This article highlights that a National Institute of Science and Technology (NIST) studies the problems in exchanging CAD models that makes at least $1 billion in profits from the US automotive industry every year. CAD has long been familiar to most engineers and designers. CAD systems design, model, and maintain engineering representations. The resulting models may describe something as simple as a screw or beam, or as complex as an automobile or aircraft. Companies now test and revise designs before building even a single physical mockup. CAD developers are discovering a demand for more interoperable programs. Private developers are rushing into the breach with a variety of programs that enable users to begin managing translation and healing. It is certainly a step in the right direction—and one that may eventually allow companies and their collaborators, customers, and suppliers to work together to design, analyze, simulate, test, manufacture, and assemble products electronically before anyone bends the first piece of metal.


Any solution carries within it the seeds of new and often flamboyantly baroque problems. No one knows this better than engineers. Many spend their days (and evenings) fighting fires ignited by both quick fixes and systematic solutions to old problems.

Nowhere is that more true than in computer-aided design. According to a National Institute of Science and Technology study, problems in exchanging CAD models pinch at least $1 billion in profits from the U.S. automotive industry every year.

An inquisitive caller asked a spokesman at NIST how that could be. After all, so many systems on the market are touted for their openness. “Around here,” he replied, “that’s what we call ‘the many flavors of open.’ ”

CAD has long been familiar to most engineers and designers. CAD systems design, model, and maintain engineering representations. The resulting models may describe something as simple as a screw or beam, or as complex as an automobile or aircraft.

Its early proponents claimed that CAD would revolutionize manufacturing and, to a large extent, it has. Early two-dimensional CAD drawings quickly replaced blueprints and schematics. They made it easy to verify part dimensions, and proved much easier to alter as changes filtered through the design process.

The revolution began in earnest with the introduction of three-dimensional CAD files. They have evolved from the glowing wire frames of the 1980s to the realistic 3-D solid representations of today.

The transformation involves capabilities as well as looks. Modern CAD systems allow users to explode views to see how parts fit together. They enable companies to create huge libraries of reusable part and component models that they can reconfigure and plug into new designs.

CAD systems have evolved to reflect how designers really work. Many high-end systems help manage collaboration. They enable teams at different sites to work on a project without getting in one another’s way. They even track design revisions and changes to ensure that everyone is working on the same current design. This is a blessing indeed for any engineer working on complex products like automobiles or airplanes.

Alan S. Brown is a writer based in Dayton, N.J.

Even more important for saving time and money, engineers can use 3-D CAD representations to run simulations. They can check a component or subassembly for mechanical and thermodynamic performance. Or check for fit, finish, and ergonomics. Or pull its components apart, then simulate the assembly process. Or simulate machining. Or use a model to create a mold or die.

As the new century dawns, electronic 3-D CAD models have begun replacing physical models in numerous engineering operations. Companies now test and revise designs before building even a single physical mockup. They send and receive CAD models from suppliers, ensuring consistency and accuracy. In a frenetic, hypercompetitive world, CAD saves big bucks while squeezing time out of product development cycles.


That’s the vision. Unfortunately, a huge barrier stands in the way of its fulfillment: CAD models are notoriously difficult to exchange. Engineers rarely exchange models without problems cropping up. Fixing them, when the errors are found, may prove as simple as a few clicks of the mouse or as complex as rebuilding most of the model from scratch. Sometimes, problems do not even show up until a part has moved into early production runs.

The March 1999 NIST study, Interoperability Cost Analysis of the U.S. Automotive Supply Chain, found all sorts of problems with CAD file interoperability.

Some of the most common are: missing, collapsed, or inverted faces; surfaces and edges that fail to connect; improperly oriented features; lines that do not meet at corners; lines that cross at corners; curves or lines drawn as multiple short segments; multiple occurrences of the same figure at the same location; lines or surfaces coincident with other lines or surfaces; surfaces that do not meet at lines; untranslated geometries; incorrect layering of geometry, dimensions, and notes; out-of-plane planar features; geometrical features not drawn to scale.

