Instrumentation defines a sensors network on a process. Hardware sensors indeed allow one to get different information sources that can be often cross-checked to provide reliable data. However, each of these sources of information contains some uncertainties, either due to the hardware sensors' measurement principles, to their possible fouling, to the estimated parameters of the models used in software sensors and/or to the specific structures of the software sensors. This paper demonstrates that, in this context, the evidence theory is a very well suited formalism for fault detection and diagnosis. This theory indeed allows one to take into account the exact knowledge supported by each source of information and to combine them in order to detect the occurring faults. Moreover, this combination guarantees the best fault isolability from a practical point of view and is suitable for multiple faults occurring at the same time. Finally, the evidence theory is a highly modular formalism since new information sources can be very easily added and old ones can be removed. Validation is performed using real-life experiments from a 1 m3 anaerobic digestion fixed bed process used applied to the treatment of winery wastewaters.

This content is only available as a PDF.