Optimization of a nearest neighbors classifier for diagnosis of condition of rolling bearings

Summary The paper concerns classification of technical condition state of rolling bearings. A methodology of optimization of a k-NN classifier with regard to selection of the symptom observation space has been proposed. The symptoms carrying the most information allowing identification of a class of technical condition were selected. The applied methodology enabled to develop a classifier which in the set of available data achieved the efficiency of 97.5%. In the set of considered symptoms the r.m.s. and peak values of vibration acceleration in the broad frequency band and the energy of acoustic emission pulses turned out to be the best for identification of arising fracture of a bearing outer ring.