One of the significant tasks in remaining useful life (RUL) prediction is to find a good health indicator (HI) that can effectively represent the degradation process of a system. However, it is difficult for traditional data-driven methods to construct accurate HIs due to their incomprehensive consideration of temporal dependencies within the monitoring data, especially for aeroengines working under nonstationary operating conditions (OCs). Aiming at this problem, this article develops a novel unsupervised deep neural network, the so-called times series memory auto-encoder with sequentially updated reconstructions (SUR-TSMAE) to improve the accuracy of extracted HIs, which directly takes the multidimensional time series as input to simultaneously achieve feature extraction from both feature-dimension and time-dimension. Further, to make full use of the temporal dependencies, a novel long-short time memory with sequentially updated reconstructions (SUR-LSTM), which uses the errors not only from the current memory cell but also from subsequent memory cells to update the output layer's weight of the current memory cell, is developed to act as the reconstructed layer in the SUR-TSMAE. The use of SUR-LSTM can help the SUR-TSMAE rapidly reconstruct the input time series with higher precision. Experimental results on a public dataset demonstrate the outstanding performance of SUR-TSMAE in comparison with some existing methods.

Remaining useful life (RUL) prediction can provide critical information for complex equipment health states (HSs) assessment. Historical long-term HS degradation trends and current short-term HS changes are two key factors affecting RUL prediction. However, most existing deep learning-based RUL prediction methods only consider learning short-term HS change features but ignore learning long-term HS degradation trend features, which limits to improvement of RUL prediction performance. To address this problem, this paper develops a RUL prediction framework based on a combination of time-series auto-correlation decomposition (TSACD) and convolutional neural network (CNN), which can learn both long-term and short-term features of mechanical equipment, so that achieves more robust and accurate RUL prediction. First, a novel TSACD method is proposed to extract historical long-term features from collected long-term monitoring data. The advantage of TSACD is to highlight the true signal by reinforcing periodic features through the Auto-Correlation mechanism and to separate pure trend components using a deep time-series decomposition architecture. Second, the long-term features are mapped to the same space as the short-term HS monitoring data using a group linear layer, which is intended to be aligned and fused with short-term monitoring data. Third, the fused features are fed into a CNN for RUL prediction. Finally, a series of comparison experiments on the C-MAPSS dataset and the XJTU-SY dataset validate the outstanding prognostic performance of the proposed method. The experimental results show that the proposed method outperforms the other RUL prediction methods.