Date of Award

Fall 2025

Project Type

Dissertation

Program or Major

Civil and Environmental Engineering

Degree Name

Doctor of Philosophy

First Advisor

Yashar Y Eftekhar Azam

Second Advisor

Erin E Bell

Third Advisor

Philippe P Kalmogo

Abstract

In the context of digital twins (DT) for structural health monitoring (SHM) of large civil infrastructure, a new set of computational tools contributing to the literature is presented. A novel linear minimum variance unbiased recursive Bayesian estimator for systems with direct feedthrough (RBE) is proposed and its gains and formula for state mean and covariance are derived. A key feature of the derivation lies in the rejection of the input to the system from the structure of the algorithm; this means that the input to the system is not present in any of the algorithm stages. Rejecting the input allows for a more computationally efficient algorithm when compared to other state-of-the-art filters by reducing the necessary steps required for the estimation, and at the same time makes it suitable for implementation in the civil engineering context, where the input to the system is often unknown. Using the structure of RBE as a base, a nonlinear extension of it is derived to obtain a novel nonlinear recursive Bayesian estimator, NLRBE. To the best of the authors’ knowledge, literature lacks a nonlinear input rejection minimum variance unbiased filter for systems with direct feedthrough. NLRBE allows for simultaneous state-parameter estimation, in systems where there is no information about the input or its statistics. The nonlinear extension is performed through linearization of the state transition function. Augmentation of the state vector and Jacobian calculation of the state matrix, in the context of structural dynamics, is explicitly provided. The practical implementation of online estimators in SHM context, and some of its current limitations, is investigated. In this regard, different stochastic estimation approaches are explored, including filtering and smoothing schemes. Filtering, the main component of this dissertation, refers to the recursive estimation of a variable relying only on the available information (from predictions and observations) at the current time step. In contrast, smoothing schemes use a window of information to obtain an estimate. It is shown that smoothing can lead to more precise estimates with a smaller covariance. The trade-off for this improved accuracy is computational cost, as for longer windows of observation, the algorithm becomes less efficient. A comparison between a filtering and a smoothing algorithm is caried out in the context of simulated experiments for input-state estimation for an open deck truss girder railway bridge. Modern methodologies to obtain measurements from a system are explored by using computer vision (CV) techniques to measure displacements and velocities from a system. With the objective of simultaneous input-state-parameter estimation, the measurements are later fed into a nonlinear algorithm for system without direct feedthrough, allowing for the observation of the entire system with a single video record obtained from a low-cost camera. It is shown that despite poor sampling frequency, the CV measurements let the algorithm accurately estimate the requested parameters. Each chapter of this dissertation includes a validation section. Chapters 2, 3, and 5 incorporate laboratory experiments, while all Chapters include validation through simulated experiments. The dissertation concludes with a discussion of the main contributions of each Chapter and potential directions for future research.

Available for download on Thursday, November 19, 2026

Share

COinS