Abstract
In this paper, a fault prognosis approach for continuous processes with hidden faults is proposed based on the principal component analysis structure and multivariate time series prediction. It is assumed that the actual fault is a slowly time-varying autocorrelated process and the fault can be completely reconstructed. Fault magnitude is estimated via reconstruction first, then predicted by a vector ARMA model. A new fault detection policy is proposed and the denoising effect on prediction modeling is studied. The case study of CSTR demonstrates the efficiency of the approach and the validity of the analysis. © 2009 IFAC.
Original language | English |
---|---|
Pages (from-to) | 1019-1024 |
Number of pages | 6 |
Journal | IFAC Proceedings Volumes |
Volume | 42 |
Issue number | 8 |
DOIs | |
Publication status | Published - Jun 2009 |
Externally published | Yes |
Event | 7th IFAC Symposium onFault Detection, Supervision and Safety of Technical Processes (SAFEPROCESS'09) - Barcelona, Spain Duration: 30 Jun 2009 → 3 Jul 2009 |
Bibliographical note
ISBN: 9783902661463Keywords
- Fault prognosis
- Fault reconstruction
- Principal component analysis
- Vector ARMA