A framework for asset prognostics from fleet data

Abstract

Prognostics of a specific asset based on data from a fleet of same assets, but operated in different environmental and operational conditions is an important and common problem in Prognostics and Health Management (PHM). Traditional data-driven models trained on all fleet data provide only a general degradation trend, without capturing the specificity of the degradation process of the different assets. A two-step data-driven framework is here proposed to tackle this problem. A general model is trained traditionally on all fleet data and a correction model is built to estimate the deviation of the general model outcome from the degradation process of the specific asset of interest. The proposed framework is tested on a case study concerning the failure of a pneumatic valve in a nuclear power plant. The experimental results show the effectiveness of the proposed two-step, data-driven framework.