The cooling system of a nuclear reactor is essential for ensuring the safety and efficient operation of the TRIGA PUSPATI Reactor (RTP). However, conventional fault detection methods often struggle with the nonlinearities and complexity of the system. This study proposes a data-driven Fault Detection and Diagnosis (FDD) framework for the RTP cooling system using a Graph Neural Network (GNN). In this approach, the system is modelled as a graph, where sensors are represented as nodes and their interactions as edges, enabling the GNN to capture component relationships and improve anomaly detection accuracy. A scaled-down control rig was developed to replicate the RTP cooling system for experimental data collection and validation. The model was trained and evaluated using experimental data, and its performance was assessed using evaluation metrics including accuracy, precision, recall, and F1 score. For comparison, an Artificial Neural Network (ANN) model was also implemented. The results show that the GNN outperforms the ANN in fault detection performance. Both models were evaluated under offline conditions, while real-time validation was conducted for the GNN. In the online implementation, a Wi Fi module was used to transmit live sensor data to the cloud for continuous monitoring. The control rig operates independently in a laboratory environment and is not connected to any operational nuclear reactor system, with cloud communication used strictly for experimental monitoring without control actions. The results demonstrate that the proposed GNN achieves accurate and reliable fault detection in both offline and real-time applications, highlighting its potential to enhance reactor safety and operational reliability.