Coevolutionary Structural Design for Robustness

This paper proposes a new nature-inspired method for achieving robustness of structural designs. In particular, biological processes of coevolution encoded in coevolutionary algorithms are proposed here to concurrently search the representation spaces of structural designs and of their loading cases, hi the proposed model, coevolutionary algorithms use two populations which competitively coevolve. The first population contains structural designs, while the second one contains loading cases. The quality of each individual structural design in the first population is determined by measuring how well it performs against the loading cases from the second population. On the other hand, the goodness of each loading case depends on the number of designs it "defeated," i.e., how many designs didn't succeed in satisfying design requirements (e.g., maximum internal forces, maximum displacement, etc.). This approach encourages generation of robust structural designs performing well against various loading cases. The proposed coevolutionary design method is illustrated by a simple demonstration problem of a welded beam design. The paper also presents initial research conclusions and discusses the most promising paths for future research.