Steps toward fault prognostics of gears

Condition-based maintenance requires the estimation of the remaining useful life of the system, i.e. prediction of the remaining useful life of each component. While the existing published studies suggest various signal processing methods to isolate and identify gear faults mainly based on synchronous average, these methods, in some cases, are not able to accentuate the effects of the fault and therefore do not provide enough information for assessing the type and severity of the damage as required for prognostics. The current research is aimed at improving signal processing algorithms, which are efficient at separating the effects of local faults in gears. The approach utilizes a dynamic model of the gear system. The dynamic model can simulate the vibration signature of healthy and damaged gears with different types and sizes of faults, and different levels of unbalance, misalignment, eccentricity, teeth irregularities and backlash. The insights obtained from the dynamic model simulations were used to develop and validate the improved signal processing algorithms. The capability to extract more accurate information from the signal contributes to isolating the effects of local faults. This paper presents two algorithms: a modified synchronous average and an adaptive difference. The modified synchronous average is designed to overcome the sensitivity of the regular synchronous average to cumulative phase errors. The adaptive difference is a signal derived from the synchronous average and it is designed to separate effects of local faults in one tooth of the gear from the dominating effects of distributed faults. The algorithms' efficacies were examined and demonstrated on simulated signals generated using the dynamic model.