Adaptive order spectrum analysis method for wind turbine drive train fault diagnosis

Abstract

The application provides an adaptive order spectrum analysis method for wind generator transmission chain fault diagnosis, and relates to the field of rotating machinery state monitoring and fault diagnosis. The method comprises the following steps: step S1, data acquisition and preprocessing, obtaining vibration acceleration signals and rotation speed pulse signals; step S2, performing adaptive instantaneous phase calculation on the rotation speed pulse signals to obtain an accurate instantaneous phase curve; step S3, based on the accurate instantaneous phase curve, performing angular domain resampling on the vibration acceleration signals to obtain an angular domain stationary signal sequence; step S4, performing autocorrelation analysis on the angular domain stationary signal sequence to adaptively select and generate an optimal window function sequence; and step S5, using the optimal window function sequence to perform order spectrum calculation on the angular domain stationary signal sequence to output the order spectrum. The adaptive order spectrum analysis method for wind generator transmission chain fault diagnosis is used, and the accuracy and robustness of vibration signal order analysis are enhanced.

Description

Technical Field

[0001] This invention relates to the field of rotating machinery condition monitoring and fault diagnosis, and in particular to an adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain. Background Technology

[0002] Currently, with the widespread use of major equipment such as wind power generation and aero engines, their core transmission components (such as gearboxes and bearings) often operate at non-stationary speeds, resulting in vibration signals exhibiting strong non-stationary characteristics. This poses a significant challenge to spectral analysis based on traditional Fourier transform. To address this issue, order tracking technology has emerged and become the mainstream analytical method. Its core idea is to resample the non-stationary time-domain signal into a stationary angular-domain signal before performing spectral analysis.

[0003] However, existing order tracking methods still face significant technical bottlenecks. First, in the instantaneous phase estimation stage, conventional methods (such as phase-locked loops and pulse counting interpolation) are prone to cumulative errors or jitter when the rotational speed fluctuates drastically or the pulse signal contains noise, leading to inaccurate reference values ​​for angular domain resampling. Second, in the angular domain signal analysis stage, fixed window functions (such as the Hanning window) are commonly used for windowed Fourier transforms. However, fixed windows cannot adapt to different signal characteristics (such as strong periodic impulses and weak periodic signals), resulting in poor spectral leakage control. This leads to either decreased amplitude accuracy or insufficient frequency resolution, making it difficult to simultaneously extract weak fault features and accurately measure strong periodic components.

[0004] In particular, for wind turbine drivetrains, the gearboxes operate under complex variable speed and load conditions for extended periods, resulting in high background noise and extremely weak fault characteristic signals that are strongly modulated. When using traditional order tracking methods to analyze gearbox vibration signals, inaccurate phase estimation and severe spectral leakage often make it difficult to extract fault characteristic orders (such as meshing order and its sidebands), hindering accurate diagnosis of early faults.

[0005] Therefore, there is an urgent need for a high-precision, adaptive order spectrum analysis method that can adapt to the complex operating conditions of wind turbines in order to effectively extract the vibration signal characteristics of typical transmission chain faults. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain, in order to solve the problems of insufficient instantaneous phase estimation accuracy and fixed and singular selection of spectrum analysis window function in the existing technology, improve the accuracy and resolution of order spectrum calculation, and enhance the accuracy, robustness and reliability of vibration signal order analysis.

[0007] To achieve the above objectives, this invention provides an adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrains, comprising the following steps: Step S1: Data acquisition and preprocessing to obtain vibration acceleration signals and rotational speed pulse signals; Step S2: Perform adaptive instantaneous phase calculation on the rotational speed pulse signal to obtain an accurate instantaneous phase curve; Step S3: Based on the accurate instantaneous phase curve, the vibration acceleration signal is resampled in the angular domain to obtain a stationary signal sequence in the angular domain; Step S4: Perform autocorrelation analysis on the stationary signal sequence in the diagonal domain, and adaptively select and generate the optimal window function sequence; Step S5: Calculate the order spectrum of the stationary signal sequence in the angular domain using the optimal window function sequence, and output the order spectrum.

