Continuous Time Bayesian Networks in Prognosis and Health Management of Centrifugal Pumps

Abstract

This paper presents a novel method for performing risk-based prognosis and health management (rPHM) on centrifugal pumps. We present the rPHM framework and apply common modeling tools used in reliability and testability analysis---dependency (D) matrices and fault tree analysis---as a basis for constructing an underlying predictive model. We then introduce the mathematics of the Continuous Time Bayesian Network (CTBN), which is a probabilistic graphical model based on a factored Markov process that is designed to capture system evolution through time, and we explain how to apply a CTBN derived from D-matrices and fault trees to consider the impact of a set of faults common to centrifugal pumps on emerging hazards in the pump system. We demonstrate the utility of using CTBNs for rPHM analysis with two experiments showing the descriptive power of our modeling approach.

Work-in-progress: Reliability prediction of API centrifugal pumps using survival analysis

“…On the other hand, Li X. et al in [19] analyse the prognosis techniques of rotating equipment using system-level models, the purpose in this case is the prediction of the RUL at a system level. Examples of proposals in this direction, also applied to centrifugal pumps, can be found in [20][21][22][23][24][25][26][27][28]. Finally, a review of the statistical methods applied to the estimation of RUL of assets is presented in [29], which is a good guide to know the different existing methods and the literature published on this subject.…”