Predicting the Remaining Useful Life of Oscillating Bearings via Recurrent and Convolutional Neural Networks Trained on Rotating Bearings

Abstract

A Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN) are employed to predict the Remaining Useful Life (RUL) of fully rotating bearings using acceleration data. The CNN utilizes frequency domain features, while the RNN incorporates both time and frequency domain features. Initially tested on a public dataset, both models are further applied to new test bench data of oscillating bearings.The study highlights the importance of time information in RUL prediction, evidenced by the RNN’s good performance compared to the CNN’s poorer results for oscillating bearings.