Prediction of sound insulation at low frequencies using artificial neural networks

Low frequency noise, generated by outdoor and indoor sources, is an increasing problem and there is a need for methods of measurement and prediction of low frequency sound transmission, particularly between dwellings. This work presents a new approach to the low frequency sound transmission problem. Artificial Neural Networks, ANNs, are information processing systems that can store knowledge from their environment through a learning process and thus can be used in developing models for prediction. ANNs were applied to existing field data, in the standard frequency range, to confirm the classical mechanisms of sound transmission between dwellings. They then were applied to Finite Element Method modelled data for the frequency range 40-100 Hz. Results show the influence of parameters in sound transmission at low frequency that are not important at higher frequencies. Of particular significance is the main room dimension, normal to the plane of the party wall, and the edge conditions of the party wall. The results of applying ANN to sound transmission at low frequencies are promising but there remains a need for larger data sets, particularly to improve the prediction of equal room configurations where maximum acoustic coupling takes place.