This paper describes an automated system to address disconnect between the design and analysis programs. The disconnect between computer-aided design (CAD) and analysis occurs because one system is powered by geometrical information and the other is not. One system is meant to handle geometry, while the other creates a mesh and does not rely on geometry in the same way that CAD does. Engineering software developers coming up with a design-analysis workaround by building design automation into their CAD packages. This method gives engineers vital information about the part from the get-go. The information is gleaned from the past designs for similar parts and can be automatically included in the new model. Such automated systems capture expert knowledge and reapply it to other models. The system might ask engineers to answer several questions about what they intend to design before they even start. The automated system would help engineers build parts upfront that would meet criteria for phenomena from varied disciplines, including fluid, structural, or thermal analysis.
You would imagine that Sandia National Laboratories would employ a number of expert analysts to study every aspect of potential designs, and you'd imagine right. Sandia, in Albuquerque, N.M., is responsible for national-security hardware and software, like space and nuclear-protection technology. Officials there obviously have a stake in ensuring that Sandia's wares are as safe as possible.
But what you might not know and what top Sandia officials didn't even know until they studied the issue is that analysts at the lab spend around 75 percent of their time cleaning up geometry after it's been imported from computer-aided design to analysis software. Cleanup is necessary so the model can be properly meshed for analysis, said Ted Blacker, the laboratory's department manager for computational modeling sciences.
"Analysts are into physics and modeling phenomena, and they want to spend a far lesser proportion of their time sitting in front of the computer messing with geometry," Blacker said.
Not to mention that the money Sandia pays high-end analysts to tinker with imported geometry could be much better spent elsewhere.
Blacker knows how much· time analysts spend readying the job at hand because he leads the lab's DART project, which stands for "design through analysis realization team." The lab instituted it more than three years ago in an attempt to resolve some of the software issues that tie up highly skilled analysts in less-skilled tasks. One of the main issues is the significant bumps in the road that analysts hit after they import the CAD geometry models into the lab's homegrown preprocessing package, Cubit. They translate the models into a neutral format (in this case, extensible markup language, or XML) that can be read by both the CAD system and preprocessor. Geometry must be brought into the preprocessing system, then cleaned up so the preprocessing software, which readies the part for finite element analysis, can mesh it properly.
The time that analysts spend cleaning up a model's geometry before they can properly study it is time wasted. However, at least for the time being, it's also time they must spend. The disconnect between CAD and analysis occurs, generally speaking, because one system is powered by geometrical information and the other isn't, Blacker said. One system is meant to handle geometry, while the other creates a mesh and doesn't rely on geometry in the same way that CAD does, Blacker said. The lab has come up with the XML language, which can be read by both systems, as an import workaround.
Other companies turn to neutral file formats like IGES or STEP for import. Right now, nearly every company relies on some type of workaround like this to get design and analysis communicating, according to Jim Rusk and Greg Brown, who work for the analysis software developers UGS and Abaqus, respectively.
Other companies turn to neutral file formats like IGES or STEP for import ~ Right now, nearly every company relies on some type of workaround like this to get design and analysis communicating, according to Jim Rusk and Greg Brown, who work for the analysis software developers UGS and Abaqus, respectively.
At Sandia, CAD information must be readied in a way that the preprocessor can easily handle, Blacker said. This is accomplished by model cleanup and defeaturing-that is, by taking out the CAD details that would affect the mesh. Holes and fillets must be filled.
According to Brett Clark, a member of Sandia's technical staff who works with Blacker, "Another issue is getting geometry transformed into something you can mesh. That means getting rid of small features, breaking it into pieces your algorithm can handle, applying the boundary conditions, designating the materials."
In an ideal world, the engineer would hand off a CAD model specifically cleaned up for preprocessing. "But, of course, this isn't an ideal world," Brown said. He's manager of the interactive product management group at Abaqus in Providence, R.I.
It may seem that Sandia designers could avoid these issues by using one of today's many top-shelf CAD systems with analysis capabilities .built right in. The systems have intelfaces that allow engineers to toggle back and forth, essentially analyzing as they design.
"But right now those CAD systems aren't adequate for the meshing we want to do," Clark said. "We've found those systems are inadequate to do any kind of quality analysis. If you want to do a quick elastic stress problem, you're probably fine, but for anything of significant complexity there can be a lot of questions."
And they often can't solve for the variety of constraints-boundary fluid flow, thermal, vibration, shock, and the like that analysts at the laboratory need to study, he added. Makers of advanced analysis systems include Abaqus Inc., Algor Inc., Ansys Inc., and MSC Software Corp., and several large companies and other organizations support their own in-house-developed applications, as Sandia does.
Until CAD vendors can catch up, companies deal with the creaky and inefficient handoff between CAD and analysis in different ways.
Bob Williams, product manager at Algor, recommends that engineers delete as many design features, like screws and bolts, as possible before handing their designs off to analysis. Those features can be reintroduced later. Algor's software can receive geometry directly from about 10 major CAD packages, with little need for translation, he said. The system also includes third-party software that fixes geometry flaws. Should analysis need more options, users can purchase add-on software for advanced cleanup.
