Towards efficient structural inverse analysis based on AI-driven differentiable optimization method

• Proposed a differentiable inverse analysis method using gradient descent . • Estimated damage parameters with high accuracy and provided probabilistic distributions. • Accomplished inverse analysis in seconds after deployment. • Reduced full-chain time to one-tenth of traditional heuristic-based methods. • Adaptable to various structural systems and monitoring strategies. Structural deterioration caused by material fatigue and environmental factors poses significant risks to the safety and performance of engineering structures. Inverse analysis plays a pivotal role in identifying internal state parameters from observed structural responses, enabling effective maintenance and performance evaluation. However, mainstream finite element model updating (FEMU) methods face significant challenges when applied to large-scale or complex systems due to their high computational cost and time demands. To overcome these limitations, this paper introduces an AI-driven differentiable optimization framework for structural inverse analysis. The framework integrates a deep learning-based surrogate model to replace finite element method calculations, significantly enhancing computational efficiency, and employs gradient-based optimization to minimize resource requirements. Unlike traditional AI approaches that rely on deterministic end-to-end mapping, the proposed framework decouples forward calculations from inverse analysis, enabling the estimation of probability distributions for state parameters and accommodating multiple potential solutions. Validation on a damaged reinforced concrete (RC) frame-shear wall structure demonstrated the framework’s effectiveness. Damage parameter estimation achieved 83.0% accuracy in a single iteration and improved to 88.2% through iterative updates, with spatial distributions aligning closely with true values. The framework reduces full-chain time to one-tenth compared to FEMU methods and performs inverse analysis within seconds after deployment. Its flexibility in selecting state parameters and structural responses ensures adaptability across diverse structural systems and monitoring strategies. These features expand the framework’s applicability to real-time or near-real-time structural health monitoring and predictive maintenance, providing a practical, efficient, and scalable solution to modern engineering challenges.