When designing a new 3D component, researchers normally consider how well its material composition responds to heat and stresses, or how its design impacts airflow, or aerodynamic performance. Validating of a complex part can take between 2 and 5 years.
Striving to speed up things, researchers from ORNL and GE believe they can cut in half the timeline to create, test and validate new 3D part designs. “Using multi-physics enabled tools and AI, we think we can beat the timeline for some traditional manufacturing processes by automating the entire process, said Brent Brunell, leader of GE Research’s Additive efforts.
Surrogate models
Optimization of structural characteristics has already been automated but has not extended to a part’s thermal and fluid properties. On this project, researchers from GE and PARC will use AI to automatically generate surrogate models from additive producibility data and seamlessly integrate it with multi-physics design optimization techniques.
The Summit supercomputer at ORNL will create these AI-based surrogates to analyze additively manufactured components and generate data necessary for training and evaluating the new models. Built using machine learning, surrogate models can guide the optimization models in feasible regions of very high dimensional design spaces.
"This is the type of project that leverages the unique capabilities at ORNL – experimental and computational facilities – as well as expertise in computational science and additive manufacturing," said John Turner, Computational Engineering Program Director at ORNL.
The program will culminate in the demonstration of a defect-free, high-performance additively manufactured multi-functional design capable of withstanding high temperatures and stresses with improved performance vs. conventional casting.