Widespread adoption of computer-aided design and analysis tools has not eliminated the need for prototyping. CAE has allowed engineers to produce more complex designs that in many ways are more heavily dependent on prototyping to validate new design concepts. However, computer simulation of product components and systems is gaining in popularity, largely in the preliminary desing stages, due to the versatility of computer models. A computer model can be examined in simulated working conditions, something that cannot be done in clay and wood. Furthermore, computerized models can usually be constructed for less cost and in a fraction of the time needed for physical models.

Some companies are increasingly looking to computer modeling as a way to reduce the costs and time delay ssociated with prototyping. Combined with comprehensive materials data bases, computerized models can accurately predict the performance of a proposed design. Tge day when virtual prototypes eliminate the need for preproduction models is near.

From Design to Production

When Kohler Co. first implemented finite element analysis eight years ago, the Kohler, Wis.-based manufacturer of small gasoline engines spent two years of its four-year product-development cycle disigning and testing protoypes. Two prototype iterations generally were required: the first to identify flaws in the initial design, and the second to prove that those flaws had been corrected. Only then did a new design go into production. Kohler characterizes itself as a small player in a large field. In order to compete, the company decided that its product development cycle had to be shortened. Engineering decided that the best way to do this was to eliminate prototyping.

Today, the company begins tooling up for a new engine once computer analysis validates the design. Kohler does not build prototype engines anymore. Todd Gerhardt, Kohler's manager of advanced engineering systems, said the construction of prototypes has been made unnecessary engineers. This has shaved an average of two years off the product-development cycle.

"The name of the game is getting products out the door faster," Gerhardt said. "Physical prototyping is like doing the job twice. Even eight years ago it was pretty clear that computer simulation of engines was the way to go."

Kohler performs stress, thermal, and vibration analysis on components and on engines as complete systems. Kohler's CAE architecture is fairly straightforward. Solids models are created using Unigraphics from Electronic Data Systems Corp. (Dallas). Since the Unigraphics solids models are not transferable to the FEA systems, analysts remodel them for meshing. Finite element models are created using Patran from PDA Engineering (Costa Mesa, Calif.) and I-Deas from Structural Dynamics Research Corp. (Milford, Ohio). Analyses are performed in ansys from Swanson Analysis Systems Inc. (Houston, Pa.) and post processed in I-Deas. Gerhardt said it requires about two and a half months to develop an FE model.

While FEA is hardly uncommon in modern manufacturing circles, Kohler places great reliance on FEA results. Physical prototyping was not phased out lightly. Eight years of documenting FEA results and correlating them with field tests were necessary before protypying could be dispensed with. When prototypes would fail, FEA would be used to determine why they failed. In many cases, physical tests were conducted in concert with computer simulation in order to develop more accurate computer models. For example, Kohler used measurements from dynamic strain gages in running engines as boundary conditions for stress analyses in computer engine models.

Kohler is now experimenting with generating boundary conditions from computer analysis. Engineers are using the DADS package from CADSi (Oakdale, Iowa) to perform rigid and flexible body dynamics analysis on engines. The results of these analyses show all of the forces that act upon the engine points and are used to derive better boundary conditions for FEA.

Gerhardt said that in certian cirsumstances, virtual prototyping produces more accurate results than physical testing. The engines Kohler produces are made from die cast aluminum. The prototypes had been made by plaster casing as die casting of prototypes was not economical. Occasionally the plaster cast prototype would pass all of the required tests while the die cast production model would fail.

Understanding Materials

One of the requirements for developing virtual prototypes that behave like the real thing is to compile accurate data about the materials that the production version is to be made from. This can be easy or quite difficult, depending on the material and the conditions in which it is expected to perform. Determining the fatigue point of steel at room temperature is no problem. However, determining the fatigue point of aluminum at 400[degrees]F is another matter. "Building a material data base is a long process," Gerhardt said. "It's just a matter of testing to verify over the years."

At Sandia National Laboratories, (Albuquerque, N.M.), where the mechanical components of nuclear weapons are designed and assembled, researchers have compiled data bases that allow them to show how materials used in weapons' production respond to the manufacturing process. Now, as the federal government is encouraging national labs to transfer advanced manufacturing technology to the private sector, Sandia is working with Pratt & Whitney (East Hartford, Conn.) to develop a coupled thermal and mechanical FEA code and materials data bases to model welding of thin metal engine parts.

The joint research and development effort, which is scheduled to run for two to three years, is aimed at developing manufacturing techniques that reduce the amount of warpage occurring when thin-skinned parts are welded together. Results of finite element thermal analysis of welded joints will be used as input for stress analyses. Through this feedback loop, researchers hope to model the welding process and avoid trial-and-error prototype welding.

John Biffle, manager of Sandia's computational mechanics and visualization division, said the key to accurate computer analysis is understanding the materials used in man ufacturing the desired product. It takes a long time to develop and verify dependable data bases, he said. Moreover, a data base that is useful for one manufacturing process may not have broad applicability outside that product line. "It took several years for us to determine the properties of two kinds of glass," Biffle said. "And there are probably umpteen kinds of glass in the world."

Sandia has compiled data bases for foams, solder, plastics, glass, exotic steels, and electrical systems used in the manufacture of nuclear weapons. The data bases are used to conduct design studies of various aspects of weapons production. Libraries have been formed cataloging glass solidification, solder-joint reliability, welding of metal parts, metal cutting, the performance of electrical components in harsh environments, and other manufacturing issues related to nuclear weapons.

There are certain aspects of virtual prototyping of nuclear weapons that are applicable to commerical manufacturing. Because electrical components in military and industrial products are similar, Biffle said, there is a high degree of commercial transferability here. Even if specific materials data bases used by Sandia cannot be packaged for industry, its approach to research and development could serve as an example. "After all, nukes have nuts, bolts, and soldering just like most manufactured products," Biffle said.

