DESIGN OF CONTROL SCHEMES FOR DISTILLATION COLUMN

Abstract

Distillation is defined as a process in which a liquid or vapor mixture of two or more substances is separated into its component fractions of desired purity, by the application and removal of heat. The objective of present research work is to develop a control scheme for distillation column through dynamic simulations and experimentation. To fulfill this objective, an existing laboratory set-up of distillation column is used. The set-up is a continuous binary type distillation column (BDC) and contains a vertical column that has nine equally spaced trays mounted inside of it. Mixture of methanol and water is taken as feed to the column. The distillation column is interfaced with auxiliary components and transducers to facilitate monitoring and control of various parameters of the column. A reboiler is connected to the vertical shell through pipes. A controller unit is interfaced to the reboiler heat input for controlling the temperature profile of the distillation column. Due to inherent complexity of distillation process, it is difficult to achieve the simultaneous control of top and bottom composition. To overcome this difficulty Wood and Berry developed a linear model of BDC. The Wood and Berry model of distillation column is a simplified linear model and may not represent the BDC exactly therefore; a mathematical model that uses the fundamental physical and chemical laws along with valid normal assumptions has been utilized in this thesis work. Distillation column is divided into three different sections from the modeling point of view. Ist section is reboiler section, IInd section is Tray section, and IIIrd section is condenser section. The balance equations are obtained by applying material and energy conservation laws to these sections. All these equations are solved to develop the model. The manipulated variables are reflux flow and reboiler heat duty. To validate the model, results obtained from this model have been compared with the results obtained from the experimental set-up. In experimental set-up, reflux flow-rate and reboiler heat duty have been varied similarly as in case of the equations based model. It is observed from the results that this equations based model is in good agreement with the experimental outputs. The results obtained in this study prove that this equation based model could be use to represent the existing experimental set-up of BDC. This equations based model has been used for the analysis and implementation of different control schemes to hold the product composition nearly to the set point under different types of disturbances i.e. disturbance in heat input, reflux ratio and feed flow. As neural network method does not require a deep mathematical knowledge of the system to develop the system‘s model accurately, in this thesis, a neural network model is proposed for BDC. The neural model gives methanol composition as output based upon the knowledge of the six inputs namely; tray temperature, reflux flow-rate, feed flow-rate, reboiler duty, reflux drum top pressure and reboiler bottom pressure. For the training of the neural network, the data has been acquired by the operation of experimental set-up of BDC. Two ii neural network topologies namely; Feed Forward Neural Network (FFNN) and Recurrent Neural Network (RNN) are utilized for the development of the neural network model. Inferential control scheme is the technique in which secondary variables are used as the controlled variables. For the laboratory set-up of BDC, it is found by the sensitivity analysis that at constant pressure, the temperature of fourth tray is an exact indicator of the corresponding concentration of methanol output. Therefore; temperature of the tray is used as a secondary variable. By using the experimental results obtained with laboratory set-up, a relation has been established between the controller current and the tray temperature by curve-fitting method. This relation has been utilized to control the temperature of the tray. A PID controller is used to control the temperature of the tray. The parameters of the PID controller have been tuned using the Genetic Algorithm in MATLAB® /Simulink environment. This PID controller has also been implemented on the experimental set-up of BDC in the laboratory. The results obtained show that the simulated and hardware PID controller are in good agreement to each other. Model Predictive Control (MPC) is one of the main process control techniques explored in the recent past for various chemical engineering applications, therefore; the linear MPC (LMPC) scheme utilize the equations based model of BDC and the nonlinear Neural Network based Model Predictive Control (NN-MPC) scheme utilize the ANN based model of BDC to control the methanol composition. In NN-MPC scheme, a three layer feed forward neural network model has been developed which is used to predict the methanol composition over a prediction horizon using the MPC algorithm for searching the optimal control moves. The training data is acquired from the existing laboratory set-up of BDC. Two cases have been considered, one is for reference tracking and another is for feed flow disturbance rejection. The performance of the control schemes are compared on the basis of performance parameters namely rise time, settling time and MSE. NN-MPC and LMPC schemes are also compared with conventional PID controller. The results show the improvement in rise time and MSE with NN-MPC scheme as compared to LMPC and conventional PID controller for both the cases. The neural network has the ability to represent arbitrary non-linear relations and can be trained even for an uncertain system. These qualities have been exploited by many researchers in the past. In the present work, direct inverse control (NN-DIC) and internal model control (NN-IMC) schemes have been developed to control the final composition of methanol. Forward model and inverse model of BDC have been developed utilizing the ANN approach. In developing the NN-DIC scheme, inverse model is utilized. In NN-IMC scheme, both forward and inverse models are used. Two input variables, reflux flow-rate and reboiler heat duty are used as manipulated variables. The results obtained show the improvement in iii the rise time, settling time and MSE with NN-IMC scheme as compared to NN-DIC and NN-MPC schemes. Soft computing is a collection of various approaches like fuzzy system, neural networks and genetic algorithm. It is useful to tackle imprecision and uncertainty involved in a complex nonlinear chemical system. Recent reviews on soft computing around the world indicate that the number of soft computing based engineering applications is increasing. Neuro-fuzzy is one of the extensively used soft computing approaches. It is a hybridization of artificial neural networks and fuzzy inference system. Adaptive Neuro-Fuzzy inference system (ANFIS) is an example of Neuro-Fuzzy systems in which a fuzzy system is implemented in the framework of adaptive neural networks. ANFIS constructs an input-output mapping based on both the human knowledge (in the form of fuzzy rules) and the generated input-output data pairs. In this work, ANFIS controller is applied on the ANN based model of BDC. The controller controls the methanol composition in BDC by the variation of reflux flow-rate and reboiler heat duty. The performance of ANFIS controller has been compared with NN-IMC control scheme. The obtained result shows the improvement in the rise time, settling time and MSE with ANFIS scheme as compared to NN-IMC scheme. The present thesis concludes that the discussed models namely; equation based model and ANN based model have been developed and validated for existing experimental set-up of BDC. A PID controller tuned by GA is studied in simulation and, subsequently, it is implemented on the experimental set-up of BDC in laboratory. The results obtained show that the simulated and hardware PID controller are in good agreement to each other. The developed equation based model and ANN based models are utilized to control the methanol composition by the application of different control schemes namely; LMPC, NN-MPC, PID, NN-DIC, NN-IMC and ANFIS. A comparison is conducted among these control schemes based on performance parameters namely; rise time, settling time, overshoot and MSE. Overall, ANFIS control scheme shows a superior performance over the above studied control schemes by presenting a shorter rise time, shorter settling time and smaller MSE.