[0008] Preferably, in step S2, an adaptive parameter phase gradient correction algorithm is used to perform adaptive instantaneous phase calculation, specifically including: Step S201: Based on the rotational speed pulse signal, calculate the initial instantaneous angular velocity estimate; Step S202: Construct an objective function with instantaneous angular velocity as the optimization variable, the specific expression of which is: ; in, This represents the total number of pulses in the rotational speed pulse signal. For the first The time corresponding to each rotational speed pulse signal In order to be in The estimated angular velocity at time t. for Estimation of the initial instantaneous angular velocity at time t. For regularization parameters; Based on instantaneous angular velocity Smoothness constraint terms; Step S203: Dynamically calculate the regularization parameters; Step S204: Solve the objective function using the regularization parameter to obtain the optimal instantaneous angular velocity curve; Step S205: Integrate the optimal instantaneous angular velocity curve to obtain the accurate instantaneous phase curve.

[0009] Preferably, in step S203, the regularization parameter Based on the dynamic adjustment of instantaneous angular acceleration variance and signal-to-noise ratio, the specific expression is as follows: ; in, These are the base values ​​for the regularization parameters. The variance of instantaneous angular acceleration. For signal-to-noise ratio, and These are preset positive weighting coefficients.

[0010] Preferably, the variance of instantaneous angular acceleration The calculation method is as follows: Discrete sampling is performed on the initial instantaneous angular velocity estimate to obtain a discrete angular velocity sequence. ,in, For the first Each sampling time, for The estimated angular velocity at that time; right The instantaneous angular acceleration sequence is obtained by performing first-order difference. The specific expression is: ; in, For the first The initial instantaneous angular velocity estimate at each sampling time. The sampling time interval; In length Within the sliding time window, calculate variance as The specific expression is: ; in, This represents the total number of sampling points within the sliding time window. Within the sliding time window The mean, For index variables, It is a discretized instantaneous angular acceleration sequence; And the length of the sliding time window Compared with the current average speed It is inversely proportional, and the specific expression is: ; in, It is a proportionality constant.

[0011] Preferred signal-to-noise ratio The calculation method is as follows: Frequency domain analysis was performed on the rotational speed pulse signal to extract its amplitude at the fundamental frequency. Average noise amplitude of adjacent frequency bands .

[0012] Preferably, the precise instantaneous phase curve in step S205 The specific expression is: ; in, For the initial phase, for The optimal instantaneous angular velocity at time t. For integration variables, For the current moment, This is the initial time.

[0013] Preferably, in step S3, the corner domain resampling is performed using a resampling algorithm based on local second-order polynomial fitting, specifically including: Step S301: For each target angle The precise target time is calculated by inversely based on the precise instantaneous phase curve. ; Step S302, at the precise moment of the target Select at least three time-domain sampling points in the vicinity; Step S303: Using time-domain sampling points, in the form of a second-order polynomial Perform local least squares fitting to obtain the fitting function. ,in, , , These are the fitting coefficients; Step S304: Precisely time the target Substitute into the fitting function Calculations yielded Interpolated estimate of the vibration signal at time interval ; Step S305: The interpolated estimated value From the perspective of the target The amplitude at that point is obtained as a sampling point in the stationary signal sequence in the angular domain. ; Through all target angles Angular domain stationary signal sequence is obtained. ,in For the first A uniform angle sampling point, This represents the total number of sampling points in the angular domain.

[0014] Preferably, in step S4, adaptively selecting and generating the optimal window function sequence specifically includes: Step S401: Calculate the stationary signal sequence in the angular domain. autocorrelation function The specific expression is: ; in, For angle delay variables; For angular delay The next Uniform angle sampling points A stationary signal sequence in the angular domain at that location; Step S402: Extract the main peak width of the autocorrelation function. and secondary peak significance ; Step S403: Based on the width of the main peak and secondary peak significance Choose the optimal window function; Step S404: Based on the type of the selected optimal window function, generate a stationary signal sequence in the angular domain. Equal-length optimal window function sequence .

[0015] Preferably, in step S403, when and When the signal is determined to be highly periodic, a flat-top window is selected; Otherwise, if the signal is determined to contain significant transient impacts or has weak periodicity, select the Nuttor window or the Blackman-Harris window; in, The threshold value for the main peak width is preset. This is the preset threshold for the significance of the secondary peak.

