Inferential active disturbance rejection control of distillation columns

Abstract

The distillation column is an important processing unit in the chemical and oil refining industry. Distillation is the most widely employed separation method in the world’s oil plants, chemical and petrochemical industrial facilities. The main drawback of the technique is high energy consumption, which leads to high production costs. Therefore, distillation columns are required to be controlled close to the desired steady state conditions because of economic incentives. Most industrial distillation columns are currently controlled by conventional multi-loop controllers such as proportional-integral-derivative (PID) controllers, which have several shortcomings such as difficulty coping with sudden set-point jumps, complications due to the integral term (I), and performance degradation due to the effect of noise on the derivative term (D). The control of ill-conditioned and strongly non-linear plants such as high purity distillation needs advanced control schemes for high control performance. This thesis investigates the use of active disturbance rejection control (ADRC) for product composition control in distillation columns. To the author’s knowledge, there are few reported applications of ADRC in the chemical industry. Most ADRC applications are in electrical, robotics and others. Therefore, this research will be the first to apply the ADRC scheme in a common chemical processing unit, and can be considered as a first contribution of this research. Initially, both PI and ADRC schemes are developed and implemented on the Wood–Berry distillation column transfer function model, on a simulated binary distillation column based on a detailed mechanistic model, and on a simulated heat integrated distillation column (HIDiC) based on a detailed mechanistic model. Process reaction curve method and system identification tools are used to obtain the 2×2 multi-input multi-output (MIMO) transfer function of both binary and HIDiC for the purpose of PI tuning where the biggest log-modulus tuning (BLT) method is used. Then, the control performance of ADRC is compared to that of the traditional PI control in terms of set-point tracking and disturbance rejection. The simulation result clearly indicates that the ADRC gives better control performance than PI control in all three case studies. The long time delay associated with product composition analysers in distillation columns such as gas chromatography deteriorates the overall control performance of the ADRC scheme. v To overcome this issue an inferential ADRC scheme is proposed and can be considered as a second contribution of this research. The tray temperatures of distillation columns are used to estimate both the top and bottom product compositions that are difficult to measure on-line without a time delay. Due to the strong correlation that exists in the tray temperature data, principal component regression (PCR) and partial least square (PLS) are used to build the soft sensors, which are then integrated into the ADRC. In order to overcome control offsets caused by the discrepancy between soft sensor estimation and actual compositions measurement, an intermittent mean updating technique is used to correct both the PCR and PLS model predictions. Furthermore, no significant differences were observed from the simulation results in the prediction errors reported by both PCR and PLS. The proposed inferential ADRC scheme shows effective and promising results in dealing with non-linear systems with a large measurement delay, where the ADRC has the ability to accommodate both internal uncertainties and external disturbances by treating the impact from both factors as total disturbances that will then be estimated using the extended state observer (ESO) and cancelled out by the control law. The inferential ADRC control scheme provides tighter product composition control that will lead to reduced energy consumption and hence increase the distillation profitability. A binary distillation column for separating a methanol–water mixture and an HIDiC for separating a benzene–toluene mixture are used to verify the developed inferential ADRC control scheme.