On Monday, January 13, 2025, the Faculty of Mechanical Engineering of the University of Sarajevo hosted an inspiring event called " Computational Mechanics and Deep learning: raising awareness ". Hybrid event, organized within the framework of an EU project "Design of Computational Mechanics and Deep learning joint master program (DSGN COM-DL)", brought together a diverse audience of students, researchers and industry experts eager to explore the transformative potential of integrating computational mechanics (CM) and deep learning (DL). The DSGN COM-DL project was implemented in the period from November 2023 to January 2025, and the project coordinator was the Faculty of Mechanical Engineering, University of Sarajevo (project leader: Assoc. Prof. Dr. Amra Hasečić). The project partners were University College Dublin, University of Zagreb and University of Zenica.
The event aimed to highlight the groundbreaking synergy between numerical simulations and machine learning, which is driving optimization and efficiency in fields such as environmental engineering, automotive, and construction. The importance of this integration is underscored by NASA's predictions that it will become an industry standard by 2030.
The event brought together distinguished speakers from leading institutions who shared their expertise and research innovations:
- Leonardo Stella(School of Computer Science, University of Birmingham) presented a groundbreaking framework using reinforcement learning for predicting porosity in additive manufacturing, validated on the high-strength A205 Al alloy. He highlighted the importance of physically informed artificial intelligence (AI)-based approaches for high-quality predictions, thereby revolutionizing additive manufacturing.
- Zeljko Tukovic(University of Zagreb) presented an innovative approach to the design of supersonic nozzles using physically informed neural networks (PINNs) to solve PDE equations describing the flow of compressible, non-ideal gases. His method offered new solutions for the design of subsonic and supersonic nozzle sections, including applications in rocket engines, wind tunnels, and steam turbines.
- Miguel Alfonso Mendez(von Karman Institute for Fluid Dynamics) spoke about innovative data-driven approaches to fluid dynamics analysis. He highlighted their impact on modeling and control of complex systems, combining real-time data assimilation and digital twins for transformative applications.
- Seid Koric(University of Illinois) presented DeepONet, a neural network that approximates solution operators of PDE equations, enabling solutions up to 10,000 times faster than traditional numerical methods. His research has shown real-world applications in multiphysics engineering and advanced manufacturing.
- Philip Cardiff(University College Dublin) explored the integration of machine learning with finite volume methods, demonstrating its ability to emulate elastoplastic models and speed up solver iterations in OpenFOAM.
- Kevin Nolan(University College Dublin) shared observations on the impact of generative AI tools like ChatGPT on education, highlighting their potential benefits and challenges in shaping the student learning process.
Participants gained insights into innovative applications of computational mechanics and deep learning, from additive manufacturing and fluid dynamics to education and research. The event highlighted the transformative role of this synergy in solving global challenges and advancing scientific understanding.
The event represents a significant step in fostering interdisciplinary collaboration and innovation.
BA