NONLINEAR MODELING AND STATE ESTIMATION FOR THE TENNESSEE-EASTMAN CHALLENGE PROCESS

Abstract

A nonlinear, mechanistic model of the Tennessee Eastman (TE) Plant-wide Challenge Process is derived for use in on-line monitoring and control The model has 26 states, 10 unmeasured disturbances and process parameters are estimated such that an equal number of model outputs match the corresponding measurments at steady state. The model represents the steady-state behavior of the TE process over the entire operating range stipulated by Downs and Vogel [Computers Chem. Engng 17, 245-255 (1993)]. An extended Kalman Filter (EKF) provides continuous adjustments of the parameters (and states) during transients. The EKF also gives accurate predictions of both measured and unmeasured quantities-such as reaction rates-even in the presence of large, unmeasured disturbances. Possible uses of the model include fault detection, control, and optimization. Details of the rationale of the model structure, the elimination of steady-state bias, and EKF tuning provide a case study to aid in applications to other industrial problems.