Supervisor(s): Dr. Vladimir Mahalec and Prashant Mhaskar

Project Description: This works aims to achieve flooding free control for a high-purity industrial ethylene splitter. The current control strategy of the tower uses two PI controllers with the goal of maintaining their ethylene product at high purity. The tower operates at high capacity and faces flooding issues due to feed disturbances. This work develops a replacement of the current control strategy with an offset-free Nonlinear Model Predictive Control (OF-NMPC) strategy to improve the control and avoid flooding. The key idea is to develop an OF-NMPC that uses a hybrid model comprising of a Nonlinear Autoregressive Network with Exogenous Inputs (NARX) for predicting dynamics, and a first principles steady state model to capture flooding. The effectiveness of the proposed OF-NMPC is demonstrated on an Aspen simulation model which has been shown in a previous work to be able to capture the key traits of the real plant. Thus, a NARX model is first identified using data generated form the Aspen simulator by applying Random Gaussian Signal (RGS) on the manipulated variables (reflux flow and reboiler duty). To deal with the flooding issue, a steady state model that relates tower internal flow to the manipulated variables is utilized, along with the NARX model for dynamic prediction. This hybrid model (comprising the data driven dynamic model and mechanistic steady state model) is deployed in the OF-NMPC. The simulation results demonstrate the ability of the proposed hybrid model based OF-NMPC design to achieve flooding free control.