[0016] Preferably, step S5 specifically includes: Step S501: Optimal window function sequence With angular domain stationary signal sequences Multiply point by point to obtain the windowed angular domain signal. The specific expression is: ; Step S502: Analyze the windowed corner domain signal. Perform a discrete Fourier transform and calculate its order spectrum. The specific expression is: ; ; in, For the first Each order For order resolution, The imaginary unit; Step S503: Calculate the amplitude spectrum of the order spectrum. The order spectrum, as the final output, is specifically expressed as follows: .

[0017] Therefore, the present invention employs the aforementioned adaptive order spectrum analysis method for wind turbine drivetrain fault diagnosis, and the beneficial technical effects are as follows: (1) By adopting adaptive instantaneous phase calculation and dynamically calculating regularization parameters, this invention can effectively overcome the influence of drastic speed fluctuations and pulse signal noise, thereby obtaining a smooth and accurate instantaneous phase curve, laying a high-precision time-angle reference for subsequent angular domain resampling.

[0018] (2) Based on the accurate instantaneous phase curve, the present invention uses local second-order polynomial fitting for angular domain resampling. Compared with the traditional linear interpolation method, it can more accurately restore the real change trend of vibration signal between non-uniform time domain sampling points, thereby obtaining a higher quality angular domain stationary signal sequence and reducing the amplitude distortion and phase error introduced by resampling.

[0019] (3) By performing autocorrelation analysis on the stationary signal sequence in the angular domain and adaptively selecting the optimal window function (such as flat-top window, Naughton window or Blackman-Harris window) based on the main peak width and secondary peak significance, the present invention can intelligently match appropriate spectrum analysis parameters according to the periodic or transient characteristics of the signal itself, thereby achieving the optimal balance between suppressing spectrum leakage, improving amplitude measurement accuracy and enhancing frequency resolution.

[0020] (4) By integrating the above-mentioned high-precision instantaneous phase estimation, high-quality angular domain resampling and adaptive window function selection, the final output order spectrum is improved in terms of amplitude accuracy, main peak sharpness (3dB bandwidth) and weak component detection capability (sideband separation). The verification results show that it has superior extraction and identification performance for rotating machinery fault features under variable speed conditions. It is particularly suitable for scenarios with high background noise and complex operating conditions, such as wind turbine drive chains. It can effectively extract the weak feature order of typical faults such as gear peeling and pitting, providing a reliable basis for predictive maintenance. Attached Figure Description

[0021] Figure 1 This is a flowchart of the adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to the present invention. Figure 2 Flowchart for adaptive instantaneous phase calculation; Figure 3 Here is a flowchart of the corner domain resampling process; Figure 4 Flowchart for adaptive selection and generation of optimal window function sequences; Figure 5 This is a flowchart for order spectrum calculation. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0024] Example 1 like Figure 1 As shown, the adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain includes the following steps: Step S1: Data acquisition and preprocessing to obtain vibration acceleration signal and rotational speed pulse signal.

[0025] The original vibration acceleration signal and original speed pulse signal of the wind turbine gearbox are acquired synchronously using a vibration acceleration sensor and a speed pulse sensor. The original vibration acceleration signal is preprocessed by DC removal and bandpass filtering, and the original speed pulse signal is shaped and thresholded to obtain the final vibration acceleration signal and speed pulse signal.

[0026] Step S2: Apply an adaptive parameter phase gradient correction algorithm to the rotational speed pulse signal to perform adaptive instantaneous phase calculation, obtaining an accurate instantaneous phase curve. For example... Figure 2 As shown, it specifically includes: Step S201: Based on the rotational speed pulse signal, the initial instantaneous angular velocity estimate is calculated using the pulse counting method.

[0027] Detecting speed pulse signal Extract the rising edge time sequence of the rotational speed pulse. ,in, This represents the total number of pulses in the rotational speed pulse signal. For the first The time corresponding to each rotational speed pulse signal.

[0028] Within the time interval between adjacent speed pulses, the speed is assumed to be approximately constant, and a piecewise constant estimation is used, the specific expression of which is: ; in, The number of pulses per revolution for Estimation of the initial instantaneous angular velocity at time t. For the first The time corresponding to each rotational speed pulse signal.

[0029] Step S202: Construct an objective function with instantaneous angular velocity as the optimization variable, the specific expression of which is: ; in, In order to be in The estimated angular velocity at time t. For regularization parameters; Based on instantaneous angular velocity Smoothness constraint terms.

[0030] Step S203: Dynamically calculate the regularization parameters.

