Automatic Classification of Gear-Units by Neural Network Technologies

Abstract

Noise emitted during test runs of mitre-gear units was analyzed using a neural network based tool. Gears can be classified dependent on how noise levels vary with changes of revolution speed. Faults can be detected and necessary adjustments can be identified. The kernel of the work is the appropriate selection, reduction and preprocessing of input data, parameterization of the clustering (unsupervised learning) algorithms, and the building of the classification model (supervised learning).