Artificial Neural Network-Based Constitutive Modeling of Mullins-Type Stress Softening
Please login to view abstract download link
Rubber-like materials often exhibit pronounced stress-softening under cyclic loading, commonly associated with the Mullins effect. This contribution investigates a physics-augmented artificial neural network formulation for Mullins-type stress softening in incompressible rubber-like materials under finite-strain loading. Synthetic cyclic loading data are generated using a classical Mullins-type model combined with a Neo-Hookean base response. The artificial neural network is embedded within a thermodynamically motivated constitutive framework and trained using stress-stretch data. The approach is evaluated on cyclic loading paths with increasing maximum stretch and tested on unseen loading histories. The results demonstrate the ability of the proposed formulation to reproduce Mullins-type stress softening while maintaining a physically interpretable and constraint-based constitutive structure.
