AEM Solid Mechanics Seminar: Nihar Moghe
Recovering Full-Field Data from Noisy and Incomplete Measurements: A Multi-Fidelity Information Fusion Approach
Full-field measurement techniques such as digital image correlation (DIC) and digital volume correlation (DVC) provide increasingly rich, localized, and heterogeneous information from individual experiments, enabling efficient material characterization and data-driven constitutive discovery. However, the growing reliance on such information also makes imperfections in the data, such as noise and missingness, increasingly consequential. While substantial effort has focused on improving measurement protocols and correlation procedures, post-measurement approaches for recovering incomplete fields remain limited and often make little use of additional physical information available from the experiment.
This work presents MFINDER (Multi-Fidelity Imputator and Denoiser), a physics-informed, constrained multi-fidelity framework that treats field recovery as an information-fusion problem. MFINDER combines noisy and incomplete measurements with mechanically informed spatial trends from calibrated forward simulations and incorporates additional local and global observations from the same experiment as physical constraints, enabling simultaneous imputation and denoising. A modular, stage-wise probabilistic formulation allows these information sources to be incorporated and examined separately, improving interpretability and providing diagnostics of inconsistencies and potential recovery failures. Using DIC as a representative full-field technique, MFINDER is evaluated on two mechanically distinct synthetic problems---wedge-indentation and notched-tension---across different noise levels and missingness patterns, and on an experimental wedge-indentation case with imperfect model and constraint information. The framework demonstrates robust recovery across the investigated cases and outperforms the baseline approaches considered. The formulation remains readily extendable to other full-field measurement settings with similar information structures.