Feasibility study on a full‐scale wind turbine blade monitoring campaign: Comparing performance and robustness of features extracted from medium‐frequency active vibrations

The present work investigates the performance of different features, extracted from vibration-based data, for structural health monitoring of a 52-meter wind turbine blade during fatigue testing. An active vibration monitoring system was used during the test campaign, providing periodic excitation of single frequencies in the medium-frequency range, and using accelerometers to measure the vibration output on different parts of the blade. Based on previous work from the authors, data is available for the wind turbine blade in healthy state, with a manually induced damage, and with progressively increasing damage severity. Using the vibration data, different signal processing methods are used to extract damage-sensitive features. Time series methods and time-frequency domain methods are used to quantify the applied active vibration signal. Using outlier analysis, the health state of the blade is classified, and the classification accuracy through use of the different features is compared. Highest performance is generally obtained by auto-regressive modeling of the vibration outputs, using the auto-regressive parameters as features. Finally, suggestions for future improvements of the present method toward practical implementation are given.