Multi-fault detection and failure analysis of wind turbine gearbox using complex wavelet transform
Introduction
Wind energy becomes an attractive power generation form among the renewable energies because of its cleanness and reproducibility. In 2013, the accumulated installed capacity in China was 91412.9 MW, which still ranks at the top of the world [1]. Behind the vigorous development, wind turbines are prone to be failure due to harsh operation conditions, which cause unexpected downtime and economic loss. Wind turbine gearbox transfers mechanical energy from the rotor hub with low rotational speed to the generator with high speed, meanwhile suffers alternating loads from varying wind speed and transient impulses of frequent brakes. Therefore, wind turbine gearbox is one of the most fragile parts in wind turbine [2], [3].
Many researches were done to monitor condition and diagnose faults for wind turbine gearbox, in order to detect defective gears or bearings incipiently and form a predictive maintenance system [4], [5], [6], [7]. Generally, methods of condition monitoring and fault diagnosis can be classified into two approaches: model based using supervisory control and data acquisition (SCADA) and vibration signals based using condition monitoring system (CMS). In model based approach, Kusiak applied various data-mining algorithms to develop models predicting possible faults of wind turbine [8]. Zhang and Verma used the derivative of the acceleration signals as feature patterns and analyzed the correlation coefficient between multi-channel signals to recognize gearbox fault [9], [10]. Yang and Court [11] developed an effective method to process raw SCADA data, and realized the quantitative assessment of the health condition of a turbine under varying operational conditions. Model-based approaches can detect faults in gearbox, even reflect fault deterioration tendency. However, they cannot localize concrete faults accurately because of the long sample intervals of SCADA.
CMS provides abundant vibration signals for people to monitor the health condition of wind turbine gearbox [12], [13]. On the basis of vibration analysis, Lei et al. [14] applied an adaptive stochastic resonance (SR) method to choose the sensitive measurement locations and found weak fault feature of planetary gearbox. Li and Chen [15] proposed a noise-controlled second-order enhanced SR model to enhance the weak feature that exhibited the advantage in detecting coupling looseness. Empirical mode decomposition was adopted to detect multi-harmonic components representing gear pitting faults intelligently [16]. Feng et al. focused on the fault feature representation of planetary gearbox, and presented energy separation and iterative generalized synchrosqueezing transform to extract planetary gearbox faults with considering the varying operation speed, which obtained good results [17], [18].
Wind turbine gearbox consists of lots of parts such as gears and bearings. The vibration energy is dominated by the mesh of the high speed gear pair. So when bearing defects arise, the corresponding fault features are inevitably concealed by intensive gear meshing energy, especially when there are also defects existing in gears. Many approaches were studied to diagnosis multiple faults in rotating machinery. To overcome the fact that the signature of a defective bearing is spread across a wide frequency band, discrete wavelet transform was used to decompose the vibration signals considering single and multiple point defects on inner race, outer race, ball fault and combination of these faults [19]. In order to detect multiple defects on one component of the bearing, Mohammadi and Safizadeh [20] presented a new method based on the high frequency resonance technique, and adopted a time constant in the envelope detector to find the pattern of the amplitude of defect frequency harmonics in the frequency domain. Ensemble empirical mode decomposition (EEMD) can decompose vibration signal into a collection of intrinsic mode functions adaptively by adding noise to the original signal and calculating the means of IMFs repeatedly, which can be referred as a powerful tool to diagnosis multiple faults. Jiang combined improved EEMD and multiwavelet packet to identify weak multi-fault features at early stage in rotating machinery [21]. Wang and Han [22] decomposed multi-fault signals from wind turbine gearbox twice using EEMD, and applied cyclic autocorrelation to each IMF to obtain single fault feature. Based on the Hermite splines interpolation, an adaptive redundant lifting multiwavelet was developed by Chen and Zi taking the minimum envelope spectrum entropy as the optimization objective [23], which can detect compound faults at one time.
