A cooperative game for automated learning of elasto-plasticity knowledge graphs and models with AI-guided experimentation

Abstract

We introduce a multi-agent meta-modeling game to generate data, knowledge, and models that make predictions on constitutive responses of elasto-plastic materials. We introduce a new concept from graph theory where a modeler agent is tasked with evaluating all the modeling options recast as a directed multigraph and find the optimal path that links the source of the directed graph (e.g. strain history) to the target (e.g. stress) measured by an objective function. Meanwhile, the data agent, which is tasked with generating data from real or virtual experiments (e.g. molecular dynamics, discrete element simulations), interacts with the modeling agent sequentially and uses reinforcement learning to design new experiments to optimize the prediction capacity. Consequently, this treatment enables us to emulate an idealized scientific collaboration as selections of the optimal choices in a decision tree search done automatically via deep reinforcement learning.

A configurational force for adaptive re-meshing of gradient-enhanced poromechanics problems with history-dependent variables

“…In the case of initial OCR = 1.25, the global responses are not sensitive to the mesh size regardless of whether micromorphic regularization method is used. This result is expected as the material exhibits associative plastic flow and strain hardening [90,[101][102][103][104][105]. Here the initial preconsolidation stress ( p ′ c 0 ) is set to −1000 kPa while the initial confining pressure ( p ′ 0 ) is −800 kPa.…”

Sobolev training of thermodynamic-informed neural networks for interpretable elasto-plasticity models with level set hardening

“…Plastic deformation of materials is a history-dependent process manifested by irreversible and permanent changes of microstructures, such as dislocation, pore collapses, growth of defects, and phase transition. Macroscopic constitutive models designed to capture history-dependent constitutive responses can be categorized into multiple families, such as hypoplasticity, elastoplasticity, and generalized plasticity [1][2][3][4]. For example, hypoplasticity models often do not distinguish the reversible and the irreversible strain [5][6][7].…”

“…While there are existing regularization techniques such as dropout layers [16], cross-validation [17,18], and/or increasing the size of the database that could be helpful, it remains difficult to assess their credibility without the interpretability of the underlying laws deduced from the neural network. Another approach could involve symbolic regression through reinforcement learning [3] or genetic algorithms [19] that may lead to explicitly written evolution laws, however, the fitness of these equations is often at the expense of readability.…”

Mechanical MNIST: A benchmark dataset for mechanical metamodels

“…The motivation for developing improved strategies for constructing metamodels relevant to mechanical data is twofold. First, meta- models enable unprecedented exploration of the model parameter space [21,22] and enhanced metamodel performance will lead to improvements in the computational methods that rely on them [23]. Second, similar to synthetic datasets in computer vision [24], the synthetic datasets generated by mechanical models are a proxy for real world data.…”