Application of the intrinsic time-scale decomposition method to fault diagnosis of wind turbine bearing

Abstract

A fault diagnosis method of wind turbine bearing based on intrinsic time-scale decomposition (ITD) is put forward. In the proposed method, the vibration signal of the main bearing is decomposed into several proper rotation components by the ITD method. The frequency centers of the proper rotation components that contain predominant energy are computed and considered as fault feature vectors. The nearest neighbor algorithm is applied to identify the fault types of the wind turbine bearing. The experimental data of the wind turbine spherical roller bearing in four conditions (normal, outer race fault, inner race fault and roller fault) are applied to evaluate the performance of the proposed method. The results demonstrate the feasibility and accuracy of this approach for the diagnosis of the wind turbine bearing faults under uncertain conditions.

Keywords

Fault diagnosis frequency center intrinsic time-scale decomposition nearest neighbor algorithm spherical roller bearing wind turbine

1. Introduction

The development of renewable energy has attracted widespread attention due to the energy crisis and environmental issues. Renewable energy development is booming worldwide. In particular, the exploitation of wind power has become very rapid, and plays an important role in the energy structure. The wind turbine faults often occur because most wind farms are in poor areas of the environment. As a result, it is important to monitor the condition and detect the fault of wind turbines to ensure their safety and stability (Amirat et al., 2009; An and Jiang, 2010a , b; Barszc and Randall, 2009; Hameed et al., 2009). The bearing faults make up a high percentage of rotating machinery failures (Guo et al., 2009; Soylemezoglu et al., 2010; Trendafilova, 2010), especially for wind turbines that operate in bad weather conditions (storm, ice, high tides, etc.) (Zhao, 2010). The vibration signal of a roller bearing fault is non-stationary and nonlinear due to the instability of wind conditions and the complexity of roller bearing faults. So it is significant to choose appropriate signal processing and fault diagnosis methods to analyze wind turbine faults.

The intrinsic time-scale decomposition (ITD) method can decompose a complex signal into several proper rotation components and extract the dynamic features of non-stationary signals accurately (Frei and Osorio, 2007). It has high decomposition efficiency and frequency resolution. It is appropriate for dealing with the complex and non-stationary signals, and is promising for the application to the real-time data analysis. Thus, a fault diagnosis model of wind turbine spherical roller bearing using ITD is proposed.

2. Intrinsic Time-Scale Decomposition Method

The ITD method decomposes a vibration signal into a series of proper rotation components and a trend component. The instantaneous amplitude and instantaneous frequency of the proper rotation components are analyzed in frequency spectrum. Through analyzing the spectrum, the amplitude modulation and frequency modulation of the vibration signal can be derived, respectively. For the signal X_t, the ξ is defined as the baseline extraction operator. After extracting a baseline from the signal X_t, the rest signal becomes an inherent rotation. The first decomposition of the X_t as follows (Frei and Osorio, 2007):

(1)

whese L_t = ξ Xt, is the baseline signal, H_t = (1 - ξ) X_t, is the proper rotation component.

Supposing {τ_k, k = 1, 2, … } are the local extrema values of the X_t, let τ₀ = 0. To simplify the symbol, the X(τ_k) and L(τ_k) are expressed by using X_k and L_k, respectively. Based on the assumption that the L_t and H_t have been defined in [0, τ_k], and the X_t is also defined in t∈[0, τ_{k + 2}]. The subsection linear baseline extraction operator, ξ, is defined in contiguous extrema interval (τ_k, τ_{k + 1}]:

(2)

In equation (2), the L_{k + 1} can be expressed as

(3)

where α∈(0, 1), usually α = 0.5. A decomposition can get a baseline signal and a proper rotation component, the latter denotes the local relative high frequency of the decomposed signal. The baseline signal is regarded as the input signal, the decomposition continues till a monotone signal is obtained. Using this method, the original signal is decomposed into several proper rotation components whose frequency range is from high to low and a monotone trend component. The whole process can be expressed as:

(4)

where ψξ^k X_t is the k + 1 layer of the proper rotation, ξ ^p X_t is either the monotonic trend or the lowest frequency baseline. The baseline is extracted if stop decomposing the signal before the monotone trend is obtained.

