Wind turbine generator gearbox output end bearing comprehensive diagnosis method based on multi-model cooperation

Abstract

The invention relates to the technical field of wind turbine generator system fault diagnosis, and discloses a wind turbine generator system gear box output end bearing comprehensive diagnosis method based on multi-model collaboration, and the method comprises the steps: predicting the bearing temperature trend through a dynamic power-bearing temperature model, and carrying out the temperature early warning through a power-temperature difference model; confidence calculation is carried out in cooperation with multi-dimensional data such as mechanical vibration characteristics (time domain / frequency domain impact energy reflects bearing mechanical damage), bearing temperature trend and temperature difference analysis (eliminating environment and load interference to position real fault temperature rise), and mechanical faults such as roller peeling and inner ring cracks and operation faults such as heat dissipation failure can be distinguished; and a traditional'single-point alarm 'mode is broken through, a'root analysis-unit power reduction / trigger shutdown' whole-process closed loop is constructed, and the safety, reliability and operation and maintenance efficiency of unit operation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to a wind turbine generator set fault diagnosis technology, in particular to a wind turbine generator set gearbox output end bearing comprehensive diagnosis method based on multi-model collaboration. Background Art

[0002] The domestic wind power industry is rapidly developing. A large number of wind turbines, installed at high altitudes in open areas, are subject to long-term exposure to harsh conditions such as wind and sand erosion and extreme weather conditions, leading to a high incidence of bearing failures at the output end of the gearbox. Existing research has mostly relied on traditional bearing overtemperature early warning and diagnostic methods or single-parameter monitoring models. These methods are unable to accurately distinguish the root cause of overtemperatures (such as mechanical failure or operational failure) and also hinder the development of targeted operation and maintenance strategies. For example, overtemperatures caused by mechanical wear require immediate shutdown and replacement of spare parts, while overtemperatures caused by operational failures (such as lubrication abnormalities or heat dissipation failures) require priority power reduction to investigate the cause.

[0003] The core shortcomings of existing technologies are:

[0004] 1. Lack of multi-source data correlation analysis: Traditional diagnostics rely on a single temperature threshold or vibration spectrum, failing to link temperature with operating parameters such as power and speed. This makes it impossible to dynamically and adaptively adjust the temperature threshold. For example, a fixed temperature threshold cannot track power changes. When a 2MW unit's power drops sharply to 1MW, the bearing temperature prediction based on the operating model should decrease accordingly. However, the fixed threshold, without dynamic adjustment based on power changes, may still misinterpret a normal temperature drop as an anomaly.

[0005] 2. Inadequate fault type identification: Vibration analysis focuses solely on a single frequency band (such as BPFO / BPF I) and fails to incorporate temperature trends. In the case of an early bearing failure, the vibration energy in the resonant frequency band increases slightly, but the temperature does not change significantly. This single vibration model can easily miss a fault. Conversely, if temperature anomalies are not correlated with power (for example, vibration energy should be lower at low power), the fault level or type may be misjudged, leading to inappropriate O&M strategies.

[0006] Therefore, for the scenario of overheating of the bearing at the output end of the gearbox, it is urgent to develop a comprehensive diagnostic method that integrates multi-dimensional data and accurately distinguishes overheating caused by mechanical failure and operational failure, so as to solve the core problems of "fuzzy identification of overheating causes and improper operation and maintenance decisions" in the existing technology. Summary of the Invention

[0007] The purpose of the present invention is to provide a comprehensive diagnosis method for the output end bearing of the wind turbine gearbox based on multi-model collaboration. The method performs temperature warning on the acquired temperature data through an early warning model, and at the same time coordinates the temperature warning model with the mechanical fault diagnosis model, thereby solving the problem of inaccurate fault cause identification under a single model.

