Nonlinear state estimation of constrained chemical systems

State estimation has great importance in chemical engineering systems. State estimation finds its applications in filtering, online control, inferential control, data reconciliation, model predictive control and parameter estimation in diagnosis. Most of the chemical engineering systems are modeled as ordinary differential equations and differential algebraic equation systems. Some of the challenges that need to be addressed in these systems are that they are large, nonlinear and have constraints and bounds imposed on the process.;Kalman filter and its variants have been used for the state and parameter estimation of dynamical systems. However, Kalman filters cannot handle constraints and bounds imposed on the system. Hence the Moving Horizon Estimator (MHE) has been a popular method for the estimation of nonlinear systems with constraints, but it is computationally very demanding. Recently, Recursive Nonlinear Dynamic Data Reconciliation (RNDDR) is developed in extended Kalman filter (EKF) framework to address the nonlinear state and parameter estimation of systems with constraints. But RNDDR uses linearization for the covariance propagation. To overcome this, Unscented Recursive Nonlinear Dynamic Data Reconciliation (URNDDR) is developed which combines the merits of Unscented Kalman filter (UKF) and RNDDR, and it can be applied for constrained nonlinear state estimation. The main disadvantage with URNDDR is that it needs to solve 2n+1 optimization problems (n is the dimension of the state) and the computational complexity is a problem. In this work, Constrained Unscented Recursive Estimator (CURE) is developed in UKF framework which gives valid estimates in a single optimization step.;The above described methods are developed for the state and parameter estimation of ordinary differential equation systems. Extended Kalman filter has been used for the state estimation of differential algebraic equation systems (DAE) with differential states measurements being available to the estimator. This thesis describes the estimation of nonlinear systems described by differential algebraic equation systems with measurements being differential states as well as algebraic states. RNDDR can also be extended to index one differential algebraic equation systems with bounds and constraints imposed on the system. The formulation of Unscented Kalman filter for DAE system is also discussed in this thesis.;Several case studies from chemical engineering are used to demonstrate the estimation techniques described above. These estimation methods also find applications in Reactive Distillation (RD) systems where the equations describing the model are ordinary differential equation systems or differential algebraic equation systems. Several case studies of RD systems are used to demonstrate different estimation techniques.