Wear Prediction of Petrochemical Granulator Gearbox Using Multidimensional Transformer Network Via Online Oil Monitoring

Large granulator is used for chemical production in petrochemical enterprises, and its gearbox needs intelligent wear prediction to ensure the safety of equipment operation and avoid unexpected shutdown. In this paper, we proposed a Multidimensional Transformer Network (MDTF) via online oil monitoring to achieve wear prediction. First, random forest algorithm is used to rank the importance of online oil monitoring features, and then the data is normalized. In addition, multi-time scale preprocessing is used to obtain redundant and complementary wear characteristics. Second, the multi-head attention mechanism is designed to combine the attention features of different subspaces and reduce the dimension of each attention vector. Third, the positional encoding and encoder-decoder module is designed to embed positional information into feature vectors to improve the model's ability to learn sequence samples. Finally, various error indexes are used to evaluate the prediction performance of the proposed model and realize intelligent early warning in advance. We used continuous online monitoring data of the main gearbox of a large granulator at a petrochemical plant in Guangdong Province, China, from July 2018 to May 2023 for validation. The experimental results show that the MDTF model can accurately predict wear particles of different sizes, with Root Mean Square Error (RMSE) of 0.00604 and Mean Absolute Error (MAE) of 0.00439%, and has good generalization performance. By comparison, the proposed method is obviously superior to the existing Long Short-Term Memory Network (LSTM) and Temporal Convolutional Network (TCN) methods.