On Machine Learning Control in LPBF Additive Manufacturing

Laser Powder Bed Fusion (LPBF) remains constrained by the absence of closed-loop control systems capable of preventing porosity and cracking during fabrication. This study introduces a physics-informed, feed-forward control framework that integrates synchronized acoustic emission and pyrometer data for defect detection and predictive process adjustment. Machine learning models trained on induced and natural defect datasets achieved up to 95% classification accuracy and produced consistent thermal responses across multiple alloys and build geometries. The framework combines variational autoencoders with physics-based control logic to enable near real-time monitoring and adaptive parameter optimization. The results demonstrate a viable pathway toward scalable, data-driven control in LPBF, advancing the development of autonomous additive manufacturing systems.