Look, Don't Touch

Perhaps the main drawback of the virtual approach to prototyping is that there is nothing to touch. This deficiency is most plain in modeling a product's interface to the human element. "This is the downfall of the CAE image," said Andrew Sant. in, president of Santin Engineering (Beverly, Mass.), a company that specializes in computer and fabricated modeling of products from concept through production. "Will the new jug slip out of your hand when it is wet? It is difficult to tell from an image on a screen." Santin said that while virtual prototyping can take the place of many conceptual and presentation models, especially in light of the photo-realistic rendering packages now available, it cannot fully supersede fabricated prototyping in every application. "You are never going to know how the new car door sounds until you actually close it."

One of the projects Santin Engineering has been engaged in since the company was rounded in 1954 has been to model new jet engines for General Electric Corp. Santin builds full-scale models complete with all components, subassemblies, and wire harnesses. While it requires six months to fabricate such an elaborate model, GE is not prepared to go into production without seeing the nonworking prototype first. "We have to take the component images on the screen and prove that they do actually fit together by building a physical model," Santin said.

The objective of virtual prototyping, Santin maintains, should be to eliminate unnecessary physical protoryping prior to assembling the models that are actually required. Computer prototyping can be used to show the thought processes that occur during the different stages of product development. Santin uses the Acis solids modeling system in the Concept Station package from Aries Technology Inc. (Lowell, Mass.) to build appearance models.

Software systems designed to model the human interface are beginning to appear. In June, Altia Inc. (Colorado Springs, Colo.) announced its Avid Design human interface software for simulating instrument and control systems. The system replicates the look and behavior of dials, meters, sliders, buttons, CRT displays, strip charts, and custom components of instrument front panels. The user can link external software for data feeds and modeling of component behavior in simulations. Because the model control panels created in Avid Design can be programmed to function like actual instrumentation, they are designed to take the place of physical prototypes.

The Computer Graphics Research Laboratory at the University of Pennsylvania (Philadelphia) is making its Jack human-factors modeling software available commercially. Jack has been used to simulate seat positioning for earth movers by John Deere Co. (Moline, Ill.). Engineers at John Deere evaluated various design proposals to see which best satisfies operator visibility requirements without forcing the company to build physical prototypes. The software has also been used to evaluate systems designed for NASA's Space Station and for the AH-64 Apache attack helicopter. Jack's virtual person has 88 articulated joints, a complete vertebral torso, and correct shoulder-clavicle joints. Jack can be used with a number of CAD and rendering systems.

Virtual Mechanical Systems

Fatigue testing of mechanical systems was one of the first areas where computer analysis was applied to reduce prototyping costs. According to Angelo Assimakopolous, associate engineer at Schwinn Bicycle Co. (Chicago), performing repeated fatigue tests on fabricated prototypes is an expensive and time-consuming affair. Prototype bicycle configurations are welded together, tested to destruction, and then another prototype is built. "You can't build the next prototype unless you know where the last one failed," Assimakopolous said.

While this sort of physical prototyping is still pursued at Schwinn, computer modeling is being brought to bear on a number of newer projects. Using the Applied Motion package from Rasna Corp. (San Jose, Calif.), Assimakopolous has been performing sensitivity studies related to the redesign of Schwinn's Aerodyne exercise bicycle. Assimakopolous indicated that computer modeling has uncovered flaws in proposed Aerodyne designs that were not intuitive and would have been difficult to spot from physical tests.

One of these problems was where to place a reinforcement spar. The redesign of the 15-year-old Aerodyne included restyling the body, which changed the allocation of stress on the frame. The beam that supports the rider is connected to the main horizontal beam at the halfway point and slopes toward the rear of the cycle. Under load, the seat beam tends to bend, so reinforcing spars that connect it to the rear of the main beam, forming a triangle, had to be added. As the twin beams were mounted closer to the seat, the amount of stress on the beams decreased. Three Rasna models had to be constructed and 15 tests ran on each before the optimum placement for the beams was found. The normal Schwinn method of physically building and testing three prototypes would have taken nearly a month and would not have located the exact point of minimum stress, according to the company.

Advanced Simulations Edge Out Hard Prototypes

Certain processes cannot readily be understood through the testing of fabricated prototypes. This is especially true of nonlinear phenomena like fluid mechanics. While wind tunnels and water tanks have long been used to test prototypes, new generations of vehicles are being designed to operate in conditions that are impossible to recreate in the laboratory. Designers of hypersonic aircraft, advanced-technology helicopters, and modern submarines all rely on supercomputers to create virtual prototypes rather than physical models.

Peyman Givi, associate professor of mechanical and aerospace engineering at the State University of New York at Buffalo, is using supercomputers to model turbulence in fluids to better understand turbulent combustion. Givi said that it is harder to extract information from actual experiments than from computer simulations. Conditions like homogeneous and isotropic flow are difficult to control and reproduce in the laboratory. Mathematical models of fluid flow are more reliable. Furthermore, the numerical data behind the computer models are easier to record than having to take physical measurements of actual flow.

"In practice, it is hard to fabricate a model that will reveal the fluctuations and variations of fuel mixing with air in an engine," Givi said. The Computational Fluid Dynamics Laboratory at the Buffalo university has access to Cray Y-MP and Cray 2 supercomputers through NASA and the National Science Foundation.

This is not to suggest that physical experimentation can be divorced from the product development cycle. Givi said that a third numerical component has to be inserted into the process alongside theory and experimentation. Supercomputer validation of advanced design concepts is a way to reduce unnecessary physical modeling. However, data from hard models is good if .it is economically obtainable. "Computer simulations are cheaper than fabricating prototypes but do not replace laboratory data." Givi said.