Wind turbines are complex electromechanical systems that require continuous monitoring to ensure operational efficiency, reduce maintenance costs, and prevent critical failures. Despite advancements, challenges remain in adapting structural and condition monitoring techniques to highly variable environmental conditions while maintaining reliable fault detection. This study proposes a data-driven model based on a Graph Neural Network (GNN) architecture that utilises experimental wind turbine data, including acceleration measurements and environmental parameters, as input to classify operational states and identify structural patterns without relying on predefined labels. The framework first extracts features from raw multi-sensor data and constructs a graph based on the similarity between samples. The GNN propagates and fuses feature Information across graph nodes, enhancing representation learning for downstream classification tasks. The embedded features obtained from the graph are then used as input to deep learning classifiers, which are performed using a baseline dense neural network layer to assess classification accuracy. The proposed model is applied to an onshore wind turbine and evaluated across various operational conditions, including normal and coupled failure operations. Results show that the proposed model achieves good accuracy in classification tasks using multiple sensors, demonstrating its potential as an effective and scalable approach for structural health and condition monitoring of wind turbines. These findings support the development of open-source tools for robust, data-driven structural monitoring in renewable energy systems.