Elsevier

Advanced Powder Technology

Volume 32, Issue 11, November 2021, Pages 4141-4148
Advanced Powder Technology

Original Research Paper
Modeling of particle sizes for industrial HPGR products by a unique explainable AI tool- A “Conscious Lab” development

https://doi.org/10.1016/j.apt.2021.09.020Get rights and content
Under a Creative Commons license
open access

Highlights

  • Using EAI for constructing a CL could be a strategic approach toward the digitalization.

  • SHAP-XGBoost as a robust EAI could translate industrial datasets to human basis information.

  • HPGR particle size productions could accurately model with a SHAP-XGBoost system.

  • As a benchmark, the SHAP-XGBoost could digitize powder technology industries.

Abstract

High-Pressure Grinding Rolls (HPGR), as a modified type of roll crushers, could intensively reduce the energy consumptions in the mineral processing comminution units. However, several problems counted for their operational modeling, especially in the industrial scales. Expanding a conscious laboratory (CL) as a recently developed concept based on the recorded datasets from the HPGR operational variables could be tackled those complications and fill the gap. Moreover, constructing such a CL base on explainable artificial intelligence (EAI) systems would be an innovative point for the digitalizing powder technology industries. Using a robust EAI model as a strategic approach could significantly improve system transparency and trustworthiness to convert any complicated black-box machine learning to a logical human basis system. This study introduced the SHapley Additive exPlanations (SHAP) and extreme gradient boosting (XGBoost) as the latest powerful EAI tool for the CL modeling of the particle sizes produced by an industrial HPGR (P80) in the Fakoor Sanat iron ore processing plant (Kerman, Iran). SHAP precisely assessed multivariable relationships between the monitored operational variables and correlated them with the HPGR P80. SHAP values showed relationship magnitudes among variables and ranked them based on their effectiveness on the P80 prediction. The working gap demonstrated the highest importance for the P80 prediction. XGBoost could precisely predict the P80 and showed higher accuracy than typical machine learning methods (random forest and support vector regression) for constructing the CL of HPGR. These significant outcomes would open a new window for robust consideration of the EAI models within powder technology.

Keywords

Working gap
Roller speed
Particle properties
Machine learning
XGBoost

Cited by (0)