Data-Driven Discovery of Multi-Dimensional Breakage Population Balance Equations

TL;DR

This paper introduces a novel data-driven sparse regression method, mPBE-ID, for discovering multi-dimensional breakage population balance equations directly from data, overcoming limitations of previous one-dimensional approaches.

Contribution

The paper develops the first systematic framework for identifying multi-dimensional breakage equations using data-driven techniques, incorporating DMD insights and robust handling of noisy data.

Findings

Successfully discovers various forms of mPBEs from limited/noisy data.

Integrates DMD for effective candidate library construction.

Demonstrates robustness and applicability to complex particulate phenomena.

Abstract

Multi-dimensional breakage is a ubiquitous phenomenon in natural systems, yet the systematic discovery of underlying governing equations remains a long-standing challenge. Current inverse solution techniques are restricted to one-dimensional cases and typically depend on the availability of a priori system knowledge, thus limiting their applicability. By leveraging advances in data-driven sparse regression techniques, we develop the Multi-Dimensional Breakage Population Balance Equation Identification (mPBE ID) algorithm for discovering multi-dimensional breakage population balance equations (mPBEs) directly from data. Our mPBE-ID enables tractable identification of mPBEs by incorporating several key strategies, namely, a breakage-informed constrained sparse regression, targeted candidate library functions construction via insights from Dynamic Mode Decomposition (DMD), and robust handling of noisy/limited data through ensembling (bagging/bragging). Notably, we demonstrate how the DMD is indispensable for distilling dominant breakage dynamics which can then be used to facilitate the systematic inclusion of candidate library terms. We showcase the ability of the mPBE-ID to discover different forms of mPBE (including those with discontinuous stoichiometric kernels) even when tested against noisy and limited data. We anticipate that the mPBE-ID will serve as a foundational framework for future extensions to generalize the discovery of multi-dimensional PBEs for various high-dimensional particulate phenomena.