Health Assessment and Fault Classification of Roller Element Bearings

Report No. ARL-TR-6080
Authors: Andrew J. Bayba, David N. Siegel, Kwok Tom, and Derwin Washington
Date/Pages: July 2012; 24 pages
Abstract: Feature extraction, health assessment, and fault classification algorithms were evaluated for ball bearings with three different fault types and multiple levels of damage. Data was analyzed for five healthy bearings, and seeded fault bearings with five levels of damage for each fault type (ball fault, inner race fault, and outer race fault). A variety of fault analysis techniques were used to calculate properties (features) of the data sets, which were then fused together to form the best feature sets for fault evaluation. Self-organizing maps were used for health assessment, and a Naïve Bayes classifier was used to determine fault type. The results indicate a very good distinction between healthy and faulted bearings, and a good classification of fault types. For health assessment, there were good general trends with increasing damage. There was, however, a significant amount of scatter, thereby making it difficult to ascertain the precise health of an individual bearing. Although our feature set is substantial, it is by no means exhaustive, and one consideration is to seek additional features that may produce a higher level of confidence in individual bearing health.
Distribution: Approved for public release
  Download Report ( 0.470 MBytes )
If you are visually impaired or need a physical copy of this report, please visit and contact DTIC.

Last Update / Reviewed: July 1, 2012