SEMTE researcher receives ASME/Boeing Structures and Materials Award for AI-driven aerospace materials research 

Since 1979, the ASME/Boeing Structures and Materials Award has been given to outstanding papers presented at the annual American Society of Mechanical Engineers’ Aerospace Structures, Structural Dynamics and Materials Conference (SSDM). 

Held in Long Beach, California, SSDM gathers top researchers and engineers from industry, academia and government agencies to convene topics related to the field of structural engineering. 

Yongming Liu, a professor and graduate program chair of mechanical and aerospace engineering in the School for Engineering of Matter, Transport and Energy, housed in the Ira A. Fulton Schools of Engineering at ASU, has received the ASME/Boeing Structures and Materials Award for research, advancing the use of Artificial Intelligence (AI) in aerospace materials engineering. 

The award recognizes Liu’s paper, “Probabilistic Physics-Guided Machine Learning with Missing Data: Applications in Additive Manufacturing,” co-authored with Virginia Tech assistant professor Jie Chen, an ASU alumnus and former doctoral student of Liu’s. ASME selected the paper for its originality and significance to the field of aerospace structures and materials. 

The paper addresses a growing challenge in engineering: using machine learning reliably in the absence of complete datasets. 

“In plain language, our research focuses on making machine learning more reliable for aerospace materials and structures when the available data are imperfect,” Liu says. 

Making machine learning work with incomplete data

Machine learning and AI have become increasingly powerful tools for materials and structural analysis, but many existing models depend on large, complete datasets. In aerospace materials research, however, experimental data are often collected across multiple publications and tests. As a result, important manufacturing parameters may be missing or inconsistent. 

To address this issue, Liu and his collaborators developed a probabilistic, physics-guided neural network that can learn from both complete and incomplete datasets, while incorporating established knowledge regarding material fatigue behavior. 

Fatigue refers to the gradual weakening of a material caused by repeated loading over time. Small cracks can form and grow under cyclic stresses, eventually leading to failure. Understanding and predicting fatigue life is critical to aerospace applications, where safety and performance are dependent on component reliability. 

The team’s work focused on predicting the fatigue life of additively manufactured Ti-6Al-4V, a titanium alloy widely used in aerospace applications. Additive manufacturing allows engineers to create complex components that would be difficult to produce through conventional manufacturing methods, making it an important technology for aerospace design.

Liu’s model predicted both the expected fatigue life and the uncertainty surrounding the prediction. This provides engineers with additional information for evaluating risk, reliability and material performance. 

“The model predicts not only an expected fatigue life, but also the uncertainty around that prediction,” Liu says. “This is important because fatigue behavior naturally has significant scatter, and reliable engineering decisions require an understanding of both performance and uncertainty.” 

The research also demonstrated that incomplete datasets can provide valuable information. Rather than discard records with missing parameters, Liu’s model learns from both complete and incomplete sources. The results showed an improvement in prediction reliability compared to relying solely on complete datasets.

Supporting safer aerospace design

A sensitivity analysis identified heat-treatment temperature and heat-treatment time as especially influential factors affecting fatigue. The model also identified other influential parameters, including scanning speed, laser power, hatch spacing and layer thickness, all of which affect the additive manufacturing process. These findings can help guide future research while optimizing processes for additively manufactured aerospace components. 

“What excites me most about this work is its potential to support safer and more efficient aerospace design,” Liu says. “Additive manufacturing gives engineers tremendous design freedom, but qualification and reliability remain major challenges. By combining physics-based understanding with machine learning, we can make better use of existing experimental data and develop tools that are more trustworthy for engineering applications.” 

Looking ahead, Liu plans to continue developing uncertainty-aware machine learning methods to help engineers make safer, data-informed decisions for advanced manufacturing and aerospace systems. His long-term goal is to bridge the gap between experimental data and computational modeling for advanced manufacturing and aerospace systems. 

“Receiving this award is a tremendous honor. It is especially meaningful because the award recognizes technical papers for originality and significance in aerospace structures and materials, which are exactly the areas this research aims to advance,” Liu says.