Niet-intrusieve parametrische modelreductie voor niet-lineaire structurele mechanica met behoud van de onderliggende fysica

High-fidelity Finite Element (FE) models enable accurate simulation of nonlinear structural systems but remain computationally expensive for many-query and Digital Twin applications. Model Order Reduction (MOR) provides efficient Reduced Order Models (ROMs), although conventional nonlinear and parametric approaches often require intrusive access to the underlying FE solver. Conversely, non-intrusive data-driven methods offer greater flexibility but generally lack guarantees on the preservation of physical and numerical properties. This PhD develops Physics-Augmented, non-intrusive MOR methodologies for nonlinear and parametric structural mechanics. The proposed POD-TD-pICNN frameworks combine Proper Orthogonal Decomposition (POD) with Physics-Augmented Neural Architectures (PANNs) that learn the reduced potential energy while enforcing relevant physical properties by construction. Internal forces and tangent operators are obtained through differentiation, preserving important properties such as equilibrium consistency, symmetry, positive semi-definiteness, and hyperelastic structure. The methodologies are validated on nonlinear academic benchmarks and a large-scale industrial structural application, demonstrating the potential of Physics-Augmentation to bridge classical physics-based MOR and data-driven Machine Learning.