[0031] Regularization parameters Based on the dynamic adjustment of instantaneous angular acceleration variance and signal-to-noise ratio, the specific expression is as follows: ; in, These are the base values ​​for the regularization parameters. The variance of instantaneous angular acceleration. For signal-to-noise ratio, and These are preset positive weighting coefficients.

[0032] In practical applications, preset parameters can be optimized through offline simulation or experimental calibration. As a typical configuration scheme: the base value of the regularization parameter... Usually in to The range of values ​​is used as a baseline level to balance the fit and smoothness. Values ​​are typically taken in the range of 0.01 to 0.1, and are used to enhance smoothing constraints when the rotational speed changes rapidly. Values ​​are typically taken in the range of 0.1 to 1.0 to suppress oscillations in results when measurement noise is high. For large rotating machinery such as wind turbine gearboxes, an initial value can be set. , , The parameters are then fine-tuned based on actual speed fluctuations and signal quality. The above parameter ranges are preferred implementations calibrated according to typical wind turbine drivetrain operating conditions. In practical applications, adjustments can be made adaptively based on the rotational inertia, speed fluctuation range, and signal quality of the specific equipment.

[0033] instantaneous angular acceleration variance The calculation method is as follows: Discrete sampling is performed on the initial instantaneous angular velocity estimate to obtain a discrete angular velocity sequence. ,in, For the first Each sampling time, for The estimated angular velocity at that time.

[0034] right The instantaneous angular acceleration sequence is obtained by performing first-order difference. The specific expression is: ; in, For the first The initial instantaneous angular velocity estimate at each sampling time. This represents the sampling time interval.

[0035] In length Within the sliding time window, calculate variance as The specific expression is: ; in, This represents the total number of sampling points within the sliding time window. Within the sliding time window The mean, For index variables, It is a discretized instantaneous angular acceleration sequence.

[0036] And the length of the sliding time window Compared with the current average speed It is inversely proportional, and the specific expression is: ; in, It is a proportionality constant.

[0037] proportionality constant The length of the sliding time window is determined by the moment of inertia of the rotating component and the time scale of the dynamic process of interest. For most industrial rotating machinery, the length of the sliding time window should cover at least one to three rotational cycles to effectively smooth random fluctuations and preserve true dynamics. Therefore, Typically set to Between radians, that is, the phase angle corresponding to 1 to 3 revolutions. For example, it can be set (Corresponding to 2 revolutions). Current average speed. It can be calculated in real time from the mean angular velocity estimate within the current sliding window.

[0038] In addition, frequency domain analysis was performed on the rotational speed pulse signal to extract its amplitude at the fundamental frequency. Average noise amplitude of adjacent frequency bands The signal-to-noise ratio was calculated. The specific expression is: .

[0039] Step S204: Solve the objective function using the regularization parameter to obtain the optimal instantaneous angular velocity curve. The specific solution process is as follows: The instantaneous angular velocity to be determined Represented as a linear combination of cubic B-spline basis functions, the specific expression is: in, For cubic B-spline basis functions, For the first The spline coefficients to be determined The number of control points (usually taken as...) ), It is a node vector.

[0040] The objective function Discretize into matrix form, the specific expression is: ; in, For the spline coefficient vector, For the initial estimation vector, To design the matrix, It is a second-order difference matrix.

[0041] The above problem is a linear least squares problem with a quadratic regularization term. It has a closed analytical solution, and the optimal spline coefficients can be obtained by solving it. The specific expression is: ; The optimal spline coefficients Substituting into the formula for the linear combination of cubic B-spline basis functions, we obtain the continuous optimal instantaneous angular velocity curve. The specific expression is: ; in, for The Each component.

[0042] Step S205: Integrate the optimal instantaneous angular velocity curve to obtain the accurate instantaneous phase curve. The specific expression is: ; in, For the initial phase, for The optimal instantaneous angular velocity at time t. For integration variables, For the current moment, This is the initial time.

[0043] Step S3: Based on the accurate instantaneous phase curve, the vibration acceleration signal is resampled in the angular domain using a resampling algorithm based on local second-order polynomial fitting to obtain a stationary signal sequence in the angular domain. For example... Figure 3 As shown, it specifically includes: Step S301: For each target angle The precise target time is calculated by inversely based on the precise instantaneous phase curve. .

[0044] Precise instantaneous phase curve Discretize into sequences .in, For the first A discrete phase point, For the first Each discrete phase value.

