Gears are common power transmission elements and are frequently responsible for transmission failures. Since a tooth crack is not directly measurable while a gear is in operation, one has to develop an indirect method to estimate its size from some measurables. This study developed such a method to estimate the size of a tooth transverse crack for a spur gear in operation. Using gear vibrations measured from an actual gear accelerated test, this study examined existing gear condition indices to identify those correlated well to crack size and established their utility for crack size estimation through index fusion using a neural network. When tested with vibrations measured from another accelerated test, the method had an averaged estimation error of about 5%.

Gears are common mechanical components used in power transmissions and frequently responsible for transmission failures. The aim of this study is to establish a gear crack prognostic methodology to predict the residual life of a cracked spur gear by integrating a fracture mechanics-based failure model, a gear dynamic simulator, and an existing gear crack diagnostic algorithm that employs an artificial neural network to estimate crack size from measured gear vibration by fusing a number of selected gear condition indices. The estimated crack size and predicted gear residual life were validated with experimental data. Experimental results showed that the method had an averaged error of 12.94 per cent in its prediction of residual life.