A new spatial-temporal reduced-order model (STROM) has been developed by researchers from Central South University in China and Texas A&M University at Qatar to predict the temperature field in zinc fluidized bed roasters with high precision and efficiency. This advancement, published in
Engineering, aims to address the challenges of accurately and promptly monitoring the temperature field in large-scale industrial roasters, which is crucial for improving the quality of zinc calcine and other products.
The zinc fluidized bed roaster is a key piece of equipment in the zinc smelting industry. Its internal temperature field directly affects the quality of zinc calcine and other products. However, accurately perceiving the temperature field in real-time has been a significant challenge due to the roaster’s large spatial dimensions, limited observation methods, and complex multiphase, multifield coupled reaction atmosphere. Traditional computational fluid dynamics (CFD) models, while accurate, are computationally expensive and time-consuming, making them impractical for real-time applications.
To tackle these issues, the research team proposed the STROM, which combines several innovative methods to achieve fast and accurate temperature field prediction based on sparse observation data. First, they introduced an initial field construction based on data assimilation (IFC-DA) method to match the initial physical field with sparse observation data. This method reconstructs the initial conditions of the model with high resolution, ensuring that the model’s starting point accurately reflects the actual operational state.
Second, to reduce the high computational cost of comprehensive CFD simulations under multiple working conditions, the team proposed a high uniformity (HU)-orthogonal test design (OTD) method. This method ensures high information coverage of the temperature field dataset under typical working conditions by considering multiple factors and levels of component, feed, and blast parameters. By introducing the centered
L2 deviation, the HU-OTD method achieves a more uniform distribution of experimental designs, covering the parameter ranges more effectively.
Finally, the researchers developed a spatial-temporal predictive model (STPM) that considers the spatial correlation between observed temperatures and the temperature field, as well as the dynamic correlation of observed temperatures over time. This model enables rapid prediction of the entire temperature field through sparse observation data, significantly improving computational efficiency.
The effectiveness of the proposed STROM was validated through extensive experiments. The results showed that the method could achieve high-precision and fast prediction of the roaster temperature field under different working conditions. Compared to traditional CFD models, the prediction root-mean-square error (RMSE) of STROM was less than 0.038, and the computational efficiency was improved by 3.4184×10⁴ times. Notably, STROM also demonstrated good prediction ability for unmodeled conditions, with a prediction RMSE of less than 0.1089.
This research represents a significant step forward in the intelligent transformation of process manufacturing. By leveraging advanced data assimilation techniques, optimized experimental design, and machine learning, the STROM provides a practical solution for real-time temperature field prediction in industrial roasters. This not only enhances the efficiency of zinc smelting processes but also opens new possibilities for applying similar models in other large-scale industrial applications.
The paper “STROM: A Spatial-Temporal Reduced-Order Model for Zinc Fluidized Bed Roaster Temperature Field Prediction,” is authored by Yunfeng Zhang, Chunhua Yang, Keke Huang, Tingwen Huang, Weihua Gui. Full text of the open access paper:
https://doi.org/10.1016/j.eng.2025.04.013. For more information about
Engineering, visit the website at
https://www.sciencedirect.com/journal/engineering.