Work

AI-vision based real-time in-situ monitoring of the metallic deposition in μ-plasma transferred arc directed energy deposition process

Year: 2026

Type: article

Abstract: In metal additive manufacturing, real-time in-situ monitoring is essential for ensuring deposition quality, minimising defects, and achieving consistent part geometry. This research proposes a unified AI-driven vision-based framework for real-time monitoring of powder-fed metallic deposition during the μ-Plasma Transferred Arc Directed Energy Deposition (μ-PTADED) process. The framework integrates High Dynamic Range imaging, Kandinsky-based arc suppression, and segmentation strategies such as supervised YOLOv11 instance segmentation and zero-shot Grounding DINO with Segment Kandinsky Anything Model (SAM), to extract deposition regions under complex visual conditions. Using the extracted regions, real-time deposition height monitoring is accomplished through calibrated pixel-to-millimetre conversion along fixed reference lines in each frame. Height estimations were validated against manual measurements. Furthermore, deposition quality classification is performed using four deep learning models: YOLOv11n, ResNet-50, VGG-16, and VGG-19 trained. The outcome revealed that the Grounding DINO with SAM model achieved superior segmentation with Intersection over Union (IoU) scores above 0.95 across a range of experimental conditions. The measured deposition heights showed exceptional accuracy, with average deviations consistently below 0.06 mm compared with manual measurements which validated the real-time height measurement approach with good agreement. Among the deep learning models used for deposition quality classification, YOLOv11n achieved the best performance with a classification accuracy of 98.2%, F1-score of 0.983, and an inference speed of 250 FPS. The results validate the framework's capability for real-time in-situ monitoring required for intelligent feedback control in arc-based additive manufacturing environments.

Source: Journal of Manufacturing Processes

DOI: 10.1016/j.jmapro.2026.05.060