8 - A waveform-based feature extraction model for AE signal analysis for structural health monitoring of composite wind turbine blade

Abstract

In the last chapter, we performed bending fatigue tests and damage mode identification of a 59.5 m composite wind turbine blade. This chapter further deals with health monitoring of blade. One major limitation of acoustic emission (AE) technique lies that AE signals are constantly processed based on several AE parameters or indexes, which fail to reconstruct original signals. Here, a waveform-based feature extraction model for AE signal analysis based on the wavelet packet decomposition is developed for structural health monitoring of the blade. The waveform-based model covers all features for reconstructed signals in the frequency domain. AE waveform is acquired after AE attenuation calibration and sensor arrangement are performed. Without the requirements for signal preprocessing including denoising, feature correction, and feature selection, the developed model is proved to be effective and feasible for periodic health condition monitoring of blade by analyzing original sampling signals. In addition, two hyperparameters in the developed model, including the scatter number for the definition of original signals and the selection of wavelet basis function, are demonstrated to show no effect on the decomposition results, indicating the robustness of the model.