Machine Learning for Materials Property Prediction

With the increased use of data-driven approaches and machine learning-based methods in material science, the importance of reliable uncertainty quantification (UQ) of the predicted variables for informed decision-making cannot be overstated.

Accordingly, various approaches for UQ have been developed, either through a quantified measure of the variance in the target variable, via confidence intervals, or by other means. Among the conventional UQ methods for multivariable regression, Gaussian Process Regression (GPR) has been generally adopted as the state-of-the-art approach that provides accurate single-point predictions and reliable uncertainty estimates. However, GPR has also important limitations, since the commonly used isotropic covariance kernels, such as Gaussian and Matern kernels, are less suitable for modeling functions with spurious covariates or anisotropic smoothness.

In addition, UQ in material property prediction poses unique challenges, including difficulties in quantifying uncertainties due to the multi-scale and multi-physics nature of advanced materials, intricate interactions between numerous factors, limited availability of large curated datasets for model training, substantial variability in material properties due to test conditions and inherent factors, measurement errors, and others.

Recently, Bayesian Neural Networks (BNNs) have emerged as a promising approach for UQ, offering a probabilistic framework for capturing uncertainties within neural networks. In this work, we introduce an approach for UQ within physics-informed BNNs, which integrates knowledge from governing laws in material modeling to guide the models toward physically consistent predictions. Our findings indicate that BNNs based on Markov Chain Monte Carlo approximation of the posterior distribution of network parameters is a promising framework for creep life prediction.

 

Publications

1. L. Li, j. Changs, A. Vakanski, Y. Wang, T. Yao, and M. Xian, "Uncertainty Quantification in Multivariable Regression for Material Property Prediction with Bayesian Neural Networks," [arXiv], 2024.