A Method of Fault Diagnosis for Rotary Equipment Based on Deep Learning

In traditional fault diagnosis, the effect of fault diagnosis is determined by the features directly, and almost features are depended on expert's experience and knowledge. For a new object or field, it's very difficult to extract features manually, while it's indispensable. With the development of digital information and the internet of everything, the data of equipment monitoring such as rotor machinery will tend to mass. In the face of mass data, the traditional feature extraction is more difficult to adapt. However, with deep learning driven by big data, the mapping from the input to output without feature extraction can be achieved. And deep learning has the transfer learning ability, which can be applied for imbalance samples. In this paper, a brief introduction for deep learning is described firstly. Secondly, a deep learning application for fault diagnosis is illustrated. Thirdly, the good performance of deep learning is verified by bearing test data. Finally, the application of transfer learning capability of deep learning on fault diagnosis of rotary equipment with scarce fault condition samples is represented.