Multimodal Data Processing for Building Material Property Predictions

Abstract: Building materials heavily influence a building’s environmental footprint. Effective material-related performance assessments demand knowledge of existing buildings’ characteristics, including material types, physical properties, and environmental indicators. However, structured data on material properties is scarce and often confined to closed repositories. While unimodal prediction models provide machine learning pipelines for predicting material properties, they overlook the intricacies of real-world scenarios requiring contextual insights from diverse data sources. This study introduces a multimodal building material property prediction method that incorporates building characteristics as contextual information. A residential building serves as the case study, focusing on brick facade to validate the proposed approach.