Ayush Jain
Advisor: Prof. Rampi Ramprasad

will defend a doctoral thesis entitled,

Multiscale Machine Intelligence Tools to Accelerate Polymer Additive Manufacturing Design

On

Tuesday, December 2nd  at 1 p.m.
MRDC Room 3515

and Virtual via Teams: Link

Committee

Prof. Rampi Ramprasad (Advisor) – School of Materials Science and Engineering, School of Computational Science and Engineering

Prof. H Jerry Qi – School of Mechanical Engineering

Prof. Aaron Stebner – School of Materials Science and Engineering, School of Computational Science and Engineering

Prof. Victor Fung – School of Computational Science and Engineering

Dr. Ehsan Haghighat – Head of Machine Learning, C-infinity; Software Research Scientist, Carbon3D

Modern manufacturing is shifting toward mass customization and sustainable production of complex, multi-material structures. Additive manufacturing (AM), particularly polymeric 3D printing, offers a promising solution by building products layer by layer, allowing for reduced material waste and greater design freedom. However, the vast design space in AM poses a significant optimization challenge across material chemistries, processing conditions, and component design. My work addresses this challenge by developing a set of computational tools that use machine learning and materials informatics to accelerate AM design through three strategies. 

The first strategy is embedding intelligence into the physical lab. I develop an end-to-end thermoset optimization pipeline that integrates several data sources, informatics models, and optimization for discovery. My experimental collaborators and I validate this to find optimal material formulations for targeted properties. Second, I introduce a Physics-Enforced Neural Network (PENN) for linear polymer melt viscosities. The PENN predicts viscosity from chemistry while encoding the relationships of molecular weights, shear rates, and temperatures directly into the model. This strategy provides a pathway for chemical optimization across the physical spectrum and can be applied to other problems in polymer informatics that blend chemistry and known physics. Third, I address the need for faster simulations by developing graph neural network surrogates that accelerate computational data generation. On the molecular scale, I introduce a framework, polyGen, which establishes a generative method for polymer structure. At the component scale, I develop LatticeGraphNet, a graph neural operator that predicts the nonlinear compressive response of lattice structures. These surrogates learn the underlying physical behavior from simulations, allowing them to explore design spaces that would otherwise be computationally intractable.

Collectively, these tools lay a foundation for multi-scale informatics-driven AM design, paving the way for innovative, customized products in an advanced manufacturing ecosystem.