Rough inference system for automated failure prediction in cooling devices

This article presents a versatile model of failure prediction in control systems of objects with large time constants. The basic model of a control unit consists of deterministic finite automaton (DFA) which states correspond to states of the system where system variables satisfy a given set of constraints. System failures can be predicted on the basis of time series of the selected signals. Further diagnosis is performed on the basis of rough set theory and original algorithms for attributes’ reduction and decision rule generation. A versatile failure prediction algorithm is constructed that integrates all involved techniques. Cooling systems are application examples of the proposed methodology.