An automated fault diagnosis method for rotating mechanical components of a wind turbine generator set

By using adaptive signal decomposition and fault feature search algorithms to process empirical mode component signals, the problem of low automation in fault diagnosis of rotating mechanical components of wind turbine generators is solved, and efficient and accurate fault detection is achieved.

CN118855646BActive Publication Date: 2026-11-10BEIJING NENGGAO PUKANG MEASUREMENT & CONTROL TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410941040.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2026-11-10
Estimated Expiration
2044-07-15

AI Technical Summary

Technical Problem

Existing fault diagnosis methods for rotating mechanical components of wind turbine generators have low automation levels, and condition monitoring systems are not sensitive to component faults, relying on the location of measuring points to set alarm indicators, which leads to false alarms.

Method used

An adaptive signal decomposition method is adopted, and an empirical mode component signal is processed by a fault feature search algorithm to match composite fault indicators of various types of faults, thereby achieving quantitative diagnosis.

Benefits of technology

This improves the automation level of fault diagnosis for rotating mechanical components of wind turbine generators, enhances the sensitivity and accuracy of fault detection, and reduces false alarms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118855646B_ABST
    Figure CN118855646B_ABST
Patent Text Reader

Abstract

This invention discloses a fault diagnosis method for rotating mechanical components of wind turbine generators. The fault diagnosis method of this invention adaptively divides the frequency domain signal of the vibration signal of the rotating mechanical component by extracting the envelope of the frequency domain signal, thus improving the problem of aliasing of modal component fault feature information caused by imperfect frequency domain signal division in traditional empirical wavelet transform. Simultaneously, it filters empirical modal components containing rich fault information, combines them with a fault feature search algorithm, and matches the state evaluation index corresponding to each type of fault, thereby realizing fault diagnosis of rotating mechanical components of wind turbine generators.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of wind power generation, specifically to an automated fault diagnosis method for rotating mechanical components of a wind turbine generator set. Background Technology

[0002] Wind energy is a clean and renewable energy source that is rapidly emerging globally, and wind power generation technology, especially wind turbines, is developing and growing rapidly.

[0003] Because wind turbines operate in extreme environments for extended periods and are subject to low-speed, heavy-load conditions, their transmission systems are prone to damage, which can result in significant economic losses for wind turbine operators.

[0004] In the transmission system of wind turbines, bearings and gears are key components that often fail due to alternating heavy loads.

[0005] Vibration signal-based diagnostic methods are currently the most widely used fault diagnosis methods for rotating mechanical components of wind turbine generators because vibration signals have obvious characteristics, mature detection methods, and are easy to implement online monitoring.

[0006] However, current research on diagnostic methods for rotating mechanical components of wind turbine generators faces some challenges, such as low automation levels and high requirements for the number and expertise of analysts.

[0007] Furthermore, most current alarm mechanisms in online condition monitoring systems for wind turbine generators based on vibration signals are set according to the location of the measuring points. The relationship between the measuring point location and the monitored components is one-to-many, which may lead to the condition monitoring system being insensitive to component failures.

[0008] Therefore, we propose an adaptive signal decomposition method that processes empirical mode component signals through a fault feature search algorithm and matches composite fault indices corresponding to various types of faults, thereby achieving quantitative diagnosis of faults in rotating mechanical components of wind turbine generators. Summary of the Invention

[0009] To address the above technical problems, this invention provides an automated fault diagnosis method for rotating mechanical components of a wind turbine generator set, the method comprising the following steps:

[0010] S100: Vibration signals are collected by vibration sensors installed on the transmission system of the wind turbine generator set, and the signals are preprocessed to obtain the processed vibration signals.

[0011] S200: Performs Fourier transform on the vibration signal to a frequency domain signal.

[0012] S300: Segment the frequency domain signal and extract local peak values, then interpolate the local peak values ​​to obtain the upper envelope of the frequency domain signal.

