Reducing defect rates in smart manufacturing processes: deep learning-based quality control for automotive tow hooks
| dc.contributor.author | Canlı, Hikmet | |
| dc.contributor.author | Varıcı, Sena | |
| dc.date.accessioned | 2026-04-16T13:44:53Z | |
| dc.date.available | 2026-04-16T13:44:53Z | |
| dc.date.issued | 2026 | |
| dc.department | Fakülteler, Mühendislik Fakültesi, Bilgisayar Mühendisliği Bölümü | |
| dc.description.abstract | In the automotive industry, quality control in the production of critical components is of great importance. One such critical component is the tow hook, a specialized connecting element mounted on the rear of motor vehicles, enabling them to safely tow another vehicle or load. Defective tow hooks can lead to both safety hazards and economic losses. This study aims to compare the performance of deep learning-based models for the classification of defective parts in industrial production processes. The constructed dataset includes defects such as burrs, dents on the thread, non-filling and dents on the surface. Additionally, a class for intact parts was created, resulting in a total of five categories. These categories were used to train and comprehensively evaluate models with different architectures, including residual neural network (ResNet), Mobile Convolutional Neural Network (MobileNet), Visual Geometry Group (VGG), and conventional convolutional neural networks (CNNs), using validation metrics. Experimental results indicate that the ResNet34 model achieved the highest performance, with a 100% accuracy rate and precision, recall, and F1-score values of 1.00 across all classes. The MobileNetV2 model achieved 96% accuracy, with class-wise validation metrics exceeding 0.90, demonstrating strong performance. These findings suggest that the proposed models can effectively distinguish between defective and intact parts with high accuracy, offering a reliable solution for industrial quality control applications. Future work will focus on enhancing the models' generalization capabilities through the evaluation of larger datasets and diverse production conditions. | |
| dc.identifier.doi | 10.7717/peerj-cs.3533 | |
| dc.identifier.issn | 2376-5992 | |
| dc.identifier.scopus | 2-s2.0-105033745697 | |
| dc.identifier.scopusquality | Q1 | |
| dc.identifier.uri | https://doi.org/10.7717/peerj-cs.3533 | |
| dc.identifier.uri | https://hdl.handle.net/11501/2689 | |
| dc.identifier.volume | 12 | |
| dc.identifier.wos | WOS:001708100400001 | |
| dc.identifier.wosquality | Q2 | |
| dc.indekslendigikaynak | Scopus | |
| dc.indekslendigikaynak | Web of Science | |
| dc.institutionauthor | Canlı, Hikmet | |
| dc.institutionauthor | Varıcı, Sena | |
| dc.institutionauthorid | 0000-0003-3394-7113 | |
| dc.institutionauthorid | 0009-0006-0749-4623 | |
| dc.language.iso | en | |
| dc.publisher | PeerJ Inc. | |
| dc.relation.ispartof | PeerJ Computer Science | |
| dc.relation.publicationcategory | Makale - Uluslararası Hakemli Dergi - Kurum Öğretim Elemanı | |
| dc.rights | info:eu-repo/semantics/openAccess | |
| dc.subject | Deep Learning | |
| dc.subject | Defect Classification | |
| dc.subject | Quality Control | |
| dc.title | Reducing defect rates in smart manufacturing processes: deep learning-based quality control for automotive tow hooks | |
| dc.type | Article |
