Detrended fluctuation analysis for gear fault diagnosis

Abstract

The paper presented a method of detecting long - range correlation in a signal-detrended fluctuation analysis (DFA) and applied it to diagnose gear faults. Two key problems, i. e. scaling range and polynomial order of fitted trend, were discussed. Scaling exponent was extracted to be as a detector for diagnosing five types of gear faults, their were normal, spalling, tooth broken, wearing and cracking. The correct diagnosis ratio is large than 96%. Further, compared with the kurtosis detector I a traditional statistic parameter, the results demonstrate DFA is an effective method in machinery fault diagnosis.