A Novel Bearing Health Indicator Construction Method Based on Ensemble Stacked Autoencoder

Abstract

In the area of data-driven bearing prognostic, the construction of health indictor from condition monitoring data is important. This paper presents a novel bearing health indicator construction method based on ensemble stacked autoencoders. Firstly, the proposed ensemble stacked autoencoders extract features directly from the FFT results of raw vibration signals. Then, a deep neural network which serves as a non-linear transformation is trained to map the multi-dimensional learned features to a one-dimensional health indicator. Finally, the proposed method is validated using the IEEE PHM2012 Challenge dataset. To show the superiority of the proposed method, its performance is evaluated and compared with other methods. The results demonstrate that the proposed method can automatically and effectively build high-quality health indictor from raw data without any signal processing and manual feature engineering.