A Learning Shell for Iterative Design (L’SID): Concepts and Applications

Abstract

A software shell called “Learning Shell for Iterative Design,” L’SID, has been developed in conjunction with a simple data matrix, the “learn table.” Histories of design are utilized in aiding the acceleration of routine design problems. The class of problems addressed are non-convex, noninvertible and with multiple performance criteria. The design parameters can be of any definable type; continuous, integer or nonordered feature based. L’SID is domain independent and highly modular. The ability of L’SID to aid deterministic methods is shown statistically with two example problems (extrusion die and airfoil). Results also show the ability of the technique to surmount nonconvexity in design space and computational noise related to roundoff.