Machine learning for predicting fiber-reinforced polymer durability: A critical review and future directions

Abstract

Fiber-reinforced polymer (FRP) composites offer significant advantages for civil infrastructure, like lightweight, high strength, and corrosion resistance. However, their broader implementation is limited by uncertainties regarding their durability under various environmental conditions. These challenges stem from the inherent complexity of predicting FRP performance, as their degradation involves multiple mechanisms. Traditional methods, mainly depending on empirical correlations and accelerated aging tests, struggle to generalize across real-world conditions and isolate individual degradation mechanisms, undermining the reliability of their predictions. Machine learning (ML) presents a compelling alternative, with the ability to manage non-linear relationships among numerous variables. Recent advancements in FRP durability testing have produced extensive data, creating opportunities for ML-driven predictive modeling. While studies have shown the great potential of ML, current research focuses primarily on algorithm selection, yielding limited practical insights for FRP design and field application. This study conducts a systematic assessment of ML techniques for FRP durability prediction, identifying key factors governing model performance and highlighting current gaps. Building on these insights, the paper proposes future research directions, aiming to improve the practical utility of ML-based durability predictions for FRP composites.