Building Models of Process Systems: Applications in Control and Design

Traditional or new methods for building models of process systems can combine physical understanding with statistical analysis so that models can be developed for use in many fields of engineering. This work focusses on developing models that can be used for system design (e.g., cementing of hydrocarbon wells), process control (e.g., in manufacturing of chemicals or pharmaceuticals), and simulation purposes (e.g., to study the robustness of automatic control system). Availability of large amounts of data that is recorded during hydrocarbon well construction makes it possible to apply data-driven modeling methods to address the problem of natural gas leakage from cemented section of hydrocarbon wells – a problem that is too complex to investigate using the existing modeling and experimental methods. The problem has significant operational and environmental implications as leaking wells pose serious pollution threats to both groundwater and the atmosphere. Design of wellbore preparation and cementing recipes involve decisions on over two dozen design variables that affect the quality of hydrocarbon well cementing. Given historical data of these design variables, data-driven models are developed to successfully classify wells into “leak” and “no leak” categories in cross-validation tests. In addition, these models identify and rank the most important factors for preventing leakages. The quality of data generated in experiments can also be improved by designing experiments in such a way that the resulting model identified from the data fulfills certain requirements of the application for which the model is identified. Controller design for multivariable system is one such application where design of experiments can be modified to generate good quality data. The controller designed using the model built from the resulting data can be used to better control the processes, such as production of chemicals. The model building method consists of two steps, namely estimation of (a) model order i.e., structure of model, and (b) values of model parameters. Of the two steps, the first one is far more challenging and more strongly dependent on the quality of data generated in the identification experiment. A new design of experiments (DOE) is proposed that accurately estimates the model order even for the most challenging systems. Comparison of the proposed DOE with other currently practiced DOEs using simulation studies on two process systems validates the effectiveness of the proposed design to accurately estimate model order when other DOEs fail to do so. A rigorous mathematical analysis is also provided to further confirm the claim. Simulating process systems before the actual installation of the systems in the field is a useful and cost-effective way to foresee the problems that may occur after installation and analyze the effects of these problems in advance. Automatic control system of a novel directional drilling technology is simulated and effects of changing drill bit torque on controller performance are studied.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Chemical Engineering

Grantor

University of Houston

Year dc:date.issued

2017

Author dc:creator

• Misra, Shobhit

Advisor dc:contributor.advisor

• Nikolaou, Michael

Committee members dc:contributor.committeemember

• Grabow, Lars C.
• Palmer, Jeremy C.
• Qin, Guan
• Darby, Mark L.

• Modeling
• Controls

Statement dc:rights

• The author of this work is the copyright owner. UH Libraries and the Texas Digital Library have their permission to store and provide access to this work. Further transmission, reproduction, or presentation of this work is prohibited except with permission of the author(s).

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/10657/2863

OAI identifier oai:identifier

oai:uh-ir.tdl.org:10657/2863