PHYSICS-INFORMED NEURAL NETWORKS FOR STRUCTURAL DYNAMICS: A COMPREHENSIVE REVIEW
Keywords:
Physics-informed neural networks, structural dynamics, inverse problems, parameter estimation, surrogate models, uncertainty quantification, machine learningAbstract
Physics-Informed Neural Networks (PINNs) represent a paradigm shift in computational mechanics by seamlessly integrating physical laws directly into neural network architectures. This review examines the state-of-the-art in PINN applications for structural dynamics, covering theoretical foundations, methodological advances, and practical implementations. We synthesize 50+ key studies across forward/inverse problems, parameter identification, and real-time monitoring applications. Results demonstrate that PINNs achieve 5-300,000× speedup over conventional finite element methods while maintaining comparable accuracy. Primary challenges include spectral bias in high-frequency problems, convergence difficulties with non-convex loss landscapes, and scalability to large-scale systems. Future directions emphasize hybrid frameworks integrating PINNs with finite elements, automated loss weighting strategies, and neuromorphic hardware implementation for energy-efficient structural health monitoring systems.
