Controle de colunas de destilação de alta pureza: um caso industrial.

Abstract

Distillation columns usually constitute a significant fraction of capital investment and production cost in chemical plants. Therefore, they need to be properly designed and controlled to achieve optimum operating conditions. A composition control strategy for a Distillation column, which is part of a vinyl chloride commercial plant, was developed. Some special properties of this column and of its control system requirements make this case study exhibit unusual aspects. This process is a high purity multicomponenl distillation column, with a very high reflux ratio, where the mixture shows an azeolropie behavior. The open loop responses to a disturbance in reboiler heat duty are characterized by huge overshoot and inverse response. Control system requirements are uncommon for including two components concentration control in the same stream, where components concentration is in parts per million and one of the components has intermediate volatility in the mixture. The column has difficulties tor keeping key component concentration steady and this is shown through current operational conditions evaluation. Designing a control system I'or such a process may represents a huge challenge. A brief literature review on Distillation columns control structures is presented and, using singular values decomposition and relative gains matrix tools, it was established a second controlled variable and a pairing that provides best controllability. An eight transfer function linearized dynamic model was identified from a rigorous dynamic model open loop responses discrete data, it was demonstrated that transfer functions fitting provided accurate results when compared to rigorous model data, even for the strongly inverse response. Decentralized feedback control was used. For purpose of tunning and synthesis of physically realizable decoupling and feedforward controllers, two out of eight transfer functions had their order decreased. For ProporcionalIntegral controllers tunning, performance of five first order more dead time model based tunning methods were compared. Using a commercial simulator, performance of chose tuning parameters were checked against rigorous model asymmetry and non linearity in a very wide range of setpoint changes. Most cases simulated with both control loops closed do not show satisfactory responses due to very strong interaction. Explicit decouplers were evaluated to reduce the degree of interaction between the loops. They do not provide satisfactory performances for this process. This case study fits in the category of those processes where recommended pairing by methods based on steady-slate gains is not suitable. It was shown that use of individually designed feedforward controllers takes this process to instability, while jointly designed feedforward controllers exhibit much superior performance responses. Finally, this case study conclude thai decentralized feedback control strategy do not perform well if both loops are closed in this coiumn and recommends the use of composition control only for the key component, in a single loop fashion with feedback and feedforward controllers.