ARTIFICIAL INTELLIGENCE FOR WIND-INDUCED VIBRATION PREDICTION IN STRUCTURES: A CRITICAL REVIEW OF DEEP LEARNING, PHYSICS-INFORMED, AND HYBRID APPROACHES

Abstract

Wind-induced vibrations pose critical challenges to modern infrastructure, including long-span bridges, high-rise buildings, and wind turbines. Traditional physics-based methods for predicting these vibrations are computationally intensive and often limited in capturing complex nonlinear fluid-structure interactions. Artificial intelligence and machine learning approaches have emerged as transformative alternatives, significantly enhancing prediction accuracy while reducing computational burden. This comprehensive review synthesizes recent advances in AI-based wind vibration prediction, examining deep learning architectures including Long Short-Term Memory networks, Convolutional Neural Networks, and hybrid ensemble methods. We analyze 45+ peer-reviewed studies spanning diverse applications from bridge flutter prediction to wind turbine aeroelastic response forecasting. Key findings reveal that hybrid CNN-LSTM models achieve prediction accuracies exceeding 96%, with computational times reduced by 95% compared to traditional CFD approaches (Meng et al., 2025; Zhou et al., 2023). Physics-informed neural networks demonstrate superior generalization across operational conditions, while ensemble methods exhibit enhanced robustness under uncertain environmental loading (Ijaz & Manzoor, 2024; Mostafa et al., 2022). Critical challenges including data scarcity, model interpretability, and real-world validation gaps are identified and addressed. This review provides practitioners with evidence-based guidance for implementing AI-driven solutions in wind engineering applications, identifies future research directions, and establishes benchmarks for model performance evaluation across structural typologies