ABSTRACT
Anyone actively trying to find an approach that might be appropriate for their specific, practical application are advised, firstly to identify, clearly, their requirements (what can be modelled/measured, what are the uncertainties etc.) and secondly, to ensure that they remember these requirements when considering some of the options available. An anomaly can be viewed as a misconception of how the plant is operating: a particular pump might not have been switched-on, a partial blockage might be developing, something more localized like a sensor fault might have occurred and so on. Any computer simulation, constructed to predict what is expected, is unlikely to mimic this situation because any simulation is largely a manifestation of what the modeller perceives to be going on. When solved in parallel with the plant (Figure 1 ) , any physical model (obtained by applying laws of chemistry and physics) will fail to predict plant measurements accurately if an anomaly has arisen (A). There are two fundamentally different ways of proceeding, either the correct anomaly must be hypothesized by classifying its properties, and hence symptoms, (B) or the model can be adapted until the correct symptoms are predicted (C). The choice depends, very much, on the application.
