Structural health monitoring aims to ensure the reliable operation of structures through the use of sensors and data analysis. It is important to develop methods that not only detect damage, but also provide a measure of uncertainty, allowing engineers to make well-informed decisions about maintenance. Recently, deep learning-based amortized Bayesian inference methods have emerged as promising tools for probabilistic structural health monitoring. One such method is BayesFlow, which can theoretically enable efficient damage detection, including uncertainty quantification, directly from sensor data without requiring explicit likelihood functions. However, its practical application is hindered by the high computational cost of generating a large and diverse set of training simulations. This work proposes a method to address the problematic simulation data requirements, thereby increasing the feasibility of BayesFlow as a viable tool for structural health monitoring. A physics-informed approach is developed that enhances the summary network component of BayesFlow by integrating knowledge about the structure and sensor network. Specifically, a Graph Summary Neural Network is designed, leveraging the sequential nature of sensor measurements and the spatial topology of the sensor network. Through a progressive case study based on a model truss bridge, it is demonstrated that the GSNN surpasses baseline networks in terms of both efficiency and performance, thus bringing BayesFlow-based SHM closer to real-world deployment.

The global energy transition towards Renewable Energy Sources (RES) introduces significant challenges to grid stability due to the inherent intermittency of sources like solar and wind. Hybrid Energy Storage (HES) systems, particularly those integrating Power-to-Hydrogen-to- Power (P2H2P) technologies, are critical for mitigating these issues. However, the effective planning and control of these systems depend on accurate models of both the HES components and the Low Voltage (LV) distribution grids they support. This thesis addresses this dual modeling challenge, which is often hindered by complex component dynamics and limited grid data.
This research develops and validates two distinct frameworks. First, a control-oriented, hybrid dynamic model for an Anion Exchange Membrane (AEM) electrolyzer is proposed. This model integrates a physics-based Equivalent Circuit Model (ECM) to capture primary electrochemical dynamics with a shallow neural network trained to correct for residual, unmodeled thermal dynamics. Second, a complete, data-driven pipeline is developed to construct "digital twins" of LV distribution grids from limited static data. This pipeline uses an Integer Linear Programming (ILP) algorithm to reconstruct the most probable network topology, which is then parameterized for power flow analysis.
The hybrid electrolyzer model was validated using operational data from a commercial Enapter EL 2.1 unit, demonstrating high fidelity with a 92.03 % fit, drastically outperforming the 26.00% fit of the standalone ECM. The grid modeling pipeline was validated in a case study of a 160kVA substation in Bonaire. Despite incomplete GIS data, the resulting model accurately simulated aggregate load profiles, achieving a low Root Mean Square Error (RMSE) of approximately 6% for both active and reactive power when compared to real-world measurements.
In synthesis, this thesis delivers a robust and validated toolset for analyzing HES-grid inter- action: a high-fidelity, computationally efficient model of an AEM electrolyzer and a viable method for accurately modeling limited-data LV grids. These components enable advanced co- simulation and control design to study the role of AEM electrolyzers in providing distribution grid support.

Over the years, the conventional open-loop basal-bolus regiment has proven to be inadequate for long-term glucose management of Type-1 diabetes mellitus (T1DM) patients. The ’artificial pancreas (AP)’, which is characterized by a closed-loop control system that typically relies solely on subcutaneous glucose measurements as a feedback to deliver insulin corrections, is widely regarded as a viable alternative to the former approach. An earlier literature survey showed that model predictive control (MPC) and proportional-integral-derivative (PID) control are the two most commonly employed approaches in APs to treat T1DM, especially the former in recent years, due to its ability to incorporate flexible safety-critical constraints and account for inherent system delays. With most state-of-the art MPC controllers designed to exert direct control on insulin corrections over the patient however, the extent to which closed-loop performance can be maximised greatly depends on how accurately the plant is modelled. In addition to the immense difficulty in identifying certain physical parameters and non-linearities that describe an individual patient’s insulin-glucose pharmacokinetics, practical challenges arising from the collection of open-loop data from the patient hinder the identification process of a prediction model. The resulting model-plant mismatch is detrimental to the closed-loop performance, especially during postprandial periods when hypoglycaemic episodes are most likely to occur. As a countermeasure for the model-plant mismatch, insulin delivery rates are often tightly enforced, which may increase robustness of the MPC controller but comes at a price of reduced, sub-optimal performance due to worse-case conservative assumptions.

This thesis presents a dual-layer control scheme: a lower layer comprising of a control architecture that regulates the rate of insulin delivery for corrections against disturbances, and an upper layer that represents a performance-driven tuning framework using Bayesian Optimisation (BO). The proposed control architecture is hierarchical, consisting of an inner-loop PID controller, and an MPC which serves as the outer-loop controller that governs the reference input signal to the inner-loop control system. The upper layer tuning framework is implemented in two stages, the first on a set of clinical-related parameters, and the second a controller and model parameter, as part of the individualized control objective. In both stages, the BO framework optimises the parameters via a daily query of one parameter set and its corresponding performance value, which is computed based on experimental data. The proposed control scheme is designed and evaluated using a cohort of 10 adult virtual patients of the U.S. Food & Drug Administration (FDA)-accepted University of Virginia/Padova simulator through several in silico scenarios. Numerical evaluation and results show that the proposed hierarchical controller outperforms its baseline counterpart, and that glycaemic performance of all the tested subjects improved using the proposed tuning approach compared to under nominal settings. Notwithstanding the absence of a probabilistic safety-critical constraint in the BO problem, the employment of a barrier-like penalty in the objective function, subjected to a deterministic domain-based constraint that is augmented using a rule-based approach, demonstrated to be an efficient safeguard against hypoglycaemia during the parameter tuning process.