Through Sandia's DART program, Blacker, Clark, and crew seek to get around the XML translation problem by linking their CAD programs to Sandia's analysis system via the design modeling engine. The laboratory mainly runs Pro/Engineer from PTC of Needham, Mass., and SolidWorks from SolidWorks Corp. of Concord, Mass. The link would cut translation time dramatically because the same modeler powers both CAD and analysis. This way engineers could carry out preliminary analysis as they design.
Now, the trick for Blacker and his team is to give engineers an easy way to work in Cubit, which is considered an advanced tool. One thought is to hide Cubit's complexity behind a relatively simple interface that engineers can easily learn and work with, Clark said.
A group of engineers began testing that in September, and rollout to all Sandia engineers is slated for the near future.
Of course, that doesn't mean analysts still won't spend time cleaning up geometry. An additional tool that DART will test to address that issue is something called virtual topography software, which defines a part broadly, not minutely, for analysis.
"Even if you link CAD and Cubit, it only solves the geometry issues," Clark said.
Getting rid of translation errors should reduce analysis cleanup right there, Blacker said. And when engineers turn to Cubit on a daily basis, cost savings should be significant.
Engineering software vendor UGS of Piano, Texas, gets around the translation problem by running the company's CAD and analysis programs in JT, the company's product visualization format. The format supports both geometric representation and visualization, so engineers using UGS NX software can pass models between analysis and design without translation, said Jim Rusk, UGS's vice president of digital simulation.
While these systems that toggle between design and analysis, mainly used by designers, are popular, they probably won't eliminate translation any time soon, Brown said.
"Those systems offer a thin layer of analysis that doesn't let you go deep," Brown said. "There's definitely a move to bring analysis up front and move into the same environment as design, but the vast amount of advanced analysis is still done by different people in different areas of the company, especially something sophisticated."
Rusk said the joint NX environment doesn't address the handoff between CAD and the kind of advanced software that specialists in analysis use. For that, he predicts that neutral file formats like STEP will persist into the future and will, in fact, be improved upon. He also thinks STEP will be the format of choice for importing information from one analysis system to another, which is yet another problem.
Software vendors realize that engineering companies seek a much tighter design and analysis marriage to reduce product cycle time, said Michel Vrinat, senior analyst at Collaborative Product Development Associates LLC, a business and ·technology consulting company in Stamford, Conn.
He predicts that vendors will find a way to merge design with high-end analysis programs within the next two years. Demand is simply too great to ignore, he said. Vrinat sees engineering software developers coming up with a design-analysis workaround by building design automation into their CAD packages. This method gives engineers vital information about the part from the getgo. That information is gleaned from past designs for similar parts and can be automatically included in the new model. Such automated systems capture expert knowledge and reapply it to other models, according to Vrinat.
The system might ask engineers to answer several questions about what they intend to design before they even start. They would also input design parameters. Based on those answers and numbers, the automation system generates a very rough, preliminary part that already meets desired parameters and necessary tolerances. It draws on past part information to generate that rough part. No need to recreate the wheel each time.
Engineers can then refine the preliminary design. Because designers know that the part already meets certain specifications they don't need to send it to analysis to learn that. The part will still have to be analyzed, of course, but several rounds of the back-and-forth engineer- analyst handoff can be saved. And with that, naturally, comes cost savings.
The automated system would help engineers build parts up front that would meet criteria for phenomena from varied disciplines, including fluid, structural, or thermal analysis, Vrinat said.
The Recycled Numbers
Parker Hannifin Corp.'s control systems division in Irvine, Calif., is already implementing an automated design process. The division makes actuators, which Brian Prasad describes as the muscles that control aircraft wings.
Although each actuator is different-they're custom designed-they're bare-bones similar enough to take advantage of automated design. The division calls its method knowledge design automation, or KDA for short.
The move toward the new method, which has been around for about three years, started because the technical staff was frustrated with the CAD-to-FEA process. The division's engineers and analysts translated CAD data into an open-format STEP model. That model still had to be simplified and defeatured before meshing and analysis, said Prasad, who leads the knowledge engineering team at the division.
The translation, meshing, and analysis steps were repeated again and again. The design changed according to the results of analysis and then went back to be reanalyzed, often to be changed yet again.
"So we thought, do we really need to create a new FEA model every time we have a few changes? Why couldn't we mine the knowledge from a number of our experts here and put them into some kind of automated fashion?" Prasad said. "Now, instead of engineers making the decisions, we let the program make some of those early-on decisions.
"If we had 10 actuators and did FEA on all 10, we'd spend a lot of time doing that," he added. "Instead, we'll take areas of the design that are more generic and do an analysis that should satisfy all 10."
Parker Hannifin engineers designed and analyzed in tandem on the company's Catia CAD system, but Jeff Rogers said that solution didn't make use of best-inclass FEA tools. Rogers is the division's technology team leader.
"We needed something more, which drove us to look at Ansys and Cosmos," he said. And to come up with the KDA system to ensure that the analyst software wasn't analyzing' the same design again and again.
With the KDA system, only when engineers have a design configured based on the sizing and performance constraints they enter does the design move on to analysis for complete verification, Rogers said.
"You can iterate so much more quickly, then drill into really specific things for FEA verification," he added.
And, for all companies that are competing in today's demanding product development environment, quick is where it's at.