Real‐time multirate state estimation in a pilot‐scale polymerization reactor

Abstract

Real

Unscented Transformation‐Based Filters: Performance Comparison Analysis for the State Estimation in Polymerization Processes with Delayed Measurements

“…Two case studies of different mathematical complexity, taken from the literature, were used to evaluate the performances of the two‐timescale UKF, URNDDR, and RCUKF filters. The first case study is a styrene solution polymerization in a continuous stirred‐tank reactor (CSTR) 26. In this case the mathematical model is simple, involving six state variables which include reactant concentrations and average molecular weights of the polymer.…”

“…Polystyrene, unreacted monomer, solvent and initiator, compose the exit stream. The reactor temperature is tightly controlled independently of other process conditions 26…”

“…This model has been validated experimentally. The values of the model parameters are found in the literature 26…”

Application of MHE to large-scale nonlinear processes with delayed lab measurements

“…Due to the computational burden required for the recalculation step, we demonstrate situations in which utilizing delayed lab data is worthwhile. Lab measurements should be used when some states are unobservable using only the fast measurements, but these states become observable when lab measurements are included (Mutha et al, 1997;Tatiraju et al, 1999;Zambare et al, 2002). Lab measurements may help improve the transient behavior of the estimator by recovering quickly from an inaccurate prior (Tatiraju et al, 1999;López-Negrete and Biegler, 2012).…”

“…As a rough approximation, the missing data of slow measurements can be predicted by polynomial extrapolation such that the nominal estimation method can be applied (Tatiraju et al, 1999;Zambare et al, 2003). If a deterministic observer is available, it can also be adjusted to handle multi-rate measurements (Zambare et al, 2002). Most efforts on the estimation with multi-rate sampling measurements are Extended Kalman Filter (EKF)-based, including parallel filters (Larsen et al, 1998), fixed-lag smoothing (Gudi et al, 1994;Mutha et al, 1997) which augments the states to cover the measurement delay, and filter recalculation starting from the time the slow measurement was taken (Prasad et al, 2002).…”