Multivariable control and online optimization of an ethylene plant

Dow Chemical is one of the largest ethylene producers in the world, with ten plants in six different countries. Most of these plants are controlled with Dow's proprietary digital control system, and before 1990 no multivariable control had been implemented in any of the plants. In the early 1990's multivariable controllers were installed on the demethanizer and ethylene fractionator in our LHC2 plant in Plaquemine, Louisiana, and on several cracking furnaces in the LHC3 plant located on the same site, with encouraging results. In the same time period, complete steady state simulation models of both plants were developed and validated against plant data [1]. In 1995, we began a project to complete the installation of multivariable controllers in the LHC3 plant, and develop and commission an online closed-loop optimization model. The project was completed in late 1997. This application was the first to use Aspen Technology's RT-Opt (TM) optimization software, together with an online version of KTI's SPYRO (TM) furnace simulation model, on an entire ethylene plant. In this paper we will describe the architecture and operation of the multivariable controllers and online optimizer. Rather than the typical emphasis on the large size of the control and optimization problems and the profitability improvement, we intend to focus on some of the interesting technical challenges encountered in the development and implementation of the technology. Some examples are the estimation of coke layer thicknesses in the cracking furnace tubes using tube wall temperature measurements, the prediction of furnace run length, and the calculation of variable profit per furnace and its use in furnace scheduling.