Parallel Autoregressive Modeling as a Tool for Diagnosing Localized Gear Tooth Faults

One of the standard approaches widely used in the field of localized gear tooth fault diagnosis is the creation of residual signals i.e. signals obtained after removing the deterministic frequency components from a Time Synchronously Averaged vibration signals. Most of the time these components are removed based on the knowledge of the characteristic gearbox frequencies. Sometimes however such information is not available. AR modeling, a type of time series modeling, has been found to be capable of faithfully estimating the deterministic content of the signal allowing meaningful residual signals to be created. An improvement to the classic AR modeling approach is proposed in this text. The method is applied to experimental data taken from a gearbox in both healthy and faulty condition. The improvement derived from the new method is quantified through a comparison with results obtained by applying Time Synchronous Averaging and the classic AR modeling method to the experimental data.