A collaborative team of researchers led by Lehigh University is pioneering new artificial intelligence (AI) techniques to revolutionize materials science. Their project, titled “Harnessing Nonnegative Matrix Factorization for Advanced Computational Materials Modeling,” is supported by an $800,000 grant from the U.S. Department of Energy. This research focuses on developing advanced scientific machine learning (SciML) algorithms to help scientists analyze vast amounts of complex material data. These cutting-edge AI tools will accelerate the discovery of new materials with applications in energy, manufacturing, and beyond.
Driving this project is a challenge that has long puzzled scientists — how can we accurately predict the properties of materials before they are even created? Traditional methods rely on trial-and-error experiments, which can be costly and time-consuming. By combining mathematical modeling with AI-powered learning, this research aims to decode the fundamental relationships between a material’s structure and its properties, enabling scientists to design new materials with specific functionalities – such as stronger, lighter, or more energy-efficient compounds — entirely in the digital space.
The research is led by Chinedu Ekuma, assistant professor of physics at Lehigh, and brings together experts in machine learning, physics and materials science. The team is developing a new class of interpretable AI models, ensuring that scientists can understand and trust the decisions made by machine learning algorithms.
The project is built on four major innovations:
- Physics-Guided Machine Learning — Researchers are designing non-negative matrix factorization (NMF) models that incorporate real-world scientific principles, such as crystal symmetries and atomic interactions, to improve accuracy and interpretability in material predictions.
- Scalable AI for Predicting Material Properties — Using deep learning-powered algorithms, the team is creating scalable AI models that can analyze massive datasets from material experiments and simulations. These models will enhance predictive accuracy and guide experimentalists toward promising new materials.
- Integrating AI with Diffusion Models — The research will merge diffusion models (used in AI image generation) with materials science datasets, uncovering hidden relationships and identifying new material candidates for high-tech applications.
- Open-Source AI Tools for Global Use — To democratize access to these advancements, the team will develop an open-source computing platform that will allow scientists worldwide to run advanced AI models on their own data. This tool will be compatible with Windows, Linux, and Mac, enabling seamless deployment on cloud and high-performance computing systems.
By developing AI systems that are not only powerful but also explainable, this project marks a major step toward making AI a trusted tool in scientific discovery. The results of this research could accelerate innovations in material design, leading to advancements in:
- Next-generation semiconductors for energy-efficient computing,
- High-performance materials for aerospace and automotive industries,
- Breakthroughs in battery technology for renewable energy storage environmental changes, and
- Healthcare applications, including AI-driven drug discovery and precision medicine.
The research team is comprised of:
- Chinedu Ekuma, Lehigh University (Principal Investigator)
- Lifang He, Lehigh University (Co-Investigator
- Akwum Onwunta, Lehigh University (Co-Investigator)
- Bao Wang, University of Utah (Co-Investigator)