A review of dynamics-based failure modeling and diagnosis techniques for gear systems

• Reviews dynamics-based modeling methods for gear system failures. • Covers various gear faults, including wear, pitting, and breakage. • Evaluates traditional signal processing and machine learning diagnostics. • Integrates dynamic fault models for early failure detection. • Highlights challenges and future research directions. Gear systems are critical in industrial and military applications but prone to failures under harsh conditions, leading to economic losses and safety risks. This paper provides a comprehensive review of dynamics-based failure modeling and diagnosis techniques for gear systems. It systematically evaluates three modeling approaches: analytical methods (efficient but limited to simple systems), lumped parameter models (balance efficiency and multi-DOF dynamics), and finite element methods (high accuracy but computationally intensive). Hybrid strategies integrating these methods and machine learning are highlighted to enhance computational efficiency and accuracy. Common failure mechanisms, including cracks, pitting, and wear, are analyzed, emphasizing their effects on time-varying mesh stiffness (TVMS) and vibration characteristics. Signal processing and machine learning techniques are discussed for fault feature extraction and diagnosis, with advanced methods like variational modal decomposition and AI-augmented models demonstrating superior performance. Challenges in real-time diagnostics, model generalizability, and coupled failure analysis are identified. Future directions propose hybrid AI-physics models, digital twins, and multi-scale frameworks to improve predictive maintenance. This review bridges theoretical insights and practical applications, offering a foundation for advancing gear system reliability and intelligent fault diagnosis.