In this paper the effects of damage in a three-blade wind-turbine system are investigated theoretically. First, the dynamic model of a three-mass oscillator is studied in order to predict how stiffness variations can affect the natural frequencies. Then, experimental modal analysis is carried out on PLA replicas of an NREL 5 MW blade, validating these predictions from the model dynamics through hammer-impact tests at multiple pitch angles. We extract damage-sensitive features from the resulting frequency response functions (modal frequencies and damping ratios of the first few modes) to train a multiclass support vector machine. The classifier predicts the crack type and the blade in which the damage is inserted achieving 96.97 % classification accuracy, demonstrating that vibration-based monitoring can reliably detect and localize the damage in the blade. Moreover, it is shown that effects of the damages can also be observed in the features of the neighboring blades, enabling detection by neighbouring-blade sensors. This concise framework offers a low-cost, high-accuracy solution for structural health monitoring in wind-energy systems.