MICROSURGEONBOT INC — Department of Defense SBIR Phase II: AF203-DCSO3

Description

It’s not easy to “eCreate before you aviate.” High-performance multiphysics simulation tools must be called up and orchestrated. For example, to build a digital twin of a jet engine, scientists and engineers will run turbomachinery aerodynamics computational fluid dynamics (CFD), structural analysis (FEA), rotor dynamics simulation, combustion simulations, and more, simultaneously. And because simulation is specialized, it's hard to own the tech stack in this arena. One of the best turbomachinery aerodynamics packages is FINE/Turbo from Numeca. Top structural analysis comes from Ansys FEA. The best rotordynamics software is made by Dyrobes. A preferred package for combustor CFD is Siemens’ Star-CCM+. It is also common to include custom / research codes. A variety of file formats exist to exchange data from one code to the next including: Parasolid, IGES, STL, CGNS, and many, many more. In other words, digital twins usually call up a multitude of packages that must speak to each other. It’s complicated. Understanding, setting up, running, and post-processing such simulations is so specialized that each of the types of phenomena listed above can be considered a PhD-level specialty. For this reason, the Air Force needs a new way to stitch together simulation codes — and a way to allow generalist engineers to run them, alongside the current cadre of elite specialists. Plus everyone involved would benefit if manual, arduous, repetitive tasks in this process could be automated — both in terms of setup time savings, as well as in a reduction in potential for error that can result from such a complex multistep process — thereby improving reliability. MSBAI has a solution. Our cognitive AI assistant for engineering, GURU, minimizes the human workload needed to translate engineering questions into computational workflows (Fig. 1a). For more than a decade, the founders performed a series of high fidelity multiphysics simulation-driven design exploration/evolution/optimizations used to invent new technologies in aviation, power generation, and space launch; and deployed them on some of the world’s most powerful supercomputers at the Oak Ridge National Laboratory Leadership Computing Facility (OLCF) and DoD’s HPCMP facilities (Ref’s y-ff). We developed novel methodologies that enable GURU to learn procedures to set up comprehensive workflows. Critically, GURU can learn from sparse datasets. It can run entire end-to-end workflows autonomously — including even the configuration of new Docker containers, and deployment to Kubernetes clusters. The user can communicate their goal to GURU simply by voice or through a visual interface, and GURU calls up, populates, and orchestrates the appropriate simulation software.