Fault detection of offshore wind turbine drivetrains in different environmental conditions through optimal selection of vibration measurements

In this study, a vibration-based fault detection method is proposed for offshore wind turbine drivetrain based on the optimal selection of the acceleration measurements. The main aim is to find the sensor positions mounted on the drivetrain that provides the most fault-related information. In fact, this study tries to optimize the vibration sensors suggested by ISO standards in terms of their position and number in order to get accurate fault detection results. The faults are considered in a set of bearings with a high probability of failure in a 5-MW reference drivetrain high-fidelity model installed on a spar-type floating wind turbine. Different simulated shaft acceleration measurements are examined under three environmental conditions. Correlation analysis is first performed on the measurements to see how the different faults and environmental conditions affect the correlation between the measurements. Then, a combined principal component analysis (PCA) and convolutional neural network (CNN) is employed as the fault detection method through the optimal vibration measurements. The prediction findings demonstrate that only two vibration sensors, one near the main shaft and another near the intermediate-speed shaft, can fully detect the considered faulty bearings. Also, it will be shown that the axial vibration data give more promising results than the radial ones which can be used in virtual digital twin models.