Optimizing Electrostatic Precipitator Collection Efficiency: Taguchi Design, Grey Relational Analysis, and Machine Learning Integration

Abstract This study presents a modeling and optimization approach for electrostatic precipitator (ESP) collection efficiency using physics simulations, Taguchi design, gray relational analysis (GRA), and machine learning. A corona discharge and turbulent flow model simulated fluid velocity, electric field, charge density, and particle trajectories in a wire-to-plane ESP. Taguchi L27 simulation run varied electrode radius, gas velocity, and applied voltage, showing voltage as the key efficiency factor, with lower gas velocities and larger electrode radii improving performance. GRA combined multi-size particle efficiencies into a single metric and determined optimal conditions. Machine learning models, Gradient Boosting and Artificial Neural Networks (ANNs), predicted collection efficiency with different strengths such as Gradient Boosting for fine particles and ANN for micron-sized particles. Both models showed limitations for ultra-fine and coarse particles, suggesting the need for expanded datasets. This framework provides a methodology for optimizing ESP performance.