Hybrid ARIMAX–LSTM Modeling for Enhanced Vibration Prediction in Rotating Machinery: Application to a Cement Mill Fan

In intensive industrial applications, such as cement manufacturing, the reliable operation of key equipment is critical to ensure the equipment longevity and a continuous efficient production. This study focuses on forecasting the mechanical vibrations in an Induced Draft (ID) fan used in a cement mill, where abnormal vibration levels may indicate impending faults or performance degradation. Given the dynamic and nonlinear behavior of such systems, accurate prediction is both challenging and essential for condition-based maintenance. A hybrid forecasting framework is utilized in the current study, that integrates the statistical accuracy of the Autoregressive Integrated Moving Average with Exogenous Variable (ARIMAX) model with the nonlinear learning capabilities of Long Short-Term Memory (LSTM) networks. The ARIMAX component captures the linear structure and exogenous influences, while the LSTM models capture the residual nonlinearities providing a holistic approach to the vibration behavior. The proposed hybrid model is benchmarked against three standalone approaches: ARIMAX, LSTM, and the Machine Learning (ML)-based XGBoost algorithm. The experimental results demonstrate that the hybrid ARIMAX–LSTM model significantly outperforms individual models in terms of prediction accuracy, as measured by the RMSE, MAE, and R² statistical metrics. These findings highlight the potential of combining classical time series models with Deep Learning (DL) architectures for advanced prognostics in industrial rotating machinery.