Hiroya Seki

Industrial application of a nonlinear model predictive control to polymerization reactors

A nonlinear model predictive control has been developed and applied to an industrial polypropylene semi-batch reactor and a high density polyethylene continuous stirred tank reactor. The control system consists of two-tier algorithms: at the first tier, an LQI is formulated based on a successively linearized nonlinear first principles process model. At the second tier, actual control actions are determined in consideration of process constraints by solving a QP problem, which is formulated by linearizing the nonlinear model around the LQI trajectory. A simple state estimator, which is capable of providing offset-free estimates of the outputs at steady states, has been designed for each application. In the semi-batch reactor process, the controller was able to maximize the monomer feed while satisfying the heat removal constraint. In the high density polyethylene process, the performance of the grade transition control was greatly improved. r 2001 Elsevier Science Ltd. All rights reserved.

Optimal grade transition control for polymerization reactors

Abstract A nonlinear control system integrating an off-line optimizer and a nonlinear MPC controller is developed to perform optimal grade transition operations at the industrial polyolefin reactors. In this, paper, the details of the optimizer are given: The sequential nonlinear programming is performed in the optimizer by employing control vector parameterization method together with sensitivity analysis. The switching times (i.e. times of chaning in the input actions) can be also optimized by new formulation and a modification in trial functions. The simulation result illustrates the capability of the control system with the proposed optimizer.

A Nonlinear Predictive Control for Optimal Grade Transition of Polymerization Reactors

Abstract An optimal grade transition control system integrating off-line optimizer and nonlinear model predictive controller is developed for polymerization reactors. The optimal trajectory of inputs and outputs is off-line computed by using first principles model. The nonlinear model predictive controller using the same first principles model together with a linearized state-space model ensures that the real process response does not stray away from the optimal trajectory. The proposed algorithm can provide attractive computational efficiency, nominal stability and good steady-state performance. An application of grade transition control on a slurry polyethylene reactor is presented to illustrate the capability of the algorithm.

Retuning PID Temperature Controller for an Unstable Gas-Phase Polyolefin Reactor

Abstract Temperature control for a gas-phase polyolefin reactor is prone to be unstable when operating conditions and/or catalyst are changed. A practical retuning method of PID temperature controller, which can be performed at the onset of closed-loop instability, is developed. The frequency domain analysis using a simple first principles model shows that plant nonlinearity introduces extra phase delay to PID controller and causes closed loop instability. Thus, temperature controller is retuned so as to increase phase margin. Effectiveness of the proposed method is demonstrated through simulations.

Industrial Application of a Nonlinear Model Predictive Control to a Semi-Batch Polymerization Reactor

Abstract A nonlinear model predictive control has been applied to an industrial semi-batch polypropylene reactor where the control objective is to maximize production rate (i.e., monomer feed) while satisfying a capacity constraint on the heat removal. The NMPC is formulated by LQI with the successive linearization of a nonlinear model. The real operation illustrates the capability of the developed system. Some issues specific to the semi-batch process control are also addressed.

A Sensitivity Analysis of Real-Time Optimizer in Industrial Constrained Predictive Control

Abstract At some particular control executions of the industrial predictive control, [DMC]™, the operating targets given by real-time LP optimizer become unstable because of the sensitiveness of the optimal LP-solutions to process noise and disturbances. This paper proposes a practical method of analyzing the sensitivity of LP-solutions of the predictive control, and providing an appropriate stability for the operating targets by either relaxing the control limits or LP coefficients. The proposed method is applied to an industrial ammonia plant of Mitsubishi Chemicals where DMC control is performed.