Inference and Uncertainty Propagation of Atomistically Informed Continuum Constitutive Laws, Part 2: Generalized Continuum Models Based on Gaussian Processes

Abstract

Constitutive models in nanoscience and engineering often poorly represent the physics due to significant deviations in model form from their macroscale counterparts. In Part 1 of this study, this problem was explored by considering a continuum scale heat conduction constitutive law inferred directly from molecular dynamics (MD) simulations. In contrast, this work uses Bayesian inference based on the MD data to construct a Gaussian process emulator of the heat flux as a function of temperature and temperature gradient. No assumption of Fourier-like behavior is made, requiring alternative approaches to assess the well-posedness and accuracy of the emulator. Validation is provided by comparing continuum scale predictions using the emulator model against a larger all-MD simulation representing the true solution. The results show that a Gaussian process emulator of the heat conduction constitutive law produces an empirically unbiased prediction of the continuum scale temperature field for a variety of time scales, which was not observed when Fourier's law is assumed to hold. Finally, uncertainty is propagated in the continuum model and quantified in the temperature field so the impact of errors in the model on continuum quantities can be determined.

Uncertainty Quantified Parametrically Homogenized Constitutive Models for Microstructure-Integrated Structural Simulations

“…Bayesian inference has also been extensively used to quantify the uncertainties related to model calibration data. For example, the Gaussian process emulator with Bayesian inference has been used to calibrate and quantify uncertainties in thermal conduction constitutive law from molecular dynamics simulations [36]. Recently, a Bayesian hierarchical model accounting for parameter uncertainty, material property variability and measurement noise has been proposed to calibrate Voce hardening parameters of a visco-plastic self-consistent crystal plasticity model and to propagate the uncertainty to stress-strain response [34,35].…”

Accelerating multiscale modelling of fluids with on-the-fly Gaussian process regression

“…Machine learning has also been used to aid hybrid methods in fluid dynamics: to quantify the uncertainty propagating from the micro- to the macro-model as a function of time-averaging window and the amount of sampled data (Salloum et al. 2012 ); constructing a constitutive relation for a continuum model that is applicable to nanoscale physics (Salloum and Templeton 2014 ); and building a surrogate model to replace MD simulations using neural networks (NNs) (Asproulis and Drikakis 2013 ) and Gaussian processes (GPs) (Salloum and Templeton 2014 ). However, all such approaches, bar that in Asproulis and Drikakis ( 2013 ) are limited by the training data used to fit the ML model; i.e.…”

Uncertainty quantification in LES of channel flow

“…In the context of this work, we explore the predictive capabilities of f in a channel flow configuration, when employing the calibrated parameters for the k s g s sub‐grid scale model. We note that Bayesian estimation has been successfully used to infer model parameters within multiscale settings in other applications, such as molecular dynamics , porous media flows , and Carbon cycle models . Because Bayesian calibration has been successfully applied in these diverse areas, it is a reasonable to assess its applicability to turbulence modeling.…”