Engineers Create AI-driven Material Modeling Program to Fight Fatigue Failure

Johns Hopkins engineers have developed a powerful computational modeling platform that promises to improve the performance, reliability, and durability of complex metals made through 3D printing. The new modeling system, developed by a team led by Somnath Ghosh, the Michael G. Callas Professor of Civil and Systems Engineering at the Whiting School of Engineering, could change how manufacturers design and test these advanced materials.

The team’s results appear in Nature Communications: https://engineering.jhu.edu/case/news/engineers-create-ai-driven-material-modeling-program-to-fight-fatigue-failure/

“Conventional tools for generating digital models aren’t able to capture the precise details in materials with complex structures,” Ghosh said. “Our new platform uses a specific machine learning method known as generative adversarial networks, GANs for short, due to its ability to create new, realistic data from raw experimental data.”

Additive manufacturing techniques, like cold spray forming (CSF) and laser powder-bed fusion (LPBF), offer unmatched design flexibility for producing new metallic materials but their complex structures leave them highly prone to fatigue failure. Ghosh’s team set out to optimize the design of these materials to improve their durability. However, the researchers quickly realized the limitations of existing modeling tools to achieve this, such as conventional statistical and image-based modeling methods.

Instead, Ghosh’s team combined integrated computational materials engineering (ICME) and multiscale modeling techniques with machine learning tools, like GANs—a type of machine-learning algorithm that learns from real data to generate realistic virtual structures—to overcome the limitations of conventional computational modeling tools.

The new platform provides users with high-quality, synthetic representations of complex, multiscale materials that can be used to more accurately predict a material’s form, behavior, and performance life during different stages in challenging environments. Material design can then be optimized according to these predictions and applied to a wide range of engineering applications, from aerospace to biomedical engineering, where durable, high-performance materials are essential.

“This was a major learning experience in terms of creating simulation tools for predicting life and reliability of additively manufactured materials. In addition to alloys, the platform can be used for other material classes and enables enhanced material design for superior performance and performance life. We believe that it will go a long way in material design and the discovery of new materials,” said Ghosh.

Study co-authors include Johns Hopkins University doctoral student Joshua Stickel, Los Alamos National Laboratory postdoctoral research associate and former Johns Hopkins University postdoctoral researcher Brayan Murgas, and Professor Luke Brewer from the Department of Materials Science & Engineering at the University of Alabama.

This work was supported by the Office of Naval Research.
engineering.jhu.edu/case/news/engineers-create-ai-driven-material-modeling-program-to-fight-fatigue-failure/