Back

 Industry News Details

 
GE Says It's Leveraging Artificial Intelligence To Cut Product Design Times In Half Posted on : Mar 06 - 2019

Artificial intelligence is helping computers drive cars, recognize faces in a crowd, and hold life-like conversations. General Electric engineers now say they've used the data-intensive technology to develop tools that could cut the industrial giant's design process for jet engines and power turbines in at least half, speeding up its next generation of products.

Today, it might take two days for engineers to run a computational analysis of the fluid dynamics of a single design for a turbine blade or an engine component. Scientists at General Electric’s research center in Niskayuna, New York, say they’ve leveraged machine learning to train a surrogate model so that it can evaluate a million different variations of a design in just 15 minutes.

“This is, we think, a huge breakthrough,” says Robert Zacharias, technology director of thermosciences at GE Research. It typically takes GE six months to a year to design a part or a new product. Zacharias says surrogate modeling could cut the design cycle time in half or more, and allow the company to do much more design work in a given period of time.

The researchers expect the technique to be put to widespread use within two to five years at GE, which is in the middle of a painful restructuring to narrow its focus to aero engines and power equipment.

Surrogate modeling has been done on a smaller scale for some time, and many manufacturers are working on creating what’s being called a “digital thread,” in which product designs are virtualized and made shareable across the business. But the scale of what GE appears to have achieved is impressive, says Karthik Duraisamy, an engineering professor at the University of Michigan who directs its center for data-driven computational physics. "GE is among the industry leaders, if not the leader in this area," he says.

One catch: Ultra-smart computer processes can't replicate the human element in product design yet. “Ideas emerge when groups of people with different expertise get together and talk about it,” Duraisamy says. “There is something about human intuition and input that is hard to formalize digitally.”

Computer modeling is a tradeoff between speed and accuracy. The highest-fidelity method currently known, direct numerical simulation, would require years of run time on the world’s most powerful computer system to evaluate the aerodynamics of an aircraft wing. A faster approach requires approximations that can reduce accuracy. GE’s AI surrogate model is an attempt to get the best of both worlds. View More