Authors:
Ladislav Jirsa
and
Lenka Pavelkova
Affiliation:
Czech Academy of Sciences, Czech Republic
Keyword(s):
Sensor Faults, Bayesian Statistics, Gaussian Mixture, Dynamic Weights.
Related
Ontology
Subjects/Areas/Topics:
Informatics in Control, Automation and Robotics
;
Nonlinear Signals and Systems
;
Signal Processing, Sensors, Systems Modeling and Control
;
System Identification
;
System Modeling
Abstract:
The paper describes a method of sensor condition testing based on processing of data measured by the sensor using a Gaussian mixture model with dynamic weights. The procedure is composed of two steps, off-line and on-line. In off-line stage, fault-free learning data are processed and described by a probabilistic mixture of regressive models (mixture components) including a transition table between active components. It is assumed that each component characterises one property of data dynamics and just one component is active in each time instant. In on-line stage, tested data are used for transition table estimation compared with the fault-free transition table. The crossing of given level of difference announces a possible fault.