A significant proportion by amount, and a much larger and increasing proportion by value, of the products obtained from the process industries are produced using finite-time or batch processes. These trends in the chemical industries toward high value-added products have increased the interest in the optimal, model based control of batch processes. However, it is well established that the potential advantages of model based control can be lost due to model uncertainties. Several methods to enhance robust performance of the model based controller will be described. The approach presented combines closed loop formulation of the nonlinear model predictive control (NMPC) in the framework of an on-line robust optimal control problem. Both the min-max and distributional robust NMPC approaches will be assessed. Efficient numerical solution strategies will be described to cope with the significantly increased computational complexity of the formulated robust NMPC strategies, to make the approaches feasibly from a practical perspective. Most of the ideas will be corroborated via two case studies: (i) thin film deposition from microelectronics industry, and (ii) pharmaceutical crystallization. An alternative practical approach to deal with uncertain model parameters is to adaptively identify these. A real-time industrial application of parameter adaptive moving horizon estimation in combination with NMPC of a batch polymerization reactor will be presented. The presented case-studies emphasizes via simulations and experiments on laboratory and industrial scale, the importance of considering parametric uncertainties in the controller design.
Back to Control Seminars Page