This master thesis looks at gallium Supported Catalytically Active Liquid Metal (SCALM) systems which show great promise in different catalytic applications, specifically propane dehydrogenation is taken as its main focus. Machine learning interatomic potential models (MLIP) provide a way to accurately simulate large atomic systems at long time scales. This new and improved speed provides the possibility to investigate dynamic systems. A selection of the best currently available models were benchmarked on computational efficiency and accuracy to make an informed choice of MLIP model architecture. Models were finetuned further to improve accuracy with DFT data. Using the best model, MD simulations were performed on gallium nanoparticles of different sizes and solutes. Pt, Pd and Ag were observed to equilibrate in the subsurface as was seen in previous literature. Breaking from this trend, gold was observed to reside in the surface and bismuth was seen to migrate to the surface of the nanoparticle. Clustering MACE descriptors showed it could accurately discern 4 groups: solute atoms, surface atoms, high and low coordination bulk. Furthermore, propane was added to a GaPt SCALM system to capture the dynamic site formation using MD, but none were recorded. The entire computational workflow was designed in Python and can serve as a basis for (multi-architectural) MLIP nanoparticle computational workflows.

This thesis explores the techno-economic feasibility of a decentralised autonomous green methanol plant powered by renewable energy sources. The study examines the potential for renewable energy harvesting in remote areas, where grid connectivity is absent and autonomous operation is essential. Green methanol, produced from renewable feedstocks such as carbon dioxide (CO2) and water (H2O), offers a promising alternative to fossil-based fuels and chemicals due to its versatility, transportability, and role as a credible, easy-to-handle green energy vector. The concept developed in this research integrates solar photovoltaic (PV) with a Direct Air Capture (DAC) for CO2, Atmospheric Water Harvesting (AWH), and the indirect CO2 hydrogenation process. A Multi-Criteria Decision Analysis (MCDA) was conducted to select the most suitable technologies for each subsystem. To assess the techno-economic viability of the plant, a comprehensive energy and mass balance was constructed, followed by a cost modelling framework. The Specific Energy Consumption (SEC) for methanol production was calculated to be 2119 [kJ/mol], highlighting the energyintensive nature of the process. Duqm (Oman) showed the most favourable conditions in terms of the techno-economic feasibility due to its consistent solar irradiance and low solar PV and storage costs.
To develop the autonomous concept in detail, a custom simulation framework was built using
SciLab (open-source version of Matlab). The optimised Base Case configuration resulted in a
LCOEproduction of 153.34 [$/MWh] which aligns with projected e-fuel costs in 2030. Additional simulations were conducted to explore varying operational strategies and control mechanisms. A process analysis was also performed to identify the relevant timescales for autonomous
operation, revealing that hourly and daily intervals are most critical for mitigating intermittency and managing energy supply to ensure a continuous and stable performance. A Model Predictive Control (MPC) framework was implemented to manage energy allocation and methanol production levels under variable solar conditions. The uncertainty-aware MPC variant demonstrated the plant’s ability to adapt dynamically, even during periods of unexpected weather changes and fluctuating irradiance.
The study concludes that decentralised autonomous methanol production is technically feasible
and economically competitive in well-selected locations. It emphasises the importance of integrated control strategies, component sizing, and predictive modelling in achieving stable and efficient operation. The findings contribute to the advancement of autonomous renewable energy supply chain development and offer a scalable blueprint for future deployment in remote regions. Future research should focus on improving component maturity, integrating holistic system design, and exploring machine learning-based autonomy to further enhance system resilience and cost-effectiveness.

Recent studies emphasise the importance of assessing battery degradation under real operational conditions instead of under laboratory-controlled, often isolated, accelerated ageing conditions. Therefore, this thesis investigates degradation in a grid-scale LFP/graphite battery energy storage system (BESS) subjected to highly dynamic real-world load profiles. Degradation over two years was characterised by a 10% increase in internal resistance and a 9% loss in lithium inventory (LLI), through the first method of data-driven ageing analysis. Key indicators supporting this include internal resistance trends, quasi-constant voltage segments, Tafel-like slopes, and voltage relaxation behaviour.

Currently, BESS owners have to rely on annual warranty checks set by the BESS suppliers and checked by the suppliers through annual self-conducted capacity tests, while system owners often lack insight into internal degradation processes. This highlights the need for independent analysis and improved diagnostic tools.

To quantify degradation, both empirical equivalent circuit models (ECMs) and physics-based models (SPM, SPMe, DFN) were applied. SPMe and DFN show good agreement with measured voltage responses under constant power capacity test simulations. A SEI growth model was fitted to explain calendar ageing, reproducing the observed trends in resistance and LLI, and confirming the assumption of SEI growth as the dominant ageing mechanism. The ageing behaviour showed strong SOC dependence. This work demonstrates how combining data-driven qualitative analysis with empirical and electrochemical modelling enables tracking of cell-level degradation in real-world BESS.