Reducing the gap between experts' knowledge and data: the TOM4D methodology

Abstract

Dynamic process modelling is generally accomplished from experts' knowledge through Knowledge Engineering (KE); however, the obtained models are sometimes deficient for interpreting the input data flow coming from the real process evolution perceived through sensors. This shortcoming lies in specialists' tacit knowledge, difficult to elicit, and in that certain process phenomena are unknown or unforeseen to experts. An alternative to complement the modelling task is to resort to a Knowledge Discovery in Database (KDD) process. Nevertheless, most KE approaches do not address the processing of knowledge obtained from data. This work proposes a KE methodology called Timed Observation Modelling For Diagnosis (TOM4D) which allows building dynamic process models from experts' knowledge and data where the obtained models can be compared and combined with models obtained through a KDD process in order to define a model more suitable to the dynamic process reality.