Studies in nonlinear state and parameter estimation

Effective monitoring and tight control of a process are often hindered by (a) incomplete or infrequent and delayed process state measurements, and (b) insufficient information on important process parameters. In such a situation, continuous estimates of the inaccessible state variables and parameters of the process can be calculated using state/parameter estimation methods, and their applications to chemical and biochemical reactors. A nonlinear state estimation method is used for a methylmethacrylate polymerization reactor to estimate the initiator and solvent concentrations and the leading moments of the molecular weight distribution (MWD) of the polymer, under three measurement scenarios. The global convergence of the estimator is proved theoretically for one of the measurement cases. A multi-rate nonlinear state estimator is developed. The estimator uses for state estimation both frequent and infrequent measurements of a process, with a designer-adjustable reliance on each type of measurement. It can use a nonlinear model of the process directly in the estimation algorithm, without any linear approximation, and is easy to design and implement. The performance of the multi-rate nonlinear state estimator is demonstrated by numerical simulations in a polymerization reactor and via real-time implementation on a bioreactor. A model-inversion-based parameter estimation method is also developed for on-line estimation of unknown process parameters in a class of nonlinear processes. The estimator calculates the least-squared-error estimate of the parameters at each time instant, using readily availably on-line measurements. The superior performance of estimator over that of a state estimation-based parameter estimator is illustrated by numerical simulations using a jacketed chemical reactor example. The nonlinear state and parameter estimation methods presented in this dissertation are powerful tools that allow for efficient monitoring and control in nonlinear processes in the presence of measurement noise and plant-model mismatch. The application studies not only demonstrate the performance characteristic of the state and parameter estimation methods, but also indicate their applicability to various nonlinear processes.