Akademik Veri Yönetim Sistemi
IV. International Applied Statistics Conference , Sarajevo, Bosna-Hersek, 26 - 29 Eylül 2023, ss.168-177
An Application of Deep Learning Algorithms for Defect Type Identification in Steel Manufacturing
Zeki Umut Cindiloğlu1*, Bahar Özyörük1
1 Gazi University, Faculty of Engineering, Department of Industrial
Engineering, 06570, Ankara,
Türkiye
*Corresponding author e-mail: umutcindiloglu@gazi.edu.tr
In the manufacturing sector, product quality and overall reliability of
business hold significant importance for customers. Many businesses employ
various quality control methods throughout the production process to meet these
quality and reliability requirements. However, existing processes are
predominantly carried out manually, leading to issues such as human error, time
inefficiencies, and reduced accuracy. To overcome these challenges and prevent
the recurrence of errors, the establishment of continuously learning automated
control systems has become indispensable for contemporary businesses. In this
study, deep learning techniques were utilized with the aim of eliminating
common errors encountered in manual quality control processes and automating
the procedure. Within the scope of this study, an analysis was conducted to
identify various types of errors present in the final product, using deep
learning algorithms. The dataset used for this research was obtained from the
Kaggle platform. This publicly accessible dataset contains a total of 7095
steel plate surface images, each featuring various types of defects. The types
of defects are categorized into four groups, namely Pitted Surface, Crazing,
Scratches, and Patches, and all images have been pre-labeled accordingly. In
this study, Convolutional Neural Networks (CNN) were employed for the automated
identification of the types of defects present in the dataset. Various
predictive models were developed using the CNN algorithm, and to determine the
final model to be employed, several modifications were made to the dataset,
including pre-processing, artificial image augmentation, and hyperparameter
tuning. The results obtained from the various developed models are presented in
a comparative manner. The study demonstrates that when artificial intelligence
and deep learning techniques are effectively employed in the industrial
production sector, error types can be accurately identified, and this will
serve as a foundation for future research endeavors.
Key words: Deep Learning, CNN,
Steel Manufacturing, Quality Control