Stiction Detection Based on Cross-correlation and Signal Shape

Abstract

Among the methods that aim to detect stiction, quite a few focus on analysing the specific signal shapes that loops exhibiting stiction usually show. The idea to use cross-correlation to investigate the signal shape for control loop data has triggered quite some further research in this area. Cross-correlation-based detection is introduced and shown to robustly indicate a stiction behaviour in many typical industrial cases. The chapter then presents a theoretical explanation why cross-correlation can robustly detect more or less typical stiction patterns and also states where it should not be used. There are, however, cases where cross-correlation will give misleading results, the most important being processes that are of an integrating nature (level control). For such loops, an alternative to cross-correlation needs to be offered. Such an approach that is signal-shape based is presented afterwards. Interestingly, the same approach can also be used for self-regulating (i.e. non-integrating) loops as well.