Machine Learning Guidance of Manufacturing Process of Solid Oxide Cells

Solid Oxide Cells (SOCs) are promising energy conversion devices with applications in both electricity generation and chemical fuel production. Fuel Electrode-Supported Cells (FESCs) particularly offer a robust architecture, but their performance and reliability are heavily dependent on the manufacturing process of a porous substrate material. Traditional manufacturing parameter optimisation primarily relies on trialand- error experimental methods, which are resource-intensive, time-consuming, and difficult to scale for industrial production. This PhD dissertation addresses these challenges by combining physics-based and data-driven modelling to link manufacturing parameters, microstructure, and physical properties of the fuel-electrode substrate. These novel approaches are developed to investigate how the manufacturing parameters affect the substrate microstructure, offering the digital tools to optimise the fabrication process of SOCs. Specifically, a coarse-grained physics-based modelling framework was developed to simulate the slurry and dried microstructures of fuel-electrode substrates for the tape casting and drying stages of fuel cell substrate materials. The microstructures generated by this model provide digital representations of the coarse substrate material, which can be used as input for multi-scale simulation methods. Concurrently, the thesis focuses on applying Machine Learning (ML) techniques to optimise SOC manufacturing processes. By integrating advanced data collection through electronic laboratory notebooks and on-site characterisation, ML models have been trained to predict the substrate properties at each manufacturing stage. This data-driven approach helps to identify fundamental relationships between key manufacturing parameters and substrate properties, enabling the optimisation of the manufacturing routes for achieving target characteristics of the fuel-electrode substrate. These physics-based and data-driven strategies form a complementary framework, bridging experimental production and predictive simulation. The contribution of this research lies in providing novel modelling tools, integrating experimental, numerical and data-driven methods, and supporting the systematic, reproducible, and energyefficient scaling up of SOC production.