Title: EVALUATION OF DEEP LEARNING-BASED MODELS TO CLASSIFY MULTI-SPECIFICATIONS IN 3-POINT BEND FATIGUE-FAILED WOVEN GFRP LAMINATES

Authors: Mukesh Kumar Singh, Anand Gaurav

DOI: https://doi.org/10.33599/GL.2026.INCOMAT.TP26-0011

Abstract: The introduction of artificial intelligence and machine learning (AI-ML) in the field of composite science and technology and added new dimensions to their property prediction and damage analysis. Machine learning method comprises of multiple models that can be selected based on the problem description. Computer vision (CV) networks are powerful deep learning methods that are increasingly used to detect defects and state of damage in composite materials. Thus, in this work we used transfer learning method and evaluated four different convolution neural networks to make predictions on multi-specifications of GFRP samples failed under 3-point bend fatigue loading conditions. The predictors were fiber orientations, view of the TLP images (top or bottom) and the normalized fatigue life, whereas the tested models were DenseNet201, ResNet18, ResNet50 and VGG 16. The models were applied on as obtained images without any preprocessing to eliminate additional costs. DenseNet201 presented the highest accuracy and F1 score of 100 percent on fibre orientation, which is unacceptable for any ML model; while presented the same of 87.5%, 87.43%, 12.5% and 10.83% on view and normalized fatigue life. Among the acceptable models ResNet50 presented the best accuracy of 97.5% on orientation followed by ResNet18 and VGG16 models. The accuracy and F1 score for prediction of normalized fatigue life remained poor across the models, with all of them presenting the similar metrics of 12.5 and 10.8% respectively, indicating preprocessing is an important aspect to integrate deep learning methods in composite science and technology.

References: [1] Gaurav, Anand, and K. K. Singh. “Fatigue Behavior of FRP Composites and CNT-Embedded FRP Composites: A Review.” Polymer Composites 39 (2018): 1785–1808. [2] Gaurav, Anand. “Data‐driven machine learning regression methods to predict the residual strength in FRP composites subjected to fatigue.” Polymer Composites 46(12) (2025): 10713-10733. [3] Sorini, Andrea, Esteban Pineda, John Stuckner, and Peter A. Gustafson. “A Convolutional Neural Network for Multiscale Modeling of Composite Materials.” In AIAA Scitech 2021 Forum, 1–8. Reston, VA: American Institute of Aeronautics and Astronautics, 2021. [4] Azad, Md. Mahmud, A. ur Rahman Shah, M. N. Prabhakar, and H. S. Kim. “Deep Learning-Based Microscopic Damage Assessment of Fiber-Reinforced Polymer Composites.” Materials 17 (2024): 5265. [5] Mezeix, Lionel, Ana S. Rivas, Antoine Relandeau, and Christophe Bouvet. “A New Method to Predict Damage to Composite Structures Using Convolutional Neural Networks.” Materials 16 (2023): 7213. [6] Xu, Dong, Peng-Fei Liu, and Zhi-Peng Chen. “A Deep Learning Method for Damage Prognostics of Fiber-Reinforced Composite Laminates Using Acoustic Emission.” Engineering Fracture Mechanics 259 (2022): 108139. [7] Raaghav, V., D. Bikos, A. Rago, F. Toni, and M. Charalambides. “Explainable Prediction of the Mechanical Properties of Composites with CNNs.” 2025. [8] Ejaz, Faisal, Lee K. Hwang, Jun Son, Jae-Sung Kim, Dong S. Lee, and Byung K. Kwon. “Convolutional Neural Networks for Approximating Electrical and Thermal Conductivities of Cu-CNT Composites.” Scientific Reports 12 (2022): 13614. [9] Gavallas, Panagiotis, George Stefanou, Dimitrios Savvas, Christophe Mattrand, and Jean-Marc Bourinet. “CNN-Based Prediction of Microstructure-Derived Random Property Fields of Composite Materials.” Computer Methods in Applied Mechanics and Engineering 430 (2024): 117207. [10] Li, Jie, Wei Yao, Yu Lu, Jie Chen, Yong Sun, and Xue Hu. “High-Fidelity FEM Based on Deep Learning for Arbitrary Composite Material Structure.” Composite Structures 340 (2024): 118176. [11] Lee, Kyung-Ho, Hyun J. Lim, and Gi-Joon Yun. “A Data-Driven Framework for Designing Microstructure of Multifunctional Composites with Deep-Learned Diffusion-Based Generative Models.” Engineering Applications of Artificial Intelligence 129 (2024): 107590. [12] Gálvez-Hernández, Pablo, and Julian Kratz. “The Effect of Convolutional Neural Network Architectures on Phase Segmentation of Composite Material X-Ray Micrographs.” Journal of Composite Materials 57 (2023): 2899–2918. [13] Mizuno, Yusuke, Akira Hosoi, Hiroki Koshita, Daiki Tsunoda, and Hiroshi Kawada. “Fatigue Life Prediction of Composite Materials Using Strain Distribution Images and a Deep Convolutional Neural Network.” Scientific Reports 14 (2024): 25418. [14] Nguyen, H.-Q., B.-A. Le, B.-V. Tran, T.-S. Vu, and T.-L. Bui. “Deep Artificial Neural Network–Powered Phase Field Model for Predicting Damage Characteristics in Brittle Composites under Varying Configurations.” Machine Learning: Science and Technology 5 (2024): 025062. [15] Simonyan, Karen, and Andrew Zisserman. “Very Deep Convolutional Networks for Large-Scale Image Recognition.” 2015. [16] He, Kaiming, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. “Deep Residual Learning for Image Recognition.” 2015. [17] Huang, Gao, Zhuang Liu, Laurens van der Maaten, and Kilian Q. Weinberger. “Densely Connected Convolutional Networks.” 2018. [18] Ansari, Md. T. A., K. K. Singh, and Anand Gaurav. “Three-Point Bending Behavior of Angle-Ply GFRP Laminates under Static and Low-Cycle Fatigue Loading: Experimental and Statistical Analysis.” Journal of Reinforced Plastics and Composites (2025).

Conference: INCOMAT 2026

Publication Date: 2026/03/13

SKU: INCOMAT.TP26-0011

Pages: 11

Price: $22.00