| Abstract Scope |
This study presents a supervised machine learning approach employing a Multi-Layer Perceptron (MLP) to identify porosity during laser-directed energy deposition (DED-L). In this process, we employed in-situ thermographic imaging, for melt pool monitoring. The collected image datasets were acquired for M300 maraging steel and Inconel (IN718) under varying process parameters, such as: laser power, scanning speed, and scan patterns. Subsequent image processing facilitated the extraction of spatial-temporal features, correlated with ex-situ computed tomography (CT) scans to validate porosity. For M300 steel, the MLP model was trained on 10,000 images categorized into non-defective melt pools, defective melt pools, and voids, achieving a classification accuracy of approximately 92%. For IN718 was used a binary classification model trained on 5,000 images attained an accuracy of around 97%. These results highlight the efficacy of MLPs in real-time defect detection within additive manufacturing processes that is ideal for in-situ/in-operando process. |