Department of Pattern Recognition

(eng): Industrial condition monitoring and fault detection rely first of all on measured signals which are - like all information coming from the real world - burdened by pervasive uncertainty. A small international consortium is developing a hierarchical condition monitoring framework, which takes this uncertainty into account at any level of the observed control system hierarchy: from individual components up to the entire system. The framework adopts the so-called subjective logic - a kind of probabilistic logic enhanced by the notion of uncertainty. The main question is how to evaluate condition of every piece of input information using this theory. There were developed several methods of quantification of uncertainty for this purpose, based on relation between allowed signal range and noise of the signal, occurrence of outliers and investigation of spectral density, respectively. A specific method based on model switching was developed for piecewise stable signals which change their working points more or less abruptly. The system entered the phase of its validation in a pilot metal processing plant.