Gaussian processes in ball bearing prognostics

Abstract

In this work, vibration analysis and Gaussian Processes techniques are used in useful life prognostics of ball bearings. The database is provided by The Prognostics Data Repository from NASA, and shows the failure evolution in ball bearings. The data basis also provides training and validation data sets for ball bearing useful life prediction. Several time and frequency characteristics are extracted from ball bearing vibration signals for trending analysis, and finally one of these is taken as input for the Gaussian process and describe, with a probabilistic strategy, the failure evolution system. No dimensionality reduction algorithm is used in this paper, only the evaluation of trends in failure evolution is taken for decision. This data basis was used in 2012 IEEE classification contest. Several participants used classification techniques based on time-frequency transformation and Artificial Intelligence algorithms but none of them used Gaussian Processes in a classification scheme. Although, the present work does not have the best results in classification it does show a major simplicity in formulation and implementation than most of the classification schemes.

Keywords

• Mechanical vibrations
• ball bearings
• gaussian processes
• prognostics