FPCA-SETCN: A Novel Deep Learning Framework for Remaining Useful Life Prediction

The accurate prediction of remaining useful life (RUL) can serve as a reliable foundation for equipment maintenance, thereby effectively reducing the incidence of failure and maintenance costs. In this study, a novel deep learning (DL) framework that incorporates functional principal component analysis (FPCA) and enhanced temporal convolutional network (TCN) is proposed for RUL prediction. Precisely, FPCA is employed to capture the changing patterns in multistream degradation trajectories. Subsequently, the reconstructed signals from FPCA are fed into a convolutional block for extracting deep-level features. An enhanced squeeze-and-excitation (ESE) block is then incorporated into the network for adaptive feature recalibration, enhancing the network’s ability to focus on the most relevant information. The framework includes a TCN module augmented with hybrid attention mechanisms, comprising ESE and spatial attention (SA) blocks, to optimally capture forward and backward sequence information of the feature tensor. The efficiency and feasibility of the proposed approach are demonstrated through case studies on both the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) and Center for Advanced Life Cycle Engineering (CALCE) battery datasets. The proposed method achieves the lowest root-mean-square error (RMSE) of 15.56 on the C-MAPSS dataset and 0.03 on the CALCE dataset. The comparative studies highlight the superiority of the proposed network over existing DL algorithms.