Monitoring wind turbine blades (WTB) is an important aspect when assessing the health of wind turbines. Structural Health Monitoring (SHM) systems enable continuous monitoring of the condition of WTB during operation. When SHM is coupled with advanced data analysis techniques, damage detection can be improved by customizing the methodologies to the structure being monitored. The work presented in this manuscript introduces an SHM methodology based on a Semi-supervised damage detection algorithm that uses preliminary findings to reinforce the selection of features used to identify damage. The methodology proposes a novel technique for feature extraction by sorting the acceleration values in each vibration response. Then, an adaptive feature selection algorithm is applied to identify the most sensitive characteristics of the feature for damage detection. This technique enhances the correlation between measurements of the same blade status and therefore the performance of the proposed SHM methodology. The methodology was implemented on real acceleration measurements on an operational Vestas V27 WTB. The results were compared with those from an alternative Semi-supervised methodology that considers only the measurements from the undamaged WTB. The comparison of the results demonstrated that the proposed adaptive feature selection algorithm enhances damage diagnosis.