[0008] The present invention provides the following technical solutions:

[0009] The present invention provides a comprehensive diagnosis method for the output end bearing of a wind turbine gearbox based on multi-model collaboration, which includes:

[0010] Step S1: obtaining the wind turbine generator set power, and the temperature signal and vibration signal of the wind turbine generator set gearbox bearing;

[0011] Step S2: Generate temperature residual ΔT and steady-state temperature T using the pre-built temperature warning model pred and dynamic temperature difference R ΔTP Steady-state temperature T pred and dynamic temperature difference R ΔTP Used to determine if a temperature rise has occurred;

[0012] Step S3: If temperature rise occurs, the vibration signal is filtered based on a pre-established mechanical fault diagnosis model, and the reconstructed signal after filtering is subjected to bandpass filtering and Hilbert transform to extract the envelope spectrum signal. The spectral kurtosis value K is calculated at the theoretical characteristic frequency of the envelope spectrum signal, and the envelope spectrum amplitude ratio R is determined.

[0013] Step S4: Coordinating the mechanical fault diagnosis model with the temperature early warning model, dynamically adjusting the envelope spectrum amplitude ratio R of the envelope spectrum signal according to the temperature residual ΔT; and defining the cross-correlation coefficient ρ between the spectrum kurtosis value K and the temperature residual ΔT;

[0014] Step S5: Based on the pre-built over-temperature diagnosis model, determine the over-temperature fault confidence level according to the spectral kurtosis value K, the envelope spectrum amplitude ratio R, the temperature residual ΔT, and the normalized value of the cross-correlation coefficient ρ; and determine the cause of the temperature increase according to the over-temperature fault confidence level.

[0015] Optionally, the temperature residual ΔT, steady-state temperature T pred and dynamic temperature difference R ΔTP The calculation formula includes:

[0016] ΔT=|T real -T pred |

[0017] T pred =k·P+b·T env +c

[0018]

[0019] Where, P is the real-time active power (kW); T env is the cabin ambient temperature (°C); k, b, c are the coefficients fitted by historical steady-state data, ΔT f-rIt is the difference between the front bearing temperature and the rear bearing temperature of the gearbox output shaft.

[0020] Optionally, the steady-state temperature T pred and dynamic temperature difference R ΔTP Methods used to determine if a temperature increase has occurred include:

[0021] Calculate R in historical data ΔTP The mean μ and standard deviation σ of

[0022] When R ΔTP When the temperature exceeds μ+2σ and lasts for a preset time, it is determined that the temperature has risen and a level 1 warning is triggered;

[0023] When R ΔTP When the temperature exceeds μ+3σ and continues for the preset time, it is determined that the temperature has risen and a secondary warning is triggered;

[0024] When T real <0.95T pred And R ΔTP When the value is less than μ+1.5σ for the preset time, the warning is lifted.

[0025] Optionally, filtering the vibration signal includes:

[0026] The original vibration signal contains the gear meshing frequency f mesh Mixed vibration signal of bearing fault signal;

[0027] The LMS (least mean square) algorithm is used to iteratively update the filter weights to minimize the meshing frequency component in the output signal.

[0028] Optionally, dynamically adjusting the envelope spectrum amplitude ratio R of the envelope spectrum signal according to the temperature residual ΔT includes:

[0029] If ΔT≤5℃, the envelope spectrum amplitude ratio R>14.0;

[0030] If 5℃<ΔT≤10℃, the envelope spectrum amplitude ratio R>12.0;

[0031] If ΔT>10℃, the envelope spectrum amplitude ratio R>10.0.

[0032] Optionally, the cross-correlation coefficient ρ between the spectral kurtosis value K and the temperature residual ΔT is calculated as follows:

[0033]

[0034] Where t represents the time and μ represents the mean.

[0035] Optionally, the calculation formula for the over-temperature fault confidence is:

[0036] Confidence=ω1f K +ω2f R +ω3f ΔT +ω4f ρ (ω1+ω2+ω3+ω4=1)

[0037] Among them, ω1, ω2, ω3, ω4 are weight coefficients, f K is the normalized value of spectral kurtosis; f R is the normalized value of the envelope spectrum amplitude ratio; f ΔT is the normalized value of temperature residual; f ρ is the normalized value of the cross-correlation coefficient.

