A computational kernel for supporting generative and evolutionary design

Abstract

Evolutionary Computation techniques have been used in design systems for exploring and generating design solutions in recent years. However, most of the current evolutionary design studies concentrate on analysis and optimisation of design solutions for problems at the stage of detailed design. There has been comparatively less research on the synthesis and generation of design solutions through a dynamic process of evolution and refinement, at conceptual stage of design process. Furthermore, many conventional studies on evolutionary design do not support multiple representations of design objects at different levels of abstraction, which are essential for exploring design solutions in an incremental and evolutionary manner. To overcome the above problems, a computational kernel is developed in this thesis for the development of design supporting system applications, based on a Generative and Evolutionary Design (GED) model. With this kernel, design objects can be dynamically evolved in a specialisation process in which design solutions are developed from abstract levels to detailed levels. Generative mechanisms are integrated with this multiple representation scheme to manipulate and generate new design solutions from basis and abstract design objects in an interactive manner which involves users in making design selections. This study focuses on the three important aspects of this kernel, 1) modelling design object and design process in a generative and evolutionary manner within an integrated computational platform; 2) adapting and capturing the knowledge of how design objects are generated within this platform; and 3) enhancing the exploration ability of generative and evolutionary design applications with the use of a number of different evolutionary and generative computing techniques, including Genetic Algorithms and Cellular Automata. Three examples of applying the GED kernel to design tasks are tested and evaluated. The results show that it is feasible and applicable to use the kernel as the core architecture of computational design systems for supporting generative and evolutionary design applications, with improved generative, explorative and adaptive ability in producing potential design solutions effectively and efficiently.