Efficient Automatic Engineering Design Synthesis Via Evolutionary Exploration

Citation

Lee, Cin-Young (2002) Efficient Automatic Engineering Design Synthesis Via Evolutionary Exploration. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/5HRP-ND58. https://resolver.caltech.edu/CaltechTHESIS:05032011-082523919

Abstract

The evolution of designs in nature has been the inspiration for this thesis, which seeks to develop a framework for efficient automatic engineering design synthesis based on evolutionary methods. The design synthesis process is equated to an evolutionary process. Because of this, the same formalization for evolution, the evolution algorithm, is used as a design synthesis formalism. Implementation of the evolution algorithm on a computer allows evolution of non-biological systems, and, hence, automatic engineering design synthesis. The early and canonical versions of such evolutionary computation are bare bones evolution tools that neglect several key aspects of evolutionary systems. Some universal aspects of good designs are identified, three of which are dealt with in this thesis. These are variable complexity, modularity, and speciation. Framed in an evolutionary context, each of these characteristics are requisites for being able to evolve in correspondence with a dynamic environment. Those that are most evolvable will survive. After all, if a species cannot evolve quickly enough with changes in the environment, it will perish. In a design context, this indicates that the characteristics are vital for efficiency and shorter design cycles. An integrated framework is developed to address all three aspects individually or in any combination thereof, which has not been done heretofore. Because of the poor theoretical foundations of evolutionary computation, the effectiveness of the developed approach is determined through computer experimentation on several test and design problems. Results are promising as all three aspects were successfully achieved in comparison to canonical evolutionary computation.

Thesis Files