Predicting Mechanical Behavior of 3D-Printed Fiber-Reinforced Nylon Composites Through Image-Based YOLOv8 and CNN Models

Mechanical characterization of fiber-reinforced composites is a cornerstone of materials engineering but typically relies on destructive and time-consuming experimental protocols. To address this limitation, we present a novel deep learning framework that leverages image-based analysis for automated prediction of tensile behavior in 3D-printed fiber-reinforced nylon composites. The proposed system integrates a YOLOv8n object detection model with a Convolutional Neural Network (CNN), using scanning electron microscopy (SEM) images as input. The YOLOv8n model was trained to detect and quantify deformation regions before and after tensile testing, achieving an accuracy of 0.93, precision of 0.895, recall of 0.832, F1-score of 0.899, and a Matthews correlation coefficient (MCC) of 0.851 for deformation classification. Complementarily, the CNN was employed to predict the deformation rate directly from raw SEM image data. These outputs were further processed to estimate maximum deformation rate and ultimate tensile load, yielding near-perfect agreement with experimental measurements (R² = 0.9995, Pearson r = 0.9998). Model interpretability was enhanced through Gradient-weighted Class Activation Mapping (Grad-CAM), which highlighted the localized image features most influential in CNN predictions. By combining high-resolution imaging, robust object detection, and predictive modeling, this framework provides an accurate, explainable, and scalable solution for virtual mechanical testing. The proposed approach has potential to reduce reliance on destructive testing, accelerate composite material design, and facilitate the development of digital twins for advanced manufacturing applications.