An intelligent prognostic approach considering multi-mode degradation processes

Accurate remaining useful life (RUL) prediction is of great importance for predictive maintenance of industrial equipment. The multi-mode of degradation processes and the relations among historical monitoring data are not taken into account comprehensively in the existing studies. To address this issue, a novel intelligent RUL prediction approach for multi-mode degradation processes considering the bidirectional memory is proposed in this paper. The mode classification is achieved via a universal change-point detection technique. The features are selected and fused by the distance correlation (DC) coefficient to obtain a clear representation of health condition. Bidirectional long short-term memory (BiLSTM) and gated recurrent unit (GRU) are introduced to enhance features and to establish the RUL prediction model. The effectiveness of the proposed method in accuracy is demonstrated by a case study of blast furnace.