Inverse Design Of A Mixed Convection Problem Via Proper Orthogonal Decomposition

Abstract

An inverse design approach is developed for a mixed convection problem in a partially open cavity with a heat generating board using Proper Orthogonal Decomposition (POD). The numerical database analyzed by POD consists of distinct CFD simulations varying with a combination of governing parameters determined by Latin Hypercube Sampling (LHS). POD's expansion coefficients together with governing parameters are interpolated by Kriging to predict unknown CFD simulations at off-design points. An inverse design problem of finding a target flow and/or temperature field is solved where the POD model for velocity outperforms the temperature model.