The result adds up to $1 billion in unnecessary expenses for the automobile industry alone, said Smita B. Brunnermeier and Sheila A. Martin of the Research Triangle Institute’s Center for Economics Research in North Carolina, who wrote the report for NIST.

“The majority of these costs,” they wrote, “are attributable to the time and resources spent correcting and recreating data files that are not usable by those receiving them.”

One automaker interviewed by the researchers estimated that as many as half of all CAD files received are not constructed properly for such downstream uses as rapid prototyping, finite element analysis, or computer numerical control programming. “These estimates,” they add portentously, “are conservative because they don’t include elements of cost our industry contacts could not quantify”


Even models created within the same company using the same software often have problems, according to a survey released by Prescient Technologies Inc., a Boston-based developer of data quality-control software.

In a recent survey, Prescient conducted data quality audits involving 3,000 separate computer models compiled from aerospace, automotive, consumer products, and electronics manufacturers. Companies in the survey ranged from multinationals with 160,000 employees to smaller firms with as few as 100. Prescient took approximately two weeks to complete each company’s audit.

Of the 3,000 models it analyzed, only 225 fully met defined model quality standards of the company creating the model. The standards included industry and company conventions and best practices. They ranged from straightforward naming conventions to interoperability guidelines that ensure downstream usability of the model.

According to Prescient, a full 70 percent of the models failed standards that the companies described as critical. The numbers are not all that different from those seen by Robert Rothacher, an Oceanside, Calif., consultant. As the former director of engineering operations for the Space Shuttle at Boeing Space Systems, he remembers experiencing about a 60 percent reject rate at the release desk.

“As I traveled around in my consulting role, I found that number was not uncommon,” he said. “Some reasons for rejection were ticky-tacky, like forgetting to put the right number in the corner of a drawing. Others were more serious. But all caused iteration and delay in cycle time.”

According to Bruce L. Jenkins, a vice president at Cambridge, Mass., consultant Daratech Inc., “Our research among auto companies shows it’s very costly to reuse designs from one car program to another. Moving CAD data downstream to tooling and production engineers, especially between OEMs and suppliers, incurs lots of costs.”

The lack of interoperability solutions creates additional expenses. Automakers increasingly outsource design and manufacture. Before the first component is ever produced, they want vendors to give them CAD models to simulate assembly, coordinate measurement inspection, robotic handling, and all the other details of modern high-tech manufacture.

Because no one really trusts file translations, automakers have asked Tier 1 suppliers to use the same CAD systems on which the factories have standardized. Naturally, each of Detroit’s Big Three has picked a different CAD standard. That leaves many vendors, who have their own CAD standards, with one or more expensive CAD workstations devoted to a single customer. “I heard one supplier refer to his new CAD station as his $50,000 modem,” said Jenkins. “It’s not uncommon.”

High reject rates, expensive iterations, and wasted assets are not surprising, given the CAD world’s diversity. When it comes to CAD, said Bob Bean, president and CEO of system developer CADKey Corp. of Marlboro, Mass., it’s not a Coke and Pepsi market. The market is diverse and complex.

There are five major high-end systems and many more midrange products. At the high end of the market, for example, Dassault Systèmes’ CATIA is the system of choice at DaimlerChrysler and most large aerospace manufacturers. Ford thinks I-Deas by SDRC of Milford, Ohio, is a better idea, while General Motors prefers Unigraphics from Unigraphics Solutions in St. Louis. BMW likes Pro/Engineer, marketed by Parametric Technology Corp. of Waltham, Mass.

CATIA has a strong foothold in Japan, where it competes with systems developed by Toyota and various Japanese software companies. Consumer product companies typically use both high-end systems and a myriad of midrange systems.

The problem is exacerbated by the waves of mergers, acquisitions, spinoffs, and asset swaps that swept through global industry during the 1990s. Even companies determined to standardize on a single CAD system often found themselves with islands of dissimilar software.

