Contextual AI for Materials: Integrating PINNs, Generative Models, and NLP for Data-Scarce Discovery - College of Engineering - Purdue University

We propose an explainable and contextual AI framework to accelerate the design of advanced materials—such as battery and energetic systems—by targeting the microstructure–property nexus in data-scarce regimes. The core is a Physics-Informed Neural Network (PINN) that encodes governing equations, thermodynamic/kinematic constraints, and microstructural symmetries to deliver physically consistent predictions with calibrated uncertainty. We couple this predictor to generative models (VAEs/GANs/diffusion) augmented with physics-based priors, transfer learning, and active learning to propose novel, manufacturable microstructures constrained by feasible process windows. To enrich learning when data are limited, natural language processing mines and structures knowledge from the scientific literature into context embeddings and a materials knowledge graph that condition both prediction and generation. To address model opacity, we develop an explanation layer that integrates attribution analysis, counterfactuals, sensitivity to physics constraints, and language-based rationales, yielding human-understandable, mechanism-grounded explanations that improve trust and decision-making. The end-to-end system will be trained and validated through multiscale simulation and targeted experiments using laser-based manufacturing, closing the loop with Bayesian optimization for sample-efficient data acquisition. Expected outcomes include: a reproducible platform for physics-consistent, explainable microstructure design; experimentally validated microstructures with improved performance; and a generalizable AI co-pilot that integrates knowledge, simulation, and experimental data to accelerate materials discovery and deployment in data-scarce settings. 

Start Date

April 1, 2026

Postdoc Qualifications

PhD in mechanical engineering, chemical engineering, aerospace engineering and realted fields. Exposure to machine learning and solid mechanics courses during PhD. Published machine learning articles is a plus. 

Co-advisors

Vikas Tomar, AAE, Purdue
https://engineering.purdue.edu/AAE/people/ptProfile?resource_id=55699

Gary Cheng, IE, Purdue
https://engineering.purdue.edu/IE/people/ptProfile?resource_id=34940

Bibliography