A role for simple, robust `black-box´ optimisers in the evolution of engineering systems and artefacts

Simple evolutionary and adaptive search algorithms exist which can be effective in addressing a wide range of real-world problems in engineering design, with little or no need for application-specific tuning of their control parameters. These include a version of the genetic algorithm with high performance and exceptionally broad applicability (CHC), a simple multistart random bit-oriented hill-climber, and population-based incremental learning (PBIL). Wider awareness of these robust and user-friendly algorithms could stimulate their use within the community of design practitioners and result in the further exploitation of some of the powerful advantages of evolutionary and adaptive search in engineering design. These include the ease with which complex and multifaceted design requirements can be handled, and the possibility of incorporating constraints and real-world complications which frustrate conventional synthesis procedures. The paper outlines experiences with these algorithms at the University of Cape Town