Problem-Specific Representations for Heterogeneous Materials Design

This paper investigates the use of problem-specific data structures and operators in evolutionary optimization for a specific class of combinatorial design problems. The problem consists of finding the optimal distribution of two or more phases of a sound absorbing material on a three-dimensional network, in order to maximize sound absorption properties. The natural structure of the problem is by the way very far from the linear chains classically used by evolutionary algorithms (EAs). Special operators exploiting the three-dimensional structure are proposed and compared with other operators that are working on a linear chain representation. The formers are potentially useful since the natural neighborhood relationships are lost in a linear representation.