In the early stages of ship design a considerable amount of experience and knowledge is used to build and evaluate empirical models with known design relationships. However, computer-based systems which aim to assist this stage have tended to concentrate on the analytical aspects of the process and have not been successful in integrating with this expertise and benefitting from it. This paper presents some of the results of a program of research into methods and representing knowledge of empirical numerical relationships used in these early stages of the design process. The work is based on an experimental system, DESIGNER, described in earlier papers. The DESIGNER system is used to carry out a series of evaluations of design sessions, using a warship design model. By examining the progress toward a set of design goals and the classes of interactions used, an improved understanding of the requirements of an interactive numerical design system is developed. As a consequence, methods have been developed to handle approximate values and relationships, to include design margins, and to represent explicitly in the system the definition and use of goals, or design requirements. Using a design model representing a bulk carrier, the paper then presents a worked example to illustrate the use of the new numerical knowledge techniques. It is concluded that the techniques could make a useful contribution to any interactive numerical design system which aims to provide improved use of expertise.
Certified Roundoff Error Bounds Using Bernstein Expansions and Sparse Krivine-Stengle Representations
