An investigation on the use of hybrid deterministic-statistical approaches to predict building vibrations

Suitable prediction models for ground-borne vibration and noise in buildings induced by human activities should be accurate, have reasonable computational costs, and require to define input parameters that are known or, at least, feasible to determine. However, vibration and noise levels at specific building locations may be influenced by small design features and/or imperfections of the building structure. In such situations, even very sophisticated models based solely on deterministic approaches may be less practical than models that account for these uncertainties by considering statistical approaches. The aim of this work is to investigate when it may be more convenient to use hybrid deterministic-statistical methods for predicting vibrations in buildings than to rely on predictions made by deterministic models. The study focuses both on the importance that non-parametric uncertainties have on the predicted results, as well as the practical benefits of using statistical approaches for developing this type of prediction models.