LSTM and ELM based transfer learning approach for online bearing remaining useful life prediction

Abstract

Rolling bearings are important components for mechanical equipment. Approximately 30% of rotating machinery failures are caused by rolling bearing damage according to statistics. These failures may cause serious consequences that are difficult to estimate. In recent years, deep learning has been widely used in bearing remaining useful life (RUL) prediction, and has achieved remarkable results. However, as bearing performance varies dramatically in the late degradation stage, most existing deep learning-based RUL prediction methods have some drawbacks regarding solving this problem. In this paper, a nonlinear fusion transfer learning method based on long-short term memory networks (LSTM) and extreme learning machines (ELM) is proposed to realize online bearing RUL prediction, especially in the late degradation stage. Specifically, since it is easier to obtain offline bearing vibration signal data, we choose it as the source domain data and the online bearing data as the target domain data. Time-frequency domain features in the source domain are extracted firstly. Subsequently, these features are fed to the LSTM and ELM respectively. Then, the prediction results of the two networks are trained by non-linear fusion based on ELM network. Afterwards, the network is transferred to the online bearing to predict its RUL. Comparison experiments are designed and implemented with a public dataset. The experimental results verify the feasibility and effectiveness of the proposed method. Keywords-transfer learning; long-short term memory ne