Intelligent prognostics of bearings based on bidirectional long short-term memory and wavelet packet decomposition

Bearing is one of the most sensitive components widely used in rotary machines and main cause for unexpected breakdown in rotating machinery. Bearing failure can lead to a lengthy downtime of the machine. Accurately predicting the damage trend of bearing is essential for planning maintenance, avoiding machine shutdowns and improving systems reliability. To reduce the maintenance cost of machine downtime, it is desirable to perform fault prognostics to enable predictive health management for bearings. This paper proposes a new data-driven approach for bearing prognostics based on wavelet packets decomposition and bidirectional long short-term memory, for preprocessing and tracking degradation process to estimate the remaining useful life. The proposed approach has two steps. The first step is to detect bearing's degradation process by learning the historical data and the second step is to predict the remaining useful life with the aid of a degradation model. The proposed approach is validated by bearing's life data obtained from a run-to-failure experiment. Results show that the proposed approach can detect the bearing degradation process successfully and predict the remaining useful life.