Wear Prediction of Petrochemical Granulator Gearbox Using Multiscale Temporal Convolutional Network via Online Oil Monitoring

As a key component of petrochemical granulator, gearbox is used for long-term continuous operation under complex working conditions, and its failure may lead to unexpected shutdown of equipment. Therefore, the wear prediction is quite significant. In this paper, we propose a wear prediction method using a Multiscale Temporal Convolutional Network (MTCN) via online oil monitoring. First, we use multiscale data preprocessing to capture the comprehensive information of the signals collected by various sensors for online oil monitoring. Second, we conduct correlation analysis to screen out several groups of online monitoring variables that can best reflect the wear status of the gearbox. Third, we construct a temporal convolutional network consisting of two residual modules and one convolution layer to train historical data. Finally, we use evaluation indicators to measure actual data and forecast data to achieve early warning. The presented method is tested on the online oil monitoring data of a petrochemical granulator in Guangdong Province, China from 2018 to 2022. The experimental results show that our method can realize the prediction of historical data with an RMSE of 1.2614 and MAPE of 0.3626%. In comparison, it is significantly better than the existing methods, such as CNN, and LSTM.