A Comprehensive Survey on Physics-Informed Neural Networks for Structural Digital Twins and Structural Health Monitoring

Physics-informed neural networks (PINNs) and broader physics- informed machine learning (PIML) methods are increasingly used in structural health monitoring (SHM) and structural digital twins (SDTs) because they incorporate governing equations, constitutive laws, and boundary conditions directly into the learning objective. This enables physically consistent inference from sparse and noisy sensor data, combining the flexibility of data-driven models with the regularization power of physics. Unlike high-fidelity simulation, accurate but computationally expensive and difficult to update online, and unlike black-box learning, which often fails under extrapolation and limited data, PIML offers a practical middle ground. This survey follows PRISMA ۲۰۲۰ and reviews ۶۰ studies published between ۲۰۲۱ and ۲۰۲۶ from ۴,۷۰۵ identified records. The selected works cover forward response prediction, inverse parameter identification, damage detection. and localization, and structural state estimation across bridges, buildings, composite structures, and benchmark systems. The survey contributes a three-axis taxonomy (task type, physics-coupling strategy, learning architecture), a comparative synthesis of performance trends and failure modes, and a critical assessment of key open challenges: loss balancing, model mismatch, uncertainty quantification, scalability, and benchmark standardization.

کلیدواژه ها:

PINN ، physics-informed machine learning ، structural health monitoring ، digital twin ، PRISMA ، civil infrastructure