An ensemble decision tree methodology for remaining useful life prediction of spur gears under natural pitting progression

Abstract

This article presents an ensemble decision tree–based random forest regression methodology for remaining useful life prediction of spur gears subjected to pitting failure mode. The random forest regression methodology does not require an elaborate statistics background knowledge and has an inbuilt health indicator selection capability compared to other existing data-driven remaining useful life prediction approaches. A correlation coefficient parameter based on the residual vibration signal is used for monitoring and detecting the pitting progression in spur gears. The effectiveness of the correlation coefficient of the residual vibration signal is assessed over the other existing health indicators for pitting fault progression. To show the inbuilt best health indicator selection capability of the random forest regression model, initially, eight indicators (existing seven and the correlation coefficient of the residual vibration signal) were used for model training. In addition, the effect of fusing the vibration sensor data from multiple positions on the gearbox on prediction accuracy of the random forest regression model is also evaluated. The accuracy in the remaining useful life prediction is found to increase after fusing the correlation coefficient of the residual vibration signal based health indicator derived from the accelerometers located at multiple positions on the gearbox in comparison to data from a single accelerometer. Furthermore, the accuracy of the proposed methodology is tested and proven using five accelerated run-to-failure experimental data collected from the specially built test rig.