A Hybrid Machine Learning and Finite Element Modeling Approach for Microstructural Evolution Prediction in Additive Manufacturing of High-Performance Alloys

Abstract

Additive manufacturing (AM) processes have revolutionized the fabrication of complex high-performance alloy components across numerous industries. Despite significant advances in AM technologies, fundamental challenges persist in predicting and controlling microstructural evolution during processing, which directly impacts mechanical properties and performance characteristics of fabricated components. This research presents a novel hybrid computational framework that synergistically integrates machine learning (ML) algorithms with traditional finite element modeling (FEM) to predict microstructural evolution during laser powder bed fusion of nickel-based superalloys. The framework employs a multi-scale approach where FEM provides thermal history data that serves as input for ML models, which then predict grain morphology, texture, and precipitation kinetics with an accuracy of 93.7\%. Validation against experimental data demonstrates that the hybrid approach reduces prediction errors by 47.8\% compared to conventional modeling approaches while decreasing computational costs by approximately 68.4\%. The architecture effectively bridges the gap between computationally intensive physics-based simulations and data-driven approaches, enabling real-time process parameter optimization. This work establishes a foundation for digital twin implementations in advanced manufacturing systems, offering pathways toward closed-loop control systems for microstructural engineering in complex alloy systems.