Wei Li, Fahri Heltha, Aulia Rahman, Mahmud Iwan Solihin
This paper presents a deep learning-based approach for detecting surface defects on automotive engine injector valve seats. Current inspection processes typically rely on manual visual assessment or traditional machine vision; however, both methods often fail to meet the necessary requirements for speed and accuracy. While conventional machine vision offers high throughput, it remains susceptible to environmental factors—such as variations in lighting, camera positioning, and background interference—which limits its robustness and feature extraction capabilities. To address these limitations, this paper proposes an optimized deep learning algorithm designed to improve defect detection performance in manufacturing environments. We implement and compare improved Faster R-CNN and YOLO algorithms for surface defect detection. Although the enhanced Faster R-CNN achieves high coverage, it still suffers from false negatives. Conversely, the YOLO-tiny and YOLO-CBAM-tiny models provide better detection accuracy, especially for scratch and black area defects. Specifically, the inclusion of the Convolutional Block Attention Module (CBAM) allows the YOLO-CBAM-tiny network to significantly outperform the standard YOLO-tiny model in identifying black defects. By addressing these aspects, this study aligns with Sustainable Development Goal, which prioritizes the advancement of resilient and sustainable manufacturing infrastructure. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2026.
Faculty of Engineering, Technology and Built Environment, UCSI University, Kuala Lumpur, 56000, Malaysia; School of Computer Science and Technology (School of Software), Guangxi University of Science and Technology, Guangxi, 545006, China; Department of Electrical and Computer Engineering, Universitas Syiah Kuala, Banda Aceh, 23111, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock