From Melt-Pool Signals to Defect-Aware Control in Metal Additive Manufacturing: An In-Situ Monitoring Review

Metal powder bed fusion can produce complex high-value components, but the relationship between online process signatures and final part integrity remains difficult to establish. This review examines thermal imaging, optical melt-pool sensing, layer-wise vision, acoustic signals, and data-driven anomaly detection in laser powder bed fusion. It separates process anomaly detection from defect classification, dimensional acceptance, and closed-loop control. The synthesis emphasizes confounding from scan strategy, alloy, geometry, calibration, and machine type, together with the need for destructive or volumetric post-build ground truth. A comparison matrix identifies the evidence required before a sensor output can serve as a quality-assurance claim or control signal. Proposed validation procedures cover cross-build transfer, false alarms, defect localization, and intervention stability. This paper offers a technical narrative and test agenda without claiming new printed parts or experimental results.