A Constraint-based Maximum Entropy Sampling Method for Kriging Models in Fuel Cell Applications

Abstract

A metamodel replaces the simulation model with an approximation model to make design optimization computationally achievable. The accuracy of a metamodel depends highly on the choice of sampling points. This article proposes a constraint-based maximum entropy sampling method that locates most sampling points within a feasible constraint domain represented as a complex nonlinear function. As a robust measure of information, a maximum entropy criterion is used to select sampling points for constructing the Kriging model. The violation ratio from the feasible domain is incorporated into the covariance function in the Kriging model. The constraint-based maximum entropy sampling method is applied to reduce the weight of a bipolar plate in a vanadium redox battery by optimizing its channel design. The proposed sampling method rapidly approximates the boundary of the feasible domain with a relatively small number of sampling points. Final optimal design results for the plate channel using the proposed method indicate a significant reduction in the plate weight compared with the existing bipolar plate design.