MODELING AND DESIGN OF OPTIMAL CONTROL OF BATCH TYPE DISTILLATION COLUMN SYSTEM WITH LINEAR QUADRATIC TRACKING APPROACH

• Main Author: Novitarini Putri, Adi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/60925
• Institution: Institut Teknologi Bandung

Today, the chemical industry is quite numerous and massive. One application of the chemical industry that is still widely used today is distillation. Distillation is a process of separation or separation of a solution. One type of distillation based on the mode of operation is batch distillation. Batch distillation is the process of separating the solution only once (not repeated). The mathematical model of this system does not meet the theory of superposition, or is nonlinear. The application of control theory that is currently being carried out is applied to linear systems. One of the challenges in performing operations on the distillation column is the optimization problem. The optimization problem on the batch distillation column is divided into two, namely column optimization and optimal control. In this thesis, we will implement Linear Quadratic Tracking (LQT) to fulfill the optimal control problem. Therefore, system modeling was carried out using several methods, namely the nonlinear function approach of ARX using the Neural Network (NARXNN), approximation to the linear form of ARMA using the Neural Network (ARMANN), and the direct approach (ARMA-direct method). Based on the test results on the model obtained from the identification process, LQT is proven to be able to produce optimization in the batch distillation process. This is based on the resulting output response, because it has been able to track the given setpoint. In addition, the control parameters obtained from solving the Riccati Equation have succeeded in ensuring the stability of the system because all closed-loop eigenvalues are located within the unity circle. As for system modeling, the best identification process is using NARX-NN based on the MSE value generated during testing with the parameters of the number of input and output delays selected. However, for control purposes, the approximation to the linear form of ARMA using a Neural Network gives a smaller MSE value than the direct approach for the number of input and output delays of 2 and 3. obtained using the ARMA-NN method is used as a mathematical model.