Analyzing the impact of charge particle rebound on spherical collector efficiency via explainable machine learning

Abstract

This study investigates the collection efficiency of charged spherical collectors while explicitly accounting for charged particle rebound using a three-dimensional Lagrangian particle tracking model integrated with the Konstandopoulos rebound criterion, applied for the first time to charged particles. The numerical framework is validated against experimental measurements and subsequently coupled with an interpretable LightGBM-based machine learning model to investigate how particle dynamics, oblique impacts, and rebound mechanisms influence collection performance.

Explainable Artificial Intelligence (XAI) techniques, including SHAP (SHapley Additive exPlanations) and Partial Dependence Plots (PDP), are employed for the first time to analyze charged particle rebound behavior. The analysis demonstrates that low flow velocities (< 0.25 m/s) substantially suppress particle rebound and improve collection efficiency. Smaller particles (≤ 3.5 μm) require consistently higher collection voltages to minimize rebound, whereas larger particles (> 5.5 μm) exhibit improved performance at relatively lower voltages.

The study further shows that particle charges in the range of approximately 1 × 10⁻¹⁵ to 1 × 10⁻¹⁴ C significantly enhance collection efficiency for particles between 2.0 and 10.0 μm. In contrast, higher flow velocities (> 0.25 m/s) and particle charges exceeding 1 × 10⁻¹⁴ C promote rebound. PDP analysis reveals the functional relationship between design parameters and both collection efficiency and rebound fraction, while SHAP-based feature attribution quantifies the contribution of each parameter and identifies favorable operating conditions that maximize collection efficiency while minimizing rebound.

BibTeX

@article{srivastava2025analyzing,
  title={Analyzing the impact of charge particle rebound on spherical collector efficiency via explainable machine learning},
  author={Srivastava, Abhishek and Kandpal, Sagar and Ray, Bahni and Kumar, Mayank and Dasgupta, Debabrata},
  journal={Powder Technology},
  pages={121803},
  year={2025},
  publisher={Elsevier}
}

Recommended citation: Srivastava, A., Kandpal, S., Ray, B., Kumar, M., & Dasgupta, D. (2025). Analyzing the Impact of Charged Particle Rebound on Spherical Collector Efficiency via Explainable Machine Learning. Powder Technology.
Download Paper | Download Bibtex