A statistical approach to multivariate measurement validation

Abstract

The fusion of uncertain sensory information into multivariate measurement systems is explored from a classical measurement perspective. The issue of sensor validation in multivariate measurement systems using data-driven statistical uncertainty models is investigated. An online approach is motivated for the detection, isolation and rectification of measurement anomalies within a class of redundant process measurement systems. The procedure nominally allows for the control of validation false-positive alarms whilst detecting incipient sensor anomalies within redundant process sensor networks. Experimental results from a set of thermocouples are presented.