Cluster analysis as a basis for local masonry typology

Abstract

Detailed knowledge of local construction features plays a remarkable role in examining and modelling historic buildings, both in the field of mechanical and energy performances. This study proposes a standard procedure for local masonry typology and explores the use of a statistical tool – cluster analysis – to define historic masonry types in local areas. The purpose of using cluster analysis as a tool for local masonry typology is to reduce the subjective influence of the observer. Consequently, the accuracy of local context analysis can be maintained, but using a homogeneous typology structure, intended as a general instrument for the detailed thermal and mechanical analysis of historic buildings. The proposed method was applied to four local contexts, namely the historic centers of four small cities in Sicily: Castel di Lucio, Patti, Santo Stefano di Camastra, and Tusa. All masonry walls with visible arrangement were examined in the case studies, thus collecting a dataset of 157 walls. Cluster analysis was carried out through the R software, considering each examined wall as an observation. Gower distance was selected as the distance metric. Partitioning Around Medoids algorithm (PAM) and the average silhouette width were used. Clusters have been identified both analyzing each case study and the entire dataset. In the latter, the analysis resulted in three homogeneous clusters, with average silhouette width equal to 0.46. Distribution of relevant construction features (average dimensions of masonry units and mortar joints, MQI) in the three clusters of the overall dataset suggest classification based on cluster analysis is appropriate to the technical examination of masonry.