Data-Driven Breakage Mechanics: Predicting the Evolution of Particle-Size Distribution in Granular Media

This paper presents a model-free data-driven framework for breakage mechanics. In contrast with continuum breakage mechanics, the de facto approach for the macroscopic analysis of crushable granular media, the present framework does not require the definition of constitutive models and phenomenological assumptions, relying on material behavior that is known only through empirical data. For this purpose, we revisit the recent developments in model-free data-driven computing for history-dependent materials and extend the main ideas to materials with particle breakage. A systematic construction of the modeling framework is presented, starting from the closed-form representation of continuum breakage mechanics and arriving at alternative model-free representations. The predictive ability of the data-driven framework is highlighted and contrasted with continuum breakage mechanics on different boundary value problems. Moreover, an application to a real experimental test in crushable sand is presented, where the data is furnished by high-fidelity grain-scale simulations, indicating that the proposed framework provides an accurate prediction of the mechanics of crushable materials including the state of comminution.