Effect of the number of design variables in thickness optimization of concrete barrel roof shells using evolutionary algorithms

Abstract

This study looks into some practical implications of using evolutionary algorithms for optimization of free-form concrete shells in search of methods for automating design representation and determining the number of design variables. This study reports the insights and learnings from a set of numerical experiments, by changing the number of parameters, on thickness optimization of a barrel roof shell subjected to self-weight, an additional snow load, and an earthquake response spectrum. The practical advantages and challenges of two methods, parametric and direct representations, are analysed, and a spectrum of methods between these two extremes are investigated through experiments. The results demonstrate how changing the number of variables affects the quality of the design and the performance of the algorithm and why a systematic problem-dependent method for finding the best design representation and number of variables outside this spectrum can be beneficial.