Layer-wise porosity prediction in LPBF process using optical tomography and deep learning

• A deep learning framework is proposed to predict layer-wise porosity in LPBF. • In-situ Optical Tomography (OT) images are registered with CT-measured porosity. • A ResNet50 regression model achieves MAE of 0.0875% and R 2 of 0.7478. • The model detects high-porosity layers with an F1 score of 0.8426 and recall of 0.9055. • Contextual frame inputs improve stability by capturing cross-layer thermal effects. Laser Powder Bed Fusion (LPBF) is a key metal additive manufacturing technology widely used in aerospace and biomedical fields. However, process-induced porosity remains a major challenge, compromising part quality and performance. This study presents a data-driven framework for layer-wise porosity prediction using Optical Tomography (OT) and deep learning. Multiple LPBF samples were fabricated under varied process parameters. In-situ OT grayscale images were collected during printing, and post-build X-ray computed tomography (CT) was used to quantify porosity. A ResNet50-based regression model was trained to predict layer-wise porosity directly from OT images. The model achieved a mean absolute error (MAE) of 0.0875% and a coefficient of determination (R 2 ) of 0.7478. For binary classification (threshold = 0.10%), it reached an F1 score of 0.8426 and recall of 0.9055, enabling effective detection of high-risk layers. Additional experiments demonstrated that incorporating adjacent frame information improved the model’s ability to capture cross-layer thermal effects. The proposed approach offers a real-time, non-destructive solution for porosity monitoring and quality control in LPBF, with strong potential for industrial implementation.