Design Software That Makes Complexity Look Simple
These days, most products are born on computer screens. But that doesn't mean they come even close to perfection. Ford Motor Co., for example, estimates that finding the best engineering design for a new car would mean hunting through more than a billion billion billion possible solutions. That would take centuries--even with all of the world's computers. So designers and engineers must content themselves with solutions they judge to be "good enough."
Management wants more. That's because the decisions made at the so-called fuzzy front end of product development can lock in 75% to 85% of subsequent costs in manufacturing and field support. So, for more than a decade, managers have been clamoring for better approaches to product development.
ONE OF A KIND. A group of computer scientists and artificial-intelligence experts formerly of the University of Delaware claim they've got the ticket. It's called Quantum Leap--a program that uses "adaptive optimization" to get closer to ideal design solutions. Developed by Quantum Development Corp. (QDC) in Claymont, Del., Quantum Leap recently solved a longstanding challenge by NASA's Langley Research Center to analyze the design for an airliner in one fell swoop, instead of picking away at separate pieces, such as the fuselage and engines. "Our software can also do a whole-car design synthesis. Nothing else can," boasts QDC President Joseph B. Elad.
Companies are reluctant to test that claim, mostly because it would entail a risky departure from the way cars and planes have always been developed. But another reason is that Quantum Leap is devilishly complicated, and some potential users are reluctant to rely on something they don't fully understand. That severely hampered sales when the software made its debut.
But companies are giving the program limited try-outs. Ford is putting it to work on a new engine. DuPont Co. is using it to solve a complex production-scheduling problem. And the software helped Texas Instruments Inc. optimize its capital-spending plans, saving at least 5% of a multibillion, five-year budget.
Because such benefits translate into significant competitive advantages, these companies are wary of divulging details. But Joseph D. Sparks, Ford's manager for competitive analysis and value management, says that every time Ford's designers have handed a problem to Quantum Leap, "it has produced a surprise"--a solution unexpectedly better than previous techniques. The results of these surprises won't show up for a couple of years. The first major application is a new engine for 2001 model-year pickup trucks.
At DuPont's Cape Fear plant in North Carolina, operations specialist Charles M. Lamberson says Quantum Leap "allowed us to do things we weren't able to do otherwise" to improve the efficiency of synthetic-fiber production. Making Dacron polyester fiber is more complicated than just melting plastic. It involves mixing polymers, colorants, and chemical additives, then maintaining tight controls on such conditions as extrusion rate and temperatures. Now that the software has proved itself for the past six months, Lamberson says, it is being groomed for other plants: "We put it in the most complicated environment first, so we're confident it will do a similar job in other situations."
What makes the front end of design projects so fuzzy is a plague of unquantifiable data. "Designers have to wrestle with soft market-research data that say consumers like this feature a little more than that one," says John C. Carter, head of Product Development Consulting Inc. in Menlo Park, Calif. "How do you put numbers on things like that?" Carter is impressed with Quantum Leap because it can deal with such fuzziness and then "explain why a choice is good."
BANDWIDTH GAP. "It's driven by the data," says Apperson H. Johnson, Quantum Development's chief technology officer and co-founder, "so the technology automatically adapts to the problem," guided by a supervisory program based on artificial intelligence (AI). This is important, says Carter, because both products and design-optimization software have grown so complicated that "almost nobody has the bandwidth" to understand them. So engineers rely on "rules of thumb and approximations--and settle for answers that are pretty good," says Karl T. Ulrich, an associate professor of operations management at the University of Pennsylvania's Wharton School.
The idea for Quantum Leap was sparked by published reports of research at Nissan Motor Co. and General Electric Co. A decade ago, Nissan, working with Ulrich, experimented with various design-optimization approaches. To limit the difficulty, the project considered only 19 performance parameters, such as fuel economy and braking distance. Otherwise, says Ulrich, "the mathematical complexity is just intractable." The research showed that better designs could be achieved by blending the results of a battery of programs, and Ulrich foresaw how AI could coordinate such a multifaceted attack.
Around the same time, GE hatched Engeneous for internal use. This design system is based on two AI techniques: an expert system and genetic algorithms. The expert system is stuffed with rules derived from successful designs of the past. The genetic algorithms tackle previously unexplored solutions by taking individual design ingredients--such as shape, material, manufacturability, and cost--and "breeding" new solutions. Engeneous was used to create a more efficient fan blade for the jet engines on Boeing Co.'s 777 plane.
Quantum Leap brings Ulrich's vision to life. It harnesses 30 different computer programs for dealing with complex problems. Each method competes with the others to find a better solution. The AI supervisor keeps tabs on all 30 programs and enables them to compare their progress and swap data--or even start over with some other program's data. The final answers are invariably combinations of several techniques, says Quantum's Elad. Several solutions can be displayed on a so-called radar graph that makes comparing them a snap (illustration).
TOO ADVANCED? The 30 optimization programs are among the tools developed for finding better shortcuts to "good-enough" answers. Each performs admirably at solving certain tasks, says Steven D. Eppinger, a member of Massachusetts Institute of Technology's Center for Innovation in Product Development. Combining those 30 tools inside an artificial-intelligence wrapper, Eppinger adds, was ingenious. He plans to run tests on the program this summer.
The program's main shortcoming, says Ulrich, is that it may be too advanced. "It requires a fairly sophisticated user just to appreciate what it can do," he explains. "I think it's going to be kind of a long sell."
QDC's Elad knows that only too well. Quantum Leap first came to market five years ago as shrink-wrapped software aimed at solving almost any kind of complex problem. Elad was certain it would be the next Lotus 1-2-3. But Wall Street analysts, money managers, and individual investors--the early targets--were leery of using "black-box" software whose inner workings were a mystery. "We wasted a lot of time trying to educate them," says Elad.
In 1995, QDC switched strategies. It went after large companies, offering to install Quantum Leap to tackle specific problems. That led to contracts with Bell Atlantic, Boeing, and Tyco International--and the first meager profits for the privately owned company. Last year, QDC chalked up margins in the 15% range, says Elad. He won't divulge revenues except to say they were under $10 million for 1997. Quantum Leap Innovations Inc., a consulting subsidiary formed in January, generated a 40% net in the first quarter. QDC's workforce has now grown to 25 people.
So if the former academics in Delaware stick with this strategy, Quantum Leap may finally live up to its name.