Deep Learning for Microstructure Segmentation and Defect Detection in Additive Manufacturing Systems.

Microstructure analysis plays a crucial role in additive manufacturing (AM) processes as it provides significant insights into material properties and print quality. Detecting melt pool boundaries and porosity defects within microstructure samples is a complex task due to inherent image processing and data analysis challenges. To address this, we propose a deep learning-based approach that employs state-of-the-art deep learning models with convolutional neural network architectures (e.g., U-Net and FPN). This approach enables automatic segmentation and detection of melt pools and porosity in AM microstructure images. Visualization results demonstrate significant potential in accurately identifying microstructures using limited data. A comparative evaluation of various deep learning network architectures indicates that U-Net paired with EfficientNet b7 is better suited for melt pool segmentation, and the FPN with the DenseNet 201 backbone attains the highest accuracy for porosity. This work and the generated results demonstrate great promise in enhancing AM quality control through deep learning-powered microstructure analysis.