Distillation Column Control Using the Whole Temperature Profile

Industrial distillation column control typically relies on one (or maybe two) carefully located tray temperature sensors. Here our aim is enhanced control by responding to movements in both the position and shape of the whole measured temperature profile. The control objective is to maintain end product specifications against feed composition disturbances using the measured temperatures and without recourse to composition sensors. While recent approaches to this make use of non-linear wave theory (Han and Park 1993, Roffel et al, 2003) we instead use a mechanistic dynamic state space column model to encapsulate the profile behaviour and an adapted version of the Minimum Error Profile (MEP) strategy to exercise control. Proposed for general profile control by Wahl et al (2002), use of the MEP strategy in controlling fixed-bed catalytic reaction systems has shown advantages (ChewHernandez et al, 2004 and Wilson et al, 2005). Here, at each sample period the measured temperature profile allows an Extended Kalman Filter (EKF) to estimate the current feed composition and a Linear Quadratic Regulator (LQR) to set the control moves that will drive the column optimally towards the MEP target steady state, this being the closest match possible to the desired operating condition with the current feed composition. In re-calculating the optimal sequence of future control moves at each measurement sample and taking the first move in the sequence, our approach may be termed a Non-linear Model Predictive Profile Control strategy. Significant performance gains over conventional dual composition control were found in three column case study simulations. While above the reflux ratio and the reboiler steam valve were manipulated for control, the ease of extending the MEP strategy offers the chance to look in principle at including feed enthalpy as a third manipulated variable. Henry and Mujtaba (1999) looked its use for feedforward control against measured feed composition swings (i.e. using a composition sensor) with success limited by the size of allowable feed enthalpy changes. Here, for the MEP strategy including feed enthalpy for true feedback control we found significant further performance gains and the MEP strategy also provided an explicit way to limit the size of control moves called for in feed enthalpy.