Year: 2018
Type: dissertation
Abstract: Fault diagnosis is a challenging problem, particularly for a large-scale industrial process. An up-to-date and accurate process topology – the causal relationships between system components of an industrial process – is crucial to understanding the fault propagation pathways and in diagnosing a fault. Data-driven techniques are more reliable for reconstructing the process topology, as compared to using expert knowledge. Many of the established algorithms for fault diagnosis impose strict assumptions that limit their applicability to industrial applications. As such, two novel algorithms from outside the field, Granger Net and Extended Convergent Cross-Mapping, were investigated. The classical Granger Causality algorithm was used as a performance benchmark. The effectiveness of the two novel algorithms for process topology reconstruction and fault diagnosis was investigated using five simulated processes and the Eastman industrial process data.
Source: TSpace (University of Toronto)
Author: Fu Yuan Tee
Institution: University of Toronto
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Funder: Eastman Chemical Company
OpenAlex ID: W3146141377
