No description available.

The growing complexity and data in modern engineering and physical systems require robust frameworks for real-time decision-making. Data-driven models trained on observational data enable faster predictions but face key challenges—data corruption, bias, limited interpretability, and uncertainty misrepresentation—which can compromise their reliability. Propagating uncertainties from sources like model parameters and input features is crucial in data-driven models to ensure trustworthy predictions and informed decisions. Uncertainty quantification (UQ) methods are broadly categorized into surrogate-based models, which approximate simulators for speed and efficiency, and probabilistic approaches, such as Bayesian models and Gaussian processes, that inherently capture uncertainty into predictions. For real-time UQ, leveraging recent data instead of historical records enables more accurate and efficient uncertainty characterization, making it inherently data-driven. In dynamical analysis, the Koopman operator represents nonlinear system dynamics as linear systems by lifting state functions, enabling data-driven estimation through its applied form. By analyzing its spectral properties—eigenvalues, eigenfunctions, and modes—the Koopman operator reveals key insights into system dynamics and simplifies control design. However, inherent measurement uncertainty poses challenges for efficient estimation with dynamic mode and extended dynamic mode decomposition algorithms. This dissertation develops a statistical framework to propagate measurement uncertainties in the elements of the Koopman operator. This dissertation also develops robust estimation of model parameters, considering observational data, which is often corrupted, in Gaussian process settings. The proposed approaches adapt to evolving data and process agnostic— in which reliance on predefined source distributions is avoided. / Doctor of Philosophy / Modern engineering and scientific systems are increasingly complex and interconnected— operating in environments with significant uncertainties and dynamic changes. Traditional mathematical models and simulations often fall short in capturing the complexity of largescale real-world, ever-evolving systems—struggling to adapt to dynamic changes and fully utilize today's data-rich environments. This is especially critical in fields like renewable integrated power systems, robotics, etc., where real-time decisions must account for uncertainties in the environment, measurements, and operations. The growing availability of observational data—enabled by advanced sensors and computational tools—has driven a shift toward data-driven approaches. Unlike traditional simulators, these models are faster and learn directly from data. However, their reliability depends on robust methods to quantify and manage uncertainties, as corrupted data, biases, and measurement noise challenge their accuracy. This dissertation focuses on characterizing uncertainties at the source using recent data, instead of relying on assumed distributions or historical data, as is common in the literature. Given that observational data is often corrupted by outliers, this dissertation also develops robust parameter estimation within the Gaussian process setting. A central focus is the Koopman operator theory—a transformative framework that converts complex, nonlinear systems into simpler, linear representations. This research integrates measurement uncertainty quantification into Koopman-based models, providing a metric to assess the reliability of the Koopman operator under measurement noise.

The restructuring of the electric energy market and the proliferation of intermittent renewable-energy based power generation have introduced serious challenges to power system operation emanating from the uncertainties introduced to the system variables (electricity prices, congestion levels etc.). In order to economically operate the system and efficiently run the energy market, a statistical analysis of the system variables under uncertainty is needed. Such statistical analysis can be performed through an estimation of the statistical moments of these variables. In this thesis, the Point Estimate Methods (PEMs) are applied to the optimal power flow (OPF) problem to estimate the statistical moments of the locational marginal prices (LMPs) and total generation cost under system uncertainty. An extensive mathematical examination and risk analysis of existing PEMs are performed and a new PEM scheme is introduced. The applied PEMs consist of two schemes introduced by H.P. Hong, namely, the 2n and 2n+1 schemes, and a proposed combination between Hong's and M. E Harr's schemes. The accuracy of the applied PEMs in estimating the statistical moments of system LMPs is illustrated and the performance of the suggested combination of Harr's and Hong's PEMs is shown. Moreover, the risks of the application of Hong's 2n scheme to the OPF problem are discussed by showing that it can potentially yield inaccurate LMP estimates or run into unfeasibility of the OPF problem. In addition, a new PEM configuration is also introduced. This configuration is derived from a PEM introduced by E. Rosenblueth. It can accommodate asymmetry and correlation of input random variables in a more computationally efficient manner than its Rosenblueth's counterpart. / Master of Science

In part to search for a possible critical point (CP) in the phase diagram of hot nuclear matter, a beam energy scan was performed at the Relativistic Heavy-Ion Collider at Brookhaven National Laboratory. The Solenoidal Tracker at RHIC (STAR) collected Au+Au data sets at beam energies, √sNN , of 7.7, 11.5, 19.6, 27, 39, 62.4, and 200 GeV. Such a scan produces hot nuclear matter at different locations in the phase diagram. Lattice and phenomenological calculations suggest that the presence of a CP might result in divergences of the thermodynamic susceptibilities and correlation lengths. The statistical moments of the identified-particle multiplicity distributions directly depend on both the thermodynamic susceptibilities and correlation lengths, possibly making the shapes of these multiplicity distributions sensitive tools for the search for the critical point. The statistical moments of the multiplicity distributions of a number of different groups of identified particle species were analyzed. Care was taken to remove a number of experimental artifacts that can modify the shapes of the multiplicity distributions. The observables studied include the lowest four statistical moments (mean, variance, skewness, kurtosis) and some products of these moments. These observables were compared to the predictions from several approaches lacking critical behavior, such as the Hadron Resonance Gas model, mixed events, (negative) binomial, and Poisson statistics. In addition, the data were analyzed after gating on the event-by-event antiproton-to-proton ratio, which is expected to more tightly constrain the event trajectories on the phase diagram.

This research study presents a new vibration-based non-destructive global structural damage identification and condition monitoring technique that can be used for detection, localization and quantification of damage. A two-stage damage identification process that combines non-model based and model-based damage identification approaches is proposed to overcome the main difficulties associated with the solution of structural damage identification problems. In the first stage, performance assessment of various response parameters obtained from the time-domain, frequency-domain and spectral-domain analysis is conducted using a non model-based damage detection and localization approach. In addition, vibration response parameters that are sensitive to local and global damage and that possess strong physical relationships with key structural dynamic properties are identified. Moreover, in order to overcome the difficulties associated with damage identification in the presence of structural nonlinearity and response nonstationarity, a wavelet transform based damage-sensitive parameter is presented for detection and localization of damage in the space domain. The level of sensitivity and effectiveness of these parameters for detection and localization of damage are demonstrated using various numerical experimental data determined from one-dimensional and two-dimensional plate-like structures.