Rolling element bearings are essential components of most
machinery and widely used in various rotating machines.
Faulty bearings cause the majority of problems in rotating
machinery. Their operating conditions influence directly the
operation of the whole machinery and the production quality.
For both economic and safety reasons in recent years, bearing
condition monitoring and diagnosis have received further
attention especially in modem automatic equipment. The
typical fault of rolling element bearings is localized defect.
Many researches focus on this failure mode, because it
frequently brings catastrophic mischance or significant
damages. Localized defect occurs when a sizeable piece of
material on the conduct surface is dislodged during operation,
mostly by fatigue cracking under cyclic contact stressing [1-2].
The type of measurement performed approximately categorizes
the existing rolling element fault detection schemes. In these
schemes, because of the rapid development of signal
processing techniques and the moderate price and excellent
adaptability, vibration method is commonly used [1-4]. Time
domain analysis of rolling bearing faults is one of the simplest
and cheapest detection and diagnosis approaches, but it can
only detect the existence and severity of bearing defect, and has
not accurate diagnosis capability. In contrast to time domain
analysis, the frequency domain analysis method, which
includes the envelope analysis, can indicate the detailed
accurate diagnosis results and is commonly used in
engineering. Due to the periodic nature of phenomena
involved, time invariance and consequently, the notion of
stationary is violated in practice [4]. Although time-frequency
analysis method has been used in many explorations of rolling
The research is partially supported by: National Natural Science
Foundation (No. 50405023).