Damage localization using differentiable physics and displacement-based structural identification

As the built environment ages and incurs natural wear and tear, it becomes imperative to employ measures that allow cost-effective, rapid, and automated inspection methods to ensure human safety. Displacement-based damage localization methods can meet these needs through non-contact monitoring techniques such as digital image correlation that provide high-resolution displacement measurement across large-scale civil structures rapidly and with great accuracy. In this work, we explore the use of displacement information as the input to an inverse solver that reconstructs the spatially varying material properties of the target system. We demonstrate the usefulness of the displacement-based data and our inverse solver on a collection of simple case studies: a fixed-fixed beam, single-story, single-span frame, and multi-story, multi-span frame. We systematically study the sensitivity of the system to damage and its location, the ability of the model to identify and localize multiple simultaneous damage, the damage severity, and the robustness of the optimizer with respect to initialization. We also study challenges likely to face the method in real-world applications such as noisy measurements, reconstruction without full-field data, extension to different system sizes and geometries, and sensitivity of the algorithm to mismatch in resolution. Investigation of the single-span frame shows the solver can achieve localization accuracy of 83.8% when there is no measurement noise and 71.3% with 2% noise. Additionally, the proposed method achieves 82.9% accuracy using surface displacement-only which is slightly lower than 84.2% localization accuracy when the full-depth displacement is available. Results demonstrate the effectiveness of the proposed solver in identifying subsurface damages with high accuracy using surface displacement only.