Collaborative prognosis using a Weibull statistical hierarchical model

Abstract

Collaborative prognosis is a concept that aims at identifying sub-fleets of similarly deteriorating assets and enable learning across these sub-fleets to improve the overall prognosis performance. This paper presents a Weibull statistical hierarchical model to achieve collaborative prognosis by modelling the times-to-failures of sub-fleets, or clusters, of similarly deteriorating assets comprising a fleet. The procedure for real-time collaborative prognosis using the proposed hierarchical Weibull model is described, and demonstrated for a fleet of simulated turbofans. The experiments presented herewith analyse the prediction accuracy along the life of an asset and the effect of clustering on the prediction accuracy. More specifically, a large cluster comprising of similarly operating assets shall enable accurate predictions with high confidence and vice versa. The results show that a hierarchical model mitigates the problem of high variance encountered while independently modelling the times-to-failures observed in the asset clusters with sparse data. Additional advantages include enabling the operator with control over the extent of learning, and showing similarities across the asset clusters comprising the fleet.