Model-based Design Optimization Taking into Account Design Viability via Classification

Abstract

Design optimization of real-world industrial products is usually a challenging high-dimensional task with several multi-modal objectives. Therefore, the solution has to be found by global optimization algorithmswhichrequire fast surrogate models to realize a large number of design evaluations. However, approximating the original optimization criteria by surrogates may mislead the optimization by offering solutions in the entire design domain, even if designs are not viable in reality. Therefore, a classification model should be used as additional optimization constraint to guide the optimizer to viable results.