04/21/2026
Join us on 4/22/2026 to support Ximena Peregrino for her Master Defense!
Excited to share my thesis research on advancing the reliability of Laser Powder Bed Fusion (LPBF)! 🚀
The Challenge: While nominal Volumetric Energy Density (VED) is used to set machine parameters, it doesn’t capture the actual, local thermal conditions materials experience during a build. This creates a gap between expected settings and actual melt-pool behavior.
The Solution: I developed a "VED Proxy"—a monitoring-oriented ML framework that maps melt-pool features from in-situ coaxial imaging to track the effective process state.
Key Highlights of the Work:
🔹 Methodology: Introduced a "Reliable Zone" logic to train models on stable, least-confounded conditions. Melt-pool features were extracted and aggregated layer-wise.
🔹 The Model: Evaluated multiple supervised-learning models and selected Linear Regression for its optimal balance of predictive performance, interpretability, and layer-wise stability.
🔹 The Results: The VED Proxy successfully captured complex variations—including short-feeds, geometry-driven changes, and thermal-history effects (build-height/interlayer-time).
🔹 Cross-Material Transferability: Excitingly, a model trained on Inconel 718 retained useful interpretive value when applied directly to Haynes 282!
Ultimately, this establishes a physically grounded foundation for translating rich sensor data into a single, process-centered metric for better monitoring and future control in metal AM.
-
