Component-level fault diagnostics of a bevel gear using a wavelet packet transform

Abstract

This article presents a novel diagnostic algorithm based on wavelet packet transform. Although existing fault diagnostic techniques may detect the occurrence of failure or excessive malfunction in a system, they rarely reveal such detailed information as fault position, not to mention their reduced effectiveness in a system whose fault feature is not so pronounced. The proposed algorithm takes advantage of time-frequency localized features of wavelet packets in order to classify the system condition and pinpoint the fault position. Fisher's linear discriminant is used to classify system conditions, all the wavelet packets are exhaustively searched in order to unravel the optimal base (or packet) that discriminates different system conditions based on the Fisher value. It is the time-frequency localized feature of the wavelet packet that further makes it possible to detect fault position and even faulty component in the bevel gear. The proposed algorithm is validated using a machine fault simulator that consists of a motor, a bevel gear, and a bearing, where the condition of the bevel gear may be varied to simulate normal, cracked, and broken conditions using vibration signal. It turns out that the proposed algorithm precisely locates fault position as well as identifies system conditions.