Dynamic forces are introduced into the ground by passing rail vehicles, vibration-intensive construction works, or geothermal plants. These forces propagate through the soil as structure-borne sound waves and are transmitted via foundations and building structures to enclosing surfaces. When sufficiently strong, these dynamic loads can be perceived as vibrations or heard as sound radiated from surfaces (so-called secondary airborne sound) (see figure).

Preventing disturbances to residents, as well as damage to buildings, is in the interest of both residents and operators of transport systems and construction sites. Mitigation measures may typically be applied either on the emission or the immission side and generally require detailed knowledge of the entire transmission chain — from force input through soil and structural dynamics to the sound radiation characteristics of building surfaces. A wide range of predictive methods exists, employing partially different input parameters and exhibiting significant variability in methodology and accuracy. In particular, no standardized calculation methods are available for vibration-intensive construction works in close proximity to buildings. As a result, practice often relies on very conservative mitigation measures or extensive continuous monitoring:

Within the framework of the research project, a model is therefore being developed on the basis of extensive measurement data, with the specific aim of predicting vibrations and secondary airborne sound generated by near-field sources. To this end, newly acquired measurement series as well as historical datasets are collected, processed, and analyzed. The focus lies on identifying relevant parameters, correlations, and temporal relationships within the data. Methodologically, various approaches are investigated, including classical statistical methods as well as machine learning techniques such as neural networks and deep learning models.

The selection and design of the models follow a data-driven and iterative process, depending on data quality, data volume, and application requirements. The objective is to develop a scalable and extensible predictive model that can be flexibly adapted to new datasets and research questions while providing robust predictions, particularly for near-field construction sites.

The research project is kindly supported by Müller-BBM Industry Solutions GmbH.