Improved Prediction of Settling Behaviour of Solid Particles through Machine Learning Analysis of Experimental Retention Time Data

TL;DR

This paper introduces a machine learning approach to predict the settling behavior of particles crossing density-stratified interfaces, providing new correlations and insights from an extensive experimental dataset.

Contribution

The study presents a novel ML-based method for analyzing particle settling data, revealing key dimensionless parameters influencing delay time and offering improved predictive capabilities.

Findings

Delay time depends on six key parameters identified by ML.

Reynolds and Froude numbers show a strong correlation.

Symbolic regression indicates Froude number alone predicts behavior.

Abstract

The motion of particles through density-stratified interfaces is a common phenomenon in environmental and engineering applications. However, the mechanics of particle-stratification interactions in various combinations of particle and fluid properties are not well understood. This study presents a novel machine-learning (ML) approach to experimental data of inertial particles crossing a density-stratified interface. A simplified particle settling experiment was conducted to obtain a large number of particles and expand the parameter range, resulting in an unprecedented data set that has been shared as open data. Using ML, the study explores new correlations that collapse the data from this, and previous work Verso et al. (2019). The ``delay time,'' which is the time between the particle exiting the interfacial layer and reaching a steady-state velocity, is found to strongly depend on six dimensionless parameters formulated by ML feature selection. The data shows a correlation between the Reynolds and Froude numbers within the range of the experiments, and the best symbolic regression is based on the Froude number only. This experiment provides valuable insights into the behavior of inertial particles in stratified layers and highlights opportunities for future improvement in predicting their motion.