A new system of artificial intelligence reduces the time of preparation for complex space maneuvers 100 times - from months to seconds. The development concerns the calculation of the jet effect of engines: phenomena, which can damage the device's own structures during the operation of traction systems.
When a rocket engine turns on in a vacuum, the gas instantly expands in all directions — this creates complex thermal and mechanical loads on the structure of the device. To predict such effects, engineers traditionally ran millions of simulations, and this work took months.
The new system differs from conventional computing accelerators. It is based on NVIDIA PhysicsNeMo, a framework for developing AI models, combining physical laws with simulation data. It embeds the laws of physics directly into the model architecture. Because of this, AI does more than just process data, and understands natural laws and acts within their limits. Flexcompute president and co-founder Vera Young said, that they are transforming the role of simulation—not just accelerating it, and expand the range of tasks, which engineers can confidently solve.
A separate advantage of the system is a built-in assessment of its own accuracy. For each forecast, the model calculates a probability level, which allows engineers to understand the reliability limits of the response. This is extremely important in real conditions, because when docking in orbit or performing maneuvers of space robotics, an error of a few centimeters can put the mission in jeopardy.
Acceleration of calculations opens wider engineering possibilities. More accurate modeling allows you to design structures with a smaller safety margin, And therefore, and with less mass. At the same time, a more detailed calculation of the interaction of jets makes it possible to save fuel and increase the duration of missions. Director of AI at Northrop Grumman Fahad Khan noted, that the partnership with Flexcompute and NVIDIA makes it possible to speed up the implementation of projects, which customers need.
Source: https://universemagazine.com