Rolling Bearing Initial Fault Detection Using Long Short-Term Memory Recurrent Network

Abstract

The complete failure of the rolling bearing is a deterioration process from the initial minor fault to the serious fault, it is meaningless for guiding maintenance when the serious fault is alarmed. This work presents a novel initial fault diagnosis framework based on sliding window stacked denoising auto-encoder (SDAE) and long short-term memory (LSTM) model. In this approach, multiple vibration value of the rolling bearings are entered into SDAE by sliding window processing. Then, multiple vibration value of the rolling bearings of the next period is predicted from the signal reconstructed by the trained SDAE in the previous period using LSTM. For the given input data, the reconstruction errors between the next period data and the output data generated by trained LSTM are used to detect initial anomalous conditions. The proposed method not only utilizes the ability of SDAE to learn the inherent distribution of data, but also ensures that LSTM can extract timing relationships between data cycles, and the model is built using only normal data. The initial fault detection as a key difficulty in the operating condition monitoring and performance degradation assessment of the rolling bearing is effectively solved. Experimental and classic rotating machinery datasets have been employed to testify the effectiveness of the proposed method and its preponderance over some state-of-the-art methods. The experiment results indicate that the proposed method can effectively detect the initial anomalies of the rolling bearing and accurately describe the deterioration trend with strong robustness, and have high significance for maintenance guiding.