Rule-based Identification of Bearing Faults using Central Tendency of Time Domain Features

Abstract

Vibration-based time domain features (TDFs) are commonly used to recognize patterns of machinery faults. This study exploits central tendency (CT) of TDFs to develop a Rule-based Diagnostic Scheme (RDS), which identifies localized faults in ball bearing. The RDS offers an accurate and efficient diagnostic procedure, and purges the requirement of expensive training of conventional classifier. A test rig is used to acquire vibration data from bearings having localized faults, and various TDFs are extracted. It is worth mentioning that fluctuations in random vibration signals may alter the feature values. Therefore, each of the TDFs is processed statistically to approximate its reliable central values (CVs) against the respective faults. In this way, every feature provides a set of CVs, which are equal in number to that of faults. Separating distances among normalized CVs (NCVs) in a set provide the criteria to select or discard that particular feature before further processing. The selected sets of NCVs are finally used as references to generate rule-set for testing the unknown vibration samples. The results are evident that the proposed RDS may be an effective alternative to the existing classifier-based fault diagnosis, even if the vibration signals are contaminated with considerable background noise.