QUIESST database of European NRD

Cluster analysis for the in-situ method for sound insulation for NRD posts (EN 1793-6)

In order to identify NRD families based on the frequency spectra of the collected data a cluster analysis has been performed. An implementation of the k-means algorithm, designed to work specifically on longitudinal data, was used. Firstly, the optimal number of clusters was chosen according to the Calinski-Harabasz criterion. The definitions of clusters A, B and C are shown in the top diagram. The second step was to analyse in detail which material types are present in the identified clusters, considering also the roughness of the barrier surface (if available) as an additional parameter for the classification of the spectra. For each available material the identification with one or more clusters can be seen. A perfect correspondence between cluster and material type was not always found. In the present case 3 clusters have been identified. Cluster A represents 53% of the data, containing different material types like metal (SM), concrete (SC) and timber (ST). Cluster B with around 39% of the data represents mainly timber barriers (ST) or metal barriers (SM). Cluster C with about 7% of the data represents only few barriers made of concrete (SC), particularly those barriers with small roughness. Metal and timber barriers are equally divided between clusters A and B.

figure figure