Unsupervised Material Fingerprinting: Ultra-fast hyperelastic model discovery from full-field experimental measurements

2026-01-21

AI Summary

This work proposes an unsupervised “material fingerprinting” approach to address the high computational cost and susceptibility to local minima associated with conventional optimization methods in identifying constitutive models for hyperelastic materials. By constructing an offline simulation database that encompasses full-field displacement and net reaction force responses, the method enables rapid online identification of the optimal constitutive model through nearest-neighbor matching between experimental full-field measurements and database entries. Crucially, this framework operates without requiring labeled stress–strain data or solving non-convex optimization problems. The study presents the first experimental validation of this unsupervised paradigm on real-world data, demonstrating significant improvements in efficiency, stability, and robustness for material model identification.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Learning with Manifolds
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Models for Web evolution
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Material Fingerprinting is a lookup table-based strategy to discover material models from experimental measurements, which completely avoids the need to solve an optimization problem. In an offline phase, a comprehensive database of simulated material responses, so-called material fingerprints, is generated for a predefined experimental setup. This database can then be used repeatedly in the online phase to discover material models corresponding to experimentally measured observations. To this end, the experimentally measured fingerprint is compared with all fingerprints in the database to identify the closest match. The primary advantage of this strategy is that it does not require solving a continuous optimization problem. This avoids the associated computational costs as well as issues of ill-posedness caused by local minima in non-convex optimization landscapes. Material Fingerprinting has been successfully demonstrated for supervised datasets consisting of stress-strain pairs, as well as for unsupervised datasets involving full-field displacements and net reaction forces. However, to date, there is no experimental validation for the latter approach which is the objective of this work.

Problem

• Unsupervised Material Fingerprinting
• hyperelastic model discovery
• full-field experimental measurements
• material model identification
• experimental validation

Innovation

• Unsupervised Material Fingerprinting
• Hyperelastic model discovery
• Full-field measurements
• Lookup table-based identification
• Experimental validation