The aforementioned methods can help people to diagnose gear or bearing faults conveniently. However, no one can always exceed others in any case. In this paper, the vibration signals originated from a real multi-fault wind turbine gearbox with catastrophic failure are analyzed. Firstly, the conventional narrow-band filtering and Hilbert transform are applied to exact conspicuous fault features (gear fault), then the cepstrum method is adopted to enhance the identification accuracy of multiple frequency components. Next, the complex Gaussian wavelet is introduced to decompose the vibration signals at different scales and the multi-scale enveloping spectrogram (MuSEnS) is obtained simultaneously. At some scale slice of the MuSEnS, the weak features (bearing fault) buried in intensive vibration energy emerge distinctly. Finally, the failure mechanism with multiple faults in wind turbine gearbox is discussed to provide guidelines for the design, manufacture and field installation.
Section snippets
Demodulation analysis using Hilbert transform
Gear or bearing fault can cause carrier frequency convoluting with modulating frequency in vibration signals. Demodulation analysis is an effective method to separate modulating components from original signals. There are many demodulation methods applied to rotating machinery fault diagnosis, such as generalized detection-filtering demodulation, cyclic autocorrelation demodulation and Hilbert transform [24], [25] et al.
Hilbert transform is one of the most common demodulation methods, and the
Drive train of wind turbine
Typical drive train of wind turbine is shown in Fig. 2, which consists of rotor hub, main shaft (bearings), gearbox and generator. Rotor hub with blades transforms the wind energy to rotating mechanical energy, then the mechanical energy is transferred from main shaft to generator via gearbox with a transmission ratio of dozens even one hundred. The generator converts the mechanical energy with high speed to power energy. The blades and rotor hub are supported by main bearings.
To monitor the
Testing and fault analysis
The rated power of the tested wind turbine is 1.5 MW, the rated rotational speed of generator is 1800 rpm, and the total transmission ratio of the gearbox is 98.26. The numbers of gear teeth are shown in Table 1.
The CMS used in this test is offline. The vibration signals were measured one point after another on the surface of the gearbox, shown as in Fig. 2. The test was carried out under normal rotational speed of generator. According to the Eq. (11), Eq. (12) and Eq. (13), the multi-stage
Conclusion
A real wind turbine gearbox with multiple faults in the HSS has been analyzed via vibration signals. The signal is processed using conventional narrow-band filtering and Hilbert transform, and the broken teeth fault of gear pair in the HSS is detected distinctly. The cepstrum analysis is adopted to distinguish the approximate frequency components. However, the fault features of HSS rear bearing cannot emerge in the demodulation and cepstrum analysis. Using the complex Gaussian wavelet
Acknowledgment
The research presented in this paper was supported by National Natural Science Foundation of China (No. 51305135), Beijing Higher Education Young Elite Teacher Project (No. YETP0701), the Fundamental Research Funds for the Central Universities of China (No. 2015ZD15), Technical Project from China Huaneng Corporation (HNKJ13-H20-05) and Science and Technology Plan Projects of Hebei (15214307D).
References (33)
- et al.
Status and problems of wind turbine structural health monitoring techniques in China
Renew. Energy
(2010) - et al.
Condition based maintenance optimization for wind power generation systems under continuous monitoring
Renew. Energy
(2011) - et al.
The prediction and diagnosis of wind turbine faults
Renew. Energy
(2011) - et al.
Wind turbine condition monitoring by the approach of SCADA data analysis
Renew. Energy
(2013) - et al.
Planetary gearbox fault diagnosis using an adaptive stochastic resonance method
Mech. Syst. Signal Process.
(2013) - et al.
A new noise-controlled second-order enhanced stochastic resonance method with its application in wind turbine drivetrain fault diagnosis
Renew. Energy
(2013) - et al.
Fault diagnosis for wind turbine planetary gearboxes via demodulation analysis based on ensemble empirical mode decomposition and energy separation
Renew. Energy
(2012) - et al.
Iterative generalized synchrosqueezing transform for fault diagnosis of wind turbine planetary gearbox under nonstationary conditions
Mech. Syst. Signal Process.
(2015) - et al.
An improved EEMD with multiwavelet packet for rotating machinery multi-fault diagnosis
Mech. Syst. Signal Process.
(2013) - et al.
A novel procedure for diagnosing multiple faults in rotating machinery
ISA Trans.
(2015)