Gear failure prediction using multiscale local statistics

In this work, multiscale local statistics are introduced as a tool for gear failure prediction. Experimental data from gears with localized defects in the form of bending fatigue cracks were analysed. It was shown that second order central moments increased with defect magnitude. The estimation procedure was carried out in the range of scales where the defect is more prominent. An empirical law that relates variance at various scales to crack magnitude was established. Compared to the wavelet transform, the multiscale statistical scheme is an attractive option due to its low complexity, high sensitivity and inherent robustness.