Complexity as a measure for machine fault detection and diagnosis

Conventional techniques used for the detection and diagnosis of machine defects, such as spectral analysis and time-frequency analysis, are based on the assumption that a physical system possesses linear transfer functions. However, these techniques cannot truthfully identify fault features when the actual behavior of the physical system is far from linear due to the change of its operating conditions, and involves nonlinearity. This paper presents a nonlinear dynamics method called complexity, which has been investigated to extract feature parameters from raw vibration signals measured from a bearing system. The results demonstrated that complexity presents a good measure for detecting machine defects.