[0038] Optionally, when Confidence>0.7, it is confirmed that the cause of the temperature increase is a failure of the high-speed shaft bearing;

[0039] When 0.4<Confidence≤0.7, the cause of the temperature increase is confirmed to be a suspected failure of the high-speed shaft bearing, and the high-speed shaft bearing should be continuously tested;

[0040] When Confidence ≤ 0.4, confirm that the temperature rise is not caused by mechanical failure, and check whether the unit has abnormal heat dissipation or operating failure.

[0041] Compared to existing technologies, this invention offers the following significant advantages: Addressing the difficulty of overtemperature warning and diagnostic methods in distinguishing between overtemperatures caused by unit operation and overtemperatures caused by mechanical failure, this invention constructs a comprehensive diagnostic system for wind turbine gearbox bearings based on multi-model collaboration. Based on vibration signal characteristics and temperature signal data, an overtemperature fault confidence algorithm is introduced to quantitatively analyze bearing status from a global perspective, accurately identifying overtemperatures caused by bearing mechanical failures, and effectively eliminating the influence of external interference factors such as ambient temperature fluctuations and load changes. This improves accuracy and reliability, providing efficient protection for the safe and stable operation of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the process of the gearbox condition monitoring method provided in Example 1;

[0043] Figure 2 is the shaft temperature curve predicted by the power-shaft temperature model in Example 1;

[0044] Figure 3 is the power-temperature difference ratio curve predicted in Example 1;

[0045] Figure 4 The outer ring fault envelope spectrum of the gearbox output end bearing before filtering provided in Example 1;

[0046] Figure 5This is a fault-free envelope spectrum of the gearbox output end bearing before filtering provided in Example 1;

[0047] Figure 6 The fault envelope spectrum of the outer ring of the gearbox output end bearing after bandpass filtering provided in Example 1;

[0048] Figure 7 The fault-free envelope spectrum of the bearing at the output end of the gearbox provided in Example 1 after bandpass filtering;

[0049] Figure 8 The fault envelope spectrum of the outer ring of the gearbox output end bearing after adaptive resonance frequency band filtering provided in Example 1;

[0050] Figure 9 The fault-free envelope spectrum of the gearbox output end bearing after adaptive resonance frequency band filtering provided in Example 1;

[0051] Figure 10 This is a line graph of the fault-free peak factor, envelope spectrum amplitude ratio, and spectrum kurtosis value for overtemperature root cause diagnosis in Example 1;

[0052] Figure 11 This is a line graph of the outer race fault peak factor, envelope spectrum amplitude ratio, and spectrum kurtosis value for overtemperature root cause diagnosis in Example 1;

[0053] Figure 12 Schematic diagram of the synergistic effect of the power-axis temperature model and the power-temperature difference ratio model in Example 1;

[0054] Figure 13 Schematic diagram of the collaborative diagnosis process in Example 1. DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0056] Example 1

[0057] Combine Figure 1 The present invention provides a comprehensive diagnosis method for the output end bearing of a wind turbine gearbox based on multi-model collaboration to solve the problem. The method includes:

[0058] Step S1: The temperature of the front / rear end bearings of the gearbox output shaft is collected through a temperature sensor, and the ambient temperature is obtained through an external sensor of the unit.

[0059] The sampling frequency of the vibration sensor in step S1 is 25600 Hz, the unit of the speed data is r / min, and the temperature can be directly displayed on the PLC.

[0060] Multi-source data is synchronized with the vibration data clock via the PLC to calculate real-time power. Outliers (such as temperature spikes >10°C / min) are eliminated based on the 3σ principle. Short-term missing data (<5s) is supplemented using linear interpolation, while long-term missing data (≥30s) is eliminated, triggering sensor alarms.

