Fault Detection in Gears Using Stochastic Resonance

Investigations carried out so far on the application of Stochastic Resonance (SR) to mechanical system faults indicate that SR shows great promise as an advanced vibration-based condition-monitoring tool. However, majority of these studies only focus on faulty systems and thus, fail to adequately treat healthy systems. It is a well-known fact that some methodologies for fault detection give off false alarms when applied to a healthy system. With a view to addressing this problem, efforts are continuously made to either modify these methodologies or develop other methodologies that are more advanced. In addition to experimentally validating the use of SR as a vibration based condition-monitoring procedure, this paper attempts to address the issue of false alarms associated with SR and experimental data complexity by applying SR to pre-processed signals and raw signals, and comparing their results. The pre-processed signal could be either a residual signal, which is obtained by removing selected frequencies from the Time Synchronous Average (TSA) signal, or filtered signal, which is acquired by passing the raw signal through a high-pass filter with proper cut-off frequency. Furthermore, it is shown that kurtosis and other statistical features can be used as fault indicators when SR is applied to a signal.