Structured Enhancement of VGG16-Based PCB Fault Diagnosis via Template-Matching Feature Maps and Component-Level Prior Injection
To meet the demand for rapid inspection of unmanned aerial vehicle (UAV) onboard circuit boards in large-scale low-altitude economy operations, and to address the insufficient generalization ability of end-to-end deep learning diagnostic methods under varying imaging conditions, a structured-enhancement method for printed circuit board (PCB) fault diagnosis based on template-matching feature maps is proposed. A three-stage structured-enhancement PCB fault diagnosis framework is constructed, consisting of structured thermal feature extraction, lightweight template matching, and a VGG16-based fault classification network adapted to component-level localization results. Conditioned on the spatial and category priors from the matching results, the VGG16 fault classification network decouples component localization from fault classification through a spatial prior injection module and a category prior injection module. Experimental results show that, for a DC/DC buck converter circuit, the thermal-feature-based template-matching method achieves an average relative localization error of 2.69% across components. For a two-phase drive (TPD) circuit, when the imaging angle and distance vary, the proposed framework improves fault-recognition accuracy from 92.7% to 95.8% compared with the conventional VGG16 network. The proposed method provides a technical solution that balances accuracy and generalization for rapid fault detection of circuit boards in low-altitude equipment.
Authors
- Chengxin Liu (ORCID: https://orcid.org/0000-0003-3723-4209)
- Yongming Wen (ORCID: https://orcid.org/0000-0003-2595-0734)
- Zhangwei Wang (ORCID: https://orcid.org/0000-0003-3700-4534)
Institutions
- Xi'an University of Technology (CN)
Publication Details
- Journal
- Big Data and Cognitive Computing
- Published
- 2026-11-17
- DOI
- https://doi.org/10.3390/bdcc10110392
- Primary Topic
- Industrial Vision Systems and Defect Detection
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China