Data-Driven Optimization of Process Parameters in Laser Powder Bed Fusion Through Deep Reinforcement Learning Techniques

Abstract

Laser Powder Bed Fusion (LPBF) has emerged as a promising additive manufacturing technique for producing complex metallic components with high precision and customizability. Despite its advantages, LPBF processes are characterized by numerous interdependent parameters that significantly impact the final part quality, mechanical properties, and production efficiency. This research presents a novel framework for real-time optimization of LPBF process parameters using deep reinforcement learning (DRL) algorithms coupled with high-fidelity multiphysics simulations. Our approach integrates thermal, fluid dynamic, and metallurgical models with advanced DRL architectures to create a robust optimization methodology that adaptively adjusts process parameters during fabrication. The proposed system demonstrates a 27\% reduction in porosity defects, 18\% improvement in surface roughness, and 34\% enhancement in dimensional accuracy compared to conventional parameter optimization approaches. Experimental validation conducted across three distinct metal alloys (Ti-6Al-4V, Inconel 718, and AlSi10Mg) confirms the generalizability of our methodology. The framework's ability to continuously refine parameters without human intervention represents a significant advancement toward fully autonomous LPBF systems capable of producing consistently high-quality components while minimizing material waste and energy consumption. This research establishes a foundation for next-generation intelligent additive manufacturing systems that can dynamically respond to processing anomalies and material variations.