Analysis for Design Engineers

[Procision Screenshot]

Are design engineers doing analysis? Flip through any popular CAD publication and you'd certainly think so. In a recent issue of one engineering magazine, I found at least six references to the use of analysis software by design engineers, couched in phrases like "Integrated Analysis," "PushButton Analysis," "Moving Engineering Up Front," and "Integrated Analysis of Solid Models."

Of course, these terms could also be interpreted as injunctions to incorporate analysis, even if performed by specialists, into the design process. But there's no question that there's currently a big push to put analysis tools into the hands of design engineers. By giving designers the ability to evaluate function early in the product cycle, so the thinking goes, better products will result.

But there's a catch. Analysis is not easy. It's possible to get a wrong answer that looks correct to the untrained eye. (For the purposes of this article, unless otherwise noted, the term analysis refers to finite-element analysis.) Until recently, FEA has been delegated to full-time analysis specialists with advanced degrees. Is this something design engineers, who have little or no training in this arena, should be doing?

"For the last 30 years or so, traditional FEA software has been used successfully only by experts," says Dan Smith, executive vice president, Rand Technologies. "You have to make a lot of assumptions and interpreting results is difficult. So many things can go wrong, even with the right user, and good code. And a wrong answer from a finite element system is probably worse than no answer, because it gives the engineer a false sense of security about his design."

Since Rand Technologies has recently begun to market an analysis product that is not based on finite elements, Smith's comments might be suspect if they weren't backed up by people in the FEA business. Says Dick Russell, director of analysis products at SDRC, "You have to understand about the method to make good decisions and get good results. There is still some judgment involved."

[MARC Screenshots]

Lou Crain, president and CEO of MARC Analysis Research Corp., takes an even stronger position. "I'm not sure we're going to make the analysis process so automatic that, over time, designers do analysis. It may well be that the proper solution is for analysts to take a greater role in design," he says.

David Weisberg, publisher of the Engineering Automation Report, titled an article in his December 1996 issue "Merging Analysis With Design - Some Adult Supervision Required." In the article he wrote, "At EA Report, we believe that companies need to proceed with caution when integrating analysis with design."

Given all these caveats, why are we seeing such efforts to have designers do analysis? Some might see it as an effort on the part of analysis vendors to expand their markets. After all, there are many more design engineers than fulltime analysts. As Weisberg noted in his article, "Many of the software vendors see this as a great opportunity for selling substantially more copies of their analysis software - potentially tens or hundreds of thousand of copies, vs. hundreds or low thousands of copies."

Nevertheless, it does make sense to have designers perform analysis as part of the what-if, creative process. In a traditional scenario where only specialists perform analysis, the use of this resource is limited and analysis is used primarily to verify a final design. If designers could also perform this function, more parts could be evaluated, and analysis could be performed early enough in the process to have a guiding effect on the design. "Marginally safe components [could be] identified, and grossly over-designed parts [could be] quickly improved," wrote Weisberg.

The challenge, then, is to deliver goof-proof analysis programs that designers can use without fear of getting misleading results. Although many analysis vendors are making efforts in this regard, some of the most interesting recent developments come from a newcomer, Rand Technologies, and a long-established analysis vendor, ANSYS Inc.

* From Separate to Integrated

Some history might help put the recent innovations into perspective. The first step toward making analysis a practical adjunct to design came several years ago when solid modeling became popular. Prior to that, when CAD models were 2D or 3D wireframes, they didn't contain enough information for an analysis.

"An analysis model needs the kind of information provided by a solid model so it knows where material is and where it isn't," explains SDRC's Russell. "It needs more than just a physical representation of the part. In fact, some of the original motivation for SDRC getting into solid modeling was to make it possible to build analysis models more productivity he adds.

When solid modeling took off, analysis vendors established links between their products and the design process by enabling the analysis programs to import CAD geometry. In addition to supporting neutral file formats such as IGES, many vendors developed direct translators that convert popular CAD file formats to their own formats. These efforts facilitated the earlier use of analysis by reducing the time needed to create the analysis model, typically a large portion of the analysis process.

The next step, which has been occurring over the last few years, has brought an even higher level of integration between analysis and CAD. Rather than just importing CAD geometry, some analysis programs now run inside the CAD system itself so that a user doesn't leave the CAD environment to perform an analysis. Boundary conditions and loads are applied directly to the CAD model.

[Matra Screenshot]

Some examples of this level of integration include Structural Analysis Research Corp.'s Cosmos/Edge, which operates within SolidWorks; MacNeal-Schwendler's (MSC) GFEM FEA embedded within EDS/Unigraphics and MSC Nastran embedded within the company's Aries CAD system; versions of ANSYS that run inside Pro/Engineer, CADDS 5, and Autodesk's Mechanical Desktop; Parametric Technology's incorporation of Mechanica into Pro/Engineer and Matra Datavision's Euclid Analyst, which runs within its modeling system, Euclid Quantum.