Each CAD system has its own unique file format and feature set. Even different releases of a single CAD program have different features. No wonder so many things can go wrong when CAD files switch hands. They often show up as problems with geometry (lines, curves) and topology (how the surfaces connect to form a closed boundary).

Some obstacles have proven relatively easy to overcome. Take approximations, for example. Every CAD system uses them. Since two surfaces cannot occupy the same space at the same time, CAD systems must set a threshold below which the parts are considered touching. One system might recognize joined parts when they are 0.001 inch away, while another sees a space until the gap measures 0.0001 inch. Connected parts on the first system appear separated on the second.

CAD developers have begun to provide solutions to the problem, allowing users to set and reset design approximations. Yet it was not an easy victory. “For a while, no one was aware of what caused the problem,” said Simon Frechette, a leading CAD researcher at NIST.

It’s difficult to determine what causes other geometric problems because CAD system developers are extremely secretive about sharing information on how their systems operate. “The way CAD developers create, edit, and manipulate geometry, the speed and elegance of their systems is a competitive advantage;” said Daratech’s Jenkins. “They keep their algorithms hidden and secret.”

Proprietary systems allow vendors to keep both their secrets and their markets. Large companies have invested millions in software and significantly more in training and the creation of reusable models. “In the eyes of most vendors, keeping their systems proprietary is the strongest way of maintaining account control," Jenkins said. “They’re in no rush to open their systems.”

The secrecy has made it extremely difficult to tell what causes a particular problem. Translating one type of CAD file into another requires reverse engineering of the highest order, especially since, as Jenkins pointed out, not all CAD systems are created equal.


“There’s no direct entity mapping from one file to another,” Jenkins explained. “Features may not have direct correlates in other systems. Then two things happen. First, some of the data in system A has no direct correlate in system B, so it winds up getting attached as a note. Second, system B may need information to construct a model, and it doesn’t get it.”

The problem is common with systems that use parametric data. Often called metadata by computer scientists, parametric information describes the relationship of the geometric data with the rest of the CAD model.

Take a through-hole, for example. Most CAD systems provide easy ways to define it. Topologically, a through-hole is a cylinder that goes from one side of the panel to the other. Parametric data defines the hole as anchored on both surfaces of the panel.

If a designer increases the width of the panel, the CAD system looks at the parametric data to determine how the hole should behave. In this case, it stretches the hole to accommodate the thicker part. A CAD system that does not understand this parametric data simply ignores the hole. A machining program looking at this new model creates instructions to drill only partway through the panel.

Other issues arise from ambiguity in design intent. “These are severe problems,” said Doug Chaney of International TechneGroup Inc., a CAD interoperability software developer in Milford, Ohio. “Their presence in a model creates an ambiguous condition. Everyone who uses this model will wonder whether the design was intentional or not.”

Chaney points to a typical problem, a thin space between a slightly angled surface and a vertical boss next to it. “When the engineer created the model, he simply used the program to extrude a boss,” Chaney said. “It went straight up at a 90 degree angle. The side wall, however, has a slight draft angle required by design.” The boss and the neighboring surface were supposed to remain touching.

“The designer didn’t account for the fact that the two features will pull apart,” Chaney said. “But how do the downstream users know that?”

Chaney asked, “Is this bad design? Most of these severe problems are unintentional. It’s humanly impossible to catch all of them. They’re made by designers working under pressure with complex geometries, using CAD systems that say nothing about this condition because they assume you know what you’re doing.”

Even when CAD designers know what they’re doing, they are prone to problems caused by geometric, parametric, and design intent errors. They’re subject to human errors as well. No wonder many observers believe that NIST’s estimate of $1 billion in lost interoperability costs may be low.

CAD interoperability is a big, complex problem. But help is on the way. One approach, favored by NIST, has been industry adoption of data exchange standards. Many CAD systems now allow users to import and export information in neutral file formats.

The first standard was IGES, scornfully called “I-Guess” by one long-time user. Originally developed for 2-D CAD drawings, it addresses communications issues but not model accuracy or approximations. “It puts up the wires, but doesn’t necessarily have the phones,” the user explained.

