Fault Diagnosis and Remaining Useful Life Prediction of the High-Speed Shaft Gear and Bearing in Wind Turbine Drivetrain

• Main Authors: Lin, Li-Hsuan, 林立璿
• Other Authors: Lin, Yih-Hwang
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/q6ruun

碩士 === 國立臺灣海洋大學 === 機械與機電工程學系 === 107 === This study investigates fault diagnosis of the high-speed shaft gear and automated condition classification and remaining useful life prediction of the high-speed shaft bearing in wind turbine drivetrain. The datasets studied are obtained from the vibration signals of the high-speed shaft gear of a 3MW wind turbine and the high-speed shaft bearing of 2MW wind turbine. The accelerometer monitoring the gear conditions may be disturbed by the vibration signals of other components and rotation of the turbine blades changes due to the variable wind speed. The time synchronous average technique is used to superimpose and average the sampled signal data sets of the gearbox vibration according to the rotational speed of the gear to be analyzed, leading to the interference elimination of the other component signals and noises. When the gear is failing, its signal sideband amplitudes are more significant as compared to those of the healthy gear. This paper calculate the sideband energy ratio (SER) to determine the severity of gear faults, and use the sideband power factor (SBPF) to avoid the disadvantage of the ratio distortion which leads to misjudgment of the gear conditions due to the low peak value at the gear mesh frequency in the calculation of the sideband energy ratio. This study shows that through proper signal processing, we can greatly reduce the interference of non-synchronous signals and noise. In addition, an automatic identification artificial intelligence in the operating state was established. Probabilistic Neural Network (PNN) was applied to perform training and testing of the neural network model with the vibration signals of high-speed drivetrain bearing. The study shows that the automatic identification of bearing fault state can be achieved very well. Based on the Exponential Degradation model, this study could perform real-time prediction of the remaining useful life of bearing in wind turbine drivetrain. The feature extraction method of the trained model is partly derived from the aforementioned fault diagnosis technique, and it can calculate the remaining life time from vibration data of the device in the early stage of the life cycle. The results of the study show that the RUL model has highly predictive performance and provides a reliable reference for the timely maintenance of the equipment.