Abstract:

In order to reduce safety accidents and economic losses caused by roller failures of underground belt conveyors in coal mines and improve the safety and transportation efficiency of workers and unit equipment, the N-BEATS prediction model with deep structure and residual network was applied to predict the life of rolling bearings for abnormal vibration of roller bearings at different positions under different working conditions. Firstly, the principle and structure of the N-BEATS prediction model were analyzed, and a life prediction model suitable for roller bearings was established based on the N-BEATS principle. Then, a vibration signal monitoring platform for roller bearings based on DVS technology was built against the actual roller operating conditions of a conveyor belt. The vibration signals of roller bearings under different working conditions were collected. Finally, the collected vibration data of roller bearings were input into the N-BEATS model, convolutional neural network (RCNN), and similarity prediction model, and they were compared with the actual values. The remaining life prediction quality of the three types of roller bearings was evaluated. The results show that the N-BEATS prediction model has an average absolute error increase of 5.3% and 4.1%, respectively, compared to RCNN and similarity prediction models. The relative root mean square error of the N-BEATS prediction model is increased by 6.3% and 5.2%.

Key words: neural basis expansion analysis for interpretable time series (N-BEATS) prediction model, roller, rolling bearing, life prediction, belt conveyor, vibration signal