Automating the analysis of complex physical systems: The virtual foundry.

摘要

The object of this thesis is to create an automated analysis capability for complex physical systems that can be integrated into software environments for design. These environments are intended to support the process of design, not to create or propose new designs. In developing these environments, the difficulties that have limited the success of traditional software environments for mechanical design are studied.; Software is developed to support the design of casting. The behavior of a casting is predicted from the underlying physics and mathematics. This system simulates a foundry for designers: they can experiment with their designs while the system provides feedback by doing the analysis. This environment is called the "virtual" foundry.; The physics of casting is complex. It involves different length and time scales. The partial differential equations are non-linear and dynamic. However, it has not been the lack of the physics, software tools or computer hardware that has inhibited the development of design environments for foundries.; Progress in the broader problem of integration of the islands of automation in design software has been modest. A hypothesis of this thesis is that in order to do significantly better than previous integration efforts, a formal framework is needed to automate the integration or analysis capability into design environments.; This thesis develops a distributed, graph based model that can represent coupled complex phenomenon as a network of communicating agents. Agents are a decomposition of the complex system. Graphs of agents can alter their topology and therefore the description of the solutions, as the problems evolve. This is a means of enabling a program to use high level components as primitives. The agent model is used as the framework of integration to develop the software for modelling continuum mechanics problems, and to develop the environment for the users.; The software and methodology that deals with the interactions between agents, i.e., everything except the internal structure of agents, are domain independent. It is argued that this generalization is a useful concept of a design environment that can be applied to other design domains.