Layer-Wise Anomaly Detection in Directed Energy Deposition using High-Fidelity Fringe Projection Profilometry

TL;DR

The paper addresses defect detection in directed energy deposition (DED) by introducing a build-height-synchronized fringe projection profilometry (FPP) system for layer-wise, in-situ 3D surface reconstruction with an accuracy of $\pm 46\,\mu\mathrm{m}$. It combines full-field 3D measurements with a geometry-based, unsupervised anomaly framework using local point density and normal-change rate (NCR) to automatically localize deposition defects without labeling. Validation shows high calibration fidelity (gauge blocks and 3D microscopy) and robust anomaly detection, including 100% success for deliberately induced defects, outperforming 2D image-based approaches. The approach links geometric deviations directly to defect formation, enabling potential closed-loop process control and certifiable quality in metal additive manufacturing.

Abstract

Directed energy deposition (DED), a metal additive manufacturing process, is highly susceptible to process-induced defects such as geometric deviations, lack of fusion, and poor surface finish. This work presents a build-height-synchronized fringe projection system for in-situ, layer-wise surface reconstruction of laser-DED components, achieving a reconstruction accuracy of ${\pm}$46 $μ$m. From the reconstructed 3D morphology, two complementary geometry-based point cloud metrics are introduced: local point density, which highlights poor surface finish, and normal-change rate, which identifies lack-of-fusion features. These methods enable automated, annotation-free identification of common deposition anomalies directly from reconstructed surfaces, without the need for manual labeling. By directly linking geometric deviation to defect formation, the approach enables precise anomaly localization and advances the feasibility of closed-loop process control. This work establishes fringe projection as a practical tool for micrometer-scale monitoring in DED, bridging the gap between process signatures and part geometry for certifiable additive manufacturing.

Figures (17)

• Figure 1: Workflow of the in-situ DED monitoring framework.
• Figure 2: Phase-to-height calibration using a $1^\circ$ angle block. A linear fit yields the calibration constant $C = 0.823$.
• Figure 3: Comparison between FPP and gauge block
• Figure 4: Comparison between FPP and focus-variation system measurements: (a) height map reconstructed from FPP; (b) corresponding height map obtained via 3D microscope; (c) pixel-wise deviation magnitude between the two reconstructions; and (d) statistical histogram of deviations.
• Figure 5: Rendering of the experimental setup.