Performance enhancement of evolutionary search for structural topology optimisation

The use of evolutionary algorithms for topological design of structures has been investigated for many years. The methods have disadvantages in that they have slow convergence rate and a complete lack of consistency. In this paper, a number of well-established evolutionary methods including genetic algorithm, stud-genetic algorithm, population-based incremental learning and simulated annealing are reviewed in terms of their philosophical bases. The effective means to deal with topological design problems and prevent checkerboards on a topology are briefly detailed. A new set of design variables employing a numerical technique named approximate density distribution is proposed. The new technique and the classical 1–0 binary variables are applied to the various evolutionary methods and they are implemented on a number of structural topology optimisation problems. The results obtained from the various design strategies are compared, illustrated and discussed. Numerical experiment shows that using the present technique can improve both convergence rate and consistency of the evolutionary algorithms.