Hybrid CFD–machine learning framework for predicting material removal rates in bonnet polishing

This study presents a novel hybrid Computational Fluid Dynamics (CFD) and Machine Learning (ML) framework to predict Material Removal Rates (MRRs) in bonnet polishing, a high-precision surface finishing technique widely used in optics and advanced manufacturing. The coupled dynamics of fluid flow and abrasive particles that govern polishing efficiency remain poorly understood, limiting accurate process control and optimization. To address this, an Artificial Neural Network (ANN) was trained on a CFD-based dataset, systematically exploring five key parameters: angular velocity, offset, jet velocity, precession angle, and the number of injected particles. CFD simulations revealed distinct flow and particle behaviors across the MRR spectrum. Low-MRR conditions (8.14 mm3/min) exhibited small recirculation zones and sparse particle clustering, favoring ultra-fine polishing. High-MRR conditions (36.12 mm3/min) showed energetic recirculation zones, intensified particle clustering, and frequent surface collisions, enabling efficient rough polishing. Although stagnation zones formed at the bonnet center, surrounding high-energy regions promoted effective material removal. The integrated CFD-ANN model achieved high predictive accuracy and revealed key mechanisms, including recirculation-driven particle clustering and impact dynamics. By explicitly linking particle behavior, fluid flow, and MRR, this work addresses a critical knowledge gap and provides a systematic, predictive framework for process optimization. The methodology offers actionable insights for high-precision manufacturing in optics, semiconductor wafers, and aerospace components.