Proceedings of the 11th International Conference on Frontiers of Design and Manufacturing
May 23~25, 2014, Nanjing, China
A comparison research on EMD and Wigner-Ville distribution for fault diagnosis of roller
bearings
Lin Li, Yixiang Huang, Cheng-liang Liu,Xiao Ling, Shuang-yuan Wang
School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
E-mail: Linli-sjtu@sjtu.edu.cn Huang.yixiang@sjtu.edu.cn
Abstract: Roller bearings are the most important parts in various
mechanical systems. Both Empirical Mode Decomposition
(EMD) method and Wigner-Ville distribution can be applied to
fault diagnosis for roller bearings. This research compares these
two methods based on experiments of fault diagnosis, and find
that EMD method could more effectively detect the defect of
roller bearings from signals with noise and identify the faulty
parts of the roller bearings, while in Wigner-Ville distribution,
severe cross-term interference might exist, which blurs the faulty
signals.
Key words: Roller bearings, Fault diagnosis, Empirical Mode
Decomposition (EMD), Wigner-Ville distribution.
1. Introduction
The failure of roller bearings in mechanical
systems could lead to series problems and cause large
economic loss. Therefore fault diagnosis for roller
bearings is one of the most important topics in machine
condition monitoring
[1]. The key procedure for
mechanical fault diagnosis is fault identification [2].
Fault diagnosis can be categorized into time domain
and frequency domain methods. For time domain
methods, the effective values of vibration signals for
bearings reflect the energy of the vibration. When
bearings break down, the amplitude of vibration will
increase. Then the effective values can be a criterion for
bearing faults [3]. For frequency domain methods, fast
Fourier transform (FFT) is a traditional and effective
method. But FFT method is suitable for global analysis.
As the vibration signals for the bearings usually change
with time, the analysis result of FFT is not accurate
enough for bearing fault diagnosis [4]. Thus neither
time domain nor frequency domain methods can
achieve satisfactory analysis result for fault diagnosis.
To solve the above problem, time-frequency
analysis methods are proposed, such as Short Time
Fourier Transform (STFT), Wigner-Ville distribution
and Wavelet Transformation (WT) [5]. Based on the
Heisenberg theory
[6], the product of time duration and
bandwidth is a constant to a given signal. Thus STFT
can hardly get the best time and frequency resolution at
the same time. Wigner-Ville distribution is another
method within the scope of time-frequency domain. It
is a quadric form distribution, and the product of time
duration and bandwidth can reach the prescribed limit
of the Heisenberg theory. So Wigner-Ville distribution
has a good time and frequency resolution.
Most of the methods mentioned above are suitable
for linear and stationary signal, but not for nonlinear or
non-stationary signals. With the development of
technology, more efficient and accurate methods for
fault diagnosis are put forward, and Empirical Mode
Decomposition (EMD) is one of the most powerful
time-frequency techniques [7]. This method was firstly
proposed by N. E. Huang etc. in 1998 [8], since then, it
has been studied in depth and applied in such many
fields as process control, modeling, surface engineering,
medicine and biology[9]. EMD method decomposes the
signal into a set of complete and almost orthogonal
components which are called intrinsic mode functions
(IMFs) [8]. Since EMD method is a self-adaptive
signal processing technique because the IMFs indicate
the natural oscillatory modes imbedded in the signal
and serve as the basis functions which are determined
by the signal itself rather than pre-determined kernels
that it is suitable for nonlinear and non-stationary signal
processes [10].
This paper aims at comparing EMD method and
Wigner-Ville distribution on fault diagnosis for roller
bearings and it is organized as follows. In Section 1, it
introduces some common methods for fault diagnosis.
Section 2 compares the theory of these two methods.
The comparison of the analysis result to the fault
diagnosis of roller bearings is in Section 3. Then
Section 4 concludes the compared results.