Dynamic Optimization of an Evaporator by a Nonlinear Model Predictive Controller for Operation at Modular Micro Rectification

Arranging micro-structured equipment to plants whole production processes can be realized with maximum efficiency in tightest space. Unit operations are thereby represented as individual functional modules in shape of micro devices. In a multi unit operation plant a correspondingly large number of manipulable variables have to be coordinated. Due to the design of micro-scaled devices plants form sophisticated systems, while for a fully optimized control still no common satisfying solutions exist. A system of modular, discontinuous phase contacting, micro rectification consists of unit operations heating, cooling, mixing and separating. Heat exchangers, mixers and cyclones for phase separation can be arranged to a counter-current rectification system with maximum mass-transfer efficiency every unit. Operating an electrical heated evaporator for modular rectification purposes a strong coupling of mass flow with the vapor fraction and the outlet temperature can be observed [4]. Operating at a predefined state for mass flow, temperature and vapor fraction may only be possible with difficulties using traditional methods of linear control technology. For dynamic optimization of the multivariable micro-structured evaporator principle of Nonlinear Model Predictive Control (NMPC) was generically formulated in C++ and implemented to LABVIEW 7. Every discrete time step an objective function is generated from nonlinear process models in the form of grouped NARX-polynomials. Optimal sequences of control actions for plant operation are evolved. The resulting constrained cost function is non-convex making detection of relative local optimum a difficult task. This obstacle can be gone around using heuristic optimization algorithm in combination with traditional techniques. Based on experimental results it was demonstrated that NMPC keeps the coupled variables mass flow and temperature energy saving with minimal control activity in the entire two-phase region on their set-points.