GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 1.2 GNGTS 2024 (Kasim et al., 2021). Specifcally, an emulator is a model where AI algorithms complement the equaton-based mathematcal representaton of physics. The emulator learns from CFD simulatons how to reproduce the CFD reference model, enabling the soluton of fuid dynamics problems in shorter tmes (Zago et al., 2023, Amato, 2023). While Eulerian methods have been extensively integrated with AI, providing high-fdelity and reliable results ( e.g., DENSE for weather predicton (Kasim et al., 2021)); the combinaton of AI and Lagrangian methods remains less consolidated. Here, we introduce an AI-based emulator designed for a SPH model. This emulator, based on an Artfcial Neural Network (ANN), is trained using SPH simulatons of complex fuids, including viscous and thermal components, such as lava. It successfully replicates the underlying physical laws and accurately predicts their spato-temporal behavior. To ensure the trustworthiness of the emulator results, it is crucial to validate its predicton ability and assess its generalizaton capability beyond the conditons encountered during the training phase. To achieve this, we conducted validaton using benchmark tests representatve of lava fows, characterized by high viscosity and temperature-dependent viscosites. We also tested the emulator’s capacity to reproduce problems and setngs with varying levels of complexity. The emulator results were then compared with the corresponding SPH simulatons, demonstratng the model’s good performance and highlightng its reliability and generalizability. References: Amato, E.; 2023: How a CFD Emulator Can Resolve the Boundary Conditons in a Viscous Flow. IEICE Proceedings Series, 76(B4L-42). Anderson, J. D., & Wendt, J.; 1995: Computatonal fuid dynamics (Vol. 206, p. 332). New York: McGraw-hill. Bonaccorso, G.; 2017: Machine learning algorithms. Packt Publishing Ltd. Bortnik, J., & Camporeale, E.; 2021, December: Ten ways to apply machine learning in the Earth and space sciences. In AGU Fall Meetng Abstracts (Vol. 2021, pp. IN12A-06). Cordonnier, B., Lev, E., & Garel, F.; 2016: Benchmarking lava-fow models. Geological Society, London, Special Publicatons, 426(1), 425-445. Del Negro, C., Cappello, A., Bilota, G., Ganci, G., Hérault, A., & Zago, V.; 2020: Living at the edge of an actve volcano: Risk from lava fows on Mt. Etna. Bulletn, 132(7-8), 1615-1625. Del Negro, C., Cappello, A., & Ganci, G.; 2016: Quantfying lava fow hazards in response to efusive erupton. Bulletn, 128(5-6), 752-763. Goodfellow, I., Bengio, Y., & Courville, A.; 2016: Deep learning. MIT press. Kasim, M. F., Watson-Parris, D., Deaconu, L., Oliver, S., Hatield, P., Froula, D. H., ... & Vinko, S. M.; 2021: Building high accuracy emulators for scientfc simulatons with deep neural architecture search. Machine Learning: Science and Technology, 3(1), 015013. Monaghan, J. J.; 2005: Smoothed partcle hydrodynamics. Reports on progress in physics, 68(8), 1703.