Contribution
Predicting the sound insulation of lightweight frame walls using an artificial neural networks approach
* Presenting author
Abstract:
Artificial neural networks (ANN) are effective in trend analysis since they can deal with complex pattern-oriented problems. As a result, ANN models have previously been applied to sound insulation predictions of floors and facades with some nice results. Thus, an ANN model is used in this paper to predict the airborne sound insulation of lightweight frame wall systems. The prediction model is based on 320 laboratory measurements, and data is gathered from 8 different frame systems, including wooden and steel studs. Input parameters to the network consist of the surface density of the panels and the insulation material, the distance and the thickness of studs and cavities, and more. Predictions from the model show satisfactory results across the frequency spectrum (50-5000 Hz), with the highest accuracy between 100-1000 Hz. The model also detects the critical frequency of the panels and the resonance frequency of the cavities. The root mean square error (RMSE) for the predictions is, as a mean value, around 3 which is within the acceptable range. The work demonstrates that an ANN approach can be used to predict airborne sound insulation of specific lightweight frame wall systems with good accuracy.