This talk explores the interplay between data, thermodynamics, and action principles in the development of thermodynamics‑aware digital twins for damage assessment in complex materials microstructures. Data is leveraged in two key ways: image data is used to represent as-built material microstructures, while measurable material state data supports data-driven computation. The link between microstructure, material state, and structural damage response is grounded in thermodynamic principles and the principle of least action. The Support Vector Machine (SVM) algorithm is employed for automatic microstructure segmentation, enabling direct model discretization from image pixels without the need for body-fitted mesh generation. Inelastic material behavior is modeled in a purely data-driven manner, bypassing traditional constitutive models that often lack generalizability across loading conditions. To capture localized damage and microstructural features with coarse discretization, we introduce neural network (NN) enrichment of the RKPM framework. The NN approximation is formulated through energy minimization, with optimal parameters encoding the location, orientation, and transition behavior of damage zones. Regularization ensures discretization-independent solutions, and convergence properties are analytically derived and numerically verified. For transient dynamics, the NN-enriched formulation is based on action minimization and symplectic integration, yielding solutions consistent with classical field theory. The effectiveness of this digital twin framework is demonstrated in modeling damage evolution in composite materials and structures, and comparison with experimental results validated the accuracy and reliability of the proposed computational framework. REFERENCES [1] Wang, Y., Chen, J. S., Casebolt, S., Network-Enriched RKPM for Dynamics based on Action Minimization, Computer Methods in Applied Mechanics and Engineering, Vol. 451, pp.118662, 2026. [2] Baek, J., Chen, J. S., A Neural Network-Based Enrichment of Reproducing Kernel Approximation for Modeling Brittle Fracture, Computer Methods in Applied Mechanics and Engineering, Vol. 410, pp.116590, 2024. [3] Baek, J., Chen, J. S., Susuki, K., INeural Network enhanced Reproducing Kernel Particle Method for Modeling Localizations, International Journal for Numerical Methods in Engineering, Vol. 123, pp.4422-4454, 2022.