Identification of rolling bearing condition by means of a classification tree

The paper deals with the problem of evaluation of technical condition of rolling bearings on the basis of synchronously measured vibroacoustic symptoms and temperature. Rolling bearings were subjected to accelerated wear under controlled conditions. The values recorded in the study were sound pressure in a broad band including ultrasound (band up to 40 kHz), vibration acceleration in a radial direction, ultrasound in a band up to 100 kHz (processed into audible band), and bearing housing temperature. The identification of the condition was carried out with the help of a supervised learning system. Two conditions were distinguished: fit - examples were obtained in the initial phase of bearing operation in temperature stability conditions, and pre-failure - examples were obtained from fragments of recording just before the occurrence of bearing failure. The CART (Classification and Regression Tree) binary tree method was used to determine the technical condition and significance of particular diagnostic symptoms.