Towards Reliable Monitoring of Laser Powder Bed Fusion Additive Manufacturing: Robust Image Processing Method for Melt Pool Segmentation

Laser Powder Bed Fusion (LPBF) additive manufacturing is used to fabricate complex, lightweight, and customized parts with high precision, making it ideal for mission-critical applications such as aerospace and healthcare. Monitoring the characteristics of the melt pool, created during the laser-melting of metallic powders, is crucial for ensuring the quality of printed parts. In doing so, melt pool images must be accurately segmented first to reliably extract and monitor the melt pool’s morphological features, which are essential for downstream analysis to detect process anomalies and identify potential defects such as lack of fusion and keyhole porosities. However, due to the dynamic nature of LPBF, melt pool images are noisy and stochastic in nature challenging conventional image segmentation methods, such as Otsu, thresholding, k-means clustering, and active contour, which lack robustness under these conditions. In response, we developed a robust multi-step image processing framework that strategically integrates edge filtering, morphological operations, and adaptive smoothing to effectively segment the melt pool region. This methodology is specifically designed to address the unique challenges posed by noise and variability in LPBF melt pool images, offering a more reliable solution compared to conventional segmentation techniques. We evaluated our method using three datasets, including both in-house and public sources, and compared its performance to conventional methods. Our method consistently achieved high segmentation accuracy (~90%), when compared to ground-truth segmentation done by expert annotators, outperforming traditional methods. This robust segmentation method will enable reliable monitoring of LPBF processes, facilitating improved anomaly detection and process optimization.