The patterns of information flow through an artificial neural network
are examined in terms of how and why a network characterises input/out
put relationships, and what insight these patterns give as to the char
acteristics of the network that could be changed to improve its descri
ption of a system. The prediction of vapour-liquid equilibrium in term
s of bubble-point conditions is used as a case study and shows that th
e network is capable of identifying the intrinsic characteristics of t
he system. However the accuracy of the prediction depends on the regio
n of the input/output data space considered, drawing attention to the
difficulties encountered by the empirical structuring of a network. Th
e ability to identify the type and strength of the relationships betwe
en process variables indicates that a priori knowledge of the system c
ould be used to relate parts of the network to dominant elements of th
e intrinsic model. This implies there could be advantages to be gained
by exploiting knowledge of the system to maximise the information con
tent captured by the network and establish a systematic way of designi
ng its structure.