The increasing need to reduce energy consumption has called for novel energy management systems in the building sector, due to its large share of global energy end use. Digital models of the controlled building are required for testing advanced control strategies and optimization of the control actions. In this context, the concept of the digital twin emerges as a promising solution to integrate simulation and monitoring, enabling the prediction, control, and optimization of energy use over time. This thesis proposes a methodological framework for the development and periodic re-calibration of a physics-based digital twin for buildings. The goal is to create a thermal model of the building capable of automatically constructing its own multi-zone structure from input data and maintaining a good level of accuracy over time through recurrent parameter updates.