Surya R・ Kalidindi”

¹ Georgia Institute of Technology, Atlanta, GA, USA

EXTENDED ABSTRACT: The dramatic acceleration of the current materials innovation cycles is contingent on the development and implementation of high throughput strategies in both experimentation and physics-based simulations, and their seamless integration using the emergent AI/ML (artificial intelligence/machine learning) toolsets. This talk presents a novel information gain-driven Bayesian ML framework [1] with the following main features: (i) explicit consideration of the physics parameters as inputs (i.e., regressors) in the formulation of process-structure-property (PSP) surrogate models needed to drive materials development workflows; (ii) information gain-driven autonomous workflows for training efficient ML surrogates to otherwise computationally expensive physics-based simulations; (iii) Bayesian design of experiments strategies that identify objectively the next experiment to be performed in order to maximize the expected information gain;

(iv)   versatile feature engineering for multiscale material internal structure using the formalism of n-point spatial correlations;

(v)     amenable to a broad suite of surrogate model building approaches (including Gaussian Process regression (GPR), convolutional neural networks (CNN)); and (vi) Markov chain Monte Carlo (MCMC)-based computation of posteriors for physics parameters using all available experimental observations (usually disparate and sparse). The benefits of this new framework will be illustrated with multiple case studies.

Keywords: Bayesian learning, AI/ML, materials data science, design of experiments, physics-based simulations REFERENCES

[1] S. R. Kalidindi, MRS Communications, 9, (2019) 518-531.