These efforts, however, are hardly revolutionary. While they integrate two types of software that formerly operated separately, they don't do anything to make analysis more suitable for design engineers. According to Scott Owens, vice president of marketing at ANSYS, analysis programs embedded into CAD systems are suitable for only a small segment of the design community. "There's some truth to the criticism about giving design engineers these tools, that they don't know how to use them and don't know how to interpret results," Owens says. "One in 10 designers would know how to use [a traditional FEA program embedded in a CAD program],".

The more difficult challenge is presenting the power of analysis technology in a way that permits its use by less-trained people. The first company to attempt this, in the early 1990s, was Rasna Corp., which has since been purchased by Parametric Technology (PTC). (Actually, Dr. K. Nematollahi, president and CEO of MCAE Inc., claims to have been "preaching this concept for 12 years," so perhaps he should have the honor of being called the first. But MCAE's marketing program has been nowhere near as aggressive as Rasna's, which is generally credited with leading this charge.)

Rasna marketed Mechanica as a designer's tool with much success, but when it came to actually delivering the technology, its results were mixed. "Rasna made a lot of progress in establishing awareness of the benefits [of analysis by design engineers], and some progress in actually delivering the technology and getting it into use," says Bruce Jenkins, vice president, Daratech Inc. (Cambridge, MA). "But it appears that a fair amount [of Mechanica licenses] are being used in the more traditional way."

As was to be expected, Rasna's acquisition by PTC promoted tighter integration between Mechanica and Pro/Engineer, while de-emphasizing its integration with other CAD packages. As for PTC's efforts to increase the usefulness to designers of Pro/Mechanica, as it is now called, Brian Shepherd, manager of functional simulation applications at PTC, says, "We're careful not to go feature-happy. We try to take a balanced approach to the functionality we deliver, so we provide something people can get accurate answers from as opposed to a mine field that lets people get wrong answers."

One company that has taken an innovative approach to making analysis suitable for designers is Pratt & Whitney, a MacNeal-Schwendler customer. Its approach has been to have its analysts write programs for designers to use. Through the Patran programming language, analysts customize the system for particular design problems, such as vibration loads on jet aircraft tubing. This complicated problem would be difficult for designers to handle on their own, but the custom program lets them enter well-defined design variables such as the stiffness and position of brackets. The program uses terminology familiar to designers and tells them, accurately, how variations in design parameters affect overall performance. "There's a lot of power in having a professional analyst create the environment for design engineers to work in," says Kelly Miller, product manager at MSC.

MARC has taken a similar approach - the difference being that MARC, rather than the end users, is doing the customizing. The first example of MARC's efforts in this arena is a program called MARC/AutoForge. It's used for the analysis of bulk metal forming processes, such as forging, extrusion, rolling, and multi-stage forming. "We're taking a tack we call verticalization," explains president Lou Crain. "We're addressing the very complicated problem of bulk metal forming, but we've made the product easy to use by tailoring the menus, algorithms, and error messages so they're intuitive to a designer in that industry."

Crain says the company will create a similar program for rubber analysis. "We'll take our general purpose capability and make a vertical product to support the rubber industry," he explains. "It's not just taking FEA and throwing it into a CAD system and saying,"Here, do your rubber analysis. That won't work. But you can take sophisticated tools and narrow the focus so you can make it intuitive to designers in that particular field."

* Replacing FEA

Rand Technologies is looking at the issue of designers doing analysis from a different perspective: if FEA is the problem, get rid of FEA. If FEA is too difficult and untrained users might get incorrect answers, give them a tool that isn't based on FEA and does a better job of conveying the validity of the results.

Rand Technologies is actually an odd choice for a company to be involved in this issue. The company is a VAR - a big one - that sells Pro/Engineer and other software from more than 50 offices around the world. Being so big has made the company a magnet for new technology inventors who want access to Rand's customer base, which is how Rand came to be involved in the analysis business. A few years ago, Rand was approached by researchers at the University of Minsk in Belarus who had developed a new mathematical approach to analysis that wasn't based on traditional H-, p-, or boundary elements.

"Their real specialty was in crack analysis and propagation, and they came up with new mathematical approaches to equation solving," explains Rand's Smith. "When analyzing a crack, you look at global characteristics of the parts but you also want extreme precision in very small detail. Solving that problem gave them a new approach that they opened up to typical analysis problems."

