Impact of particle compositions on screening performance: Modeling and prediction via data segmentation approach

Abstract

Variations in particle composition along the screen pose significant challenges for modeling screening dynamics. This study examines the influence of feeding conditions on screening efficiency through a segmental approach and investigates various machine-learning algorithms for predictive modeling. Segmentation and transformation of passing rate curves were clearly observed and discussed. A novel algorithm was developed to predict screening efficiency for screens of varying lengths, and a method for segmenting simulated data for machine learning modeling is proposed. Results show that the mass content of easy-to-sieve particles affects segmental screening efficiency oppositely to that of difficult-to-sieve particles. Specifically, increasing the mass content of easy-to-sieve particles narrows the crowded screening region while enhancing overall screening efficiency. For particles of different sizes, K-Nearest Neighbors Regression, Multilayer Perceptron Regression, and Support Vector Regression provide effective prediction models. For screens of varying lengths, Support Vector Regression outperforms the others, achieving an average deviation of less than 3%.