Computational study demonstrates enhanced failure surface prediction in concrete under triaxial compression, highlighting a robust pathway for physics-informed constitutive models.
Key Points
To develop a physics-informed Gaussian process regression framework that replaces empirical failure surfaces in the Karagozian & Case concrete model with an uncertainty-aware, data-driven surrogate.
Retained the modular elastoplastic structure of the Karagozian & Case concrete model while training a Gaussian process regression surrogate directly on triaxial compression test observables under varying confinement levels.
Integrated derivative-based physical constraints matching known material behavior into the surrogate model to evaluate generalization at confinement levels omitted from training.
Unconstrained Gaussian process regression interpolated accurately near training conditions but deteriorated and violated fundamental physical constraints during extrapolation, even with simulated data augmentation.
Physics-informed Gaussian process regression maintained physical consistency, substantially improved prediction accuracy at high confinement levels outside the training set, and tightened confidence intervals by reducing predictive variance.