Combination of KPCA and causality analysis for root cause diagnosis of industrial process fault
Authors:
- 5 May 2017
- journal article
- research article
- Published by Wiley in The Canadian Journal of Chemical Engineering
- Vol. 95 (8) , 1497-1509
Abstract
Kernel principal component analysis (KPCA) based monitoring has good fault detection capability for nonlinear process systems; however, it can only isolate variables that have a contribution to the occurrence of a fault, and thus it is not precise in diagnosing. Since there is a cause and effect relationship between different variables in a process, accordingly a network‐based causality analysis method was developed for different fault scenarios to show causal relationships between different variables and to see the causal effect between the variables most contributing to the occurrence of a fault. It was shown that KPCA in combination with causality analysis is a powerful tool for diagnosing the root cause of a fault in the process. In this paper the proposed methodology was applied to a fluid catalytic cracking (FCC) unit and the Tennessee Eastman process to diagnose root causes for different faulty scenarios.Keywords
This publication has 27 references indexed in Scilit:
- Transfer entropy—a model-free measure of effective connectivity for the neurosciencesJournal of Computational Neuroscience, 2010
- Reconstruction-Based Contribution for Process Monitoring with Kernel Principal Component AnalysisIndustrial & Engineering Chemistry Research, 2010
- Granger Causality and Transfer Entropy Are Equivalent for Gaussian VariablesPhysical Review Letters, 2009
- Causality detection based on information-theoretic approaches in time series analysisPhysics Reports, 2007
- Measuring Information TransferPhysical Review Letters, 2000
- Fault diagnosis in dynamic systems using analytical and knowledge-based redundancyAutomatica, 1990
- The Casual Causal Relationship Between Money and Income: Some Caveats for Time Series AnalysisReview of Economics and Statistics, 1979
- Estimating the Dimension of a ModelThe Annals of Statistics, 1978
- A new look at the statistical model identificationIEEE Transactions on Automatic Control, 1974
- Investigating Causal Relations by Econometric Models and Cross-spectral MethodsEconometrica, 1969