We discuss a new type of macroarchitecture of neural networks called a
HANN and how to train it for fault diagnosis given appropriate data p
atterns. The HANN divides a large number of patterns into many smaller
subsets so the classification can be carried out more efficiently via
an artificial neural network. One of its advantages is that multiple
faults can be detected in new data even if the network is trained with
data representing single faults. The use of a HANN is illustrated in
fault diagnosis of a chemical reactor.