A genetic methodology for configuration design

In an increasingly competitive world, the ability to efficiently produce viable artifact design alternatives is necessary for organizations to succeed. For millennia, engineers have been using design methodologies to assist in the configuration of new artifacts. However, as these artifacts have grown in complexity, the need for more capable design methodologies has increased. This thesis presents a design methodology to aid the designer of complex artifacts by generating viable artifact design alternatives for further consideration. This methodology, called Genetic Design (GD), uses formal grammars for artifact description and representation, evaluates the artifacts automatically and manipulates the representations with genetic programming-like operations. Human designers and optimization codes are very good at improving the performance of existing artifacts. However, due to economic constraints imposed while designing new artifacts, human designers tend to limit the range of alternative configurations considered. GD can explore a wide breadth of the available design space, though at shallow depth, and present viable alternatives to the human designer. The combination of GD's ability to explore the design space and the human engineer's ability to optimize existing configurations promotes the production of viable, new design concepts by avoiding the inefficiencies associated with trial and error methods. GD represents an attempt to devise a design methodology that can be used across a broad range of application areas. This thesis explores two applications of GD. In the first application, GD is used to determine optimal dimensions and controllers for an abstracted model of a frame-walking robot. In the second application, GD is used for the simultaneous type, number and dimension synthesis of planar mechanisms.