[0045] For each Find satisfaction of Linear interpolation is performed to obtain the precise time of the target. The specific expression is: ; in, For the first A discrete phase point, For the first Each discrete phase value.

[0046] Step S302, at the precise moment of the target Select at least three time-domain sampling points in the vicinity.

[0047] The specific selection method is as follows: Based on... Take as the center, and take forward and backward. sampling points ( ), ensuring the total number of sampling points For example, it is advisable to take Then in Two points are taken at the beginning and two at the end, for a total of five time-domain sampling points for fitting, in order to improve interpolation accuracy.

[0048] Step S303: Using time-domain sampling points, in the form of a second-order polynomial Perform local least squares fitting to obtain the fitting function. ,in, , , represents the fitting coefficient.

[0049] Step S304: Precisely time the target Substitute into the fitting function Calculations yielded Interpolated estimate of the vibration signal at time interval .

[0050] Step S305: The interpolated estimated value From the perspective of the target The amplitude at that point is obtained as a sampling point in the stationary signal sequence in the angular domain. ; Through all target angles Angular domain stationary signal sequence is obtained. ,in For the first A uniform angle sampling point, This represents the total number of sampling points in the angular domain.

[0051] Step S4: Perform autocorrelation analysis on the stationary signal sequence in the diagonal domain, adaptively select and generate the optimal window function sequence. For example... Figure 4 As shown, it specifically includes: Step S401: Calculate the stationary signal sequence in the angular domain. autocorrelation function The specific expression is: ; in, For angle delay variables; For angular delay The next Uniform angle sampling points A stationary signal sequence in the angular domain.

[0052] Step S402: Extract the main peak width of the autocorrelation function. and secondary peak significance .

[0053] The main peak is defined as the angular delay. The peak value at that location. The full width at half maximum (FWHM) of the main peak is calculated as the width of the main peak. The specific expression is: ; in, and To meet And the left and right delay values ​​closest to the origin, The autocorrelation value is zero-delay.

[0054] In the delay region excluding the main peak Within, find the significance of the secondary peak, the specific expression is: ; in, To minimize search latency, To maximize search latency, it is typically taken as... , To avoid the influence of the main peak and the boundary.

[0055] Step S403: Based on the width of the main peak and secondary peak significance Choose the optimal window function.

[0056] when and When the signal is determined to be highly periodic, a flat-top window is selected because it has a wide main lobe width in the frequency domain but high amplitude accuracy, making it suitable for accurately measuring the amplitude of periodic components.

[0057] Otherwise, if the signal is determined to contain significant transient impulses or has weak periodicity, the Nuttall window or the Blackman-Harris window should be selected. These two windows have lower sidelobe attenuation, better suppressing spectral leakage, and are suitable for analyzing signals with broadband or transient characteristics. Furthermore, if transient impulses dominate the signal, the Nuttall window, with its superior sidelobe attenuation performance, should be chosen to minimize spectral interference and highlight transient characteristics; if the signal has weak periodicity but requires high amplitude accuracy and dynamic range, the Blackman-Harris window, which achieves a better balance between sidelobe attenuation and frequency resolution, should be selected.

[0058] in, The threshold value for the main peak width is preset. This is a preset threshold for the significance of the secondary peak. In this embodiment, it is set to... , This initial reference value is set based on the statistical characteristics of vibration signals from a typical wind turbine drivetrain. In practical applications, to adapt to different equipment, operating conditions, and signal quality, it is recommended to calibrate according to the specific signal characteristics of the equipment. For example: (1) Adjustments can be made through historical data or simulation analysis. and To optimize the selection effect of window functions.

[0059] (2) An adaptive threshold strategy can be adopted, such as dynamically setting the threshold based on the average period length of the signal. .

[0060] (3) If the signal quality is poor or the operating conditions are complex, human experience or expert system can be used to assist in decision-making.

[0061] Step S404: Based on the type of the selected optimal window function, generate a stationary signal sequence in the angular domain. Equal-length optimal window function sequence .

[0062] Specifically, if a flat-top window is selected, it is generated using the discrete form of a fifth-order flat-top window; if a Naughtor window or a Blackman-Harris window is selected, it is generated using the discrete expression of its corresponding standard window function, respectively.

