DETECTION AND DIAGNOSIS OF PLANT-WIDE OSCILLATIONS USING GRANGER CAUSALITY

摘要

This paper addresses the problem of detecting multiple oscillationsand tracking down their root cause. The oscillationsare detected by two methods based on spectral analysis,and the power related to the oscillations is used to identifythe signals more likely to be the root cause. Since thismethod of finding the root cause is prone to fail when multiplefrequencies are detected, Granger causality and subsetselection methods are proposed. After a presentation ofthe methods, examples are provided using both simulatedand industrial data. The clustering of signals containing thesame frequency of oscillations followed by the methods toindicate the root cause is shown to be a good approach forrapid troubleshooting using only historical data.