This paper describes the architecture of the knowledge based system (KBS) c
omponent of Smartfire, a fire field modelling tool for use by members of th
e fire safety engineering community who are not expert in modelling techniq
ues. The KBS captures the qualitative reasoning of an experienced modeller
in the assessment of room geometries, so as to set up the important initial
parameters of the problem. Fire modelling expertise is an example of geome
tric and spatial reasoning, which raises representational problems. The app
roach taken in this project is a qualitative representation of geometric ro
om information based on Forbus' concept of a metric diagram. This takes the
form of a coarse grid, partitioning the domain in each of the three spatia
l dimensions. Inference over the representation is performed using a case-b
ased reasoning (CBR) component. The CBR component stores example partitions
with key set-up parameters; this paper concentrates on the key parameter o
f grid cell distribution. (C) 1999 Elsevier Science Ltd. All rights reserve
d.