[0013] S400: Empirical wavelet transform is performed using the minimum point of the envelope of the frequency domain signal as the frequency domain boundary point to obtain the empirical mode components.

[0014] S500: Calculate the correlation kurtosis of each empirical mode component, select the empirical mode component with the largest correlation kurtosis, and calculate its envelope frequency domain signal.

[0015] S600: Using fault feature search calculations, the fault feature frequencies and their harmonics are searched in the envelope frequency domain signal of the empirical mode component to determine the fault point.

[0016] S700: Match the fault assessment index corresponding to the fault point and output the status assessment result.

[0017] Preferably, S300 includes:

[0018] S310: Divide the frequency domain signal f(ω) into s segments, and take the maximum value point f of each segment. max (ω1),f max (ω2),…,f max (ω s ), to obtain the local peak value of the frequency domain signal.

[0019] S320: The local peak points are interpolated using the cubic spline interpolation method to obtain the overall interpolation function S(x), which is the upper envelope of the frequency domain signal f(ω).

[0020] Preferably, S400 includes:

[0021] S410: Find the local minima of the upper envelope function S(x) and sort them in ascending order, using them as the frequency domain boundary point ω. n (n = 1, 2, ...).

[0022] S420: Wavelet function ψ for constructing empirical wavelets based on frequency boundary points n (ω) and scaling function φ n (ω), expressed as follows:

[0023]

[0024]

[0025] in: β(x)=x 4 (35-84x+70x 2 -20x 3 )

[0026] S430: The inner product of the original signal and the empirical wavelet function and scaling function is defined as the detail correlation coefficient. and approximate correlation coefficient The expression is:

[0027]

[0028]

[0029] In the formula: ψ n For empirical wavelet functions, For ψ n The conjugate of complex numbers, For empirical scaling functions, for The conjugate of complex numbers, For ψ n The Fourier transform, and the inverse Fourier transform are denoted as F(·) and Finverse, respectively. -1 (·)

[0030] S440: Reconstruct the signal, the result is:

[0031]

[0032] In the formula: and They represent and Fourier transform.

[0033] S450: Empirical Mode f k (t) is defined as follows:

[0034]

[0035]

[0036] Preferably, S500 includes:

[0037] S510: Calculate the correlation kurtosis of each empirical mode component, and select the empirical mode component with the largest correlation kurtosis as the empirical mode component containing the most fault feature information.

[0038] The relevant kurtosis calculation methods are as follows:

[0039] In the formula: N is the sampling length, T is the period of the pulse signal of interest, and M is the number of offset periods.

[0040] S520: Perform Hilbert transform on the empirical mode components selected in the above steps, and combine them with the original signal to form an analytic signal, finally obtaining the envelope signal of each mode component.

[0041]

[0042]

[0043] In the formula: f k Hilbert transform of (t), A k (t) is f k The envelope signal of (t).

[0044] S530: Envelopes A of each modal component k (t) Perform Fourier transform to obtain the envelope frequency domain signal A of each empirical mode component. k (ω).

[0045] Preferably, S600 includes:

[0046] S610: Calculate the characteristic frequencies of all gear and bearing faults monitored at the measurement point locations.

[0047] S620: Set the search frequency multiplier H = 1, the search range threshold ρ, the effective search decision threshold N, and the decision threshold N. It is affected by many factors such as sensor installation location, signal transmission path, and sampling frequency.

[0048] S630: Using H times the fault characteristic frequency f fault Centered on [(1-ρ)Hf], the search range is [(1-ρ)Hf] fault ,(1+ρ)Hf fault In the envelope frequency domain of each empirical mode component signal A k Search for the maximum value in (ω) and corresponding frequency like H+1, and repeat this step. When H≥N, proceed to the next step.

