Physics Informed Machine Learning for Industrial Glass Furnace Optimization

The performance and lifetime of industrial glass furnaces strongly depend on the coupling between combustion dynamics, heat transfer, and refractory degradation. This work proposes a Physics Informed Machine Learning (PIML) approach to model the thermal behavior and material wear inside a flat glass furnace. The model integrates physical laws such as heat conduction and diffusion equations into the learning architecture, allowing it to capture temperature gradients, conductivity variations, and long-term degradation under real operational data. Using CFD-based simulations generated in Siemens STAR-CCM+ as synthetic data combined with field measurements, the PIML model predicts the evolution of refractory damage over time with improved accuracy and reduced computational cost compared to traditional solvers. The results demonstrate the potential of hybrid modeling frameworks that merge data-driven and physics-based approaches to enable digital twins for predictive maintenance and energy efficiency optimization in high-temperature industrial systems.