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

Abstract

We introduce a deep learning framework designed to train smoothed elastoplasticity models with interpretable components, such as the stored elastic energy function, yield surface, and plastic flow that evolve based on a set of deep neural network predictions. By recasting the yield function as an evolving level set, we introduce a deep learning approach to deduce the solutions of the Hamilton-Jacobi equation that governs the hardening/softening mechanism. This machine learning hardening law may recover any classical hand-crafted hardening rules and discover new mechanisms that are either unbeknownst or difficult to express with mathematical expressions. Leveraging Sobolev training to gain control over the derivatives of the learned functions, the resultant machine learning elastoplasticity models are thermodynamically consistent, interpretable, while exhibiting excellent learning capacity. Using a 3D FFT solver to create a polycrystal database, numerical experiments are conducted and the implementations of each component of the models are individually verified. Our numerical experiments reveal that this new approach provides more robust and accurate forward predictions of cyclic stress paths than those obtained from black-box deep neural network models such as the recurrent neural network, the 1D convolutional neural network, and the multi-step feed-forward model. c

Multiscale plasticity of geomaterials predicted via constrained optimization‐based granular micromechanics

“…On the other, demonstrating the constrained optimization framework's predictivity of macroscale decohesion (Figures 10 and 11) is of central importance going forward because it will provide the groundwork for more complex modes of mortar/aggregate interaction, for example, as learnt from direct numerical simulation (cf. Vlassis and Sun 83 ). We note that this is another considerable advantage of the proposed framework, at least for our own research objectives, that is, that by attaching a Lagrangian term defining the micro‐to‐macro kinematic hypothesis for tensor‐valued internal variables, a machine‐learned model can be homogenized using the same framework as the closed‐form models derived herein.…”

Predicting the Non-Deterministic Response of a Micro-Scale Mechanical Model Using Generative Adversarial Networks

“…Trained ML can computationally advantage classical physics-based numerical methods in several orders of magnitude [ 26 , 27 , 28 ]. In the geomechanics and materials fields AI (Artificial Intelligence) and ML have shown different levels of success in multiscale problems [ 29 , 30 ], material constitutive modelling [ 31 , 32 , 33 , 34 ] as well as in the study of composite in both forward and inverse design approaches [ 35 , 36 , 37 ]. Some recent applications of AI in the macro modelling of geotechnical problems include: natural hazard prediction and mitigation [ 38 ], determination of driven piles bearing capacity in sands using ANN [ 39 ], advanced ML techniques [ 40 ] and AI systems optimized by evolutionary computation [ 41 ], determination of slope stability with ANN [ 42 ], among others.…”

Bounding the Multi-Scale Domain in Numerical Modelling and Meta-Heuristics Optimization: Application to Poroelastic Media with Damageable Cracks

“…With the increasingly available computing power of classical computers, accurate discretizations of Partial Differential Equation (PDE) problems can serve the needs of researchers: a new experimental domain is born: in-simulatio experimentation [ 19 ]. In recent years artificial intelligence (AI) techniques have become very popular to solve problems which require a high level of cognition, e.g., image recognition, audio signal discrimination, autonomous driving, natural hazard mitigation [ 20 ], materials constitutive modelling ([ 21 , 22 , 23 , 24 , 25 , 26 , 27 ], among others), but also to solve physical problems which where traditionally the realm of PDEs (see [ 28 , 29 , 30 , 31 ] for example). AI paradigms are implemented in classical von Neumann computers because of their degree of developement compared to physical computing systems but those architectures may not be the optimal solution for such applications.…”