Transfer Learning Based Mechanical Reliability Prediction of Smart Sustainable Marine Composite Materials Under Cyclic Loading

Abstract

For offshore applications, smart sustainable marine composites offer excellent strength and corrosion resistance;
nevertheless, cyclic stress causes increasing deterioration. This research proposes a transfer learning-based model for predicting the mechanical reliability of smart sustainable marine composite materials under cyclic loading. Composite laminates reinforced with sustainable fibers and a bio-based polymer matrix were fabricated and experimentally characterized through tensile, flexural, and fatigue tests. Damage evolution and failure mechanisms were analyzed using scanning electron microscopy (SEM), while dynamic mechanical analysis (DMA) and digital image correlation (DIC) were employed to assess stiffness degradation and strain distribution. The optimized composite achieved 78.6 MPa tensile strength and 118.5 MPa flexural strength at 40% fiber volume fraction (FVF), with 55.3% residual strength retention after 75,000 cycles. The transfer learning model attained R² (0.98) and Root mean square error (RMSE) (2.84 MPa), enabling accurate reliability prediction with limited data. Overall, the proposed model supports durable and reliable offshore marine composite structural design.