Semi-Supervised Clustering for Architectural Modularisation

Abstract

Modular construction allows for a faster, safer, better controlled, and more productive construction process, yielding quality results with low risk and controlled costs. However, despite the potential advantages of this methodology, its adoption has remained slow due to the reasonably high degree of standardisation and repetition that projects require, inexorably clashing with the unique building designs created to meet the clients’ needs. The present article proposes performing a modularisation process after the building design is complete, reaping most benefits of modular construction while preserving the unique vision and design of the building. This objective is achieved by implementing a semi-supervised methodology reliant on the clustering of individual rooms and subsequent user validation of the obtained clusters to identify base modules representative of each cluster. The proposed methodology is applied in a case study of an existing apartment complex, in which the modularisation process was previously performed manually—thus serving as a baseline. The acquired results display a 99.6% reduction in the modularisation process’ duration, while maintaining a 96.4% Normalised Mutual Information Score and a 93.3% Adjusted Mutual Information Score, justifying the continuous development and assessment of the methodology in future works.