Wavelet Based Real-Time Planetary Gearbox Health Monitoring Under Non-Stationary Operation

Modern wind turbines employ a multistage planetary gearbox to convert the low rotation speed of the turbine blades to the speed required by the generator. Studies have shown that gearbox failures rank the highest among the contributors to an unplanned downtime. Real-time condition monitoring systems can provide useful insights to a turbine's operation there by reducing the chance of an unplanned downtime. This study focused on developing an automated real-time fault detection methodology for a miniature wind turbine planetary gearbox subjected to non-stationary loading. The data-driven multi-component fault detection methodology implements multiple scales of continuous wavelet transform to extract information from a non-stationary signal. This multi-scale approach ensures that all possible component signatures are captured and organized into a feature rich data-set. The wavelet coefficients were then abstracted using descriptive statistics to reduce size of data-set. This was done so as to minimize the computation requirements. The proposed methodology was tested using a pattern recognition algorithm based on Artificial Neural Networks and two Decision Tree algorithms. The results indicated that the proposed methodology worked well with the Decision Tree algorithm thereby ensuring that such a method could be deployed for a compact signal analyzer, where processing capability and memory capacity is premium. Further, a stand-alone application was deployed to automate the process with the trained machine learning model. The proposed method proved its capability in classifying multi-component faults under non-stationary operating conditions.