IDENTIFICATION OF PARAMETER COUPLING IN TURBINE DESIGN USING NEURAL NETWORKS

This article discusses a new technique for improving convergence In op timization problems by pruning the search space of weak variables. Suc h variables are identified by learning from a database of existing des igns using neural networks. By using clustering techniques, different sets of weak variables are identified in different regions of the desi gn space. Parameter sensitivity information obtained in the process of identifying weak variables provides accurate heuristics for formulati ng design rules. The impact of this methodology on obtaining converged designs has been investigated far turbine design problems. Results fr om a three-stage power turbine and an aircraft engine turbine design a re presented in this article.