Reduced-Order and Machine-Learning Models for Engineering Simulation

Reduce the cost of transient multiphysics simulations while retaining useful full-field predictions. Developed POD, DMD, recurrent neural networks, autoencoders, and latent-space models for electrospray and automotive thermal systems.

Achieved approximately 1% error for deposited-droplet-area prediction in one study and approximately 98% reported predictive accuracy for an automotive curing digital twin.

[More information contact]
email: silvio.candido@ubi.pt