Prediction of remaining useful life of metro traction motor bearings based on DCCNN-GRU and multi-information fusion.

A single type of sensor is susceptible to interference and limited degradation information can be extracted. Therefore, a multi-information fusion-based bearing remaining useful life prediction method using the dual-channel convolutional neural network (DCCNN) and the gated recurrent unit (GRU) is proposed. Firstly, vibration sensors are utilized as a basis for signal collection, while acoustic emission sensors are introduced as a complement to obtain more comprehensive degradation information. Secondly, the time domain, frequency domain and time-frequency domain features of the vibration signal and acoustic emission signal were extracted respectively to construct a comprehensive bearing degradation feature set. Third, multiple evaluation indicators are used to comprehensively evaluate the degradation features, and effective degradation features that are highly related to bearing degradation are selected for feature fusion. Subsequently, the DCCNN-GRU model was established, which captures and comprehensively utilizes different degradation feature information in each channel through the DCCNN network structure, and employs GRU to process the time relationships and dependencies in sequence data to solves the vanishing gradient problem. Finally, an experimental bearing test bench was constructed to collect data, and this data was used to experimentally validated the model and compared with other methods to demonstrate its feasibility and effectiveness. [ABSTRACT FROM AUTHOR]