Bayesian Framework for Calibration of Gas Turbine Simulator

Abstract

A Bayesian framework is developed for characterization of uncertainty in gas turbine performance predictions. Bayesian framework enables updating of uncertain parameters with availability of new information. Framework is demonstrated for a single spool turbojet engine with uncertainty in compressor map. A two step approach is proposed to provide robustness with respect to initially guessed compressor map. First step involves the global update of initially guessed compressor map using one data point. Second step involves representing compressor map as a Gaussian process and results in local update based on each subsequent information. Experimental data on steady state and expert judgment are used as new information. Markov Chain Monte Carlo method is used to sample from Bayesian posterior distribution. Monte Carlo method is used for propagation of posterior compressor map uncertainty to system performance. Resultant uncertainty in performance predictions is represented using Bayesian confidence bounds. Nomenclature English Symbols A Area CN Corrected Speed f Probability Density Function H Altitude M Flight Mach Number N Spool Speed P Pressure PR Pressure Ratio Pr Probability STR Scaled Temperature Ratio SW Scaled Mass Flow Rate T Simulator W Mass Flow Rate Ye Experimental Observations PhD Student, Department of Aerospace Engineering, IIT Bombay Professor, Department of Aerospace Engineering, IIT Bombay Retd. Professor, Department of Aerospace Engineering, IIT Bombay 49th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference 16t 7 10 April 2008, Schaumburg, IL AIAA 2008-1809