Inferencing Component Maps of Gas Turbine Engine Using Bayesian Framework

Abstract

A Bayesian framework is proposed for characterization of uncertainty in gas turbine performance predictions induced by poorly known component maps and uncertain model structure. Uncertainty in component maps and the model structure is characterized by independent Gaussian processes. Bayesian calibration is used to update the uncertainty whenever new information is available through experimental observations and expert opinion. The Markov Chain Monte–Carlo method is used to sample from posterior distribution. Parameters from the posterior distribution are identified that provide information about the validity of the model and experimental observations. Updated uncertainty is propagated to system response using theMonte–Carlo method. The propagated uncertainty in system response is represented using Bayesian confidence bounds. Proposed framework is demonstrated for calibration of a design intent single spool turbojet engine simulator. Compressor and turbinemaps are inferred using the Bayesian framework.