Deep learning-enhanced physical modelling for tape-casting slurry microstructures of solid oxide cell substrates

Efficient modelling of tape-cast slurry microstructures is a key requirement for accelerating the manufacturing development of solid oxide cell (SOC) fuel-electrode substrates. While physics-based discrete element method (DEM) simulations provide high-fidelity microstructural insight, their computational cost restricts rapid process exploration. This work presents a hybrid surrogate modelling framework that combines DEM simulations with deep learning to efficiently model tape-cast slurry microstructures of the fuel-electrode substrate. Our proposed framework captures how processing parameters influence the resulting slurry microstructure at the particle scale, enabling prediction of the slurry microstructure without the cost of full-scale DEM simulations. The hybrid model reliably reproduces key microstructural characteristics and microstructural metrics consistent with physics-based observations of DEM-generated data, demonstrating the robustness of the hybrid modelling strategy. Our hybrid surrogate approach minimised the computational cost versus the pure DEM simulation, reducing the calculation time from 1230 to 30 min for the whole slurry simulation process. Although certain slurry microstructures still present minor deviations, the approach reduces reliance on computationally intensive DEM simulations and supports faster validation and early-stage optimisation. As a proof of concept, our work provides a foundation for developing data-efficient and scalable digital modelling tools that could support high-throughput and guide experiments in the SOC manufacturing research field.