Fault Detection in Single Stage Helical Planetary Gearbox Using Artificial Neural Networks (ANN) and Decision Tree with Histogram Features

Drive train failures are most common in wind turbines. Lots of effort has been made to improve the reliability of the gearbox but the truth is that these efforts do not provide a lifetime solution. Majority of failures are caused by bearing and gearbox. It also states that wind turbine gearbox failure causes the highest downtime as the repair has to be done at Original Equipment Manufacturer [OEM]. This work aims to predict the failures in planetary gearbox using fault diagnosis technique and machine learning algorithms. In the proposed method the failing parts of the planetary gearbox are monitored with the help of accelerometer sensor mounted on the planetary gearbox casing which will record the vibrations. A prototype has been fabricated as a miniature of single stage planetary gearbox. The vibrations of the healthy gearbox, sun defect, planet defect and ring defect under loaded conditions are obtained. The signals show the performance characteristics of the gearbox condition. These characteristics and their number of occurrences were plotted in a histogram graph. Predominant statistical features which represent the fault condition were selected using decision tree algorithm. Using these features the Artificial Neural Network (ANN) and J48 algorithms were trained and tested to classify the faults. The accuracy of the machine learning algorithm greatly helps in deciding the optimum time to carry out the required maintenance operation.