A vision-based deep convolutional neural network framework for surface defect detection in laser powder bed fusion (LPBF)

Laser Powder Bed Fusion (LPBF) is highly vulnerable to a range of process-induced defects that can adversely part quality, structural integrity, dimensional accuracy and long-term reliability, thereby limiting broader industrial adoption. This work presents a feasibility investigation into automated detection of defect-related surface patterns in LPBF using a convolutional neural network (CNN)-based image classification framework. Classical image-processing techniques were initially found to be insufficient for reliable defect detection due to their limited robustness to variations in surface morphologies and image characteristics. To address these limitations, a CNN-based image classification framework was developed and trained using surrogate-labeled LPBF surface images to evaluate the viability of deep learning for automated surface condition classification and recognition. The proposed and developed model achieved a validation accuracy of approximately 84% while exhibiting stable convergence behavior during training, indicating that meaningful classification of LPBF surface patterns is achievable under limited labeling conditions, demonstrating the potential of deep learning for automated LPBF surface condition assessment. This work establishes a foundation for future work focused on incorporating physically verified defect labels, improved imaging fidelity, and integration with in-situ monitoring framework to enable real-time defect detection in LPBF process conforming for quality assurance.