[0063] Step S5: Calculate the order spectrum of the stationary signal sequence in the angular domain using the optimal window function sequence, and output the order spectrum. For example... Figure 5As shown, it specifically includes: Step S501: Optimal window function sequence With angular domain stationary signal sequences Multiply point by point to obtain the windowed angular domain signal. The specific expression is: .

[0064] Step S502: Analyze the windowed corner domain signal. Perform a discrete Fourier transform and calculate its order spectrum. The specific expression is: ; ; in, For the first Each order For order resolution, It is the imaginary unit.

[0065] Step S503: Calculate the amplitude spectrum of the order spectrum. The order spectrum, as the final output, is specifically expressed as follows: .

[0066] To verify the effectiveness and superiority of the method of the present invention, a comparative experiment was conducted with the traditional equal-angle interval resampling order tracking method.

[0067] A measured dataset was collected from the gearbox of a certain model of 2MW wind turbine under variable speed operation. The known fault type is localized gear spalling. This dataset includes synchronously acquired and preprocessed vibration acceleration signals and rotational speed pulse signals, with a sampling frequency of [missing information]. Pulses per revolution .

[0068] The same set of test data was input into both the comparison method and the method of this invention. This was done based on the fault characteristic order. For each order, an evaluation index is calculated based on the order spectrum obtained by the two methods, including the amplitude error of the main peak. 3dB bandwidth and sideband separation The experimental results are shown in Table 1.

[0069] Table 1. Comparison of performance indicators between the method of this invention and traditional methods

[0070] As can be seen, in terms of amplitude accuracy, the method of this invention reduces amplitude distortion caused by speed fluctuations and interpolation errors through adaptive instantaneous phase calculation and a resampling algorithm based on local second-order polynomial fitting, significantly reducing the amplitude error of the main peak. The narrower 3dB bandwidth indicates that the order spectrum obtained by the method of this invention has a sharper main peak, more concentrated energy, and higher order resolution, which is beneficial for distinguishing dense order components. Furthermore, the higher sideband separation proves that the method of this invention, through adaptive selection and generation of the optimal window function sequence (in this experiment, autocorrelation analysis determined that the signal was highly periodic, so a flat-top window was selected), effectively suppresses spectral leakage, improves the signal-to-noise ratio, and makes the weak sideband components related to the fault more clearly distinguishable.

[0071] In summary, compared with traditional methods, the method of this invention has improved in three aspects: amplitude measurement accuracy, order resolution, and weak fault feature extraction capability. It is especially suitable for condition monitoring and fault diagnosis of rotating machinery such as wind turbines under variable speed and strong non-stationary operating conditions, and can provide more reliable and accurate order analysis results.

[0072] Therefore, the present invention adopts the above-mentioned adaptive order spectrum analysis method for fault diagnosis of wind turbine drive train, which solves the problems of insufficient instantaneous phase estimation accuracy and fixed and single selection of spectrum analysis window function in the prior art, improves the accuracy and resolution of order spectrum calculation, and enhances the accuracy, robustness and reliability of vibration signal order analysis.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain, characterized in that, Includes the following steps: Step S1: Data acquisition and preprocessing to obtain vibration acceleration signals and rotational speed pulse signals; Step S2: Perform adaptive instantaneous phase calculation on the rotational speed pulse signal to obtain an accurate instantaneous phase curve; Step S3: Based on the accurate instantaneous phase curve, the vibration acceleration signal is resampled in the angular domain to obtain a stationary signal sequence in the angular domain; Step S4: Perform autocorrelation analysis on the stationary signal sequence in the diagonal domain, and adaptively select and generate the optimal window function sequence; Step S5: Calculate the order spectrum of the stationary signal sequence in the angular domain using the optimal window function sequence, and output the order spectrum.

2. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 1, characterized in that, In step S2, an adaptive instantaneous phase calculation is performed using an adaptive parameter phase gradient correction algorithm, specifically including: Step S201: Based on the rotational speed pulse signal, calculate the initial instantaneous angular velocity estimate; Step S202: Construct an objective function with instantaneous angular velocity as the optimization variable, the specific expression of which is: ； in, This represents the total number of pulses in the rotational speed pulse signal. For the first The time corresponding to each rotational speed pulse signal In order to be in The estimated angular velocity at time t. for Estimation of the initial instantaneous angular velocity at time t. For regularization parameters; Based on instantaneous angular velocity Smoothness constraint terms; Step S203: Dynamically calculate the regularization parameters; Step S204: Solve the objective function using the regularization parameter to obtain the optimal instantaneous angular velocity curve; Step S205: Integrate the optimal instantaneous angular velocity curve to obtain the accurate instantaneous phase curve.

3. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 2, characterized in that, In step S203, the regularization parameter Based on the dynamic adjustment of instantaneous angular acceleration variance and signal-to-noise ratio, the specific expression is as follows: ； in, These are the base values ​​for the regularization parameters. The variance of instantaneous angular acceleration. For signal-to-noise ratio, and These are preset positive weighting coefficients.

4. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 3, characterized in that, instantaneous angular acceleration variance The calculation method is as follows: Discrete sampling is performed on the initial instantaneous angular velocity estimate to obtain a discrete angular velocity sequence. ,in, For the first Each sampling time, for The estimated angular velocity at that time; right The instantaneous angular acceleration sequence is obtained by performing first-order difference. The specific expression is: ； in, For the first The initial instantaneous angular velocity estimate at each sampling time. The sampling time interval; In length Within the sliding time window, calculate variance as The specific expression is: ； in, This represents the total number of sampling points within the sliding time window. Within the sliding time window The mean, For index variables, It is a discretized instantaneous angular acceleration sequence; And the length of the sliding time window Compared with the current average speed It is inversely proportional, and the specific expression is: ； in, It is a proportionality constant.

5. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 4, characterized in that, Signal-to-noise ratio The calculation method is as follows: Frequency domain analysis was performed on the rotational speed pulse signal to extract its amplitude at the fundamental frequency. Average noise amplitude of adjacent frequency bands .

6. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 5, characterized in that, Precise instantaneous phase curve in step S205 The specific expression is: ； in, For the initial phase, for The optimal instantaneous angular velocity at time t. For integration variables, For the current moment, This is the initial time.

7. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 6, characterized in that, In step S3, corner domain resampling is performed using a resampling algorithm based on local second-order polynomial fitting, specifically including: Step S301: For each target angle The precise target time is calculated by inversely based on the precise instantaneous phase curve. ; Step S302, at the precise moment of the target Select at least three time-domain sampling points in the vicinity; Step S303: Using time-domain sampling points, in the form of a second-order polynomial Perform local least squares fitting to obtain the fitting function. ,in, , , These are the fitting coefficients; Step S304: Precisely time the target Substitute into the fitting function Calculations yielded Interpolated estimate of the vibration signal at time interval ; Step S305: The interpolated estimated value From the perspective of the target The amplitude at that point is obtained as a sampling point in the stationary signal sequence in the angular domain. ; Through all target angles Angular domain stationary signal sequence is obtained. ,in For the first A uniform angle sampling point, This represents the total number of sampling points in the angular domain.

8. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 7, characterized in that, In step S4, the adaptive selection and generation of the optimal window function sequence specifically includes: Step S401: Calculate the stationary signal sequence in the angular domain. autocorrelation function The specific expression is: ； in, For angle delay variables; For angular delay The next Uniform angle sampling points A stationary signal sequence in the angular domain at that location; Step S402: Extract the main peak width of the autocorrelation function. and secondary peak significance ; Step S403: Based on the width of the main peak and secondary peak significance Choose the optimal window function; Step S404: Based on the type of the selected optimal window function, generate a stationary signal sequence in the angular domain. Equal-length optimal window function sequence .

9. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 8, characterized in that, In step S403, when and When the signal is determined to be highly periodic, a flat-top window is selected; Otherwise, if the signal is determined to contain significant transient impacts or has weak periodicity, select the Nuttor window or the Blackman-Harris window; in, The threshold value for the main peak width is preset. This is the preset threshold for the significance of the secondary peak.

10. The adaptive order spectrum analysis method for fault diagnosis of wind turbine drivetrain according to claim 9, characterized in that, Step S5 specifically includes: Step S501: Optimal window function sequence With angular domain stationary signal sequences Multiply point by point to obtain the windowed angular domain signal. The specific expression is: ； Step S502: Analyze the windowed corner domain signal. Perform a discrete Fourier transform and calculate its order spectrum. The specific expression is: ； ； in, For the first Each order For order resolution, The imaginary unit; Step S503: Calculate the amplitude spectrum of the order spectrum. The final output spectrum of order is expressed as follows: 。

Images