PCBA Standard-to-Real Challenge: Cross-domain VQA for Real-world Manufacturing Inspection

Tai-I Chen, Ruyi Xu, Tsai-Yen Chen, Luyang Lin, Yen-Tzu Chiu, Rong-Sheng Lin, Frank Oppenheimer, Yung-Yu Chuang, Wen-Huang Cheng

Keywords: Industrial Anomaly Detection, Industrial inspection, Manufacturing Inspection, reasoning VQA, vision-language models, domain generalization, SMT-PCBA

Abstract:

Surface Mount Technology (SMT) is a critical stage in Printed Circuit Board Assembly (PCBA) manufacturing, where Automated Optical Inspection (AOI) systems are typically employed to make correct decisions that depend on both component-level semantics and process rules (e.g., package constraints, polarity, and handling), and on low-level texture cues.
While industrial anomaly detection benchmarks have driven progress in anomaly localization, and recent visual-text resources and MLLM benchmarks begin to probe multimodal reasoning in inspection settings, there remains a gap in standardized evaluation for \emph{actionable} defect under the significant cross-domain shift from manufacturing standards to real-world PCBA environments.
We propose the PCBA Standard-to-Real Challenge, a new grand challenge that evaluates multimodal models on cross-domain visual question answering grounded in real-world inspection imagery.
The challenge covers: (i) perception-centric recognition and detection (component identity, mount-side, defect presence, defect type), (ii) quantitative reasoning (component/pin counting), and (iii) defect cause and handling decisions.
Our goal is to accelerate research toward industrially deployable multimodal systems that can generalize from design principals and manufacturing standards, across product variations and inspection conditions, while producing interpretable, actionable answers aligned with manufacturing practice.
The challenge provides public development data, an official evaluation toolkit with robust answer normalization, and a hidden test protocol to encourage generalizable and reproducible comparisons. Our challenge website available at: https://sites.google.com/cmlab.csie.ntu.edu.tw/asus-ntu-pcba-vqa-challenge/home

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 21