an evolution curve fitting unit, and an evolution prediction unit, wherein: The damage area calculation unit is used to integrate the pixel values ​​within the boundary of the damage area to calculate the equivalent damage area. The evolution curve fitting unit is used to perform curve fitting of the equivalent damage area as a function of the number of rotation cycles, constrained by the fracture mechanics model. The evolution prediction unit is used to predict the equivalent area of ​​damage for a preset number of cycles in the future based on the fitted curve, and output the damage development prediction result.

[0016] The advantages of this invention are as follows: The intelligent analysis system for vibration signals of wind turbine blades proposed in this embodiment of the invention, through the coupling of coherent accumulation and transfer function inversion, inverts the spatial distribution of damage on the blade surface from the response at the measurement point, solving the limitation of existing technologies that can only judge the overall abnormality of the blade but cannot locate the specific spatial location of the damage; by superimposing peak detection and centroid fitting, the spatial coordinates of the damage are obtained and a trajectory sequence is generated, solving the problems of damage location drift and false peak interference; through curve fitting and extrapolation constrained by the fracture mechanics model, damage evolution prediction based on physical laws is realized; by combining aerodynamic load compensation and adaptive adjustment of operating conditions, the false damage area caused by changes in operating conditions is eliminated, solving the problem of high damage misjudgment rate under variable operating conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall module architecture of an intelligent analysis system for vibration signals of wind turbine blades provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the operation of a rotation period marking unit according to an embodiment of the present invention. Figure 3 This is a flowchart of the spatial-temporal joint module provided in an embodiment of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The main solution of this invention is: An intelligent analysis system for vibration signals of wind turbine blades includes: The signal acquisition module is used to collect the vibration time sequence of multiple measuring points on the blade and identify the rotation period to generate the raw dataset; The matrix generation module is used to divide the original dataset into single-period signal segments, and resample and stack them at equal angular intervals to generate a periodic angular two-dimensional signal matrix. The space-time joint module is used to perform coherent accumulation on the periodic angle two-dimensional signal matrix and invert the damage spatial distribution to generate a two-dimensional damage spatial distribution map of the blade surface. The damage locking module is used to perform peak detection and centroid fitting on the two-dimensional damage spatial distribution map to obtain the damage spatial coordinates and generate a damage spatial trajectory sequence. The damage evolution module is used to perform curve fitting of the equivalent area of ​​damage in the damage spatial trajectory sequence under the constraints of the fracture mechanics model, and output the damage development prediction results. The adaptive operating condition module is used to calculate the aerodynamic load influence coefficient based on real-time operating condition data in order to eliminate the pseudo-damage area caused by changes in operating conditions. The damage grading module is used to compare the adaptive damage equivalent area with a preset threshold range and output a damage level label.

[0020] This invention provides an intelligent analysis system for vibration signals of wind turbine blades. A space-time joint module performs coherent accumulation on a two-dimensional periodic angle signal matrix and uses the blade structure transfer function to invert the spatial distribution of damage, generating a two-dimensional damage spatial distribution map of the blade surface, providing a spatial domain image basis for damage localization. A damage locking module introduces a sub-pixel centroid fitting and period locking verification mechanism to solve the problems of damage location drift and false peak interference, obtaining damage spatial coordinates and generating a trajectory sequence. A damage evolution module uses curve fitting constrained by a fracture mechanics model to address the lack of physical interpretability. An adaptive operating condition module calculates the aerodynamic load influence coefficient based on real-time operating condition data, eliminating false damage areas caused by changes in operating conditions and solving the problem of high damage misjudgment rate under variable operating conditions.

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can understand it.

[0022] refer to Figure 1 An embodiment of the present invention provides an intelligent analysis system for vibration signals of wind turbine blades, including a signal acquisition module, a matrix generation module, a space-time joint module, a damage locking module, a damage evolution module, an operating condition adaptive module, and a damage grading determination module.

[0023] The signal acquisition module is used to collect the vibration time sequence and identification rotation period of multiple measurement points on the blade and generate the raw dataset.

[0024] The signal acquisition module includes a multi-point synchronous acquisition unit, a rotation period identification unit, and a data aggregation unit. The multi-point synchronous acquisition unit is used to synchronously read the original vibration time sequence of each blade's spanwise and chordally arranged acquisition points. The rotation period identification unit is used to extract the rotation period from the original vibration time sequence and identify the start and end times of each rotation period to output a vibration time sequence segment with a period number. The data aggregation unit is used to bind the original vibration time sequence with the period number, blade number, and measurement point coordinates in the vibration time sequence segment with a period number to generate the original dataset.

[0025] In the specific implementation process, during the sensor installation phase, the multi-point synchronous acquisition unit measures the three-dimensional spatial coordinates of each sensor mounting base using a total station, and then transforms them into the local coordinate system coordinates of the blade with the blade root as the origin, the span as the Y-axis, and the chord as the X-axis.

