Semester

Summer

Date of Graduation

2024

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Chemical and Biomedical Engineering

Committee Chair

Fernardo V. Lima

Committee Co-Chair

Yuhe Tian

Committee Member

Yuhe Tian

Committee Member

Mario Perhinschi

Abstract

As demand for more efficient processes keeps rising and more restrictive environmental legislation requires higher levels of both yield and purity, the process industry has now, more than ever, been looking for new ways to achieve these goals. One approach has been the implementation of Process Intensification (PI). PI aims to combine multiple unit operations into a single one, promoting efficiency gains on multiple levels. This combination, however, leads to complex process dynamics, and controlling such processes, particularly when setup as multi-input-multi-output (MIMO) systems, presents a great challenge due to highly interactive complex process dynamics and the imposition of many operating constraints. The objectives of this research are to generate first-principles and data-driven water-gas shift membrane reactor (WGS-MR) models and to create predictive control schemes to perform effective setpoint tracking under a MIMO framework. The AVEVA Process Simulation platform is chosen for simulating the first-principles model and both Gaussian Processes using a Squared Exponential kernel (GP-SE) and Bayesian Smoothing Spline Analysis of Variance (BSS-ANOVA) are explored as possible frameworks for data-driven models. A Proportional–Integral– Derivative (PID) controller is created as a benchmark and a Quadratic Dynamic Matrix Controller (QDMC) is developed for advanced control of the system. The GP model is later used as a digital twin of the plant for calculating a control action trajectory that could be applied to the physical system in case of a desired setpoint change. The results demonstrate the superiority of the BSS-ANOVA model over the GP-SE, especially under open-loop simulation conditions. Also, both PID and QDMC perform well under the tested conditions, but the QDMC is able to deliver smoother control action, specially under noisy signals. Finally, when operating under closed-loop control, the BSS-ANOVA model is capable of providing a good representation of one of the variables being studied, namely carbon capture, with Integral of the Absolute Error (IAE) values ranging from 0.39 to 12.6, while creating a fit for the other variable, namely hydrogen recovery, with IAE measurements ranging from 3.71 to 26.7. These results indicate that the proposed framework is viable, although not universal, as each variable desired to be modeled and controlled should be tested and have its viability carefully assessed.

Download

Share

COinS