Robust Estimation of Nonredundant Measurements and Equivalent Sets of Observations

Abstract

The equation-oriented estimation of measurements constitutes the initial stage of many procedures devoted to optimizing a chemical plant. Nevertheless, the estimation of nonredundant measurements and observations related to equivalent sets of systematic errors remains an unsolved problem, which deteriorates the knowledge of the actual process operation. In this work two statistical tests, which can take the part of the estimation method and deal with those problems, are analyzed. The Window Measurement Test is easily computed using an analytical expression, while the Robust Measurement Test uses robust statistics concepts to calculate the measurement adjustment and its covariance matrix. Both tests take advantage of the spatial redundancy given by the process model and the temporal redundancy provided by a window of process observations. Performance results demonstrate that the tests are able to detect atypical observation values for all kinds of measurements, and highlight the improvements achieved with the use of robust statistics.