A hybrid signal-dynamics health index fusion framework for prognostics of small-scale wind turbine blades

Ensuring the dependable operation of wind turbines is essential for enhancing energy efficiency and minimizing maintenance costs, particularly in small-scale wind turbines where blade failures can severely affect safety and performance. This paper proposes a hybrid condition monitoring and prognostic framework for small-scale wind turbine (SSWT) blades that combines signal-based features, structural dynamic characteristics, and Health Index (HI) fusion. Experimental measurements were obtained from strain gauges and accelerometers mounted on turbine blades under three operating states: healthy (C), localized degradation (R), and global degradation (L). The acquired strain and vibration signals were preprocessed using band-pass filtering, normalization, and wavelet-based denoising to mitigate noise and environmental disturbances. Features were extracted from multiple domains, including time-domain indicators (RMS, skewness, kurtosis), frequency-domain descriptors (peak frequency and spectral energy), and dynamic properties such as natural frequencies, transmissibility, and mode shape curvatures. Dimensionality reduction using PCA and t-SNE demonstrated a clear separation among the three blade conditions. Several HIs were constructed based on PCA projections, autoencoder reconstruction errors, and Mahalanobis distance measures, which were then integrated into a Combined Health Index (CHI). The results indicate that transmissibility changes and natural frequency shifts are the most sensitive indicators of blade stiffness degradation. Compared with individual HIs, the CHI shows improved robustness to noise and operational variability. In addition, Remaining Useful Life (RUL) was estimated using Weibull-based degradation modeling and LSTM-inspired prediction, effectively capturing nonlinear damage evolution and supporting predictive maintenance of SSWT blades.