#27. Uncertainty Quantification, Sensitivity Analysis, and Machine Learning in Materials Modeling


  • Philippe Geubelle, University of Illinois at Urbana-Champaign, USA
  • Lori Graham-Brady, Johns Hopkins University, USA
  • James Kermode, University of Warwick, UK
  • Jaroslaw Knap, Army Research Laboratory, USA
  • Marisol Koslowski, Purdue University, USA
  • Maryam Shakiba, Virginia Tech, USA
  • Michael Shields, Johns Hopkins University, USA (michael.shields@jhu.edu)
  • Xiang Zhang, University of Wyoming, USA


It is increasingly recognized that computational modeling of materials requires a careful accounting of uncertainties that exist at various length-scales and in the bridging between length and time-scales. For example, heterogenous materials inherit random properties which affect their fracture, damage, and failure response. Moreover, manufacturing procedures produce uncertainties in material properties. This process of uncertainty quantification (UQ) is often aided by the application of advanced machine learning (ML) algorithms (e.g. Gaussian process regression, artificial neural networks, Bayesian inference, etc.) and sensitivity analysis (SA). SA assists estimating the relative importance of uncertain parameters either through gradient calculation or variance decomposition. Consequently, the advancement of UQ, SA, and ML in materials modeling have come to be strongly coupled. The aim of this symposium is to highlight recent advancements in the use of ML methodology for UQ/SA in materials modeling and to provide a platform for researchers to present, discuss, and exchange the latest development in SA and UQ.

In particular, we aim to explore the UQ, SA, and ML associated with:
• The definition, calibration, learning, and sensitivity of interatomic potentials for atomistic simulations
• Coarse-graining from discrete, lower-scale models to upper-scale continuum models
• The definition, calibration, learning, and sensitivity of constitutive models for continuum simulations at various length-scales
• Simulation of composite microstructures for failure analysis, fracture modeling, and material/shape parameter sensitivity
• Model-order reduction associated with scale-bridging
• Inference of material properties, constitutive relations, and governing equations from limited experimental and/or simulation data