These researchers' work, which began in 1981, is now a product from Rand Technologies called Procision. The company describes it as "a new analysis system that requires no mesh and is completely integrated within the Pro/ENGINEER environment." Interestingly, while Procision incorporates a number of features designed to reduce the possibility of error and to make it clear to users how accurate their results actually are, Rand isn't promoting Procision as the ultimate analysis solution for design engineers - yet.

"Right now, we're not marketing Procision as exclusively or primarily a designer's tool," says Smith. "We believe we have the technology that can become a very good designer's tool, but we'll probably have to go through one or two more versions to get there. We're trying to be responsible about it because a lot of people look at the software and say, "This is fantastic. I want to give this to all my designers." We have to hold them back."

According to Smith, it's the issue of error that the company wants to address more thoroughly before promoting Procision as suitable for designers. "Procision already uses some different approaches to calculate a more exact representation of the error in a solution," Smith explains. "Procision will give you the best answer it can, but it also gives a quantitative feeling as to how good that answer is. The results include a contour plot of error, so you could see that on some sections of a surface, for instance, you might have a 15% error, in the stress results but everywhere else is right on. As long as it's right on in the area you're worried about, you're happy. At the end of February, we'll have a version that uses error calculations to modify the model and run it again to drive that error down. When we get to that point, it will be a system we can give to design engineers with a lot of confidence."

At the moment, the only CAD system that Procision supports is Pro/Engineer, which raises an interesting question: how does PTC feel Procision stacks up against its own Pro/Mechanica product? Well, the official word is that PTC isn't viewing it as competition yet. "What we see that's impressive about Procision today is its speed," says PTC's Shepherd. "But it is difficult to use because answers are dependent on how you split the model. You don't need a mesh, but you do need to break complex models down into smaller pieces. That is not intuitive. Also, users need to understand in advance where the problems are, as it says on the Rand Web site."

Shepherd is referring to this statement, found under the Procision button at www.rand.com: "Using Procision, engineers simply select features which they feel may create stress concentrations, and the system will automatically calculate very accurate stresses due to the local effects of these geometric features." In Shepherd's opinion, it is not always safe to assume that users know in advance where high stress concentrations will be.

* New Approach From ANSYS

[ANSYS Screenshot]

While it might be OK for an upstart like Rand to scrap FEA, you wouldn't expect such an approach from an established FEA vendor like ANSYS. And, indeed, it hasn't done that. But the company is developing an entirely new product line geared toward design engineers, with some products based on FEA and others built on other technologies where appropriate. ANSYS calls this development effort DesignSpace, and it has created a new division to support it. The first product built on DesignSpace technology is ANSYS/AutoFEA 3D-Validation, an engineering tool embedded in the AutoCAD and Autodesk Mechanical Desktop environments.

Says ANSYS' Scott Owens, " Designspace is a designer's tool. It just happens to be the ANSYS solver under there today [in ANSYS/AutoFEA 3D-Validation]. But in the future, there might be a spreadsheet underneath a cost estimating program. DesignSpace is an umbrella that will encompass a lot of different technologies."

ANSYS/AutoFEA 3D-validation differs significantly from the traditional version of ANSYS, called AutoFEA, which is embedded within AutoCAD. The differences are what Owens thinks will make ANSYS/AutoFEA 3D-Validation appeal to all design engineers, not just the one in 10 that might be able to use traditional ANSYS.

For instance, ANSYS/AutoFEA 3D-Validation is a subset of ANSYS limited to linear static structural and modal analysis. More importantly to the user it incorporates a good deal of intelligence for helping set up a problem, uses familiar vocabulary, and makes some of the more difficult aspects of analysis more intuitive. For example, a slider bar is used to indicate a user's preference for a quick solution or a more accurate one. The software accommodates the request by varying the density of the finite-element mesh, but the user doesn't have to know this or even look at the mesh.

Results come back in a way that ANSYS thinks will be easy to understand. "Our normal analysis results are a pretty color-coded picture that corresponds to values such as stress levels or thermal conditions," says Owens. "Ph.D.s can interpret these, but the everyday designer might have trouble. ANSYS/Auto FEA 3D-Validation gives answers in stop-light colors [red, yellow, and green]. A designer may be told 'if it's green, just pass it along. If it's yellow, you need to annotate some things so the next person knows what you were trying to do. And if it's red, go find yourself an analyst.'"

ANSYS sees ANSYS/AutoFEA 3D-Validation as more of a computerized Roark than an analysis program. It is supposed to help designers explore more options and improve the quality of their designs - a nice fit with the current push to "move analysis up front" that we're hearing so much about. This effort by ANSYS, along with those of Rand Technologies and other vendors who keep the needs of the design community in mind, should go a long way toward adding the study of function to the issues of form and fit that comprise a design.

This article is from Computer Graphics World.