Workshop Announcement

Title: EUCLID: Unsupervised Discovery of Hyperelastic Constitutive Laws (无监督超弹性本构发现)

Duration: 2 Days

Date: 15 Oct, 2026

This hands-on workshop introduces EUCLID (Efficient Unsupervised Constitutive Law Identification & Discovery), a groundbreaking framework that automatically discovers interpretable hyperelastic constitutive models directly from experimental data—without requiring any stress-labeled ground truth.

Key Topics

Physics‑Informed Sparse Regression
Automatically discover both the functional form and the parameters of the strain‑energy density function from full‑field displacement and reaction‑force measurements.
Custom Feature Library
Build a rich basis of invariant‑based candidates, including:
- I ₁ = tr ( C )
- I ₂ = tr ( C ² )
Polynomial and exponential combinations of these invariants.
Unsupervised Learning Paradigm – Core Innovation
No stress annotations are required. The method uses only displacement fields and reaction forces to recover a complete, physically consistent constitutive law.
Automatic Feature Selection via L₁ Regularization
From hundreds of candidate terms, the algorithm automatically selects the 2–5 most physically relevant strain‑energy terms, ensuring parsimony and interpretability.
Hands‑On Project
Implement the full EUCLID pipeline in Python to discover a hyperelastic constitutive law from real‑style benchmark data.
(All materials include a Chinese project introduction for bilingual reference.)

Who Should Attend

Researchers and engineers in solid mechanics, materials science, computational mechanics, and data‑driven modeling who want to move beyond black‑box neural networks toward interpretable, physics‑grounded material modeling.

Format

Lectures + live coding + guided project work. All code and datasets will be provided.