40th International Conference on Production Engineering of Serbia
ICPES 2025
Nis, Serbia, 18-19th september 2025
DEEP LEARNING TECHNIQUES FOR DEFECT DETECTION IN AUTOMATED QUALITY CONTROL SYSTEMS
A. Gandhi Manikandan, V. Sandhiya, P. Harine, R. Ajay Dhisone, M. Sathyaprakash
DOI: 10.46793/ICPES25.189GM
Automated quality control (AQC) systems have transformed modern manufacturing by enabling high-speed, consistent, and cost-efficient inspection of products. However, traditional image processing techniques often fail to detect subtle, non-conforming defects due to limitations in adaptability and generalization. This paper explores the integration of deep learning (DL) techniques—particularly convolutional neural networks (CNNs), autoencoders, and vision transformers—for enhancing defect detection accuracy in AQC environments. A comprehensive methodology is presented where training datasets are augmented using synthetic defect generation, followed by supervised and unsupervised learning approaches for feature extraction and classification. The research evaluates various model architectures using metrics such as accuracy, precision, recall, and inference time on datasets collected from electronic component inspection and surface defect analysis in metal casting. The results demonstrate that DL models significantly outperform conventional rule-based systems, particularly in detecting micro-defects and anomalies in complex textures. Moreover, the incorporation of transfer learning and model pruning techniques further reduces computational overhead, making the deployment feasible in real-time production lines. This study concludes with an outline of implementation challenges, including data imbalance, hardware constraints, and model interpretability, and proposes potential directions for future research in intelligent adaptive inspection systems. The findings aim to contribute toward the development of robust, scalable, and intelligent quality control frameworks aligned with Industry 4.0 principles
Deep learning, defect detection, automated quality control, convolutional neural networks, computer vision, real-time inspection, Industry 4.0
[1] H. Czichos, D. Klaffke, E. Santner, M. Woydt: Advances in tribology: the materials point of view, Wear, Vol. 190, No. 2, pp. 155-161, 1995.
[2] D. P. Kingma and M. Welling: Auto-Encoding Variational Bayes, in: International Conference on Learning Representations, 2014.
[3] A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, N. Houlsby: An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, in: International Conference on Learning Representations, 2021.