Failure behavior of granular materials is governed by their microstructure and its loading-induced evolution. While discrete simulations can capture these mechanisms with high fidelity, their use is limited by prohibitive computational cost and the requirement for detailed knowledge of the microstructure. Homogenization-based frameworks have been developed to address these limitations while retaining essential micromechanical features. In this context, the Granular Micromechanics Approach (GMA) provides a robust alternative to both discrete and classical continuum models by deriving macroscopic behavior through homogenization of particle interactions over all orientations [1]. Without explicitly resolving the granular microstructure, GMA is thus able to account for loading-induced anisotropic microstructural evolution and its influence on macroscopic response and failure. Nevertheless, the resulting macroscopic failure characteristics emerge implicitly from microscale assumptions and are often highly nonlinear and sensitive to parameter choices, complicating both model calibration and constitutive design. Moreover, ensuring physically admissible macroscopic responses remains a longstanding challenge. This contribution presents a data-driven multiscale strategy to systematically explore and control the mapping between microscale constitutive parameters and macroscopic failure envelopes in homogenization-based continuum models for granular solids. Surrogate models are constructed to represent this mapping efficiently, enabling both forward prediction of macroscopic failure behavior and inverse identification of microscale parameters from target macroscopic responses. The framework further allows admissibility constraints to be incorporated at the macroscopic level, providing a pathway toward constitutive models with guaranteed physical consistency. The proposed approach illustrates how surrogate models can complement classical multiscale modeling by enabling rapid parameter exploration, inverse design, and admissibility enforcement without compromising the underlying physics-based multiscale formulation. Although demonstrated for granular materials, the methodology is broadly applicable to other multiscale constitutive models exhibiting complex failure behavior. REFERENCES [1] P. Poorsolhjouy, K. Mews, A. Misra. Rock Failure Characteristics Evaluated Under True Triaxial Loading from Micro-mechanical Viewpoint. Rock Mech. & Rock Eng., 58, 5653–5672, (2025)