Gearbox degradation identification using pattern recognition techniques

Gear stiffness degrades over the life of a gearbox. In this paper stiffness degradation is identified using pattern classification techniques that rely on the spectral content of the vibration induced during the operation of the gearbox. In particular, the k-nearest-neighbor algorithm, as well as a novel neural network classifier was deployed to address this issue. The classification process was generally able to classify early signs of stiffness degradation. It was found, however, that multiple networks are essential to classification in regions of practical concern. To this end selection of features and clear understanding of the disparity among them play key roles. It was further determined that noise attenuation must be incorporated into the process for the results to be reliable. Finally, the effects of initial conditions must be well understood in order for the diagnostic process to produce reliable conclusions.