Mansur Darlington, Steve Culley and Stephen Potter
Engineering Design Centre, Department of Mechanical Engineering,
University of Bath, Bath, UK
Much progress has been made in the area of computer-aided designer support, but little has been made in that of design automation. Where progress has been made, it has been largely in the analytical aspects of the task (for example, simulation and stress analysis) - tasks for which computers are more suited than humans. Less tractable is automation of the early, conceptual, phase of design, heavily reliant as it is on the expert knowledge of the design practitioner. Emulating this computationally is the domain of Artificial Intelligence (AI) and requires a detailed understanding of the nature of the design process (Darlington et al, 1998).
This paper discusses some of the issues raised during an investigation in to the automation of the configuration phase of fluid power system design, and identifies some of the hurdles to be cleared before automation, supported by AI, becomes a reality. Two models, developed by the authors, are chosen to illustrate the way in which very different approaches can be taken to automating the same task with an emphasis on the knowledge that is used by designers, which must be acquired and used in automation.
Keywords: configuration design, design knowledge, automation, neural networks, case-based reasoning