Innovative Design as Systematic Search

Abstract

We present a new algorithm, sie, for designing lumped parameter models from first principles. Like the ibis system of Williams [ 1989, 1990 ] , sie uses a qualitative representation of parameter interactions to guide its search and speed the test for working designs. But sie's interaction set representation is considerably simpler than ibis's space of potential and existing interactions. Furthermore, sie is both complete and systematic --- it explores the space of possible designs in an nonredundant manner. Introduction A long standing concern of Artificial Intelligence has been the automation of synthesis tasks such as planning [ Allen et al., 1990 ] and design. Of the many approaches to design (e.g., library design, parameterized design, etc.) innovative (or first principles) design has seemed to present the greatest combinatorial challenge. In this paper, we extend the work of Williams [ 1989, 1990 ] on the ibis innovative design system. Like ibis we assume the lumped parameter m...