[0026] An array of accelerometer sensors is arranged on the blade surface. The spanwise direction is from the blade root to the blade tip, and the chordwise direction is from the leading edge to the trailing edge. Four cross-sections (0.3L, 0.5L, 0.7L, and 0.9L) are selected for the spanwise sensor placement, where L is the blade length. Three locations are selected for the chordwise sensor placement: the leading edge, the middle, and the trailing edge. The four spanwise cross-sections and the three chordwise locations are interleaved, forming 12 measuring points. A PCB Piezotronics 356A15 ICP accelerometer is used, with a sensitivity of 100mV / g, a range of ±50g, a frequency range of 0.5Hz to 10kHz, and a resonant frequency ≥35kHz. Each accelerometer is connected to an NI PXIe-4499 synchronous dynamic signal acquisition card via a low-noise coaxial shielded cable. An RG58 coaxial cable with a characteristic impedance of 50Ω and a shielding coverage of ≥95% is used to effectively suppress interference from the strong electromagnetic environment inside the wind turbine tower on weak vibration signals. The NI PXIe-4499 synchronous dynamic signal acquisition card features 16 channels of synchronous sampling capability, with 24-bit resolution per channel and a maximum sampling rate of 204.8 kS / s. The 24-bit resolution corresponds to a quantization step of 1.19 μV at a range of ±10V, meeting the acquisition requirements of blade vibration signals. The acquisition card is inserted into the NI PXIe-1082 chassis, which is equipped with a PXI Express backplane. The backplane integrates a 100MHz differential clock bus and multiple synchronous trigger lines. The multi-point synchronous acquisition unit inserts multiple PXIe-4499 acquisition cards into the same chassis, achieving strict synchronization across card channels through the PXI backplane clock sharing mechanism, with a sampling delay of less than 1 ns between channels.

[0027] The specific working process of the multi-point synchronous acquisition unit is as follows: the multi-point synchronous acquisition unit configures the sampling parameters of each acquisition card; each acquisition card generates a local sampling clock based on the 10MHz reference clock of the PXI backplane; each channel starts analog-to-digital conversion on the same clock edge of the local sampling clock to achieve strict synchronous acquisition of the original vibration timing of multiple measurement points; the acquired original vibration timing data is transmitted to the chassis controller via the PXI Express bus, and the chassis controller aligns the data of each channel according to the sampling timestamp to generate a synchronized original vibration timing of multiple measurement points.

[0028] In the specific implementation process, refer to Figure 2 The rotation cycle marking unit selects the 0.3L spanwise position as the reference measuring point. This measuring point is close to the connection area between the blade and the hub and is least affected by the flexible deformation of the blade.

[0029] The rotation period identification unit performs autocorrelation analysis on the reference measurement point signal, and uses the time delay corresponding to the first peak position of the autocorrelation function as the rotation period. The autocorrelation function is a mathematical tool for calculating the similarity between a signal and its time-shifted version. Its principle is as follows: the signal is shifted by different time delays, and the sum of the products of corresponding points of the shifted signal and the original signal is calculated. The autocorrelation function of a periodic signal shows a peak at integer multiples of the period. The formula is: in, For reference measurement point signal; This represents the time shift.

[0030] The rotation period identifier searches for the first significant peak of the autocorrelation function within the range τ>0. The time delay T corresponding to the position of this peak is the rotation period. At the rated speed of 15 rpm, T=4s; at the cut-in speed of 3 rpm, T=20s.

[0031] Using the peak position of the autocorrelation function as the period segmentation point, the original continuous vibration time series is divided into single-cycle signal segments. Each single-cycle signal segment starts at the peak position and ends at the next peak position. The cycle numbers are incremented sequentially by time, with the first detected cycle marked as cycle 1, subsequent cycles as cycle 2, cycle 3, and so on, up to the Nth cycle. By comparing the start time of each cycle with the timestamp of the original vibration time series, vibration data segments within each rotation cycle are extracted, and vibration time series segments with cycle numbers are output.

[0032] In the specific implementation process, the data aggregation unit receives the original vibration time series from each measuring point output by the multi-measuring-point synchronous acquisition unit and the vibration time series segment with period number output by the rotation period identification unit. It then binds the original vibration time series with the period number and blade number in the period-numbered vibration time series segment to the measuring point coordinates. The data aggregation unit stores data in HDF5 file format, with the blade number as the top-level group, the measuring point number as the subgroup, and the period number as the dataset name. Each dataset contains a timestamp and vibration acceleration value, generating the original dataset.

