Flaschel M, Moreno Mateos MA, Wiesheier S, Steinmann P, Kuhl E (2026)
Publication Status: Submitted
Publication Type: Journal article
Future Publication Type: Journal article
Publication year: 2026
Book Volume: 461
Article Number: 119256
DOI: 10.1016/j.cma.2026.119256
Institutions such as DIN, EN, ISO, and ASTM define standardized experimental protocols to ensure the reproducible and consistent characterization of mechanical material behavior. If a comprehensive database of precomputed material responses were available for these standardized tests, material characterization could instead be formulated as a pattern recognition problem, which enables significantly faster and more robust identification compared to optimization-based methods. This paradigm underlies our recently proposed Material Fingerprinting method. Material Fingerprinting is a lookup table-based strategy to infer material models from experimental measurements, which completely avoids the need to solve a continuous optimization problem. In an offline phase, a comprehensive database of simulated material responses, so-called material fingerprints, is generated for a predefined, standardized experimental setup. Although it can be extended a posteriori, the database is generated only once. Then, it is used repeatedly in an ultra-fast online phase to infer material models from experiments performed according to guidelines and standardized sample geometries. The experimentally measured fingerprint is compared with a database to identify the closest match. The main advantages of Material Fingerprinting are two: (i) ultra-fast material model inference in just a few seconds and (ii) robustness in the identification because a continuous optimization problem does not need to be solved. The method circumvents ill-posedness and non-convex landscape-related issues in traditional, usually computationally-expensive, methods. To date, there exists no demonstration of Material Fingerprinting applied to unsupervised experimental datasets (i.e., sets of full-field displacements and global reaction forces). Here, unsupervised Material Fingerprinting offers a robust material modeling framework by directly comparing precomputed simulated displacements and reaction forces with experimental counterparts. In this work, we apply this strategy to biaxial deformation tests of soft elastomer specimens (Elastosil, Sylgard, and VHB tape) with a central strain concentrator and inhomogeneity in the deformation field. We construct a single database across different materials and infer hyperelastic material models. We show that, for an already existing standardized database, Material Fingerprinting is several orders of magnitude faster than comparable optimization-based approaches for material model characterization from full-field measurements. The method provides parameter estimates that are close to the optimum identified by optimization-based approaches, while requiring only a fraction of the computational effort. We also demonstrate that the models identified through Material Fingerprinting can serve as high-quality initial guesses for optimization-based methods. The database generated in this work can be used to infer constitutive models for other unseen materials, always following the testing guidelines and standardized sample geometry in our consciously designed experimental protocol.
APA:
Flaschel, M., Moreno Mateos, M.A., Wiesheier, S., Steinmann, P., & Kuhl, E. (2026). Unsupervised Material Fingerprinting: Ultra-fast hyperelastic model discovery from full-field experimental measurements. Computer Methods in Applied Mechanics and Engineering, 461. https://doi.org/10.1016/j.cma.2026.119256
MLA:
Flaschel, Moritz, et al. "Unsupervised Material Fingerprinting: Ultra-fast hyperelastic model discovery from full-field experimental measurements." Computer Methods in Applied Mechanics and Engineering 461 (2026).
BibTeX: Download