Efficient High-Accuracy Casting Defect Detection with Attention Module

Al-Sameai H., AKAY R., SALEH R. A. A.

5th International Conference on Emerging Smart Technologies and Applications, eSmarTA 2025, Ibb, Yemen, 5 - 06 Ağustos 2025, (Tam Metin Bildiri) identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/esmarta66764.2025.11132262
  • Basıldığı Şehir: Ibb
  • Basıldığı Ülke: Yemen
  • Anahtar Kelimeler: Casting analysis, Convolutional Neural Networks (CNN), Deep learning, Hybrid model, Multi-Scale Attention, Visual inception
  • Kocaeli Üniversitesi Adresli: Evet

Özet

Defect detection in casting processes plays a critical role in manufacturing, directly impacting product quality and operational efficiency. During casting, molten metal is injected into molds to form components; however, imperfections can compromise structural integrity and production outcomes. This study proposes an advanced defect detection framework by integrating the Convolutional Block Attention Module (CBAM)—which combines the Channel Attention Module (CAM) and Spatial Attention Module (SAM) in a unified block—into the MobileNetV3-Large architecture. To benchmark performance, four state-of-the-art models—EfficientNet-B7, ResNet152, RegNet-Y 32GF, and MobileNetV3-large—were trained and evaluated on a publicly available Kaggle dataset comprising 7,348 annotated images categorized into defective and non-defective castings. Experimental results demonstrate that the proposed model outperforms competing architectures, achieving exceptional accuracy (99.33%), recall (99%), and precision (99.8%). The proposed model’s superior performance underscores its potential as a robust solution for industrial quality control, offering high precision in defect identification to enhance manufacturing reliability.