[0033] The matrix generation module is used to divide the original dataset into single-period signal segments, and resample and stack them at equal angular intervals to generate a periodic angular two-dimensional signal matrix. The matrix generation module includes a periodic slicing unit, an angle resampling unit, and a periodic stacking unit. The periodic slicing unit is used to extract the corresponding single-period signal segments from the original dataset using the period number output by the rotation period identifier unit as an index. The angle resampling unit is used to resample the single-period signal segments at equal angular intervals to generate angle-aligned single-period signals. The periodic stacking unit is used to stack multiple angle-aligned single-period signals according to their period numbers to generate a two-dimensional periodic angle signal matrix.

[0034] In practice, the periodic slicing unit locates the corresponding dataset in the HDF5 file hierarchy based on the period number, reads the vibration data of all measuring points under that period number, and forms a single-period signal segment. The single-period signal segment contains time-series data from 12 measuring points, and the duration of the time-series data from each measuring point is one rotation cycle.

[0035] In practice, the angle resampling unit performs cross-correlation analysis on the signals at each measurement point and the reference measurement point within a single-cycle signal segment to obtain the phase shift difference between each measurement point and the reference measurement point. The cross-correlation function is a method for measuring the similarity between two signals. Its principle is as follows: one of the two signals is time-shifted, and the sum of the products of corresponding points of the two signals after the shift is calculated. The peak position corresponds to the optimal time alignment point of the two signals. The formula is: in, For reference measurement point signal; The signal of the measurement point to be aligned; This represents the time shift.

[0036] The time delay corresponding to the peak value of the cross-correlation function is the phase offset difference between the measurement point and the reference measurement point. By shifting the timing waveforms of each measurement point according to the phase offset difference, the timing zeros of all measurement points are unified to obtain a phase-aligned single-cycle signal.

[0037] The duration of the phase-aligned single-cycle signal is divided into a preset number of angular intervals, of which 512 are defined. Within each angular interval, linear interpolation is performed on the phase-aligned single-cycle signal to generate equally spaced sampled values. Linear interpolation is a numerical analysis method whose principle is: given the values ​​of a function at two points, assuming the function changes linearly between the two points, the function value at any point within the interval is estimated through a proportional relationship. The linear interpolation formula is: in, From the perspective of the target; , For adjacent original sampling angles; , These are the corresponding sampled values. Original sampling angle. Calculated from sampling timestamps and rotational speed: in, This represents the offset of the sampling time relative to the start time of the period; The duration of the cycle.

[0038] The angularly spaced sampled values ​​from each angular interval are arranged in angular order to generate an angle-aligned single-cycle signal. The angle-aligned single-cycle signal contains 512 angularly spaced sampled points, with uniform angular resolution in each cycle, eliminating the difference in the number of sampled points within a cycle caused by speed fluctuations.

[0039] In the specific implementation process, the periodic stacking unit arranges the angle-aligned single-cycle signals of N consecutive cycles according to the measurement point number to generate a single-cycle multi-measurement point signal matrix. The single-cycle multi-measurement point signal matrix has a dimension of 512×12, where 512 is the number of angle intervals and 12 is the number of measurement points.

[0040] The periodic stacking unit stacks N consecutive single-cycle multi-measurement point signal matrices according to their cycle numbers, generating a periodic angle two-dimensional signal matrix. The periodic angle two-dimensional signal matrix has dimensions of N×512×12, where N is the number of cycles, 512 is the number of angle intervals, and 12 is the number of measurement points. N is set according to the signal-to-noise ratio requirement, typically ranging from 64 to 256 cycles. When N=64, the coherent cumulative signal-to-noise ratio gain is... When N=256, the gain is 24dB.

[0041] The space-time joint module is used to perform coherent accumulation on the periodic angle two-dimensional signal matrix and invert the damage spatial distribution to generate a two-dimensional damage spatial distribution map of the blade surface.

[0042] The space-time joint module includes a transfer function estimation unit, a coherent accumulation unit, and a spatial inversion unit. The transfer function estimation unit is used to calculate the transfer function matrix of the vibration response from each spatial position on the blade surface to each measuring point based on the blade structure dynamics model. The coherent accumulation unit is used to perform phase coherent accumulation on the two-dimensional signal matrix of the periodic angle of multiple periods. The spatial inversion unit is used to invert the spatial distribution of damage on the blade surface from the coherently accumulated measuring point response to generate a two-dimensional damage spatial distribution map.

[0043] In the specific implementation process, refer to Figure 3 A finite element model of the blade was established using the transfer function estimation unit, and ANSYS Mechanical APDL was selected for modeling. The finite element method is a numerical method for solving partial differential equations. Its principle is to discretize the continuous solution domain into a finite number of elements, assume an approximate function within each element, and solve a system of algebraic equations using variational principles. The blade finite element model uses the SHELL181 shell element, a four-node shell element with six degrees of freedom per node, suitable for analyzing thin to medium-thickness shell structures. The blade finite element model contains 100 spanwise elements and 50 chordal elements, totaling 5000 surface nodes.

