As modern district heating networks integrate buildings with multiple energy sources, fault detection has become increasingly relevant and critical. This study investigates the effectiveness of an unsupervised data-driven fault detection approach to identify stuck valve and faulty thermostatic radiator valve scenarios in the baseboard radiators of an office building. A baseline model for a typical Swedish office building was developed, featuring a ground-source heat pump, solar photovoltaic-thermal panel, water-based radiators, and a connection to the district heating system to support its heating demand. Multiple fault scenarios were considered in the model, involving partially stuck valves and thermostatic radiator valves that deviated from their intended setpoints. Synthetic noise was added to generate faulty scenarios. The model performed well in detecting severe stuck valve faults but showed lower performance on less severe faults and faulty thermostatic radiator valves. The insights gained from this research emphasize the importance of fault monitoring in the context of evolving buildings connected to district heating networks.