Predictive Modeling of Powder Bed Fusion Microstructures Using CFD Simulations and Machine Learning Algorithms

Abstract

The integration of Computational Fluid Dynamics (CFD), Machine Learning (ML), and additive manufacturing techniques has revolutionized the Laser Powder Bed Fusion (LPBF) process by enabling predictive modeling, process optimization, and quality control. This study explores the synergy between CFD simulations and ML algorithms to predict microstructural outcomes, such as grain size and porosity, based on process parameters like laser power, scan speed, and particle size. The proposed hybrid framework leverages CFD to generate synthetic datasets, which are then utilized to train ML models, including Random Forest Regression. These models accurately capture the non-linear relationships between process parameters and microstructural properties. The results underscore the potential of this approach for enhancing process efficiency, reducing computational costs, and minimizing experimental trials. Key findings reveal that energy density significantly influences microstructural outcomes, with higher densities reducing porosity and refining grain structures. Furthermore, the study highlights real-time process control and optimization opportunities through ML-driven predictions. By combining the physical rigor of CFD with the adaptability of ML, this framework provides a pathway toward intelligent manufacturing systems capable of delivering high-performance components with tailored properties. Future research should incorporate experimental data and advanced ML techniques to enhance reliability and generalization.