Interdigitation of approximations for design optimization

Abstract

No single approximation can be applied universally to approach all problems. In this paper, a new approach of approximation interdigitation is presented. The emphasis of this work is on establishing an approximate model, forming a principle to automatically choose and apply the best approximation for each constraint and objective during optimization process. The principle is generated based on the error comparison between the exact and approximate function values at the previous design point.. The approximation model includes three singlepoint (linear, reciprocal and conservative) approximations, and two two-point approximations (the two-point adaptive nonlinear approximation TANA and full twopoint adaptive nonlinear approximation TANA2). Numerical results show that the interdigatation of approximations make approximation-based optimization design more stable and converge faster with no extra analysis cost