A measurement system has been developed for the testing of cyanobacteria in
water, and it consists of three main stages: the odour sampling system, an
electronic nose (e-nose) and a CellFacts instrument that analyses liquid s
amples. The e-nose system, which employs an array of six commercial odour s
ensors, has been used to monitor not only different strains but also the gr
owth phase of cyanobacteria (i.e. blue-green algae) in water over a 40-day
period. Principal components analysis (PCA), multi-layer perceptron (MLP),
learning vector quantisation (LVQ) and Fuzzy ARTMAP were used to analyse th
e response of the sensors. The optimal MLP network was found to classify co
rrectly 97.1% of the unknown nontoxic and 100% of the unknown toxic cyanoba
cteria. The optimal LVQ and Fuzzy ARTMAP algorithms were able to classify 1
00% of both strains of cyanobacteria samples. The accuracy of MLP, LVQ and
Fuzzy ARTMAP in terms of predicting four different growth phases of toxic c
yanobacteria was 92.3%, 95.1% and 92.3%, respectively. These results show t
he potential application of neural network based e-noses to test the qualit
y of potable water as an alternative to instruments, such as liquid chromat
ography or optical microscopy. (C) 2000 Elsevier Science S.A. All rights re
served.