Physics-Informed Modelling of Rail Track Degradation from Sparse Inspection Data

Annual track measurements create a structural blind spot because geometry and rail defects evolve continuously, while data is captured only once per year. This limitation is even more critical on lightly trafficked lines, where inspection intervals are longer despite potential rapid local degradation at transitions or weak subgrade zones. Sparse measurements lead to unstable degradation rate estimation, missed acceleration phases, and reactive maintenance decisions. A learning-based approach, trained on historical degradation patterns and physics-based models, can reconstruct intermediate track states between inspections with quantified uncertainty. This reduces temporal under sampling, improves geometry growth estimation, and supports more reliable maintenance planning and life-cycle optimization.