"It would be impossible to get anything useful by just making manual adjustments to the design program we use," says Peter Menegay, senior performance engineer at Dresser-Rand Co., Olean, N.Y. But an optimization program called Darwin from Phoenix Integration, Blacksburg, Va., generated several useful conclusions. It provides exploration techniques such as trade-off studies and probabilistic analysis. Another optimization module lets programs work as a team by calculating values others can use.
Menegay says Darwin is a genetic algorithm for optimization that helps explore a noisy design space. "By noisy I mean the outputs are discontinuous and jump all over the place instead of varying smoothly." Users interact with the software by suggesting values for high and low points. "Test data would also guide that input, and you'd tell the optimizer to minimize the difference between test data and the prediction," adds Menegay.
"We bought the optimization software because we wanted to improve our design efficiency," he says. Engineers there learned the software from online resources and needed no formal training, "though it was useful to have someone from the software developer point out its capability. And there are other programs with similar functions but they are just more expensive," he says.