Physics-Informed Machine Learning for Structural Integrity Assessment of Critical Infrastructure
TL;DR
A generalized PIML framework is proposed to support accurate, explainable, and intelligent structural health monitoring for resilient next-generation infrastructure.
Abstract
Structural integrity assessment is essential for ensuring the safety and reliability of critical infrastructure such as bridges, buildings, dams, tunnels, pipelines, and power plants. Traditional Structural Health Monitoring (SHM) methods rely on either physics-based simulations or data-driven machine learning, each having limitations in computational cost, data requirements, and generalization. Physics-Informed Machine Learning (PIML) integrates physical laws with deep learning to improve damage detection, structural condition assessment, and predictive maintenance using limited and noisy sensor data. This paper reviews recent PIML approaches, including PINNs, hybrid finite element–AI models, digital twins, and uncertainty-aware learning, while identifying key research challenges. A generalized PIML framework is proposed to support accurate, explainable, and intelligent structural health monitoring for resilient next-generation infrastructure.