Carleton University
Robust, Sparse, and Bayesian Learning of Stochastic Nonlinear Dynamical Systems with Applications in Aeroelasticity and Infectious Disease Modelling
Abstract
dc:description.abstractThis thesis is principally concerned with Bayesian methods for calibrating models of stochastic nonlinear dynamical systems. Of particular interest are systems where the available data represent noisy, sparse, or incomplete observations of the system states. Case studies in the fields of aeroelasticity and infectious disease modelling are considered, using established Bayesian frameworks for sparse learning of nonlinear-in-parameter models, and for the concurrent estimation of time-varying and time-invariant parameters. In nonlinear aeroelasticity, the fluid-structure interaction of wing-like structures and external flow is investigated with a specific focus on the experimental set-up of an elastically mounted single degree-of-freedom rigid airfoil undergoing pitching oscillations. The epidemiological modelling aspects consider the use of compartmental models complemented by public health data for the population-level spread of infectious diseases, such as COVID-19. Although the two applications are practically quite distinct, they both provide excellent opportunities for the deployment of the two aforementioned Bayesian computational frameworks. The modelling efforts in aeroelasticity and in infectious disease modelling both involve the description of physical systems by nonlinear coupled ordinary differential equations. Furthermore, there is precedent in both fields for the use of semi-empirical models that consist of a mechanistic component and a data-driven component. The accurate description of the system dynamics may involve a combination of time-varying and time-invariant parameters, necessitating a robust framework for handling their joint estimation in a fully Bayesian setting. The data-driven component is particularly susceptible to overfitting, which may lead to undue uncertainty in predictions. The predictive capability of such models can therefore be improved by the use of sparsity-inducing algorithms for the automatic discovery of the data-optimal nested model. The computational framework involves the use of standard methods for state estimation and sampling-based algorithms for parameter inference. The final contribution of the thesis pertains to improving the numerical implementation of the sparse learning framework by investigating the use of optimization-based sampling-free methods to bypassing the reliance on computationally expensive sampling-based methods. These investigations consider simple approximate representations of highly non-Gaussian joint probability densities in the interest of the efficient computation of the data-optimal sparse model structure.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy (Ph.D.)
- Level thesis:degree_level
- Doctoral
- Discipline thesis:degree_discipline
- Engineering, Civil
- Grantor dc:publisher
- Carleton University
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Robinson, Brandon Michael George
Subjects
dc:subject × 3Rights
dc:rights- Statement dc:rights
-
- Copyright © 2025 the author(s). Theses may be used for non-commercial research, educational, or related academic purposes only. Such uses include personal study, distribution to students, research and scholarship. Theses may only be shared by linking to the Carleton University Institutional Repository and no part may be copied without proper attribution to the author; no part may be used for commercial purposes directly or indirectly via a for-profit platform; no adaptation or derivative works are permitted without consent from the copyright owner.
- Language dc:language.iso
- en