Multi-feature Fused Bidirectional Long Short-term Memory for Remaining Useful Life Prediction

Abstract

In industry, prognostic health management (PHM) is used to improve the system reliability and efficiency. In PHM, remaining useful life (RUL) prediction plays a necessary role in preventing machine failure and lowering operation cost. Recently, benefitted from deep learning technology development, many RUL prediction approaches are proposed by using long short-term memory (LSTM) or convolutional neural networks (CNN). There methods show impressive performances. However, existing deep learning based methods directly utilize raw signals. Affected by noise in the raw input, the feature representation is degraded, further degenerating the prediction accuracy. To address this issue, a multi-feature fused bidirectional LSTM (MF-LSTM) is proposed. Our proposed MF-LSTM consists of two part: multi-feature fusion (MF) module and multi-head attentive fusion (MA) module. In MF module, feature extracted by a bidirectional LSTM is combined with traditional handcrafted features. A fusion layer is proposed in MF module, which effectively combines both features and improves the feature representation. Furthermore, an attention module is proposed according to multi-head attention mechanism, which improves the performance further. To verify our MF-LSTM performance, experiments are carried out on the C-MAPSS dataset, showing a state-of-the-art performance.