A novel temporal convolutional network with residual self-attention mechanism for remaining useful life prediction of rolling bearings

Remaining useful life (RUL) prediction has been a hotspot in the engineering field, which is useful to avoid unexpected breakdowns and reduce maintenance costs of the system. Due to the high nonlinearity and complexity of mechanical systems, traditional methods cannot meet the requirements of medium-term and long-term prediction tasks, and often ignore the influence of temporal information on prediction performance. To solve this problem, this paper proposes a new deep learning framework - Temporal convolutional network with residual self-attention mechanism (TCN-RSA), which can learn both time-frequency and temporal information of signals. First, we input the marginal spectrum of vibration signals to TCN. The causal dilated convolution structure in the TCN has the ability to capture long-term dependencies and extract high-level feature representations from the time-frequency domain at the same time. Then, a residual self-attention mechanism is introduced to obtain the feature contribution degree of different moments in the bearing degradation process. Finally, an end-to-end RUL prediction implementation can be established based on TCN-RSA network. The effectiveness of the proposed method is verified by IEEE PHM 2012 Data challenge datasets and XJTU-SY datasets respectively. The comparative study indicates that the proposed TCN-RSA framework outperforms the other state-of-the-art methods in RUL prediction and system prognosis with respect to better accuracy and computation efficiency.