A Novel Knowledge Sharing Method for Rolling Bearing Fault Detection Against Impact of Different Signal Sampling Frequencies

Accurate and efficient bearing fault diagnosis method plays an important role in the modern industrial system. In practical applications, a large number of data are employed for fault analysis. However, the performance of the fault diagnosis model varies with the sampling frequency of the training data. Theoretically, higher sampling frequency (HSF) data contains much richer fault information, the model trained with HSF data has a better performance compared with that of lower sampling frequency (LSF). However, most of the intelligent methods only focused on the bearing fault diagnosis at a specific data sampling frequency. They did not take into account the impact of different sampling frequencies on fault diagnosis performance. To this end, a novel knowledge-sharing multi-task (KSMT) model is proposed in this article. The KSMT model consists of two task branches, which can automatically share useful features from the task with HSF input (HSF-task) to the task with LSF input (LSF-task) by a feature selection and fusion module (FSF-module). The effectiveness of the proposed model is verified on a rolling bearing dataset. Experimental results show that the performance of the LSF-task can be improved by the proposed knowledge-sharing mechanism. Compared with other state-of-the-art methods, the KSMT model has a better performance and the results can effectively demonstrate that the KMST model can overcome the limitation of less information on fault diagnosis performance.