Energy-efficient design of cyclone separators: Machine learning prediction of particle self-rotation velocities

The self-rotation of particles within cyclone separators has garnered significant attention due to its critical role in separation processes and mass transfer enhancement. This study investigates the complex dynamics of particle self-rotation in cyclone separators, focusing on the statistical patterns influenced by factors such as cyclone structure, particle characteristics, and operational parameters. By synthesizing data from numerous high-quality academic sources, the research classifies particle self-rotation velocities in cyclones into four distinct categories. Advanced statistical analyses, including correlation analysis and principal component analysis, were employed to uncover key internal relationships and reduce data dimensionality, identifying three principal components. Machine learning algorithms-namely SVM, Bayesian regression, KNN, and Adaboost-were applied to predict self-rotation velocities under varying conditions. Among these, Adaboost demonstrated superior predictive performance, achieving R2 values of 0.88–0.97 on training sets and 0.8–0.95 on test sets, underscoring its strong generalization capabilities. Additionally, SHAP and partial dependence plots analysis provided a comprehensive visualization of feature importance. These findings offer a cost-effective and efficient predictive framework for particle self-rotation velocities in cyclones, addressing the challenge of limited visibility within energy-related chemical equipment.