A federal energy research initiative is gaining momentum with the release of a contract award aimed at using supercomputing to harness 3D printing technology that would boost the performance of power generators.
Partners GE Research, PARC, the Xerox research arm, and Oak Ridge National Laboratory (ORNL) will investigate methods for reducing design timelines for printed components used in power systems ranging from wind to gas turbines. Under a $1.3 million contract awarded by the Advanced Research Projects Agency-Energy, (ARPA-E), the partners will use the national lab’s Summit supercomputer to develop machine learning frameworks.
The goal is to halve design timelines for 3D-manufactured components for turbomachinery applications by leveraging AI models to complete millions of design iterations. The partners will seek to reduce 3D components design times to as little as one year.
The HPC-driven research is part of an ARPA-E initiative called DIFFERENTIATE, mercifully short for Design Intelligence Fostering Formidable Energy Reduction (and) Enabling Novel Totally Impactful Advanced Technology Enhancements. The energy ML effort targets math optimization problems common to energy and other system design processes.
The newest effort takes that approach a step further by leveraging emerging 3D printing technology to hasten the design and deployment of new turbomachinery components. Among the design considerations for turbines are thermal and aerodynamic properties. Validating an individual component can take anywhere from two to five years.
“One of the keys to enabling the widespread use and benefits of 3D printing is the reduction of the time it takes to create and validate defect-free 3D component designs,” Brent Brunell, leader of GE Research’s additive manufacturing unit.
“Using multi-physics enabled tools and AI, we think we can beat the timeline for some traditional manufacturing processes by automating the entire process,” Brunell added.
The design of structural characteristics has been automated. Using ONRL’s Summit machine, GE and PARC will seek to extend design automation to thermal and fluid properties. The world’s fastest supercomputer will be used to generate precise “surrogates” built using machine learning. The researchers also will use the lab’s High Flux Isotope Reactor to analyze additively manufactured parts. The reactor will generate training data for AI models.
Among the milestones is demonstrating an accelerated generative design capability for defect-free, additively manufactured components with better thermal and structural performance than conventional component casting. That capability promises to dramatically reduce the time required to design and fabricate high-quality components for energy generation, the researchers said.
That’s in line with ARPA-E’s goals for DIFFERENTIATE, leveraging high-fidelity simulations to generate design details that can be used to improve components. Computational fluid dynamic tools can be used both to predict system performance while generating costly training data needed to improve machine learning models, the agency said.