3. Fault Diagnosis of Wind Turbine Bearing Using Itd Method

The decomposed proper rotation components using the ITD method have certain physical meanings. They can effectively reflect the characteristics of the original signal, and are suitable for analyzing the amplitude modulation and frequency modulation signals. In this paper, the vibration signal of a spherical roller bearing fault in a wind turbine is decomposed using the ITD algorithm. The frequency centers of proper rotation components that contain predominant energy are calculated and regarded as bearing fault feature vectors of the wind turbine. In order to detect and locate bearing faults, the nearest neighbor algorithm is adopted. The procedure of the proposed fault diagnosis method can be summarized as follows:

Four experiments (spherical roller bearing normal, outer race fault, inner race fault and roller fault) of a direct-drive wind turbine are carried out. The main shaft of the wind turbine is supported by two spherical roller bearings. The bearing with artificially induced defects is installed closer to the wind wheel. A total of 4N samples are obtained with N samples for each condition.

Using the ITD method, each sample signal can be decomposed into a series of proper rotation components, c₁, c₂, …, c_n, with different scales.

The frequency center u_i(i = 1, 2, 3,…, m) of predominant energy proper rotation components are calculated, the U = {u₁, u₂, u₃, …, u_m} are regarded as bearing fault feature vectors. The frequency center u is defined as:

(5)

where the s(k) is a frequency spectrum of vibration signal series, for k = 1,…, K, K is the number of spectrum lines, f_k is the frequency value of the kth spectrum line.

From the 4N datasets, L samples are used to train the nearest neighbor classifier (Duda et al., 2000), and the remaining 4N-L samples are presented to estimate the performance of the proposed method.

4. Experiment and Results

Considering that the bearing faults occur in outer race, inner race or roller, four experiments with spherical roller bearing (22206-type) – normal operation, outer race fault, inner race fault and roller fault are carried out. The direct-drive wind turbine test rig is shown in Figure 1. The bearing faults are set by grooving using a linear cutting method. The sizes of the outer race defect, the inner race defect and the roller defect are the same, with 0.2 mm width and 0.3 mm depth, as shown in Figure 2. The defect bearings are installed in the near wind wheel. The experimental rotating frequency of wind turbine is 4.25 Hz. The sampling rate is 2000 Hz, and sampling length is 8192 for four conditions.

Figure 1.

Test stand of direct-drive wind turbine.

Figure 2.

View of spherical roller bearing in good condition and with three faults.

The vibration acceleration signals of wind turbine bearing in four conditions (normal, outer race fault, inner race fault and roller fault) are acquired. Each condition has 24 samples with a total of 24 × 4 samples. The ITD method is used to decompose these samples of different conditions and the proper rotation components of each sample can be obtained. Figure 3 shows the time domain waveform and the ITD decomposed results of main bearing vibration acceleration signals with an inner race fault. It can be seen from the figure that the ITD method can decompose a vibration signal into a series of proper rotation components of different scales. The energies of the first four proper rotation components are predominant. The signals of bearing normal, outer race fault and roller fault are similar to the inner race fault. Figure 4 displays the frequency spectrum of the first four components c₁, c₂, c₃ and c₄ in four conditions, respectively. From Figure 4, it is easily seen that the decomposed components have high frequency resolution.

Figure 3.

The time domain waveform and intrinsic time-scale decomposition (ITD) decomposed results of main bearing vibration acceleration signals with inner race fault.

Figure 4.

The spectrum of the proper rotation components c₁, c₂, c₃ and c₄ of acceleration signals in four conditions.

The feature vectors of four conditions can be obtained by the procedure defined in Section 3. The energies of vibration acceleration signals of wind turbine spherical roller bearings can be intently represented with the first four components in four conditions. Therefore, the first four components based on the ITD method of four conditions are chosen to extract the features. In this paper, the frequency centers of the first four components are computed respectively, and are regarded as the feature vectors of the vibration signal. The feature vector is 4 × 1 = 4 dimension. Table 1 shows the feature vectors of four conditions (for the sake of brevity, only seven vectors for each condition are listed). In Table 1, the u₁, u₂, u₃ and u₄ are the frequency center of the first four proper rotation components, c₁, c₂, c₃ and c₄, respectively. In this paper, 24 × 4 samples are acquired from the wind turbine test rig as shown in Figure 1. These samples cover four patterns with bearing working normally, with an outer race fault, with an inner race fault and with a roller fault. Each pattern has 24 samples. For these 24 × 4 samples, 32 datasets (eight sets for each condition) are selected as training datasets randomly and the remaining 64 datasets (16 sets for each condition) are tested. It can be found that the identification results of only four test samples are not consistent with the real-world condition; the fault diagnosis accuracy of the proposed method is 93.75%.

Table 1.

