Abstract
The future landscape of ‘Smart Manufacturing’ is poised to revolutionize the industry with several groundbreaking features. Central to this transformation will be the seamless integration of shared data, models, and tools; the simplification of distributed technology and user interfaces; and the extensive use of generative AI models, autonomous robotics, and digital simulation and control for manufacturing processes. This paradigm shift aims to enhance production line monitoring, management, and optimization, particularly in terms of energy productivity and cost. Despite the advancements in additive manufacturing, existing practices fall short of being 'smart' and are hindered by notable constraints, including a limited range of printable materials, the frequent occurrence of defects, the variability of mechanical and surface properties, and the challenge of isolated equipment coupled with inconsistent data. Addressing these challenges, this talk introduces the concept of the multiphysics-resolved digital twin, which integrates multiphysics modeling across scales with data-driven learning. This advanced digital twin can self-update by reconciling discrepancies between predictions and actual data, enabling model predictive control over the physical manufacturing system. The speaker will leverage their interdisciplinary expertise in mechanics, manufacturing, and artificial intelligence to discuss multiphase multiphysics modeling of keyhole dynamics in metal additive manufacturing, revealing universal scaling laws and dimensionless numbers. The presentation will showcase newly developed computational methods for predicting crystalline grain structure and recrystallization in laser powder bed fusion and friction stir processing. Moreover, the talk will highlight the development of machine learning-based surrogate models, such as physics-embedded graph network (PEGN) and composable machine learning predictions. These innovative models significantly accelerate physics-based simulations by orders of magnitude while maintaining rigorous physical fidelity. Integrated into a JAX GPU-accelerated framework, these innovations have led to the creation of an open-source, differentiable simulation toolbox for additive manufacturing. This toolbox is designed to tackle both forward and inverse problems, marking a significant step towards real-time control and design of manufacturing processes and microstructures. Concluding the presentation, the speaker will outline ongoing projects and future research directions, including funding pursuits, to further advance the field of Smart Manufacturing.
About the Speaker
Zhengtao Gan completed a Ph.D. in Mechanics with national honors from the Chinese Academy of Sciences in 2017. Following graduation, Dr. Gan secured a Postdoctoral Fellow at Northwestern University and was quickly elevated to the positions of Research Associate in 2019 and Research Assistant Professor in 2020. By August 2022, Dr. Gan had transitioned into the role of Tenure-Track Assistant Professor at the University of Texas at El Paso. Dr. Gan’s research expertise lies in multiphase multiphysics modeling of complex mesoscopic systems, machine-learning-based design for advanced manufacturing processes, and thermo-hydromechanical-chemical processes in enhanced geothermal systems (EGS). Dr. Gan's achievements have been recognized with a solid track record of federal funding from DOE, NSF, and PNNL, alongside prestigious awards, including the Top Performer Award from the Air Force Research Laboratory (AFRL) in 2020, and 1st Place Awards at the National Institute of Standards and Technology (NIST) AM-Bench in both 2018 and 2022.
Host: Professor Iwona Jasiuk