[0044] Modal analysis was performed on the finite element model. Modal analysis is a method for solving the inherent vibration characteristics of a structure. Its principle is to solve the eigenvalue problem of the undamped free vibration equation of the structure to obtain the natural frequencies and mode shapes. The transfer function estimation element extracted the natural frequencies, mode shapes, and modal damping ratios of the first 20 modes. The first 20 modes cover the frequency range of 0Hz to 30Hz and include the main low-order bending and torsional modes of the blade.

[0045] Unit pulse excitation is applied to each node on the blade surface, and the vibration response at each measuring point is calculated. The transfer function matrix from each spatial position on the blade surface to the vibration response at each measuring point is then constructed. The transfer function matrix has dimensions of 12 × 5000 × F, where 12 represents the number of measurement points, 5000 represents the number of surface nodes, and F represents the number of frequency points corresponding to the frequency resolution. Frequency resolution. T is the period duration, and Δf = 0.25Hz at rated speed. F = 40 corresponds to the range of 0Hz to 10kHz.

[0046] In the specific implementation process, the coherent accumulation unit receives the periodic angle two-dimensional signal matrix output by the periodic stacking unit, performs phase coherent accumulation on the periodic angle two-dimensional signal matrix of multiple periods, and outputs the measurement point response after coherent accumulation.

[0047] Coherent accumulation is a signal processing method that utilizes the periodicity of a signal to enhance deterministic components and suppress random noise. Its principle is as follows: if a deterministic component exists in the signal that is synchronized with a reference phase, then superimposing it in phase over multiple periods will amplify the amplitude of this component by a factor of N, where N is the number of periods. Random noise, due to its random phase, is only amplified. The signal-to-noise ratio improvement factor is [number] times. The coherent cumulative formula is: in, For the first Periodic measurement points At angle The signal at the location; As a reference phase, the position of the period segmentation point determined by the rotation period identifier unit is used for calculation: in, This represents the harmonic order.

[0048] After coherent accumulation, the periodic modulation components related to the rotation angle are enhanced, thus improving the signal-to-noise ratio. With N=100 cycles accumulated, the signal-to-noise ratio is improved by 20dB.

[0049] In the specific implementation process, the spatial inversion unit separates the periodic angle two-dimensional signal matrix according to the measurement points, obtaining a two-dimensional array of angle responses for each measurement point. A Discrete Fourier Transform (DFT) is then performed on the two-dimensional array of angle responses for each measurement point to extract the frequency domain components corresponding to the fundamental frequency and harmonics of the rotation. The DFT is an algorithm that converts a discrete-time signal into a frequency domain representation. Its principle is to represent the signal as a linear combination of a series of complex exponential functions, where each exponent corresponds to a frequency component. The DFT formula is: in, For the first The time-domain signal value of each sampling point; For the first The complex amplitude of each frequency component; This represents the number of sampling points; It is a complex exponential basis function.

[0050] Rotating fundamental frequency The calculation formula is: ,in, This refers to the real-time speed of the fan. At the rated speed of 15 rpm... The harmonic is an integer multiple of the fundamental frequency, i.e. 3 , , .

[0051] The frequency domain components extracted from each measurement point are arranged according to the frequency of the measurement points to construct the frequency domain measurement point response matrix Y. The frequency domain measurement point response matrix Y has a dimension of F×12, where F is the number of frequency points and 12 is the number of measurement points. Least squares inversion is performed on the frequency domain measurement point response matrix Y. The least squares method is an optimal approximation method for solving overdetermined linear equation systems. Its principle is: when the equation system... When no exact solution is available, find a residual norm that satisfies the condition. The minimum solution x. For the damage inversion problem, the system of equations is: ,in, For the transfer function matrix, The damage response amplitude is to be determined. This is the response at the measurement point.

[0052] Due to the transfer function matrix Typically, these are ill-conditioned matrices, and direct inversion amplifies noise. Spatial inversion units are solved using Moore-Penrose pseudoinverses combined with Tikhonov regularization. The Moore-Penrose pseudoinverse is a generalized matrix inverse, its principle being: for a non-square or singular matrix A, the pseudoinverse A... + satisfy and The solution is calculated using singular value decomposition. Tikhonov regularization adds an L2 norm penalty term to the objective function to suppress oscillations in the solution. The regularized solution formula is: in, for The conjugate transpose of; For regularization parameters, The typical value is 0.01; It is the identity matrix; For the transfer function matrix, The damage response amplitude is to be determined. This is the response at the measurement point.

[0053] Damage response amplitude A two-dimensional damage spatial distribution map is generated by arranging the data in spanwise and chordwise coordinates. The horizontal axis of the two-dimensional damage spatial distribution map represents chordwise coordinates, and the vertical axis represents spanwise coordinates. Pixel values ​​are the normalized intensity values ​​of the damage response amplitude. The image resolution is 50×100 pixels, and each pixel corresponds to a spatial region on the blade surface.

