MAOL (Morphology-Aware Ordinal Learning) is a new framework for fine-grained industrial defect severity grading. It treats severity grading as an instance-level ordinal learning task, uses explicit morphological features to improve representation, introduces class-conditional adaptive ordinal thresholds to capture defect-specific grading boundaries, and applies prediction-aware training with localization perturbation to handle noisy predicted instances. Experiments show MAOL outperforms rule-based methods, nominal classification models, and existing ordinal baselines, especially in predicted-instance settings, and it ranked third in the IDA 2026 Challenge on Fine-Grained Severity Grading for High-Precision Manufacturing.
By Zhaoyang Wang, Haiyong Chen, Binyi Su, Kun Liu, Kun Wang, Xianen Zhou, Atik Shahariar
arXiv:2606. 08908v1 Announce Type: cross Abstract: Semiconductor lithography inspection requires reliable detection of small pattern defects such as bridge, burr, pinch, and contamination.
By Pangyun Jeong, Jiyeong Kong, Yuehua Hu, Dohee Jeong, Kyung-Tae Kang
The paper introduces the Aero-engine Blade Defect Detector (ABDD), an online adaptive detection framework that uses test-time adaptation to handle domain shifts in aero-engine blade inspection. ABDD employs a Dual-Alignment Strategy combining feature-statistics alignment with pseudo-box alignment to adapt both global visual style and local defect morphology, and incorporates an Uncertainty-aware Box Filtering mechanism to mitigate errors from noisy pseudo labels. A lightweight Sparse Dilated Mona module enables efficient parameter tuning while preventing source-domain forgetting, and the method is validated on CD-AeBD and HD-AeBD datasets, showing improved robustness and practical applicability on an industrial inspection platform.
By Zhaoyang Wang, Haiyong Chen, Dongying Li, Yining Wang, Huapeng Wu, Xinwei Lv, Atik Shahariar
The paper introduces an automated defect detection system for printed circuit board (PCB) pin insertion, combining U‑Net semantic segmentation with contour‑based feature extraction and logistic regression for board‑level pass/fail classification. Segmentation masks generate contour representations of individual pins, from which features such as average contour size are extracted to train the classifier. Evaluated on an industrial dataset and a public PCB pin‑inspection dataset, the method achieved ROC‑AUC scores of 0.990 and 1.000, outperforming PatchCore and instance‑segmentation baselines.
By Nils Rabeneck, Andr\'e Kiunke, Nicole Hoess, Wolfgang Mauerer
The paper introduces FuDU, a streaming active learning framework that enhances real‑time industrial defect detection by quantifying uncertainty at both image and box levels. It employs a Prototype-based Global Uncertainty Quantification module to assess image‑level uncertainty and a Dual‑entropy defect Uncertainty Evaluator for box‑level uncertainty. By fusing these uncertainties through fuzzy inference, FuDU enables expert‑knowledge‑driven adaptive sampling, improving reliability in tasks such as nuclear fuel rod defect detection.
By Zhaoyang Wang, Haiyong Chen, Binyi Su, Xinwei Lyu
arXiv:2608.21967v1 Announce Type: new
Abstract: Automated visual inspection in manufacturing aims to replace slow and inconsistent manual checks, but its economic value depends on whether its decisio...
By Panagiotis Sapoutzoglou, Jessy Ribaira, Martin Kanounnikoff, Bas Tijsma, Christian Gei{\ss}, Maria Pateraki