Defect localization using region of interest and histogram-based enhancement approaches in fused filament fabrication additive manufacturing

Fused Filament Fabrication (FFF), a prominent Additive Manufacturing (AM) technology commonly known as 3D printing, has revolutionized the production of complex structures across various industries. However, ensuring quality and detecting defects in 3D-printed objects remain significant challenges. This study focuses on defect detection primarily on cylindrical geometries and extends to a multi-class defect dataset through novel pre-processing techniques such as Region of Interest (ROI) selection, Histogram Equalization (HE), and Details Enhancer (DE) with Convolutional Neural Networks (CNNs), specifically the modified Visual Geometry Group (VGG)16 model. The approaches, ROIN (ROI selection with normalization), ROIHEN (ROI selection with HE and normalization), and ROIHEDEN (ROI selection with HE, DE, and normalization), produced promising results, with the best model achieving an accuracy of 1.00 and an F1-score of 1.00 on the test set. In addition, the study extends to multi-class defect classification on the imbalanced FDM dataset and achieves perfect accuracy (1.00) across five defect types (cracking, layer shift, off-platform, stringing, and warping) through class-weighted training. Following established XAI practice in additive manufacturing, the study applies LIME and Gradient-weighted Class Activation Mapping (Grad-CAM) to provide interpretable visualizations of model decision-making, complementing the novel custom localization layer introduced in the proposed architecture. Furthermore, the modified VGG16 model demonstrates superior computational efficiency with 30,713M FLOPs and 15M parameters, the lowest among the compared models. These findings underscore the role of tailored pre-processing and CNNs in defect detection in AM and establish a pathway to improve manufacturing precision, accuracy, and efficiency. This research contributes to the advancement of 3D printing technology and shows the potential of machine learning (ML) integration with AM for superior quality control. • ROI, Histogram Equalization, and Details Enhancement boost defect detection accuracy. • Modified VGG16 achieves 100% accuracy and F1-score on FFF defect datasets. • Custom separable convolution layer enables efficient and accurate defect localization. • Grad-CAM and LIME provide interpretable, physics-aligned defect heatmaps. • Class-weighted training resolves severe imbalance in multi-class FDM defect data.