Open Access
This is an unedited manuscript accepted for publication and provided as an Article in Press for early access at the author’s request. The article will undergo copyediting, typesetting, and galley proof review before final publication. Please be aware that errors may be identified during production that could affect the content. All legal disclaimers of the journal apply.
Dibya Tripathi,
Kheelraj Pandey,
Akhilesh Kumar Chauhan,
Ashwani Sharma,
- Research Scholar, Department of Mechanical Engineering, Amity University, Lucknow, Campus, Uttar Pradesh, India
- Assistant Professor, Department of Mechanical Engineering, Amity University, Lucknow, Campus, Uttar Pradesh, India
- Associate Professor, Department of Mechanical Engineering, KNIT, Sultanpur, Uttar Pradesh, India
- Assistant Professor, Department of Mechanical Engineering, Amity University, Lucknow, Campus, Uttar Pradesh, India
Abstract
The identification of microstructural defects in steel-fibre-reinforced composites is useful for the material integrity, mechanical reliability and longevity. Classic material inspection techniques face difficulties in recognizing the complex and visually similar defects as a result of heterogeneous material morphology, changing imaging conditions, and limits of manual feature extraction. To tackle those problems, this study presents a comparative deep-learning framework that integrates Convolutional Neural Networks (CNNs) with three hyperparameter optimization techniques namely Keras Tuner, Particle Swarm Optimization (PSO), and Grey Wolf Optimization (GWO). The framework focuses on automated recognition of composite defects through image pre-processing, hierarchical extraction of relevant features, and optimization-supported configuration.
Three methods are evaluated in a common comparative framework to study learning behaviour, convergence characteristic, and classification performance. The experiment assesses both training and validation accuracy as well as their associated losses over 25 epochs. The graphical comparison shows that CNN–GWO achieved the highest training and validation accuracy followed by CNN–PSO and CNN–Keras Tuner. Values derived from the given performance curves suggest that CNN–GWO achieves a validation accuracy of 99.5%, followed by CNN–PSO at 97.0% and CNN–Keras Tuner at 94.5%.
Moreover, CNN–GWO achieves high initial loss reduction and demonstrates good convergence properties, indicating its potential effectiveness in choosing discriminative CNN configurations for composite defect recognition. While these results demonstrate the advantages of CNN–GWO in the reported experimental setting, the statistical robustness and generalization of the results need to be evaluated with independent test sets and repeated experimental trials. The swarm-intelligence-assisted deep learning framework has the potential applications in automated material inspection. Moreover, it offers a comparative background for developing reliable defect classification systems for steel-fibre-reinforced composites.
Steel fiber reinforced composites (SFRCs) are widely used in the construction industry to improve strength and durability. However, the detection of defect in SFRCs is performed manually which is time consuming and does not provide accurate results. There are various automated deep learning approaches which can detect the defect in SFRCs quickly and provide reliable result. This study aims to carry out the hyperparameter optimization of
convolutional neural network (CNN) using Keras Tuner and with the help of advanced metaheuristic algorithms such as particle swarm optimization (PSO) and grey wolf optimization (GWO). The authors will develop a model which will provide better defect detection accuracy than the existing model and will make the process more automated. Results will be presented in the latter sections to validate the effectiveness of the model.
Keywords: Steel-Fibre-Reinforced Composites; Defect Detection; Convolutional Neural Network; Keras Tuner; Particle Swarm Optimization; Grey Wolf Optimization; Hyperparameter Optimization; Deep Learning.
References
[1] Zhang P, Wang C, Gao Z, Wang F. A review on fracture properties of steel fiber reinforced concrete. Journal of Building Engineering. 2023 May 15;67:105975.
[2] Ding Y, Zhou SX, Yuan HQ, Pan Y, Dong JL, Wang ZP, Yang TL, She AM. Crack identification method of steel fiber reinforced concrete based on deep learning: a comparative study and shared crack database. Advances in Materials Science and Engineering. 2021;2021(1):9934250.
[3] Yang L, Shami A. On hyperparameter optimization of machine learning algorithms: Theory and practice. Neurocomputing. 2020 Nov 20;415:295-316.
[4] O’Malley T, Bursztein E, Long J, Chollet F, Jin H, Invernizzi L, et al. KerasTuner [Internet]. 2019. Available from: github.com
[5] Kennedy J, Eberhart R. Particle swarm optimization. InProceedings of ICNN’95-international conference on neural networks 1995 Nov 27 (Vol. 4, pp. 1942-1948). ieee.
[6] Mirjalili S, Mirjalili SM, Lewis A. Grey wolf optimizer. Advances in engineering software. 2014 Mar 1;69:46-61.
[7] BELDAR P. Optimized convolutional neural networks for defect classification in fiber-reinforced composites. Sigma. 2026 Jun;44(3):2207-18.
[8] Medeghini F, Tiberti G, Guhathakurta J, Simon S, Plizzari GA, Mark P. Fiber orientation and orientation factors in steel fiber‐reinforced concrete beams with hybrid fibers: a critical review. Structural Concrete. 2025 Feb;26(1):481-500.
[9] Amin MN, Ahmad W, Khan K, Ahmad A. Steel fiber-reinforced concrete: a systematic review of the research progress and knowledge mapping. Materials. 2022 Sep 5;15(17):6155.
