LEARNING FROM EVOLUTIONARY-ALGORITHM BASED DESIGN OPTIMISATION OF AXISYMMETRIC SCRAMJET INLETS

Amit Saha
SEIT, UNSW, Canberra, Australia

*Tapabrata Ray
SEIT, UNSW, Canberra, Australia

* Hideaki Ogawa
Centre for Hypersonics, UQ, Queensland, Australia

*Russell Boyce
Centre for Hypersonics, UQ, Queensland, Australia

     Full text: Not available
     Last modified: June 1, 2011

Abstract
Optimisation is a key element in today’s design processes and there is an ever increasing emphasis on development of efficient algorithms to deal with computationally expensive optimisation problems. While surrogate assisted optimisation methods are commonly used for such problems, there are few studies that attempt to mine the solutions to decipher the reason for their optimality or sub-optimality and use this information to direct the optimisation process in the hope of faster convergence. This paper introduces a novel scheme to uncover hidden relationships among the variables in the promising regions of the search space. Such relationships can be subsequently used to separate promising and unpromising designs. The study is conducted using an axisymmetric scramjet inlet design optimisation exercise, where the optimal geometry is sought by simultaneously maximising the compression efficiency and minimising the inlet drag and the maximum adverse pressure gradient. The inlet exit temperature is restricted to be greater than 850K and was thus posed as an inequality constraint in the mentioned optimazation study. Results clearly indicate that the scheme has the potential to reduce evaluation of unpromising solutions to about 50%. The paper also provides new directions on how such schemes can be adopted on-the-fly within an Evolutionary Algorithms based optimisation framework. This has the capacity to accelerate the rate of convergence to the Pareto-optimal front for a particular design optimisation task and hence will significantly reduce the computational cost in such a scenario.