[0061] Step S2: Adopt a dual-model collaborative framework to take into account both steady-state prediction and dynamic response. The specific architecture is as follows: Figure 12 As shown:

[0062] Define the dynamic power-axis temperature regression model: T pred =k·P+b·T env +c, and use the sliding window least squares method to solve the correlation coefficient. Scroll to select the data of the last 72 hours and bring it into the corresponding objective function: At the same time, array k, b, and c and solve: in:

[0063] At the same time, in order to ensure that the model parameters are more accurate, k, b, and c are recalculated every 72 hours to adapt to seasonality and unit aging.

[0064] Define the power-temperature difference model: (∈=1kW to prevent zero), and calculate R in historical data at the same time ΔTP The mean μ and standard deviation σ of R ΔTP >μ+2σ and lasts for the preset time, it is a first-level warning. ΔTP >μ+3σ is the second level warning. real <0.95T pred And R ΔTP When the value is less than μ+1.5σ for the preset time, the warning is lifted.

[0065] like Figure 2 and Figure 3 Shown are the shaft temperature curve and power-temperature difference ratio curve predicted by the power-shaft temperature model.

[0066] Step S3: Construct a mechanical fault diagnosis model. The relevant vibration data is measured by the vibration sensor; the speed data is the gearbox output shaft speed f provided by the main control PLC. r (Hz) or estimated by the speed estimation algorithm.

[0067] When the fault characteristic frequencies of high-speed shaft bearings (such as BPFO and BPFI) are completely masked by the gear meshing frequency and its harmonics, traditional envelope spectrum analysis may fail. In this case, preprocessing of the raw vibration data is necessary.

[0068] In this patent, adaptive resonance frequency band is used to track and filter out gear meshing frequency harmonics and related noise in real time, retaining the fault frequency band of bearing resonance frequency. The core is to use LMS (least mean square) algorithm to iteratively update the filter weights to minimize the interference components in the output signal. The input signal is the gear meshing frequency f mesh The purpose of the LMS algorithm in step S3 is to update the filter coefficients to minimize the meshing frequency component in the output signal. The corresponding transfer function is: (λ is the bandwidth control parameter).

[0069] The reference signal is The purpose is to suppress the corresponding target frequency. where ω k (n) is a time-varying weight, and the error signal e(n) = d(n) - y(n).

[0070] Take the spectral kurtosis of time domain features Crest Factor Perform bandpass filtering and Hilbert transform on the reconstructed signal z[n]=x bp [n]+j·Η(x bp [n]), where H is the Hilbert transform, and H(x) = IFFT(-j·sign(f)·FFT(x)) is implemented in the frequency domain to obtain the envelope signal The envelope spectrum of the reconstructed signal is calculated to obtain its envelope spectrum: The spectral kurtosis value is calculated at the theoretical characteristic frequency in the envelope spectrum to determine whether a fault has occurred.

[0071] Finally, the fault characteristic frequency envelope spectrum amplitude ratio is obtained (taking the bearing outer ring fault as an example):

[0072] A BPFI =S[f BPFI ].

[0073] like Figure 5-10 Shown are the original and processed envelope spectra of the bearing inside the gearbox.

[0074] Step S4: The mechanical fault diagnosis model and the temperature warning model achieve deep collaboration through data sharing, dynamic threshold adjustment, and time series causal analysis, forming a collaborative system of "temperature warning triggering vibration analysis → dynamic adjustment of vibration feature threshold". The specific architecture is as follows Figure 13In step S4, collaborative diagnosis mainly refers to the Pearson correlation coefficient:

[0075] The most important thing about the multi-factor decision engine in step S4 is normalization processing, which is to divide the feature value by its threshold to obtain a normalized value. If the normalized value is greater than 1, it indicates that the feature exceeds the limit; if the normalized value is less than 1, it indicates that the feature does not exceed the limit.

[0076] Steps S2 and S3 have already introduced the data synchronization and preprocessing process in detail. This step focuses on the temperature warning model triggering vibration analysis and temperature-vibration time series causal analysis. First, define the dynamic threshold adjustment rules:

[0077]

[0078] Also define the cross-correlation coefficient: When ρ ≥ 0.7, it is determined that the temperature rise is caused by mechanical failure (the vibration abnormality precedes the temperature rise), and when ρ < 0.3, it is determined that the temperature rise is caused by an external heat source (the temperature abnormality dominates).

