Wind Turbine Mass Imbalance Detection Using Reassigned Spectrogram and CNN-LSTM Networks

This paper presents online detection of mass imbalance of wind turbine using real-time measurements from turbine's drivetrain. The drivetrain measurements are preprocessed to produce a spectrogram, i.e., time-frequency representation of the measurements that contains extracted information regarding the pertinent frequency and harmonics of imbalanced-induced vibration. More importantly, the power spectrum reassignment method is applied to obtain a sharper spectrogram, in which the spectral power is more centralized around the local time-frequency center. This would contribute to more effective and discriminating features that neural networks can use to accurately detect and classify the imbalance. A convolutional Neural Network (CNN) together with a long short-term memory (LSTM) network is then used for detecting the mass imbalance of wind turbines with help of the reassigned spectrogram images. Numerical experiments using simulated wind turbine data from FAST show that the proposed method outperforms existing methods in terms of detection accuracy.