Date of Award

27-2-2024

Document Type

Thesis

School

School of Electrical & Electroncis Engineering

Programme

Ph.D.-Doctoral of Philosophy

First Advisor

Rakesh Kumar Sidharthan

Keywords

Model Predictive Control, Fault Detection, Redundant Control System, Distillation Column, Canonical Cross-Correlation Analysis

Abstract

Model Predictive Controller (MPC) is one of the industrial-ready control systems and an ideal choice for complex, slow-varying, and energy-intense Distillation Columns (DC). However, MPCs are subjected to failures and their performance degrades with faults and disturbances. Hence, there is a need for a redundant control strategy to ensure the continuous operation of MPC-controlled DC, which is investigated in this work.

A pilot-plant binary DC used to separate methanol and water is implemented with a computer-controlled system and is subjected to Pseudorandom Binary Excitation Signals (PRBS). The output response acquired is used to develop a MIMO (Multi-Input Multi- Output) state-space model of DC, which exhibits an 85% fit percentage. MPC is designed using a state-space model and various Quadratic Problem (QP) solvers used for cost function minimization are analyzed. Among the three QP solvers considered, the Iterative Interior Point (IIP) solver provides optimal control action amidst noises and disturbances.

DC is a multi-variable process with a significant number of sensors and actuators making fault inevitable. Canonical Cross-Correlation (CCA)-based fault detection is developed to detect faults in actuators (stiction and dead band), sensors (drift and bias), and controllers of DC. Genetic Algorithm (GA) is used to tune the threshold that minimizes false detection and provides about 96.54% of fault detection accuracy on average.

A redundant control strategy is vital to ensure a reliable MPC under various abnormal operating conditions like unbounded disturbance and controller failure. Autotuned PID controller has been considered to provide redundancy for MPC failure. An on-demand GAbased controller tuning technique inspired by MPC is developed to tune PID parameters. A Graphical User Interface (GUI) is also developed to monitor and control the DC.

Finally, an experimental investigation of the proposed redundant control system is carried out and compared with conventional MPC. About 87.7% reduction in Integral Time Absolute Error (ITAE) performance metric is observed during MPC failure, which demonstrates the reliability and capability to ensure continuous operation of the DC.

Download

Share

COinS