[0079] Step S5: The overtemperature diagnosis model locates the root cause of overtemperature by coordinating multiple data sources such as temperature and vibration, combining dynamic thresholds and a multi-factor decision engine. The overtemperature diagnosis model is similar to the temperature warning model in step S2, but the biggest difference is that the confidence level is calculated through a multi-factor decision engine:

[0080] Confidence=ω1f K +ω2f R +ω3f ΔT +ω4f ρ

[0081] Among them, ω1+ω2+ω3+ω4=1, and the weight distribution is based on the fault mechanism and historical data.

[0082] In order to eliminate the dimension difference, the original features need to be standardized. i The common method for normalizing values ​​is threshold relativization: Targeting ω i The weight distribution adopts the entropy weight method, and the specific steps are as follows: Figure 11-12 Shown are line graphs of four indicators used in overtemperature root cause diagnosis.

[0083] Assume there are m samples (historical data), n features, and construct the data matrix X = [x ij ] m×n , in the standardization process:

[0084] Treat the normalized values ​​as a probability distribution: Entropy calculation: Then calculate the coefficient of difference to reflect the information content of the indicator: d j =1-E j ; Finally, determine the weight and find the normalization parameter:

[0085] When Confidence>0.7, the temperature rise is confirmed to be caused by a high-speed shaft bearing failure; when 0.4<Confidence≤0.7, the temperature rise is confirmed to be caused by a suspected high-speed shaft bearing failure, and continuous testing will be carried out; when Confidence≤0.4, the temperature rise is confirmed not to be caused by a mechanical failure, but may be caused by abnormal heat dissipation or operational failure.

[0086] The principle of this model is based on the idea of ​​weighted comprehensive scoring, combining the weights and normalized values ​​of each input feature, and finally outputting a confidence score between 0 and 1.

[0087] The cross-correlation coefficient ρ, mentioned in step S4, is crucial for determining which temperature anomaly occurred first and which occurred later, determining whether the temperature anomaly was caused by a mechanical fault. The confidence level, mentioned in step S5, is crucial for synergizing multidimensional evidence to comprehensively determine the fault type and severity. While the cross-correlation coefficient focuses on causal verification, the confidence level assessment focuses on integrating multidimensional evidence and accurately diagnosing the fault. The two complement each other, improving diagnostic accuracy and reliability.

[0088] This embodiment, based on multi-model collaboration, conducts comprehensive diagnosis of wind turbine gearbox output bearings. It predicts temperature trends using a dynamic power-shaft temperature model, issues temperature warnings using a power-temperature difference model, and calculates confidence levels using multi-dimensional data such as mechanical vibration characteristics, bearing temperature trends, and temperature difference analysis. It supports mechanical fault types such as roller spalling, inner ring cracks, and retainer fractures, as well as operational faults such as lubrication anomalies and heat dissipation failure. Furthermore, it breaks through the traditional "single-point alarm" model and establishes a closed-loop process from "root cause analysis to unit power reduction / triggering shutdown," significantly improving the safety, reliability, and maintenance efficiency of unit operations.

[0089] Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the implementation scenario description, or can be modified accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module or further split into multiple submodules.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements 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 application.

