The rolling bearing is a key component of rotating machinery. The prediction of its remaining useful life (RUL) is a critical challenge in prognosis and health management. The current data-driven rolling bearing RUL prediction approaches still require a lot of prior knowledge to extract features, construct health indicators and set failure thresholds, which are affected to some degree by anthropogenic factors. To address the issues listed above, an effective RUL prediction method based on the temporal convolutional network (TCN) with an attention mechanism is proposed. The approach includes two steps: feature extraction and RUL estimation. Firstly, the frequency spectrum of the original vibration signal is taken as the input of the stack denoising auto-encoder to obtain the depth feature representation and reduce the computational complexity. Then the depth feature is input into Multi-head attentional TCN. It can automatically extract the features closely related to the degradation state of mechanical equipment and estimate its RUL. Finally, the case study is carried on the rolling bearing data set of PRONOSTIA and the compared results with other methods are also given out. It is shown that the prediction error of the proposed method is the lowest and the score is highest, so it can provide reliable guidance for the health management of rotating machinery and equipment.

Bearing,Remaining useful life prediction,Temporal convolutional network,Multi-head self-attention