Advancing Ultrahigh Carbon Steel Characterization: A Metaheuristic-Tuned Deep Learning Approach

Authors

• Walaa Omar El-Farouk Badr, Communications and Electronics Engineering Department, Mansoura Higher Institute of Engineering and Technology, Mansoura, Egypt
• Hossam El-Din Mostafa, Faculty of Artificial Intelligence and Information, Horus University-Egypt, New Damietta, Egypt
• Rania Elbana, Basic Science Department, Mansoura Higher Institute of Engineering and Technology, Mansoura, EgyptFollow

Subject Area

Material Science and Engineering

Article Type

Original Study

Abstract

Accurate classification of ultrahigh carbon steel (UHCS) microstructures is essential for elucidating processing-structure-property relationships and enhancing material performance. While convolutional neural networks (CNNs) offer powerful automated classification, navigating their complex, high-dimensional hyperparameter spaces present a significant computational bottleneck. To advance automated materials characterization, this study proposes a metaheuristic-tuned deep learning approach for robust and efficient microstructure classification. Two architectures, VGG16 and MobileNetV2, were evaluated under both pretrained and fine-tuned configurations, with Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Ant Colony Optimization (ACO) applied for systematic hyperparameter tuning. Results show that GA substantially improves the performance of fine-tuned VGG16, achieving a validation accuracy of 99.4%, while PSO and ACO yield competitive results in fine-tuned settings. For MobileNetV2, PSO achieves the highest validation accuracy of 95.9% in the pretrained configuration, whereas ACO enhances fine-tuned accuracy from 94.4% to 94.6%. Overall, GA exhibits stable and consistent performance across evaluation metrics, particularly when combined with VGG16. These results highlight the efficacy of metaheuristic optimization for improving accuracy, generalization, and reliability of ultrahigh carbon steel characterization

Keywords

Ultrahigh carbon steel, Microstructure classification, Materials characterization, Deep learning, Metaheuristic optimization, Hyperparameter tuning.

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 License.

Recommended Citation

Badr, Walaa Omar El-Farouk; Mostafa, Hossam El-Din; and Elbana, Rania (2026) "Advancing Ultrahigh Carbon Steel Characterization: A Metaheuristic-Tuned Deep Learning Approach," Mansoura Engineering Journal: Vol. 51 : Iss. 3 , Article 15.
Available at: https://doi.org/10.58491/2735-4202.3450