Prediction of roller surface remaining life of mine Roller mill based on LSTM-transformer

Abstract As the core equipment of mineral processing, roller surface wear of mine roller mills directly affects production efficiency and maintenance costs. Therefore, accurately predicting the remaining life of the roller surface is of significant importance. This paper builds a combined LSTM-Transformer model based on the Long Short-Term Memory (LSTM) network, which introduces the parallel processing capability of the Transformer while maintaining LSTM’s ability to capture long-term dependencies. LSTM-Transformer has high efficiency in processing large-scale data and can capture both long-term and short-term dependencies simultaneously, which has a good effect on the prediction of roll surface remaining life. Through the experiment on the roller mill data of CITIC Heavy Industries, the predicted remaining life results are consistent with the actual remaining life data of the roller mill. The mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) are 0.0053, 0.0001, and 0.0072. Compared with convolutional neural network (CNN), LSTM, and single Transformer, MSE decreased by 88.9%, 66.67%, and 98.59% respectively, RMSE decreased by 77.0%, 62.50%, and 91.46% respectively, and MAE decreased by 78.97%, 65.13%, and 92.50% respectively. The experiment verified the effectiveness of the proposed method in predicting the remaining life of the roller surface in roller mills.