A comprehensive study on Structural Health Monitoring (SHM) of wind turbine blades by instrumenting tower using machine learning methods

In this article, a feasibility study has been carried out in order to detect structural faults in the blade by analyzing the tower vibration. A 5-MW onshore wind turbine was modeled using the NREL FAST code. The structural faults were created on the blade, while only the tower was instrumented by accelerometers and displacement meters. The wind grid was modeled in four Wind Turbulence Intensity (WTI) levels from a highly laminar to a turbulent wind flow by NREL TurbSim code. The tower vibrations were captured in the different WTIs and blade health conditions. Time Series Amplitude (TSA), Discrete Wavelet Transform (DWT), Fast Fourier Transform (FFT) and Statistical Feature Function (SFF) of the tower vibration were calculated and utilized to reveal a faulty blade effect on the tower vibrations. Eventually, a Convolutional Neural Network (CNN) classifier was developed to classify the tower vibrations collected in the blade's healthy and faulty conditions. The results showed that a defective blade has a considerable and detectable effect on the tower vibration. It is observed that blade fault could be precisely tracked and diagnosed mostly in tower Side-to-Side displacements. Also, a reverse relationship between WTI and classification accuracy was concluded.