A Local Weighted Multi-Instance Multilabel Network for Fault Diagnosis of Rolling Bearings Using Encoder Signal

Rolling bearings are the key components of modern machinery, and thus, the diagnosis of bearing faults plays a crucial role in ensuring the reliable operation of the equipment. Due to the limitation of cost and environmental conditions, the traditional fault diagnosis method based on the vibration signal is not always suitable for practical use. An encoder is widely used in the mechanical transmission system, and its signal contains rich information about health status and working conditions. However, the encoder signals are mostly used to monitor the rotational speed in the industrial and not considered as an alternative to the vibration signal. In this article, a framework based on a local weighted multi-instance multilabel (LWMIML) network is proposed to make full use of the information contained in the encoder to achieve the purpose of bearing fault diagnosis. This framework first constructs the instance generator to acquire feature vectors. On this basis, a local weighted strategy based on the MIML network is used to enhance the potential association between features and labels. Finally, the superiority of this method is verified by the data of the fault simulation test rig. The results indicate that the proposed method offers a promising tool, which will help to reduce the cost of future intelligent fault diagnosis systems.