Bayesian Hierarchical Models for aerospace gas turbine engine prognostics

Data-driven approach is suitable for gas turbine engine prognostics.Bayesian methods produce predictive results within well defined uncertainty bounds.Bayesian Hierarchical Model (BHM) uses optimally degradation data for prognostics.This integrates effectively multiple unit data to address realistic prognostic challenges. Improved prognostics is an emerging requirement for modern health monitoring that aims to increase the fidelity of failure-time predictions by the appropriate use of sensory and reliability information. In the aerospace industry, it is a key technology to maximise aircraft availability, offering a route to increase time in-service and to reduce operational disruption through improved asset management.An aircraft engine is a complex system comprising multiple subsystems that have dependent interactions so it is difficult to construct a model of its degradation dynamics based on physical principles. This complexity suggests that a statistically robust methodology for handling large quantities of real-time data would be more appropriate. In this work, therefore, a Bayesian approach is taken to exploit fleet-wide data from multiple assets to perform probabilistic estimation of remaining useful life for civil aerospace gas turbine engines.The paper establishes a Bayesian Hierarchical Model to perform inference and inform a probabilistic model of remaining useful life. Its performance is compared with that of an existing Bayesian non-Hierarchical Model and is found to be superior in typical (heterogeneous) scenarios. The techniques use Bayesian methods to combine two sources of information: historical in-service data across the engine fleet and once per-flight transmitted performance measurement from the engine(s) under prognosis. The proposed technique provides predictive results within well defined uncertainty bounds and demonstrates several advantages of the hierarchical variant's ability to integrate multiple unit data to address realistic prognostic challenges. This is illustrated by an example from a civil aerospace gas turbine fleet data.