[0054] The damage locking module is used to perform peak detection and centroid fitting on the two-dimensional damage spatial distribution map to obtain the damage spatial coordinates and generate a damage spatial trajectory sequence.

[0055] The damage locking module includes a damage area identification unit, a localization unit, and a period locking unit. The damage area identification unit is used to perform peak detection on the two-dimensional damage spatial distribution map and identify the damage center location and the damage area boundary. The localization unit is used to perform centroid fitting on the damage center location and calculate the damage spatial coordinates. The period locking unit is used to utilize the periodicity of blade rotation to align the damage spatial coordinates of different rotation periods to the same blade reference frame and generate a damage spatial trajectory sequence.

[0056] In the specific implementation process, the damage area identification unit performs Gaussian smoothing on the two-dimensional damage spatial distribution map. Gaussian smoothing is an image filtering method. Its principle is as follows: a two-dimensional Gaussian function is used as the convolution kernel to perform a weighted average on the image. The weights decrease exponentially with the distance from the center pixel, effectively suppressing false peaks caused by high-frequency noise. The smoothing kernel size is 3×3 pixels, and the standard deviation σ=1 pixel.

[0057] Then, the 8-neighbor connected component labeling algorithm is used to identify local maxima. The 8-neighbor connected component labeling algorithm is an image segmentation method. Its principle is: for each pixel, check its 8 neighboring pixels. If the gray values ​​of neighboring pixels are similar, they are classified into the same connected region. Region labeling is achieved through recursion or disjoint-set data structure.

[0058] Set peak threshold The threshold is set to 30% of the global maximum response amplitude, retaining only peak points exceeding this threshold. For each retained peak point, the damage region identification unit expands along the gradient descent direction using a region growing algorithm until the response amplitude drops to 50% of the peak value, forming the damage region boundary. The region growing algorithm is an image segmentation method. Its principle is as follows: starting from the peak point, it checks whether neighboring pixels meet the growth criterion (response amplitude exceeding 50% of the peak value). Pixels meeting the criterion are merged into the region, and this process is iteratively expanded until no new pixels are added. The damage region identification unit outputs the damage center location and the damage region boundary.

[0059] In the specific implementation, the localization unit uses the damage center location output by the damage area recognition unit as the center pixel and takes a 5×5 pixel neighborhood window. The first moment of the pixel values ​​within the neighborhood window is calculated. , and second moment , , : in, pixels within the neighborhood window The response amplitude. For column indexes Power of; For row indexes Powers. Moments are mathematical quantities that describe the geometric features of an image region. The first moment describes the centroid of the region, and the second moment describes the shape of the region.

[0060] Calculate the coordinate offset of the centroid of the damage response using the moment method. : in, The pixel coordinates of the damage center location; It is the zeroth moment; , This is the first moment. The damage space coordinates are obtained by adding the coordinate offset to the pixel coordinates of the damage center. .

[0061] For the damage spatial coordinates obtained from different rotation cycles, calculate the coordinate difference between adjacent cycles. : when Exceeding the preset drift threshold If the peak value is identified as a false damage point, it will be removed. A preset drift threshold is used. The typical value is 3 pixels, corresponding to 6m spanwise or 0.24m chordwise on the blade surface. Output the damage space coordinates.

[0062] In practice, the blade reference frame is based on the blade root as the origin, with the spanwise direction as the Y-axis and the chordwise direction as the X-axis. Since the damage is fixed on the blade, the coordinates of its blade reference frame should remain unchanged in adjacent cycles.

[0063] The periodic locking unit arranges the multi-period damage spatial coordinates output by the positioning unit according to the period number, forming a damage spatial trajectory sequence, and outputs the damage spatial trajectory sequence. Each element in the damage spatial trajectory sequence contains the period number, damage spatial coordinates, and damage response amplitude.

[0064] The damage evolution module is used to perform curve fitting of the equivalent area of ​​damage in the damage spatial trajectory sequence under the constraints of the fracture mechanics model, and output the damage development prediction results. The damage evolution module includes a damage area calculation unit, an evolution curve fitting unit, and an evolution prediction unit. The damage area calculation unit is used to integrate the pixel values ​​within the boundary of the damage region to calculate the equivalent damage area. The evolution curve fitting unit is used to perform curve fitting with the fracture mechanics model constraint on the change of the equivalent damage area with the number of rotation cycles. The evolution prediction unit is used to predict the equivalent damage area for a preset number of cycles in the future based on the fitted curve and output the damage development prediction result.

[0065] In the specific implementation process, the damage area calculation unit calculates the equivalent damage area using the following formula: in, The area equivalent to the damage; The set of pixels within the boundary of the damaged area; For pixels Response amplitude weights; This represents the blade surface area corresponding to a single pixel. The damage equivalent area comprehensively reflects the geometry and response intensity of the damaged region.

