Kriging surrogates for evolutionary multi-objective optimization of CPU intensive sheet metal forming applications
Abstract
The aim of this paper is to present a method to perform evolutionary multi-objective optimization of CPU intensive sheet metal forming applications using kriging surrogates. Two main ingredients are employed to achieve this goal. First of all, given a learning dataset, the kriging surrogate is designed to minimize the leave-one-out error. Secondly, during the optimization, new data points are added to the learning set to update the surrogate locally (by well chosen points on the current Pareto front) and globally (by maximum kriging variance points over the entire design landscape). The ability of the method to capture Pareto fronts with accuracy is demonstrated on the well-known ZDT test functions. The method is then tested on a real-life problem, the simultaneous minimization of springback and failure for a three-dimensional CPU intensive high strength steel stamping industrial use case.