The design of engineering components must meet increasing technological demands in ever shorter development cycles. To face these challenges, a holistic approach is essential that allows for the concurrent development of part design, material system and manufacturing process. Current approaches combine experiments and numerical simulations, but quickly reach their limits due to the large number of simulations required for finding suitable design. Therefore, data-driven machine learning methods have recently emerged as a competitive alternative for time- and resource-intensive numerical simulations. In particular, graph network simulators (GNS) have shown promising results. They enable fast and accurate predictions on unseen mesh geometries while being fully ifferentiable for optimization [1]. However, these models rely on large amounts of data, such as numerical simulations. Physics-informed neural networks (PINNs) [2] offer an opportunity to train neural networks with partial differential equations instead of labeled data, but have not been extended yet to handle time-dependent simulations of arbitrary meshes. This work introduces physics-informed graph network simulators (PI-GNS), a hybrid approach that combines PINNs and GNSs, to quickly and accurately predict the thermal evolution of unseen engineering parts in new process settings. More specifically, the model can predict non-linear and non-stationary simulations, remotely inspired by composite processing, of arbitrary geometries with inhomogeneous material distribution. Further, a model trained on small generic meshes is able to precisely predict simulations on large complex meshes, which enables a fast pre-training at early stages of the development process. REFERENCES [1] Pfaff, T., Fortunato, M., Sanchez-Gonzalez, A. and Battaglia, P. (2020, October). Learning Mesh-Based Simulation with Graph Networks. In International Conference on Learning Representations. [2] Raissi, M., Perdikaris, P. and Karniadakis, G. E. (2019). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics, 378, 686-707.