[0066] In the specific implementation process, the evolution curve fitting unit receives the damage equivalent area sequence for M consecutive rotation cycles. and the corresponding cumulative number of rotation cycles Paris's law is used as the fracture mechanics constraint model. Paris's law is an empirical formula describing the relationship between fatigue crack propagation rate and stress intensity factor amplitude. Its principle is: under fatigue loading, the crack propagation rate da / dN is proportional to the power of the stress intensity factor amplitude ΔK. in, The crack size; This represents the number of load cycles, corresponding to the number of rotational cycles. , These are all material constants. For glass fiber reinforced epoxy resin composites for blades, typical values ​​are... , .

[0067] Integrating Paris's law yields the relationship between crack size and the number of cycles. The integration process involves assuming the stress intensity factor amplitude... ,in Let the stress amplitude be the value. Substitute this value into Paris's law and separate the variables to integrate: Solving for: in, This represents the initial crack size. Assume the equivalent damage area... With crack size The equation for the evolution of the equivalent area of ​​damage is proportional to the square of the number of π / 2. in, For comprehensive parameters; This is the area shape coefficient.

[0068] Constrained by the evolution equation of the equivalent damage area, nonlinear least squares fitting is performed on the equivalent damage area sequence to optimize the parameters. , and The fitting objective function is: in, It is the area shape factor; For comprehensive parameters; Initial crack size The Levenberg-Marquardt algorithm is used for iterative optimization. The Levenberg-Marquardt algorithm is a nonlinear least squares optimization algorithm. Its principle is to combine gradient descent and Gauss-Newton's method. When the error is large, gradient descent is used to ensure convergence; when the error is small, Gauss-Newton's method is used to accelerate convergence. A damping factor is then used to further optimize the algorithm. The weights of the two methods are dynamically adjusted. The convergence condition is that the rate of change of the parameters is less than... .

[0069] In the specific implementation process, the evolution prediction unit will preset the number of future cycles. Substitute the fitted curve output by the evolution curve fitting unit to predict the future equivalent area of ​​damage. : in, It is the area shape factor; For comprehensive parameters; This represents the initial crack size. This represents the number of rotation cycles that have been completed so far. Pre-determine the number of cycles for the future; is a material constant.

[0070] Simultaneously, the predicted confidence intervals are calculated, and the Bootstrap resampling method is used to assess parameter uncertainty. Bootstrap is a statistical resampling method. Its principle is as follows: B sample subsets are randomly drawn from the original dataset with replacement, and Paris's law is re-applied to each subset to obtain B sets of parameter estimates. The 2.5% and 97.5% quantiles of the parameter estimates are used as the 95% confidence intervals.

[0071] When the predicted equivalent damage area exceeds the maximum damage area allowed by the blade structure At that time, the evolution prediction unit outputs a danger warning signal. According to the blade structure design specification IEC 61400-23, the maximum allowable delamination area at the blade root is 0.5% of the blade surface area. For an 80m blade with a 4m chord length, the surface area of ​​the blade root region is approximately 320m². The value is 1.6m² = 16000cm².

[0072] The evolution prediction unit outputs damage development prediction results, including the current equivalent area of ​​damage, the predicted equivalent area of ​​future damage, the number of cycles required to reach the danger threshold, and the 95% confidence interval.

[0073] The adaptive operating condition module is used to calculate the aerodynamic load influence coefficient based on real-time operating condition data in order to eliminate pseudo-damage areas caused by changes in operating conditions.

[0074] In practice, the adaptive operating condition module communicates with the wind turbine's main control system via the OPC UA protocol. OPC UA (OLE for Process Control Unified Architecture) is an industrial automation communication protocol. Its principle is as follows: it adopts a client-server architecture, achieving interoperability between devices from different manufacturers through a standardized data model, and supports real-time data subscription and event notification. The operating condition parameter reading unit reads real-time operating condition data once per second. The timestamps of the operating condition data and vibration data are aligned to ensure the temporal consistency of operating condition compensation. Timestamp alignment uses a nearest neighbor matching algorithm, comparing the timestamps of the operating condition data and vibration data, and selecting the operating condition data with the smallest time difference as the operating condition state at the corresponding moment.

[0075] The influence coefficient of aerodynamic load variation on the damage spatial distribution map is calculated based on real-time operating data, and the aerodynamic compensation coefficient matrix is ​​output. An aerodynamic load model for the blade is established, and the aerodynamic load distribution under various operating conditions is calculated using BEM (Blade Element Momentum) theory. BEM theory is a method for analyzing the aerodynamic performance of wind turbine blades. Its principle is as follows: the blade is divided into multiple blade elements along its spanwise direction; the momentum theorem and blade element theory are applied to each blade element, considering the influence of induced velocity, and the aerodynamic force and inflow angle of each blade element are iteratively solved.

