AI-Enhanced Heritage Building Modeling: Bridging Digital Documentation and Adaptive Reuse in Mediterranean Thermal Complexes

This study proposes an integrated approach to the conservation and revitalization of the Roman thermal complex of Hammam Mellègue, located in northwestern Tunisia. Although classified as a historic monument since 2016, the site remains structurally degraded and socially disconnected from its original therapeutic identity. Faced with the limitations of an institutional project favoring technocratic modernization, this research develops an alternative methodological framework based on Heritage Building Information Modeling (HBIM), enriched by artificial intelligence (AI), to support the analytical documentation, critical interpretation, and adaptive reuse of Mediterranean thermal heritage. The methodology combines terrestrial laser scanning (TLS), semi-automated parametric modeling, and segmentation assisted by convolutional neural networks (CNN) and generative neural networks (GAN), transforming the digital survey into a decision-making tool for heritage governance. Beyond geometric reconstruction, the HBIM model integrates historical, material, pathological, and environmental data, offering a systemic understanding of the site's architectural dynamics. The study explores how AI can assist in typological recognition and the restoration of missing elements, while ensuring interpretive transparency and human validation in accordance with international conservation charters. Preliminary results highlight the potential of this HBIM-AI framework to reconnect ancient thermal architecture with its contemporary social and therapeutic functions, proposing a model of “living conservation” rooted in community health and cultural sustainability. This interdisciplinary approach aims to renew the dialogue between heritage, emerging digital technologies, and territorial development in the Maghreb and Mediterranean context.