Universidade Federal do Ceará
Laboratório de Mecânica Computacional e Visualização

Between October 7th, 8th, and 9th, Block 728 at the UFC Pici Campus hosted the short course “Machine Learning for Multiscale Computational Modeling.” The event focused on accelerating multiscale simulations, aiming to replace high-cost micromodels with physically-informed recurrent neural networks.

Day 1 (October 7th)

The first day was led by Dr. Iuri Rocha from Delft University of Technology. The session centered on the fundamentals of machine learning, covering basic consepts of popular algorithms (such as k-NN, Neural Network and Gauss Processes), with a focus on approximating constitutive models.

Day 2 (October 8th)

The second day was taught by Dr. Marina Maia from École Nationale des Ponts et Chaussées. The topic covered was an overview of surrogate modeling strategies, presenting different approaches from recent literature for building surrogate models aimed at micro-macro analysis.

Day 3 (October 9th)

Concluding the schedule, Dr. Marina Maia led an interactive workshop with concluding remarks. The closing session consolidated knowledge through the exploration of feed-forward Neural Networks, Recurrent Neural Networks, and physics-guided hybrid models, in addition to showcasing many of the works developed by the university’s research group.

Acknowledgments and Impact

The short course stood out for its practical approach, allowing participants to quickly assimilate and apply the concepts presented. This immersion strengthened academic ties, boosted the planning of new research, and provided valuable ideas for projects already underway in the laboratory.

We express our special thanks to Delft University of Technology and École Nationale des Ponts et Chaussées for the partnership and for the visit of Dr. Iuri Rocha and Dr. Marina Maia. The event was also made possible thanks to the continuous support of LMCV, INCT-Infra, and CNPq.