[0076] Changes in aerodynamic loads cause variations in the overall vibration response of the blade. This variation manifests as a global shift in response amplitude in the two-dimensional damage spatial distribution map, rather than a peak value caused by localized damage. The aerodynamic load compensation unit calculates the aerodynamic load difference between the current operating condition and the baseline operating condition, generating an aerodynamic compensation coefficient matrix. : in, For exhibition location Aerodynamic forces at the location; These are the pixel coordinates of the two-dimensional damage spatial distribution map. The aerodynamic compensation coefficient reflects the proportion of aerodynamic load change under the current operating condition relative to the baseline operating condition. Under the baseline operating condition, V = 12 m / s and RPM = 15 rpm. .

[0077] The aerodynamic compensation coefficient matrix is ​​applied element-by-element to the two-dimensional damage spatial distribution map to eliminate spurious damage areas caused by changes in operating conditions, outputting the compensated two-dimensional damage spatial distribution map. The compensation formula is: in, A two-dimensional damage spatial distribution map output by the spatial inversion unit; This is a two-dimensional spatial distribution map of the damage after compensation. This is the aerodynamic compensation coefficient matrix. This compensation operation eliminates the global response shift caused by changes in operating conditions, while preserving the peak characteristics caused by local damage.

[0078] The damage grading module is used to compare the compensated equivalent area of ​​damage with a preset threshold range and output a damage level label.

[0079] In practice, the damage grading module establishes a four-level damage threshold range based on blade design specifications and historical damage statistics. The four-level threshold range is: Intact A e <5cm², minor injury 5cm²≤A e <20cm², moderate injury 20cm²≤A e <50cm², severe injury A e ≥50cm².

[0080] Using the nearest neighbor principle, the distance between the equivalent damage area and the center value of each interval is calculated. The equivalent damage area is assigned to the threshold interval with the smallest distance, and the corresponding damage level label is output. The damage level label is 0 intact, 1 mild, 2 moderate, and 3 severe.

[0081] Calculate the dispersion index of the damage spatial trajectory sequence, including the standard deviation of spatial location. and the coefficient of variation of response amplitude : in, The length of the trajectory sequence; For the first Damage spatial coordinates for each cycle; For the first Damage response amplitude per cycle; for The standard deviation of the damage response amplitude over each cycle.

[0082] when More than 5 pixels corresponds to a blade surface span of 10m or tangential length of 0.4m or... When the value exceeds 0.3, the damage grading module determines that the damage is false, resets the damage level marker to 0, and outputs a stability judgment alarm.

[0083] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0084] In the description of this invention, it should be noted that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0085] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0086] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0087] 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; that is, 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 according to actual needs.

[0088] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0089] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0090] Furthermore, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. An intelligent analysis system for vibration signals of wind turbine blades, characterized in that, include: The signal acquisition module is used to collect the vibration time sequence of multiple measuring points on the blade and identify the rotation period to generate the raw dataset; The matrix generation module is used to divide the original dataset into single-period signal segments, and resample and stack them at equal angular intervals to generate a periodic angular two-dimensional signal matrix. The space-time joint module is used to perform coherent accumulation on the periodic angle two-dimensional signal matrix and invert the damage spatial distribution to generate a two-dimensional damage spatial distribution map of the blade surface. The damage locking module is used to perform peak detection and centroid fitting on the two-dimensional damage spatial distribution map to obtain damage spatial coordinates and generate a damage spatial trajectory sequence. The damage evolution module is used to perform curve fitting of the equivalent area of ​​damage in the damage spatial trajectory sequence under the constraints of the fracture mechanics model, and output the damage development prediction result. The adaptive operating condition module is used to calculate the aerodynamic load influence coefficient based on real-time operating condition data in order to eliminate the pseudo-damage area caused by changes in operating conditions. The damage grading module is used to compare the adaptive damage equivalent area with a preset threshold range and output a damage level label.

2. The intelligent analysis system for vibration signals of wind turbine blades according to claim 1, characterized in that, The signal acquisition module includes a multi-point synchronous acquisition unit, a rotation period identification unit, and a data aggregation unit, wherein: The multi-point synchronous acquisition unit is used to synchronously read the original vibration time sequence of each blade's spanwise and chordwise acquisition points. The rotation cycle identification unit is used to extract the rotation cycle from the original vibration time sequence and identify the start and end times of each rotation cycle to output a vibration time sequence segment with a cycle number. The data aggregation unit is used to bind the original vibration time series with the period number, blade number and measuring point coordinates in the vibration time series segment with period number to generate the original dataset.

3. The intelligent analysis system for vibration signals of wind turbine blades according to claim 2, characterized in that, The rotation period identification unit selects the signal of the leaf root measuring point in the original vibration time sequence as the reference measuring point signal, performs autocorrelation analysis on the reference measuring point signal, and detects the time delay corresponding to the first peak position of the autocorrelation function as the rotation period. Using the first peak position of the autocorrelation function as the period segmentation point, the original vibration time sequence is divided into single-period signal segments, and the period number is incremented in time sequence. By comparing the start time of each cycle with the timestamp of the original vibration time sequence, the vibration data segment within each rotation cycle is extracted, and the vibration time sequence segment with cycle number is output.

