Predicting system degradation using Bayesian time series models

Abstract

Efficient maintenance of industrial equipment requires degradation monitoring and prediction. Currently used prediction models are mostly deterministic and cannot consider uncertainty inherent to degradation measurements. In this paper we propose using time series models obtained using Facebook Prophet algorithm to predict the evolution of degradation of turbomachinery. We illustrate our considerations with data from large scale industrial centrifugal compressors. Our predictions are promising and confidence intervals cover the predictions well.