State Variable Based Statistical Methods for Auditing Sensors of Multivariable Dynamic Processes

Abstract

Abstract A statistical process monitoring method based on a state space model of a dynamic process is introduced for auditing sensor status for bias, drift and excessive noise affecting the sensors of multivariable continuous processes. Changes in the magnitudes of means and variances of residuals between measured and predicted process variables are used to detect and discriminate sensor abnormalities. The statistical model that describes the in-control variability is based on a canonical variate state space (CVSS) model. The CV state variables obtained from the state space model are linear combinations of the past process measurements which explain the variability of the future measurements the most, and they are regarded as the principal dynamic dimensions. The method can detect and discriminate between bias change, drift, and variations in noise levels of process sensors based on the analysis of data batches. An experimental application to a high-temperature short-time (HTST) milk pasteurization process illustrates the proposed methodology.