Answers are grounded in this paper and cite their sources.
Demonstrates defect optimization in additively manufactured parts, implying enhanced production quality.
In-situ alloying during laser powder bed fusion (LPBF) offers great compositional flexibility but is prone to process-induced defects. To address this problem, we developed a machine learning framework to predict and minimize major defects such as porosity (inclusive of lack of fusion, gas pores, and keyhole-induced porosity) and unmelted Nb particles (partially and completely unmelted particles) in LPBF-fabricated in-situ alloyed Ti–45Nb alloy. For this purpose, two independent least-squares boosting (LSBoost) ensemble regressors were trained using five process parameters (part shape, laser power, scan speed, hatch spacing, and scan rotation), along with their polynomial and interaction terms, to capture nonlinear relationships. Under a restricted 4-fold cross-validation, these models achieved pooled out-of-fold R2 values of 0.672 for porosity and 0.702 for unmelted Nb, despite being trained on a small dataset. The grouped permutation importance analysis revealed that porosity is primarily governed by hatch spacing and laser power, whereas unmelted Nb particles are primarily governed by laser power and scan speed. The models were implemented in two graphical interfaces: a forward predictor for real-time defect estimation and an inverse optimizer for identifying low-defect parameter sets. Together, they establish a unified, data-driven approach for defect-aware process detection, prediction, and optimization in in-situ alloyed systems, offering a pathway towards reproducible, low-defect additive manufacturing.
No takes yet. Share an insight, caveat, or question.
Nezhad et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: