Neuro-fuzzy computing for vibration-based damage localization and severity estimation in an experimental wind turbine blade with superimposed operational effects

Fueled by increasing demand for carbon neutral energy, erections of ever larger wind turbines (WTs), with WT blades (WTBs) with higher flexibilities and lower buckling capacities lead to increasing operation and maintenance costs. This can be counteracted with efficient structural health monitoring (SHM), which allows scheduling maintenance actions according to the structural state and preventing dramatic failures. The present study proposes a novel multi-step approach for vibration-based structural damage localization and severity estimation for application in operating WTs. First, partial autocorrelation coefficients (PACCs) are estimated from vibrational responses. Second, principal component analysis is applied to PACCs from the healthy structure in order to calculate scores. Then, the scores are ranked with respect to their ability to differentiate different damage scenarios. This ranking information is used for constructing hierarchical adaptive neuro-fuzzy inference systems (HANFISs), where cross-validation is used to identify optimal numbers of hierarchy levels. Different HANFISs are created for the purposes of structural damage localization and severity estimation. For demonstrating the applicability of the approach, experimental data are superimposed with signals from numerical simulations to account for characteristics of operational noise. For the physical experiments, a small scale WTB is excited with a domestic fan and damage scenarios are introduced non-destructively by attaching small masses. Numerical simulations are also performed for a representative fully functional small WT operating in turbulent wind. The obtained results are promising for future applications of vibration-based SHM to facilitate improved safety and reliability of WTs at lower costs.