4. The intelligent analysis system for vibration signals of wind turbine blades according to claim 3, characterized in that, The matrix generation module includes a periodic slicing unit, an angle resampling unit, and a periodic stacking unit, wherein: The periodic slicing unit is used to extract the corresponding single-cycle signal segment from the original dataset using the period number output by the rotating period identifier unit as an index. The angle resampling unit is used to resample the single-cycle signal segment at equal angular intervals to generate an angle-aligned single-cycle signal. The periodic stacking unit is used to stack multiple angle-aligned single-cycle signals according to their cycle numbers to generate a two-dimensional periodic angle signal matrix.

5. The intelligent analysis system for vibration signals of wind turbine blades according to claim 4, characterized in that, The angle resampling unit performs cross-correlation analysis on the signals of each measuring point and the reference measuring point in the single-cycle signal segment to obtain the phase offset difference of each measuring point relative to the reference measuring point. Based on the phase offset difference, the timing waveforms of each measurement point are shifted, and the timing zeros of all measurement points are unified to obtain a phase-aligned single-cycle signal. The duration of the phase-aligned single-cycle signal is divided into a preset number of angular intervals, and linear interpolation is performed in each angular interval to generate equiangular interval sampling values. Arrange the angularly spaced sampled values ​​of each angular interval in angular order to generate an angularly aligned single-cycle signal.

6. The intelligent analysis system for vibration signals of wind turbine blades according to claim 1, characterized in that, The space-time joint module includes a transfer function estimation unit, a coherent accumulation unit, and a spatial inversion unit, wherein: The transfer function estimation unit is used to calculate the transfer function matrix of the vibration response from each spatial position on the blade surface to each measuring point based on the blade structure dynamics model. The coherent accumulation unit is used to perform phase coherent accumulation on a two-dimensional signal matrix of periodic angles for multiple periods; The spatial inversion unit is used to invert the spatial distribution of blade surface damage from the coherently accumulated measurement point response to generate a two-dimensional damage spatial distribution map.

7. The intelligent analysis system for vibration signals of wind turbine blades according to claim 6, characterized in that, The spatial inversion unit separates the periodic angle two-dimensional signal matrix according to the measurement points to obtain a two-dimensional array of angle responses for each measurement point. Perform a discrete Fourier transform on the two-dimensional array of angle responses at each measurement point to extract the frequency domain components corresponding to the fundamental frequency and harmonics of the rotation. The frequency domain components extracted from each measurement point are arranged according to the frequency measurement points to construct the frequency domain measurement point response matrix; The least squares inversion is performed on the frequency domain measurement point response matrix to obtain the damage response amplitude at each spatial location on the blade surface; The damage response amplitudes are arranged according to spanwise and chordwise coordinates to generate a two-dimensional damage spatial distribution map of the blade surface.

8. The intelligent analysis system for vibration signals of wind turbine blades according to claim 1, characterized in that, The damage locking module includes a damage area identification unit, a positioning unit, and a periodic locking unit, wherein: The damage area identification unit is used to perform peak detection on the two-dimensional damage spatial distribution map and identify the location of the damage center and the boundary of the damage area. The positioning unit is used to perform centroid fitting on the location of the damage center and calculate the damage spatial coordinates. The periodic locking unit is used to utilize the periodicity of blade rotation to align the damage space coordinates of different rotation periods to the same blade reference frame, thereby generating a damage space trajectory sequence.

9. The intelligent analysis system for vibration signals of wind turbine blades according to claim 8, characterized in that, The localization unit takes the damage center pixel output by the damage region identification unit as the center and takes the pixel value within the neighborhood window. Calculate the first and second moments of the pixel values ​​within the neighborhood window, and calculate the coordinate offset of the centroid of the damage response. Add the coordinate offset to the pixel coordinates of the damage center to obtain the damage space coordinates; For the damage spatial coordinates obtained from different rotation cycles, the difference between the coordinates of adjacent cycles is calculated. When the difference exceeds the preset drift threshold, it is determined to be a false peak and is removed.

10. The intelligent analysis system for vibration signals of wind turbine blades according to claim 1, characterized in that, The damage evolution module includes a damage area calculation unit, an evolution curve fitting unit, and an evolution prediction unit, wherein: The damage area calculation unit is used to integrate the pixel values ​​within the boundary of the damage area to calculate the equivalent damage area. The evolution curve fitting unit is used to perform curve fitting of the equivalent damage area as a function of the number of rotation cycles, constrained by the fracture mechanics model. The evolution prediction unit is used to predict the equivalent area of ​​damage for a preset number of cycles in the future based on the fitted curve, and output the damage development prediction result.

Images