Discovering non-associated pressure-sensitive plasticity models with EUCLID.

Area of Science:

• Computational mechanics
• Materials science
• Solid mechanics

Background:

• Automated material model discovery is crucial for predicting material behavior.
• Existing methods struggle with complex plasticity models like pressure-sensitive ones.
• Interpretable constitutive laws are needed for engineering applications.

Purpose of the Study:

• To extend the EUCLID framework for pressure-sensitive plasticity models.
• To enable discovery of arbitrarily shaped yield surfaces with convexity and non-associated flow rules.
• To achieve a balance between model accuracy and simplicity.

Main Methods:

• Utilized a data-driven framework (EUCLID) requiring only full-field displacement and boundary force data.
• Constructed a material model library using Fourier series for yield surfaces and pressure-sensitive terms.
• Implemented sparsity-promoting regularization and convexity constraints for inverse optimization.

Main Results:

• Successfully extended EUCLID to pressure-sensitive plasticity with non-associated flow rules.
• Demonstrated accurate material model selection from a library using noisy experimental data.
• Learned constitutive laws are presented as interpretable mathematical expressions.

Conclusions:

• EUCLID provides an effective approach for automated discovery of complex material models.
• The framework accurately captures pressure sensitivity, yield surface shape, and flow rules.
• This method offers a robust alternative to traditional parameter identification techniques.