Generative Models for Design and Process Intelligence

Abstract The increasing complexity of modern manufacturing systems calls for new paradigms that tightly integrate data with human expertise and scientific modeling. In this talk, I will present a few recent works that leverage generative models, including diffusion and flow-matching models, to advance design and process intelligence. I will first introduce Coupled Flow Matching, which employs optimal transport and flow-based generative modeling to enable controllable dimension reduction and design. Next, I will discuss the active learning framework enabled by generative diffusion surrogates, which merge prior knowledge with experimental measurements to guide experiment design. I will also introduce our work on diffusion calibration models that combine imperfect computer simulations with real-world measurements. These efforts highlight how generative models can serve as a unifying language to connect data, physics, and humans in the future of manufacturing intelligence.

Bio
Naichen is an assistant professor in the Department of Mechanical Engineering
and the Department of Industrial Engineering and Management Sciences. His
research focuses on integrative data analytics and science-informed AI models.
Naichen received his Ph.D. from the University of Michigan. His work has been
recognized with several honors, including the INFORMS Data Mining Best
Paper Award and the Wilson Prize.

Abstract:

We provide an overview of a recent NASA University Leadership Initiative project entitled “Development of an Ecosystem for Qualification of Additive Manufacturing Processes and Materials in Aviation”. This work demonstrated the dependence of fatigue on defect structures in AM Ti-6Al-4V. Our current NASA Science & Technology Research Institute “Institute for Model-Based Qualification & Certification of Additive Manufacturing (IMQCAM)” is creating a numerical digital twin (DT) for predicting fatigue as a function of process parameters. The key result of a systematic variation across power-velocity space was that the defect number density was anti-correlated with fatigue life. The fatigue-based process window was much narrower compared to typical reports and explainable in terms of observed variations based on spatter rates and intermittent lack of melt pool overlap. We report on recent developments in modeling microstructure, texture and pores along with variations in heat treatment. We conclude that establishing a qualified materials process is feasible via an efficient survey of a limited domain of process space. Linking models together in the DT requires substantial effort in terms of data exchange. In terms of predictability of fatigue, the limited data suggests that the scatter in life is smallest in the long crack (Paris Law) regime, intermediate for short cracks and largest for crack nucleation. This provides an important background for Uncertainty Quantification (UQ), which is a major focus for the IMQCAM team and which has already revealed a significant sensitivity to compositional variations, for example.

Support from multiple agencies is gratefully acknowledged, including NASA, DOE/BES, DOE/NNSA, ONR, NSF, OEA, Commonwealth of Pennsylvania, and Ametek. I am indebted to my many collaborators.

Bio:

I have been a member of the faculty at Carnegie Mellon University since 1995. I am also the Co-Director of the NextManufacturing Center on additive manufacturing. Previously, I worked for the University of California at the Los Alamos National Laboratory. I spent nine years in management with four years as a Group Leader (and then Deputy Division Director) at Los Alamos, followed by five years as Department Head at CMU (up to 2000). I have been a Fellow of ASM since 1996, Fellow of the Institute of Physics (UK) since 2004 and Fellow of TMS since 2011. I received the Cyril Stanley Smith Award from TMS in 2014, was elected as Member of Honor by the French Metallurgical Society in 2015 and then became the US Steel Professor of Metallurgical Engineering and Materials Science in 2017. I received Cyril Stanley Smith Award from the International Conference on Recrystallization and Grain Growth in 2019 and the Hans Bunge Award from the Intl. Conf. on Textures of Materials (ICOTOM) in 2024. I was an International Francqui Professor (Belgium) in 2022 and I received the ASM Gold Medal and was promoted to University Professor, both in 2024. My research focuses on processing-microstructure-properties relationships with interests in additive manufacturing, the measurement and prediction of microstructural evolution, the relationship between microstructure and properties, especially three-dimensional effects, texture & anisotropy and the use of synchrotron x-rays.