[0091] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A comprehensive diagnosis method for the output bearing of a wind turbine gearbox based on multi-model collaboration, characterized in that: include: Step S1: obtaining the wind turbine generator set power, and the temperature signal and vibration signal of the wind turbine generator set gearbox bearing; Step S2: Generate temperature residual ΔT and steady-state temperature T using the pre-built temperature warning model pred and dynamic temperature difference R ΔT / P ; The steady-state temperature T pred and dynamic temperature difference R ΔT / P Used to determine if a temperature rise has occurred; Step S3: If temperature rise occurs, the vibration signal is filtered based on a pre-established mechanical fault diagnosis model, and the reconstructed signal after filtering is subjected to bandpass filtering and Hilbert transform to extract the envelope spectrum signal. The spectral kurtosis value K is calculated at the theoretical characteristic frequency of the envelope spectrum signal, and the envelope spectrum amplitude ratio R is determined. Step S4: Coordinating the mechanical fault diagnosis model with the temperature early warning model, dynamically adjusting the envelope spectrum amplitude ratio R of the envelope spectrum signal according to the temperature residual ΔT; and defining the cross-correlation coefficient ρ between the spectrum kurtosis value K and the temperature residual ΔT; Step S5: Based on the pre-built over-temperature diagnosis model, determine the over-temperature fault confidence level according to the spectral kurtosis value K, the envelope spectrum amplitude ratio R, the temperature residual ΔT, and the normalized value of the cross-correlation coefficient ρ; and determine the cause of the temperature increase according to the over-temperature fault confidence level.

2. The wind turbine gearbox output end bearing comprehensive diagnosis method according to claim 1, characterized in that: The temperature residual ΔT, steady-state temperature T pred and dynamic temperature difference R ΔT / P The calculation formula includes: ΔT=|T real -T pred | T pred =k·P+b·T env +c Where, P is the real-time active power (kW); T env is the cabin ambient temperature (°C); k, b, c are the coefficients fitted by historical steady-state data, ΔT f-r It is the difference between the front bearing temperature and the rear bearing temperature of the gearbox output shaft.

3. The wind turbine gearbox output end bearing comprehensive diagnosis method according to claim 1, characterized in that: The steady-state temperature T pred and dynamic temperature difference R ΔT / P Methods used to determine if a temperature increase has occurred include: Calculate R in historical data ΔT / P The mean μ and standard deviation σ of When R ΔT / P When the temperature exceeds μ+2σ and lasts for a preset time, it is determined that the temperature has risen and a level 1 warning is triggered; When R ΔT / P When the temperature exceeds μ+3σ and continues for the preset time, it is determined that the temperature has risen and a secondary warning is triggered; When T real <0.95T pred And R ΔT / P When the value is less than μ+1.5σ for the preset time, the warning is lifted.

4. The wind turbine gearbox output end bearing comprehensive diagnosis method according to claim 1, characterized in that: The dynamically adjusting the envelope spectrum amplitude ratio R of the envelope spectrum signal according to the temperature residual ΔT includes: If ΔT≤5℃, the envelope spectrum amplitude ratio R>14.0; If 5℃<ΔT≤10℃, the envelope spectrum amplitude ratio R>12.0; If ΔT>10℃, the envelope spectrum amplitude ratio R>10.

0.

5. The comprehensive diagnosis method for the output end bearing of the wind turbine gearbox according to claim 1 is characterized in that: The calculation formula of the cross-correlation coefficient ρ between the spectral kurtosis value K in step S4 and the temperature residual ΔT in step S2 is: Where t represents the time and μ represents the mean.

6. The wind turbine gearbox output end bearing comprehensive diagnosis method according to claim 1, characterized in that: The calculation formula of the over-temperature fault confidence is: Confidence=ω1f K +ω2f R +ω3f ΔT +ω4f ρ (ω1+ω2+ω3+ω4=1) Among them, ω1, ω2, ω3, ω4 are weight coefficients, f K is the normalized value of spectral kurtosis; f R is the normalized value of the envelope spectrum amplitude ratio; f ΔT is the normalized value of temperature residual; f ρ is the normalized value of the cross-correlation coefficient.

7. The wind turbine gearbox output end bearing comprehensive diagnosis method according to claim 6, characterized in that: The causes of the temperature increase determined according to the over-temperature fault confidence level include: When Confidence>0.7, it is confirmed that the cause of the temperature increase is a failure of the high-speed shaft bearing; When 0.4<Confidence≤0.7, the cause of the temperature increase is confirmed to be a suspected failure of the high-speed shaft bearing, and the high-speed shaft bearing should be continuously tested; When Confidence ≤ 0.4, confirm that the temperature rise is not caused by mechanical failure, and check whether the unit has abnormal heat dissipation or operating failure.

Images