Investigation of advanced experimental and numerical techniques for the effective real-time health monitoring on mechanical structures

Advances in experimental and numerical methods enable increasingly precise Structural Health Monitoring, where real-time identification and localization of damage by means of hybrid digital twins has become a central objective. Achieving this requires reliable, mechanics-consistent inference from noisy measurements.This thesis develops a physics-guided inference framework that couples the modified Constitutive Relation Error (mCRE) with sequential data assimilation, yielding a Modified Dual Kalman Filter (MDKF) for stepwise model parameter identification and damage tracking. The coupling of variational and sequential paradigms enables consistent tracking of evolving parameters at each update, while targeted algorithmic optimizations improve numerical stability and computational efficiency under model uncertainty and measurement noise. The method is here applied to experimental data obtained from optic fiber sensing on real specimens in order to demonstrate applicability on representative scenarios. A comparative study against a direct sequential mCRE baseline indicates gains in robustness and assimilation rate, and qualitative validation with Digital Image Correlation supports the identified trends. Finally, a CgFEM-inspired localization strategy and an initial extension toward the sequential identification of nonlinear behaviors indicate scalability to richer physics and support operational deployment.