[0049] S640: Based on the searched fault characteristic frequency f fault Identify the location of the fault. Attached Figure Description

[0050] Figure 1 Flowchart of an automated fault diagnosis method for rotating mechanical components of a wind turbine generator set

[0051] Figure 2 Schematic diagram of vibration acceleration signal

[0052] Figure 3 Schematic diagram of the envelope of a frequency domain signal Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0054] Please refer to Figure 1 The specific operating steps are as follows:

[0055] The vibration acceleration time-domain signal f(t) is as follows Figure 2 As shown, the Fourier transform of signal f(t) yields its corresponding frequency domain signal f(ω).

[0056]

[0057] In one embodiment of the present invention, the frequency domain signal f(ω) is divided into s segments, and the maximum value point f is taken for each segment. max (ω1),f max (ω2),…,f max (ω s ), to obtain the local peak value of the frequency domain signal.

[0058] Let them be denoted as f max (ω1),f max (ω2),…,f max (ω s ) is used as an interpolation node, with boundary condition f' max (ω1),f' max (ω s ).

[0059] Calculate h j =ω j+1 -ω j (j=1,2,…,s-1).

[0060] calculate

[0061] calculate

[0062] Solve the system of equations

[0063] Substitute the solution of the above system of equations into the following equation:

[0064]

[0065] S(x) is a cubic spline interpolation function, i.e., the upper envelope of the frequency domain signal f(ω) is as follows: Figure 3 As shown.

[0066] Find the local minima of the upper envelope function S(x) and arrange them in ascending order, using them as the frequency domain boundary point ω. n (n = 1, 2, ...).

[0067] The wavelet function ψ is used to construct empirical wavelets based on frequency boundary points. n (ω) and scaling function φn (ω), expressed as follows:

[0068]

[0069]

[0070] in:

[0071] The inner product of the original signal and the empirical wavelet function and scaling function is defined as the detail correlation coefficient. and approximate correlation coefficient The expression is:

[0072]

[0073]

[0074] In the formula: ψ n For empirical wavelet functions, For ψ n The conjugate of complex numbers, For empirical scaling functions, for The conjugate of complex numbers, For ψ n The Fourier transform, and the inverse Fourier transform are denoted as F(·) and Finverse, respectively. -1 (·)

[0075] The signal is reconstructed, and the result is:

[0076]

[0077] In the formula: and They represent and Fourier transform.

[0078] Empirical mode f k (t) is defined as follows:

[0079]

[0080]

[0081] Calculate the correlation kurtosis of each empirical mode component, and select the empirical mode component with the largest correlation kurtosis as the empirical mode component containing the most fault feature information.

[0082] The relevant kurtosis calculation methods are as follows:

[0083] In the formula: N is the sampling length, T is the period of the pulse signal of interest, and M is the number of offset periods.

[0084] The empirical mode components selected in the above steps are subjected to Hilbert transform and combined with the original signal to form an analytic signal, ultimately yielding the envelope signal of each mode component.

[0085]

[0086]

[0087] In the formula: f k Hilbert transform of (t), A k (t) is f k The envelope signal of (t).

[0088] The envelopes A of each modal component k (t) Perform Fourier transform to obtain the envelope frequency domain signal A of each empirical mode component. k (ω).

[0089]

[0090] Calculate the characteristic frequency of gear failure: f r .

[0091] Calculate the characteristic frequency of bearing outer ring failure:

[0092] Calculate the characteristic frequency of bearing inner ring failure:

[0093] Calculate the characteristic frequencies of bearing rolling element failures:

[0094] Calculate the characteristic frequencies of bearing cage failure:

[0095] In the formula: f r The shaft rotation frequency, i.e., the characteristic frequency of gear failure; f outer raceway F represents the characteristic frequency of bearing outer ring failure. inner raceway f is the characteristic frequency of the bearing inner ring fault. ball f is the characteristic frequency of bearing rolling element failure; cage denoted as the bearing cage failure characteristic frequency; z is the number of rolling elements; d is the rolling element diameter; D is the bearing pitch diameter; and α is the contact angle.

