In 1850, Ralph Waldo Emerson observed that in nature, "the highest simplicity of structure is produced, not by few elements, but by the highest complexity." A century and a half later, this seems especially true in the world of computer-aided design (CAD).
Consider iSight, a computationally complex software package made by Engineous Software Inc. By hiding all the complicated math, the program gives product designers a fast way to make improvements they wouldn't have discovered using yesterday's CAD systems. Developed two decades ago at General Electric (GE) Co.'s research division, iSight was initially snapped up by auto and aerospace companies after Engineous was spun off in 1994. The basic approach has since spread to other CAD systems, so even small manufacturers are exploiting the tools. Engineous Vice-President Mike Sheh shared his outlook on the future of design technology with Senior Writer Otis Port.
What's the key to this new approach to design?
Let's start with finite element analysis. With FEA, you carve up a design or computer model into myriad geometric shapes. The result is a mesh that looks a little like you've wrapped a fishnet around [the object]. But the mesh isn't just on the surface -- it permeates the model in three dimensions. The computer uses this grid to calculate how the model will respond to pressure, temperature, and other forces. But to simulate how a system will work in the real world, you need to do many FEA repetitions to determine which features can be changed to improve the design.
How does the designer decide what to change?
That's where design of experiments, or DOE, comes in. This is a sophisticated statistical technique that can identify the most important variables, and it does this without running through the common process of changing only one variable for each repetition. For all but the simplest designs, that would take too long even on a powerful computer. DOE greatly streamlines the process of systematically evaluating many, many alternatives.
DOE actually revises the original design?
No, it identifies critical variables, such as the thickness or stiffness of the material at a given location. Then the designer can specify the range of variation that's permitted. Other mathematical techniques juggle those variables to try to find a better, more balanced solution.
Someday, will an engineer be able to feed in a list of functions, hit a button, and get a finished design?
It's already possible. But the designer must start with a very clear list of requirements and allowable ranges of changes for all the various parameters. For the 2000 America's Cup race, Italy's Team Prada used iSight and ran 30,000 simulations over a six-month period. They were able to automatically optimize part of the keel design. But the inputs included very specific factors -- not only the strict regulations concerning boat design but also sea and wind behaviors and engineering knowhow.
What about automating more mundane projects?
Some promising "CAD-less" morphing techniques are being tested by early adopters. You start with a free-form FEA mesh plus some general functional guidelines. Optimization tools explore many alternatives, and the mesh-morphing software turns out different designs that can be analyzed, using FEA, by computer-aided engineering software such as Abaqus [from Abaqus Inc. in Pawtucket, R.I.]. So far, this morphing technology is restricted mainly to exploring different shapes, but it should be possible to include other design variables.
To improve the manufacturability of a design, can these systems link up to shop-floor equipment?
We're beginning to see that. One client is including assembly procedures during the design phase. Other companies are creating designs that speed up production or reduce factory waste. But probably no more than 15% [of designers] are using even basic FEA and DOE tools. So advanced levels of design integration and automation won't be commonplace for several more years.