Statistical Approach to Diagnostic Rules for Various Malfunctions of Journal Bearing System Using Fisher Discriminant Analysis

Abstract

This research is focused on developing an efficient fault diagnosis procedure for a journal bearing system. Vibration data of journal bearing rotor simulator under four conditions (i.e. a normal condition and three anomaly conditions including unbalance, rubbing and misalignment) was used to develop the algorithm. In order to improve diagnostic performance, cycle based time-domain features and frequency-domain features were extracted after resampling process being applied to the raw vibration data. Then, the optimal feature selection was accomplished by mixture of random combination performance test and Fisher Discrimin- ant Ratio (FDR) criteria. After selecting optimal features, Fisher Discriminant Analysis (FDA) algorithm classified each abnormal conditions mentioned above. To end with, the result of classification is evaluated and verified