According to Thomas R. Chase, an interoperability researcher and associate professor at the University of Minnesota in Minneapolis, “Because people were good at exchanging drawings, they described it as doing CAD data exchange. Now we’re more serious about trying to exchange real CAD data models. In general, we still can’t take something defined using the high-end features of Pro/ Engineer and pull it up in high-end features of 1-Deas.”

An emerging standard, STEP, brings designers one step closer. STEP is actually a family of standards for exchange of electronic engineering data that range from product data management and finite element analysis to configuration management and CNC machining. CAD was one of the first STEP standards to emerge.

Unlike IGES, STEP was built to handle solid models. According to NIST’s Frechette, it supports the resolution of approximation-based problems.

“There are other data desirable to capture and transfer,” he said. “They include engineering analysis data, finite element data, and stress analysis information. If we want to test a model with a load on it, we need to transfer that information and get away from redrawing our models.” Ultimately, the STEP standard could accommodate that information.

Still, he admits STEP lags behind today’s CAD systems. “Eight to 10 years ago, its concepts were ahead of CAD systems,” he said. “CAD is now ahead of the standards.”


That has not kept private companies from forwarding varied interoperability solutions. Their diverse approaches bring to mind the story of the blind men and the elephant. The one who touched the tusks found the beast cold and smooth, while the one who skimmed the skin found it warm and bumpy. In the case of CAD, how companies view the problem defines the type of solution they offer.

Prescient Technologies, for example, starts with what it calls data quality. “The reason data interchange has so many problems is not because software is inadequate, but because the data they’re working on has a lot of problems,” said Gavin Finn, the company’s president and CEO. “CAD tools are so rich, there are many different ways to implement a feature even within the same system. That builds inconsistency into the system.

“Let’s say I want to represent a hole through a plate,” he offered. “In a Boolean operation world, I might take a plate and subtract a cylinder from it. I can also do it by using the program’s CAD hole feature to create a hole with a specific diameter and specify its width and depth.

“If you look at a screen or drawing, they both look the same. But when you examine the underlying structure of the geometry, they are fundamentally different. A program that automatically generates drilling instructions wouldn’t necessarily recognize a Boolean hole.”

Sometimes, errors occur because people simply make mistakes. They may get a feature wrong, omit part of a design, or simply put data in the wrong place.

The result is a file that requires reworking before it can be used downstream. According to Finn, Prescient helps companies assess downstream user needs. Its DesignQA software automatically scans designs to see that they comply with a company’s standards for machining tolerances, feature use, and data exchange.

Does it work? Rothacher, who now consults for Prescient, was one of the earliest users of DesignQA while at Boeing. Although he retired too early to quantify results, he estimates savings of approximately $2.3 million on the 7,000 models Boeing released annually.

“I looked at some preliminary numbers at Boeing Commercial Airplanes in Seattle that were that big,” he recalled. “Boeing Wichita came to some substantial numbers in that range, too. These savings have been borne out in other places as weh. I’ve never seen any analysis that didn’t produce 100 percent returns on investment after four to six months. It’s amazing. The numbers are actually so huge, folks are skeptical when they probably shouldn’t be.”

Prescient also makes software that analyzes CAD model problems and ranks them by occurrence so managers can resolve them through training and improved focus.

At International TechneGroup Inc., Chaney sees design intent as the key issue. Often designers include ambiguities in their designs. They range from the tiny gap caused by a surface angling away from a boss to inadvertent voids created by Boolean operations. Whatever their cause, they create models that downstream users and software have trouble interpreting.

Like DesignQA, ITI’s CAD/IQ looks over the designer’s shoulder and asks pointed questions about difficult-to-machine gaps, voids, uneven surfaces, improperly joined sections, and other unusual features.

Any developer could build those features into its CAD system, said Chaney, but they don’t because buyers, at least until now, are not interested in downstream usability.

“The feedback we’ve gotten from CAD vendors in the past and even now to some extent is that when CAD systems are pitted in benchmarks, they are judged on speed, robust performance, flexibility, and interface usability,” he said. “Notice that no one is concerned about whether anyone downstream can use the model it creates.

