Object-based design modeling and optimization with genetic algorithms

DOME (Distributed Object-based Modeling Environment) is a software framework for product design system modeling where designers are distributed geographically and make use of different software tools. Designers develop their own local software components, and distributed object technology is used to integrate their services via the Internet to form an overall system model. Designers can then explore alternatives by making changes to local models or remote services while observing how the entire model responds. This exploration is amenable to automated search, which involves both continuous parameters (changing the value of services) and discrete changes (selecting different objects to substitute entire local models). A genetic optimization object and appropriate direct representation genomes and operators were developed for this purpose. The effectiveness of several genetic algorithms was compared and a new variation of restricted tournament selection (RTS) was developed. The RTS variation, called the Struggle GA, most reliably located the global optima and the most local optima. Other crowding algorithms reliably located the global optima but were less successful identifying multiple local solutions. The global algorithms (simple and steady state) did not reliably locate the global optima in mixed variable problems. Finally, a realistic beverage container design example is presented.