Wind Turbine Planetary Gear Fault Identification Using Statistical Condition Indicators and Machine Learning

Abstract

This work describes an automated condition monitoring framework to process and analyze vibration data measured on wind turbine gearboxes. The current state-of-the-art in signal processing often leads to a large quantity in health indicators thanks to the multiple potential pre-processing steps. Such large quantities of indicators become unfeasible to inspect manually when the data volume and the number of monitored turbines increases. Therefore, this paper proposes a hybrid analysis approach that combines advanced signal processing methods with machine learning and anomaly detection. This approach is investigated on an experimental wind turbine gearbox vibration data set. It is found that the combination of physics-based statistical indicators with machine learning is capable of detecting planetary gear stage damage and significantly simplifying the data analysis and inspection in the process.