Authors:
Dani Juricic
1
;
Pavel Ettler
2
and
Jus Kocijan
1
Affiliations:
1
Jozef Stefan Institute, Slovenia
;
2
COMPUREG Plzen and s.r.o., Czech Republic
Keyword(s):
Gaussian process model, Fault detection, Statistical hypothesis test, Cold rolling mill.
Related
Ontology
Subjects/Areas/Topics:
Adaptive Signal Processing and Control
;
Industrial Automation and Robotics
;
Industrial Engineering
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Intelligent Fault Detection and Identification
;
Signal Processing, Sensors, Systems Modeling and Control
;
System Identification
Abstract:
In this paper a fault detection approach based on Gaussian process model is proposed. The problem we raise is how to deal with insufficiently validated models during surveillance of nonlinear plants given the fact that tentative model-plant miss-match in such a case can cause false alarms. To avoid the risk, a novel model validity index is suggested in order to quantify the level of confidence associated to the detection results. This index is based on estimated ‘distance’ between the current process data from data employed in the learning set. The effectiveness of the test is demonstrated on data records obtained from operating cold rolling mill.