Identifiability in Multi-Point Gas Turbine Parameter Estimation Problems

Abstract

A method for estimating performance parameters in jet engines with limited instrumentation has been developed. The technique is applied on a nonlinear steady state performance code by making simultaneous use of a number of off-design operating points. A hybridized optimization tool using a genetic algorithm to obtain an initial estimate of the performance parameters, and a gradient method to refine this estimate, has been implemented. The method is tested on a set of simulated data that would be available during performance testing of a PW100 engine. The simulated data is generated assuming realistic noise levels. The technique has been successfully applied to the estimation of ten performance parameters using six simulated measurement signals. The determination of identifiability (the property governing whether the performance parameters of the model can be uniquely determined from the measured data) and the selection of parameters in the performance model has been based on the analysis of the system Hessian, i.e. the multidimensional second derivative of the goal function. It is shown, theoretically as well as in practice, how the process of selecting model parameters can be approached in a systematic manner when nonlinear multi-point problems are studied.