Remaining Useful Life Prediction for Rolling Bearings Using EMD-RISI-LSTM

Abstract

Bearings are applied in rotating machinery commonly, and the effective prediction of remaining useful life (RUL) plays an increasingly crucial role in establishing reasonable maintenance decision-making with the goal of avoiding sudden downtime and ensuring machinery safety. The selection of representative degradation features and the prediction algorithm are the key factors of RUL prediction. In this work, based on empirical mode decomposition (EMD) and long short-term memory (LSTM) network, a novel prediction architecture is proposed to improve accuracy and robustness under different working conditions. The architecture integrates three parts. Initially, a failure vibration signal is decomposed into several intrinsic mode functions (IMFs) and a residual through EMD decomposition, meanwhile, a novel similarity measurement method based on Euclidean distance and cosine similarity, namely the representative IMF selection index method (RISI), is presented to select the IMFs with more degradation features. Then, the LSTM model is trained for each of the selected IMFs and residuals. Ultimately, the prediction results of selected IMFs are utilized to determine the output for RUL prediction. Typical examples and comparative experimental analysis illustrate that the proposed prediction architecture can provide an efficient reference to predict the RUL of rolling bearings.