Metamodelling of multivariable engine models for real-time flight simulation.

摘要

Sophisticated real-time distributed flight simulation environments may be constructed from a wide range of modelling and simulation tools. In this way accuracy, detail and model flexibility may be incorporated into the simulator. Distributed components may be constructed by a wide range of methods, from high level environments such as Matlab, through coded environments such as C or Fortran to hardware-in-the- loop. In this paper the Response Surface Methodology is combined with a hyper-heuristic (evolutionary algorithm) and applied to the representation of computationally intensive non-linear multivariable engine modelling. The paper investigates the potential for metamodelling (models of models) dynamic models which were previously too slow to be included in multi-component, high resolution real-time simulation environments. A multi-dimensional gas turbine model with five primary control inputs, six environmental inputs and eleven outputs is considered. An investigation has been conducted to ascertain to what extent these systems can be approximated by response surfaces with experiments which have been designed by hyper-heuristics as a first step towards automatic modelling methodology.