Modeling industrial hydrocyclone operational variables by SHAP-CatBoost - A conscious lab approach

Undoubtedly hydrocyclones play a critical role in powder technology, which can considerably affect the plants' process efficiency. However, hydrocyclones were rarely modeled on an industrial scale, where a model can be used to train operators and minimize potential scale-up errors and lab costs. The novel approach for filling such a gap would be using conscious lab CL as a new concept that builds based on an industrial dataset and explainable artificial intelligence (XAI). As a novel approach, this study developed a CL and explored the interactions between hydrocyclone variables by the most recent XAI method called SHapley Additive exPlanations (SHAP), and a novel machine-learning model, CatBoost. The hydrocyclone output and the particle size of the plant magnetic separator were modeled by SHAP-CatBoost. SHAP could successfully model all the relationships, and CatBoost could predict the O-80 and K-80, where outcomes had a higher accuracy (R-2 similar to 0.90) than other conventional AIs.

Automated summary

Undoubtedly, hydrocyclones are crucial in powder technology and have a significant impact on process efficiency in plants. However, there is a lack of industrial-scale modeling of hydrocyclones, which can be used to train operators and reduce scale-up errors and lab costs. This study proposes a novel approach using conscious lab (CL) and explainable artificial intelligence (XAI) to fill this gap. The interactions between hydrocyclone variables were explored using the SHapley Additive exPlanations (SHAP) method and a new machine-learning model, CatBoost. The SHAP-CatBoost model successfully captured all the relationships and achieved higher accuracy in predicting O-80 and K-80 compared to other conventional AI methods.