Title: Analysis and Optimization of Input Trajectories for Parameter Identifiability in Multi-Compartment Dynamic System Models
Author: Mahsa Doosthosseini
Date/Time: 08/17/2021 12:30 pm-2:30 pm
Examining Committee:
Dr. Hosam K. Fathy (Chair/Advisor)
Dr. Alisa Morss Clyne (Dean’s representative)
Dr. Joseph S. Friedberg
Dr. Jin-Oh Hahn
Dr. Simona Onori
Dr. Monifa Vaughn-Cooke
Zoom Link: https://umd.zoom.us/j/6260893257
Abstract:
This dissertation examines the interconnected problems of (i) analyzing and (ii) optimizing the impact of a multi-compartment dynamic system’s input history on the identifiability of its parameters. Identifiability refers to the feasibility and accuracy with which a system’s parameters can be uniquely estimated from input-output test data. The shape of a system’s input history versus time often affects identifiability. This makes it possible to optimize this input shape for identifiability, in a manner analogous to the use of a cardiac stress test to better diagnose patients with heart disease.
The research in this dissertation makes four contributions to the literature, motivated by the following four practical research questions. First, is it possible to characterize CO2 gas transport dynamics in a laboratory animal where the peritoneal perfusion of a perfluorocarbon (PFC) is used as a potential treatment for hypercarbia? Second, how does the shaping of chemotherapeutic treatment affect the accuracy with which drug resistance dynamics can be estimated in a partially drug-resistant cancerous tumor? Third, can the dynamic cycling of a lithium-sulfur (Li-S) battery be tailored to maximize the accuracy with which its parameters are estimated? Finally, can Pontryagin methods from optimal control theory yield fundamental insights into the structure of the ambient temperature cycling trajectory that maximizes the identifiability of a lithium-ion battery model’s thermal parameters?
In addressing the above practical research questions, this dissertation navigates a progression of four fundamental topics in the field of multi-compartment dynamic system parameter identification and identifiability. Specifically, the dissertation’s examination of peritoneal CO2 gas transport dynamics highlights and motivates the importance of analyzing multi-compartment dynamic system identifiability. The subsequent examination of the identifiability of drug resistance dynamics in cancerous tumors highlights the degree to which input shaping can negatively affect parameter identifiability. In contrast, the examination of parameter identifiability for Li-S batteries highlights the potential of input shaping to improve identifiability significantly for multi-compartment systems. Finally, the dissertation’s examination of thermal battery parameter identifiability highlights the degree to which the fundamental tool of Pontryagin analysis can help gain insight into optimal input shaping for identifiability. In summary, the work in this dissertation explores a progression of fundamental topics in the area of dynamic system parameter identifiability while highlighting the broad applicability of this area to different practical domains.