My doctoral research at the Indian Institute of Technology Delhi focused on developing physics-based and data-driven computational frameworks for understanding and optimizing charged aerosol particle transport and collection.
The research combined Computational Fluid Dynamics (CFD), electrostatics, Lagrangian particle tracking, multiphysics modeling, machine learning, explainable artificial intelligence, surrogate modeling, and multi-objective optimization.
A central objective of the work was to bridge high-fidelity numerical simulations with modern machine learning methodologies to accelerate engineering analysis while retaining physical interpretability.
Developed three-dimensional multiphysics numerical models in COMSOL Multiphysics by coupling:
The numerical framework was used to investigate charged particle transport and deposition around spherical collectors under varying operating conditions.
The influence of parameters including particle diameter, flow velocity, applied voltage, particle charge, and collector geometry was systematically investigated.
Investigated the physical mechanisms governing charged particle transport and collection, including:
The analysis provided mechanistic insight into how competing hydrodynamic and electrostatic effects influence particle capture.
Developed machine learning surrogate models to approximate computationally expensive multiphysics simulations and enable rapid exploration of large engineering design spaces.
The work included:
Applied Explainable Artificial Intelligence (XAI) techniques to interpret machine learning predictions and connect data-driven results with underlying physical mechanisms.
Methods included:
These approaches were used to identify dominant operating parameters and favorable conditions for improving particle collection while suppressing rebound.
Developed a data-driven multi-objective optimization framework integrating:
K-means clustering was used to identify favorable regions of the design space and provide physically meaningful bounds for subsequent optimization.
NSGA-II was then employed to determine Pareto-optimal trade-offs between maximizing particle collision efficiency and minimizing particle rebound.
The doctoral research resulted in peer-reviewed research publications and computational frameworks spanning multiphysics simulation, particle dynamics, interpretable machine learning, surrogate modeling, and engineering optimization.
The work established a computational workflow progressing from:
Physics-Based Modeling → High-Fidelity Simulation → Machine Learning → Explainable AI → Surrogate-Assisted Optimization
This research provided the foundation for my subsequent work in Scientific Machine Learning, Bayesian Inference, Bayesian Inverse Problems, and Uncertainty Quantification.