Automated optimal design of dynamic systems
The A-ODDS (Automated Optimal Design of Dynamic Systems) method proposed in this dissertation is generally represented by two iterative search processes (loops) linked by automated modeling of design topologies. The first loop is for topology generation and the second loop is for tuning the parameters for a design topology. In the first loop, a design synthesis method is proposed that combines a probability based decision making strategy and design grammars (production rules) into a “design agent” for system development. The decision making strategy can be evolved to facilitate the exploration of a multi-domain design space in a topologically open-ended manner, and still efficiently find appropriate design configurations. Probability based decision making is applied at each stage of topology development. Experimental results show that a design agent demonstrates steady performance in terms of the overall fitness of the designs it generated. A good design strategy or agent has a better chance of producing superior designs. The research in automated modeling and in the following second tuning loop is leading to a computer-aided design tool in which engineering designers can test various design concepts (topologies) in an environment equipped to automatically model the dynamics and conveniently optimize the specified components, given the evaluation criteria pre-defined by human designers. Automated modeling of design configurations is facilitated by a design representation named as CD-Graph (Conceptual Dynamics Graph) and generic models of various components. CD Graphs record the coupling formats in which not only physical components, but also their generic models are assembled topologically in the same fashion. A generic component model can accommodate various types of coupling between this component and its environment. In the second loop a systematic approach is proposed to automatically prepare a design problem for the convenient application of parameter optimization. This preparation encodes and decodes proper design variables into design genotypes, while taking into consideration of the design constraints and physical constitutive laws.