Much of engineering design can be characterised as putting together va
riants of existing mechanisms to meet novel requirements. This paper d
escribes an approach to engineering design in which fuzzy sets are use
d to represent the range of variants on existing mechanisms. Membershi
p functions are chosen to reproduce the distinctions and classificatio
ns used by experienced designers. Methods are introduced to calculate
the fuzzy performance range achievable by each component type, and a m
etric is suggested for the ranking of design candidates against design
requirements. The underlying approach to design evaluation derives fr
om that developed by Antonsson, Wood and Otto. If the components model
the design domain accurately, this architecture will always find a su
ccessful design, where one exists. However, finding this design may in
volve exhaustive search of the design. space, and the time required fo
r such an exhaustive search is typically far greater than is available
for the task. The architecture is therefore augmented by the introduc
tion of design agents, embodying heuristic rules to direct the search
intelligently, so that a satisfactory design is found within a reasona
ble length of time. As an example, the method is applied to preliminar
y design of a Stirling engine heat exchanger. (C) 1998 Elsevier Scienc
e B.V. All rights reserved.