Monitoring sensor performance in multivariable continuous processes

This paper presents two sensor auditing techniques which rely on input-output models developed from operation data. The multipass PLS (partial least squares) algorithm is developed for detecting simultaneous multiple sensor abnormalities. This sensor monitoring algorithm is only suitable for process measurements where the successive measurements are not autocorrelated. For process measurements which are strongly correlated in time a state space modeling paradigm based on canonical variate (CV) analysis is proposed. In order to detect multiple sensor abnormalities the heuristic multipass PLS algorithm is modified by replacing the PLS models with the CV state space models in generating the functional redundancy in terms of model residuals. Experimental results from a milk pasteurization pilot plant are used to illustrate the applicability of both methods