The operation and maintenance of transport infrastructure - bridges, tunnels, and road networks - requires an understanding of structural behavior across multiple modeling scales. High fidelity models capturing this interaction across scales to predict structural behavior are desirable, yet computationally expensive, motivating surrogate approaches that embed the governing physics within data driven learning frameworks to bridge these scales. Physics‑informed neural networks have been proposed to embed the governing equations directly into the training objective, thereby aiming to enforce physical consistency during learning and inference [1]. However, when modeling structural dynamics, plain PINNs often struggle in complex geometries and under realistic loading conditions, exhibiting convergence failures or physically inconsistent predictions. To combine the best of high-fidelity models and PINNs, finite-element based neural networks (FENNs) were developed and have been applied to linear [2] and nonlinear [3] problems in structural and fluid mechanics. In a FENN, the residual of a finite element discretization of the weak form of the governing equation is embedded into the loss function, leveraging the improved conditioning of the weak form-based discretization to facilitate training and accelerate convergence. The finite-element method is employed for spatial discretization, while the neural network learns the parameter dependence of the solution, enabling efficient parametric predictions. In this contribution, we extend FENNs to model beams, frames and plates. We benchmark the resulting physics constrained surrogate against PINNs on examples that illustrate its relevance for practical applications such as structural health monitoring and digital twinning. [1] Raissi, M.; Perdikaris, P.; Karniadakis, G. E. (2019): Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. In: J. Comput. Phys. 378, S. 686–707. [2] Meethal, R.E., Kodakkal, A., Khalil, M., Ghantasala, A., Obst, B., Bletzinger, K. U., & Wüchner, R. (2023). Finite element method-enhanced neural network for forward and inverse problems. Adv. Model. and Simul. in Eng. Sci. 10, 6. [3] Griese, F.; Hoppe, F.; Rüttgers, A.; Knechtges, P. (2025): Preconditioned FEM-based neural networks for solving incompressible fluid flows and related inverse problems. In: J. Comput. Appl. Math. 469, S. 116663.