Implementation challenges of covariance estimation techniques for an experimental polymerization system

In this paper, we study the estimation of covariance matrices using experimental data obtained from a laboratory emulsion polymerization reactor. Two different methods, the Autocovariance Least-Squares (ALS) and the Direct Optimization (D.O.) are considered for this purpose. The obtained covariance matrices are implemented to define the statistics of stochastic state estimation techniques. The similarities and differences between both approaches are highlighted assuming the same disturbance noise structure, initial guess for the covariance matrices, filter used to perform the covariance estimation and experimental data. The results show that the ALS method can obtain less noisy estimates for the unmeasured states when compared to the D.O. On the other hand, the ALS technique requires further mathematical assumptions on the system conditions that affect the selection of the system model and the noise disturbance structure.