Machine learning surrogates for vibro-acoustics
Neural network models that predict the acoustic behaviour of building components in milliseconds.
High-fidelity vibro-acoustic simulation is far too slow for early-stage design exploration. This project trains convolutional and graph neural networks as surrogate models for the vibro-acoustic behaviour of timber floor slabs and metamaterial arrays, and uses reinforcement-learning recommenders to guide designers through large parametric design spaces. The goal: simulation-quality feedback at the speed of design iteration.
Collaborators
- Mohammad Tabatabaei Manesh