Deep Learning Surrogate Models with OpenSource simulators apply to the IFMIF-DONES facility.

The optimization of key fluxes in the IFMIF-DONES facility is crucial for ensuring its effectiveness in material testing for fusion energy applications. Open-source software plays a fundamental role in this process by enabling high-fidelity simulations, which are integrated in DONES-FLUX project in optimization and control loops with deep learning surrogates models. Among the open-source softwares used, OPAL is utilized for simulating the accelerator beam dynamics, Geant4 models the Li(d, xn) stripping reaction, and OpenMC performs Monte Carlo neutron transport simulations. These tools generate large datasets that train machine learning models with, allowing simulation acceleration, design optimization and deep reinforcement learning training. By integrating AI with open-source physics-based simulations, this approach significantly reduces computational costs while maintaining high precision. This work highlights the synergy between open-source software and AI in advancing surrogate modeling techniques for IFMIF-DONES and future fusion facilities.