[10] Hampe M, “Micro-CT dataset of a steel fibre reinforced concrete sample,” DOI 10.5281/zenodo.10785684.
[11] Chen J, Yu Z, Jin H. Nondestructive testing and evaluation techniques of defects in fiber-reinforced polymer composites: A review. Frontiers in Materials. 2022 Oct 24;9:986645.
[12] HUANG L. Review of nondestructive testing methods for fiber-reinforced polymer composites. Acta Aeronautica et Astronautica Sinica. 2024 Mar 15;45(5):529697.
[13] Zheng Y, Lv X, Hu S, Zhuo J, Wan C, Liu J. Mechanical properties and durability of steel fiber reinforced concrete: A review. Journal of Building Engineering. 2024 Apr 1;82:108025.
[14] Daghigh V, Daghigh H, Lacy Jr TE, Naraghi M. Review of machine learning applications for defect detection in composite materials. Machine Learning with Applications. 2024 Dec 1;18:100600.
[15] LeCun Y, Bengio Y, Hinton G. Deep learning. nature. 2015 May 28;521(7553):436-44.
[16] Li L, Jamieson K, DeSalvo G, Rostamizadeh A, Talwalkar A. Hyperband: A novel bandit-based approach to hyperparameter optimization. Journal of machine learning research. 2018;18(185):1-52
[17] Lorenzo PR, Nalepa J, Kawulok M, Ramos LS, Pastor JR. Particle swarm optimization for hyper-parameter selection in deep neural networks. InProceedings of the genetic and evolutionary computation conference 2017 Jul 1 (pp. 481-488).
[18] Ye F. Particle swarm optimization-based automatic parameter selection for deep neural networks and its applications in large-scale and high-dimensional data. PloS one. 2017 Dec 13;12(12):e0188746.
[19] Kumar D. Grey Wolf Optimization Based Hyper-Parameter Optimized Convolution Neural Network for Kidney Image Classification. International Journal of Intelligent Engineering & Systems. 2022 May 1;15(3):319
[20] Cheng J, Tan W, Yuan Y, Zhao Z, Cheng Y. Research on defect detection method for composite materials based on deep learning networks. Applied Sciences. 2024 May 14;14(10):4161.
[21] Bai J, Wu D, Shelley T, Schubel P, Twine D, Russell J, Zeng X, Zhang J. A comprehensive survey on machine learning driven material defect detection. ACM Computing Surveys. 2025 Jun 12;57(11):1-36.
[22] Khanam R, Hussain M, Hill R, Allen P. A comprehensive review of convolutional neural networks for defect detection in industrial applications. Ieee Access. 2024 Jul 8;12:94250-95.
[23] Bergstra J, Bengio Y. Random search for hyper-parameter optimization. Journal of machine learning research. 2012 Feb 1;13(2).
[24] Yu X, Zhang H, Duan Y, Wang X. An optimized convolutional neural network based on multi-strategy grey wolf optimizer to identify crop diseases and pests. Displays. 2026 Jan 6:103341.
[25] Yu M, Xu J, Liang W, Qiu Y, Bao S, Tang L. Improved multi-strategy adaptive Grey Wolf Optimization for practical engineering applications and high-dimensional problem solving. Artificial Intelligence Review. 2024 Sep 5;57(10):277.
[26] Manickaraj K, Karthik A, Palanisamy S, Jayam M, Ali SK, Sankar SL, Al-Farraj SA. Improving Mechanical Performance of Hybrid Polymer Composites: Incorporating Banana Stem Leaf and Jute Fibers with Tamarind Shell Powder. BioResources. 2025 Jan 1;20(1):1998.
[27] Palanisamy S, Ayrilmis N, Sureshkumar K, Santulli C, Khan T, Junaedi H, Sebaey TA. Machine Learning Approaches to Natural Fiber Composites: A Review of Methodologies and Applications. BioResources. 2025 Jan 1;20(1).
[28] Sumesh KR, Palanisamy S, Khan T, Ajithram A, Ahmed OS. Mechanical, morphological and wear resistance of natural fiber/glass fiber-based polymer composites. BioResources. 2024, 19(2).
[29] Bachtiar, D., Mohammed, A.A., Palanisamy, S., Imran, A.I., Siregar, J.P., bin Mat Rejab, M.R., Syaubari, S., Cionita, T., Fitriyana, D.F., Al-Farraj, S.A. and Almansour, M.I., 2025. Effect of alkaline treatment on the thermal and mechanical properties of sugar palm fibre reinforced thermoplastic polyurethane composites. Scientific Reports, 15(1), p.14085.
[30] Palanisamy S, Mayandi K, Dharmalingam S, Rajini N, Santulli C, Mohammad F, Al-Lohedan HA. Tensile properties and fracture morphology of Acacia caesia bark fibers treated with different alkali concentrations. Journal of Natural Fibers. 2022 Nov 23;19(15):11258-69.
Journal of Polymer & Composites
| Volume | 14 | |
| 05 | ||
| Received | 15/10/2026 | |
| Accepted | 21/10/2026 | |
| Published | 23/10/2026 | |
| Publication Time | 8 Days |