“Imagine this,” Chaney continued. “Two CAD systems are in a benchmark test. The second runs slower because it questions your intent in order to produce a better model. The first appears better, while the second would be perceived as slow.”

Even ITI’s CAD/IQ does not run all the time. That would significantly sap system speed. Instead, customers must deliberately invoke it when they want to use it. In case they don’t, ITI sells another program, CAD/Fix, which acts as a Band-Aid for severe problems. It lets users patch problems that have already escaped.

Spatial Inc. of Boulder, Colo., has a program that translates and heals CAD files. It checks models for correct geometric information, makes sure parametric information moves from one file format to another, and keeps track of such nonrepresentational information as notes covering product identification, configuration and assembly information, and materials of construction. “A lot of models are leaky,” said company president Bruce Morgan. “Our goal is to make a leaky model watertight.”

What sets Spatial’s offering apart is not so much what it does but how it does it. It delivers fee-based translation and healing services over the Internet.

The company’s service is part of a general movement of engineering services to the Internet, a development that is the focus of “Online Design,” on page 68.

Users log onto the site, indicate their translation and healing preferences, and submit a 3-D model. Morgan said an average 5-megabyte CAD file takes about five minutes to translate and heal. When processing is complete, Spatial e-mails the customer to download the model. Customers also receive a comprehensive report of repair activities.

Users benefit from Internet delivery of transaction-based services, said Spatial’s vice president of interoperability and corporate fellow, Doug Hakala. The site makes it simple to upload models. Users do not have to stay connected while their files are translated. Most important of all, especially for the small and medium-sized firms Spatial hopes to serve, customers do not have to invest in multiple CAD software, translation programs, or workstations.

On the other hand, technical benefits accrue because all software and hardware stay in Spatial’s hands. Most importantly, Spatial is able to upgrade its software continuously.

One of the big problems in translation is that there’s an awful lot of art as well as science,” Morgan said.

Morgan 1S realistic about the service. “We’re not claiming it works 100 percent of the time, or that it works 100 percent on every model,” he said. “An awful lot of the time, it works great and all the time it works somewhat. If customers don’t like the result, they don’t pay for it.”

The site opened for business in October 1999, offering translation and healing of models based on IGES and SAT (Spatial’s 3-D CAD engine format). It added capabilities for STEP and beta translators for CATIA and Pro/Engineer in December. Additional CAD formats will follow in 2000.

CADKey is one of several vendors that sell CAD systems based on Spatial’s ACIS engine. An engine, in this case, is the core algorithms that model the design.

CADKey has combined some of Spatial’s translation and healing powers with its own proprietary technology. Bean, the company’s president, argues that the resulting system can import STEP files from CATIA, Pro/Engineer, and other high-end software, and manipulate them as if they were native files because they are all reduced to geometry.

STEP filters out intelligence in high-end CAD files, especially parametric data, and translates files as pure geometry and topology. Pure geometry is CADKey’s native format.


“If someone exports a solid as a STEP file from Pro/Engineer, we can operate on that solid though we use different editing techniques in CADKey than those found in the original editing tool,” Bean said. “CADKey doesn’t know if you imported or created it. You can manipulate it as if it were created natively.”

Bean thinks interoperability is a way to differentiate CADKey from other companies vying for the desktops of small and medium-size manufacturers.

“We’re intended to be used by the masses,” he said. “Our users are Tier 3 guys. They can’t afford to buy all these systems from companies they do business with. With us, they can read any STEP or IGES file and work on it.”

CAD developers are discovering a demand for more interoperable programs. Private developers are rushing into the breach with a variety of programs that enable users to begin managing translation and healing. Stan-dards-setting bodies are looking to include more parametric data in STEP.

It’s certainly a step in the right direction—and one that may eventually allow companies and their collaborators, customers, and suppliers to work together to design, analyze, simulate, test, manufacture, and assemble products electronically before anyone bends the first piece of metal. This is, after all, the promise of CAD.