Set representation and multi-parent learning within an evolutionary algorithm for optimal design of trusses

Abstract

Trusses are common structures that comprise one or more triangular units constructed with straight slender members connected at joints. Outlined in this chapter is a novel scheme for representation of truss geometry and an evolutionary optimization algorithm that operates on the new representation guided by inheritance and learning. Trusses are represented as a set of elements having a collection of properties (e.g. cross-sectional area, type of material). The sets with varying cardinality represent truss structures with different numbers of elements and hence different topologies. The evolutionary algorithm generates topologies by inheriting common elements from the parents and the corresponding element properties are generated via recombination. Depending on the physical problem being solved, specific recombination operators can be designed to aid the optimization process. One such mutation operator is used in this study to reduce the number of elements in an attempt to identify the smallest feasible topology. Another mutation operator is used to perturb the properties of the elements. Algorithm provided in this chapter are useful insights about learning and inheriting topologies and element properties from 3, 5 and 8 parents. A number of case studies are included to highlight the benefits of our proposed set based representation and effects of learning topologies from multiple parents. © 2008 Springer-Verlag Berlin Heidelberg.