Adaptive techniques for Evolutionary Topological Optimum Design

This paper introduces some advances in Evolutionary Topological Optimum Design, thanks to extensive use of adaptive techniques. On the genotypic side, a variable length representation is used: the complexity of the representation of each individual is evolved by the algorithm rather than being prescribed by some fixed mesh of the design domain, resulting in self-adaptive complexity. On the phenotypic side, an original adaptive mechanism is proposed that maintains both feasible and infeasible individuals, thus exploring both sides of the boundary of the feasible region, where the optimum structure is known to lie. Not only does this improves the results of past work in on Evolutionary Topological Optimum Design on standard benchmark bidimensional cantilever problems, but it also allows to address three-dimensional problems who had up to now stayed beyond reach for evolutionary algorithms