[0096] Set the search frequency multiplier H = 1, the search range threshold ρ ≈ 5%, and the effective search judgment threshold N = 4.

[0097] With H times the fault characteristic frequency f faultCentered on [(1-ρ)Hf], the search range is [(1-ρ)Hf] fault ,(1+ρ)Hf fault In the envelope frequency domain of each empirical mode component signal A k Search for the maximum value in (ω) and corresponding frequency like H+1, and repeat this step. When H≥N, proceed to the next step.

[0098] In one embodiment of the present invention, based on the searched fault characteristic frequency f fault Identify the faulty location and match the corresponding condition assessment index X for that location. fearture Output the status assessment level.

[0099]

[0100] In the formula: X std Let f(t) be the standard deviation.

Claims

1. An automated fault diagnosis method for rotating mechanical components of a wind turbine generator set, characterized in that, The fault diagnosis method includes the following steps: S100: Vibration signals are collected by vibration sensors installed on the transmission system of the wind turbine generator set, and the signals are preprocessed to obtain processed vibration signals; S200: Perform Fourier transform on the vibration signal to obtain the frequency domain signal; S300: The frequency domain signal is segmented, the local peak value of each segment is extracted, and the local peak value is interpolated using the cubic spline interpolation method to obtain the upper envelope of the frequency domain signal; S400: Locate the local minimum points of the upper envelope function and arrange them in ascending order. Use the local minimum points as frequency domain boundary points. Perform empirical wavelet transform based on the frequency domain boundary points to obtain empirical mode components. S500: Calculate the correlation kurtosis of each empirical mode component, wherein the calculation parameters of the correlation kurtosis include the period T of the pulse signal of interest and the number of offset periods M; select the empirical mode component with the largest correlation kurtosis as the empirical mode component containing the most fault feature information, and calculate the envelope frequency domain signal of the selected empirical mode component. S600: Calculate the fault characteristic frequencies of all gears and bearings monitored at the measurement point locations; use the fault characteristic search algorithm to search for the fault characteristic frequencies and their harmonics in the envelope frequency domain signal of the selected empirical mode components, and confirm the fault location based on the searched fault characteristic frequencies; S700: Match the fault assessment index corresponding to the fault point and output the status assessment result.

2. The automated fault diagnosis method for rotating mechanical components of a wind turbine generator set according to claim 1, characterized in that, The empirical wavelet transform includes: The wavelet function and scaling function of the empirical wavelet are constructed based on the frequency boundary point; the inner product of the original signal with the empirical wavelet function and the scaling function are calculated respectively to obtain the detail correlation coefficient and the approximate correlation coefficient; the signal is reconstructed based on the detail correlation coefficient and the approximate correlation coefficient to obtain the empirical mode components.

3. The automated fault diagnosis method for rotating mechanical components of a wind turbine generator set according to claim 1, characterized in that, The fault feature search algorithm includes: Calculate the characteristic frequencies of all gear and bearing faults monitored at the measurement points; The search frequency multiplier, search range threshold, and effective search determination threshold are set. The effective search determination threshold is affected by factors such as sensor installation location, signal transmission path, and sampling frequency. Centered on the corresponding fault characteristic frequency, within the search range determined by the search range threshold, the maximum value and corresponding frequency are searched in the envelope frequency domain signal of the selected empirical mode component as described in claim 1, and the frequency doubling search result is determined according to the comparison result of the maximum value and the effective search judgment threshold; the fault location is confirmed according to the searched fault characteristic frequency.

Citation Information

Patent Citations

  • Wind turbine generator bearing fault diagnosis method under variable speed

    CN105784366A

  • Rolling bearing fault diagnosis method, device, equipment and medium

    CN116358873A

  • Fault diagnosis method based on improved empirical wavelet transform and envelope spectrum energy ratio

    CN117928951A