Michal Falkowski
supervisor: Pawel Domanski
Modern industrial processes are characterized by a high degree of interconnection between individual control loops. One of the most difficult issue regarding analysis of large-scale industrial processes is to find root cause of faults. Faults that are commonly caused by inappropriate control loops operation may lead to low productivity of whole system, can increase operational costs and in the most dangerous cases, an unwanted system shutdown or its destruction. Effective and quick answer to this type of problems is Root Cause Analysis. This approach allows for non-invasive finding such errors in large-scale industrial control systems.
However, to begin this type of analysis, causality between variables of a given process should be determined first. One may find several various approaches, which have been developed in different domains. None of them are applied in industry. They are mainly used in medicine or chemistry and they are based on the model. It has been shown that building a model is a tedious process and its accuracy depends on many factors. This is a complex and time-consuming issue; thus the universality of these methods is negligible. Problem also appears because these approaches work well only for linear systems.
Work focuses on broadly understood signal analysis and use of the Transfer Entropy method - an information theoretic interpretation of Wiener’s causality definition.