The study of the vibro-acoustic properties of light-weight floor slabs has garnered increased interest, given that these structural assemblies represent a large percentage of the carbon emissions in buildings and that current designs are dimensioned principally to reduce vibrations and noise transmission. Computing the frequency-dependent radiated sound power of a floor slab with complex geometries can be computationally expensive, making their design and optimization more difficult. However, not all numerical studies relevant to the vibro-acoustic behavior of the slabs are expensive. It is known that vibration patterns and sound radiation of thin shells are highly correlated to their modal frequencies and shapes. This study proposes the use of Machine Learning methods for the prediction of the radiated sound power of cross-laminated timber floor slabs based on geometrical and modal analysis inputs. A large dataset of triangularly discretized geometries was produced, their radiated sound power was computed and used to train several ML models. The resulting models are tested for accuracy and speed by using unseen geometries. The resulting models reveal the most relevant geometric and modal attributes for the prediction of the vibro-acoustic performance as well as the potential for these methods to be used in the early design phase.
Echenagucia et al. (Wed,) studied this question.