Main bearing fault features of wind turbine based on intrinsic time-scale decomposition

No.	Main bearing condition	Main bearing fault feature
No.	Main bearing condition	u ₁	u ₂	u ₃	u ₄
1	Normal	516.3	286.3	209.1	128.4
2		513	278	206.7	134.2
3		513.9	289.7	204.8	127.5
4		519.2	291.8	205.1	132.5
5		517	278.8	208.8	130.3
6		519.8	280.9	206.6	129.8
7		514.7	284.3	203.8	132.1
8	Outer race fault	555.5	294.3	204.1	130.9
9		561.2	290.2	212.1	129.4
10		550.2	302.5	213.1	136.5
11		559.1	291	211.2	133.5
12		565.7	276.3	203	127.7
13		560.8	286.4	201.6	128.2
14		535.6	289.2	204.7	129
15	Inner race fault	594.1	301.2	209.3	139.4
16		591.4	305.7	210.2	140
17		590.4	299.3	208.7	134.5
18		584.2	297.1	210.1	139.8
19		592.2	299	209.4	137.7
20		594.4	298.7	210.6	136.4
21		590.3	305.7	209.3	139.1
22	Roller fault	497.2	269.2	204.8	125.2
23		498.6	273.9	202.5	126.3
24		509.3	276.7	207	131.7
25		496.1	270.4	203.5	127
26		499.9	272.2	204.8	138.2
27		500.7	277.8	207.3	129.5
28		495.7	271.4	202.4	127.5

5. Conclusions

The vibration signal of a wind turbine bearing fault is strongly non-stationary and nonlinear due to the instability of wind conditions and the complexity of spherical roller bearing faults. The intrinsic time-scale decomposition method is a new, suitable method for the analysis of non-stationary and nonlinear signals. It can accurately express the dynamic characteristics of non-stationary signals with high time and frequency resolution. Therefore, in this paper, the ITD is applied to decompose the vibration signal into several proper rotation components.The frequency centers of the first few primary proper rotation components are computed as feature vectors of bearing fault diagnosis. The nearest neighbor algorithm is used to identify the operation condition of wind turbine bearings based on these feature vectors. Experimental results show that the proposed ITD-based method has excellent performance for wind turbine bearing fault diagnosis.

Footnotes

Funding

This work is supported by the National Basic Research Program (973 Program) (No. 2007CB210304) and China Postdoctoral Science Foundation funded project (No. 20090460273).

References

Amirat

Benbouzid

Al-Ahmar

Bensaker

Turri

(2009) A brief status on condition monitoring and fault diagnosis in wind energy conversion systems. Renewable and Sustainable Energy Reviews 13(9): 2629–2636.

Jiang

. Correlation analysis of oil temperature trend for wind turbine gearbox. Proceedings of the ASME 2010 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference, (IDETC/CIE 2010), (2010a) August 15–18, Montreal, Quebec, Canada.

Jiang

(2010b) Chaotic characteristics identification and trend prediction of running state for wind turbine. Electric Power Automation Equipment 30(3): 15–19.

Barszc

Randall

(2009) Application of spectral kurtosis for detection of a tooth crack in the planetary gear of a wind turbine. Mechanical Systems and Signal Processing 23(4): 1352–1365.

Duda

Hart

Stork

(2000) Pattern Classification 2 New Yorki: Wiley-Interscience.

Frei

Osorio

(2007) Intrinsic time-scale decomposition: time–frequency–energy analysis and real-time filtering of non-stationary signals. Proceedings of the Royal Society A 463(2078): 321–342.

Guo

Chen

(2009) Rolling bearing fault classification based on envelope spectrum and support vector machine. Journal of Vibration and Control 15(9): 1349–1363.

Hameed

Hong

Cho

Ahn

Song

(2009) Condition monitoring and fault detection of wind turbines and related algorithms: a review. Renewable and Sustainable Energy Reviews 13(1): 1–39.

Soylemezoglu

Jagannathan

Saygin

(2010) Mahalanobis Taguchi System (MTS) as a prognostics tool for rolling element bearing failures. ASME, Journal of Manufacturing Science and Engineering 132(5): 1–12.

10.

Trendafilova

(2010) An automated procedure for detection and identification of ball bearing damage using multivariate statistics and pattern recognition. Mechanical Systems and Signal Processing 24(6): 1858–1869.

11.

Zhao

. Fault feature analysis and experimental investigation for wind turbine. Thesis for the masters of engineering degree(2010)Beijing: Tsinghua University.