A novel unsupervised learning method for intelligent fault diagnosis of rolling element bearings based on deep functional auto-encoder

Recently, several studies tried to develop fault identification models for rolling element bearing based on unsupervised learning techniques. However, an accurate intelligent fault diagnosis system is still a big challenge. In this study, a deep functional auto-encoders (DFAEs) model with SoftMax classifier was designed for valuable feature extraction from massive raw vibration signals. To maximize the unsupervised feature learning ability of the proposed model, various activation functions were applied in an effective methodology, these hidden activation functions enhance significantly the sparsity of the training data-set. The proposed method was validated using the raw vibration signals measured from the machine with different bearing conditions. The achieved results showed that the high-superiority of the proposed model comparing to standard deep learning and other traditional fault diagnosis methods in terms of classification accuracy even with massive input data sets.