DIAGNOSTICS OF PRINTED CIRCUIT BOARDS BASED ON NEURAL NETWORK MODELS

Abstract

Current article deals with an Interpretable YOLOv11-Grad-CAM Framework for Enhanced Automated Optical
Inspection of Printed Circuit Boards. The critical demand for reliability in electronics manufacturing underscores the need
for advanced quality control. Automated Optical Inspection of printed circuit boards remains a cornerstone of this process.
However, conventional methods, including template matching and manual inspection, are often inadequate in terms of
robustness and scalability for modern production volumes. Although deep learning models have demonstrated superior
performance in defect detection, their inherent lack of interpretability poses a significant barrier to adoption in high stakes
industrial environments. This study bridges this gap by introducing an interpretable deep learning framework that
integrates the state-of-the-art YOLOv11n architecture for real time defect detection with Gradient-weighted Class
Activation Mapping for model explainability. This integration provides transparent, visual justification for the model
predictions by highlighting discriminative features. The proposed framework was evaluated on the public HRIPCB
dataset, which includes 1386 images spanning six common defect classes. It achieved a mean average precision
(mAP@0.5) of 87.4 percent, significantly outperforming both a traditional SVM classifier and a YOLOv8n baseline. The
principal contribution of this work is the novel and systematic fusion of high-speed object detection with explainable AI
principles, tailored for PCB inspection. By simultaneously achieving high accuracy, real time inference, and critical
interpretability, this framework presents a viable and trustworthy solution for industrial AOI systems. The goal of this
study is to develop an interpretable defect detection framework for PCB automated optical inspection that delivers both
high accuracy and transparent decision making. To achieve this goal, the following tasks were performed: selection and
configuration of the YOLOv11n model for PCB defect detection; training and evaluation on the HRIPCB dataset;
integration of Grad-CAM into the detection pipeline for visual explanation; comparison against baseline methods
including SVM and YOLOv8n; and quantitative performance measurement using mAP, precision, and recall alongside
qualitative documentation through visual heatmaps.
Keywords: quality control; defect detection; neural networks; artificial intelligence; printed circuit boards; GradCAM; YOLO11n; deep learning.

References
1. Evgeniou T., Pontil M. Support vector machines: theory and applications. Machine Learning and Its
Applications: Advanced Lectures. 2001. Vol. 2049. P. 249-257. DOI: https://doi.org/10.1007/3-540-44673-7_12.
2. Gamini P., Yasaswini A., Lakshmi N., Kumar C., Manikanta A., Sandeep B., Students B. Defect detection in
PCB using image processing. International Journal of Advanced Science and Technology. 2020. Vol. 29, No. 4s. P. 1205–
1210.
3. Ge Y., Li Z., Meng L. YOLO-MSD: a robust industrial surface defect detection model via multi-scale feature
fusion. Applied Intelligence. 2025. Vol. 55, No. 12. Article 840. DOI: https://doi.org/10.1007/s10489-025-06739-0.
4. Huang W., Wei P., Zhang M., Liu H. HRIPCB: a challenging dataset for PCB defects detection and
classification. The Journal of Engineering. 2020. Vol. 2020, No. 13. P. 303–309. DOI:
https://doi.org/10.1049/joe.2019.1183.
5. Liu J., Kang B., Liu C., Peng X., Bai Y. YOLO-BFRV: an efficient model for detecting printed circuit board
defects. Sensors. 2024. Vol. 24, No. 18. Article 6055. DOI: https://doi.org/10.3390/s24186055.
6. Moraes A. M., Pugliese L. F., Santos R. F. d., Vitor G. B., Braga R. A. d. S., Silva F. R. d. Effectiveness of
YOLO architectures in tree detection: impact of hyperparameter tuning and SGD, Adam, and AdamW optimizers.
Standards. 2025. Vol. 5, No. 1. Article 9. DOI: https://doi.org/10.3390/standards5010009.
7. Muhammad M. B., Yeasin M. Eigen-CAM: class activation map using principal components. 2020
International Joint Conference on Neural Networks (IJCNN). IEEE, 2020. P. 1–7. DOI:
https://doi.org/10.1109/IJCNN48605.2020. 9206626.
8. Park J. H., Kim Y. S., Seo H., Cho Y. J. Analysis of training deep learning models for PCB defect detection.
Sensors. 2023. Vol. 23, No. 5. Article 2766. DOI: https://doi.org/10.3390/s23052766.
9. Purwono P., Wulandari A. N. E., Ma’arif A., Salah W. A. Understanding generative adversarial networks
(GANs): a review. Control Systems and Optimization Letters. 2025. Vol. 3, No. 1. P. 36–45. DOI: https://doi.org/
10.59247/csol.v3i1.170.
10. Ren S., He K., Girshick R., Sun J. Faster R-CNN: towards real-time object detection with region proposal
networks. arXiv. 2016. URL: https://arxiv.org/abs/1506.01497.
11. S D. A., Prathapchandra D., Ramachandra M. Fault detection in printed circuit board (PCB) using image
subtraction method. International Journal of Innovative Technology and Exploring Engineering. 2024. Vol. 13, No. 7. P.
1–6. DOI: https://doi.org/10.35940/ijitee.G9906.13070624.
12. Salman H., Kalakech A., Steiti A. Random Forest algorithm overview. Babylonian Journal of Machine
Learning. 2024. P. 69–79. DOI: https://doi.org/10.58496/BJML/2024/007.
13. Selvaraju R. R., Cogswell M., Das A., Vedantam R., Parikh D., Batra D. Grad-CAM: visual explanations from
deep networks via gradient-based localization. International Journal of Computer Vision. 2020. Vol. 128, No. 2. P. 336–
359. DOI: https://doi.org/10.1007/s11263-019-01228-7.
14. Tang S., He F., Huang X., Yang J. Online PCB defect detector on a new PCB defect dataset. arXiv. 2019. URL:
https://arxiv.org/abs/1902.06197.
15. Wang H., Wang Z., Du M., Yang F., Zhang Z., Ding S., Mardziel P., Hu X. Score-CAM: score-weighted visual
explanations for convolutional neural networks. arXiv. 2019. URL: https://arxiv.org/abs/1910.01279.
16. Xiang Z., Tan H., Ye W. The excellent properties of a dense grid-based HOG feature on face recognition
compared to Gabor and LBP. IEEE Access. 2018. Vol. 6. P. 29306–29319. DOI: https://doi.org/10.1109/ ACCESS.
2018.2813395.