A Comparison of Stationary and Non-Stationary Transforms for Early Fault Detection in the OH-58A Main Transmission

A comparison between stationary statistical fault indicators and the discrete harmonic wavelet packet transform (DHWPT) is investigated for accelerometer vibration data obtained from the OH-58A main rotor transmission. This transmission is comprised of two stages, a spiral bevel input stage and a planetary stage, resulting in an overall reduction of 17.44:1. Three loaded transmission test runs are used to evaluate the sensitivity of previously developed stationary fault metrics and a new non-stationary fault metric for early fault detection. The non-stationary fault indicator consists of a normalized energy metric which was developed to quantify the redistribution of sideband energy about the dominant mesh frequency and its harmonics in the presence of actual gearbox faults. This metric is used to qualitatively assess the presence, type and location of gearbox faults. Gearbox faults considered in this study include, sun gear spalling, spiral bevel gear scoring, planet bearing wear and and a top housing cover crack. Results from the comparison suggest that the non-stationary normalized energy metric is capable of detecting damage at a rate that is comparable to the stationary statistical fault indicators. In addition, the normalized energy metric provides more information about the nature of a fault.