Abstract
The conceptual design of a fluid power circuit from a set of customer-specified requirements is difficult, traditionally requiring the expertise of proficient designers. Attempts have been made to automate this complex task, predominantly using rule-based approaches. The limitations and problems associated with these approaches have been well documented. In response to these problems, the authors have adopted a variety of machine learning techniques to automatically extract what can be thought of as the ‘design knowledge’ from an archive of existing fluid power designs. The benefits of this approach are that this knowledge can be acquired more rapidly and consistently (i.e. without bias), and represented and used in a fashion that seems more nearly to approximate human reasoning.
This paper describes a system for the automatic configuration design of fluid power circuits. The architecture of this system is based upon a model of the configuration design process involving several distinct stages. The knowledge required for performing these stages is machine-learned from the design archive and then applied to generate novel designs in response to new design requirements. Though not without its own problems, the authors believe that their approach offers the prospect of more accurate and consistent automatic design tools.
