New Hardware Design Software

An anonymous reader writes "AScribe is reporting that mechanical engineers from Purdue University have developed software they claim will increase the efficiency of creating a wide range of industrial parts. From the article: 'The new approach integrates the design and analysis processes, which are now carried out separately. Currently, the geometry of a part is first created using computer-aided design, or CAD, software. This geometry is then converted into a mesh of simple shapes, such as triangles or rectangles, which, when analyzed using a computer, indicates the part's strength and other characteristics. The painstaking procedure, called finite-element analysis, is extensively used in industry.'"

Comments

Making a 2 stage process into a 1 stage process

[BadAnalogyGuy]

The gist of it is that when one polygon in a mesh is changed, that the entire mesh does not need to be redesigned. The updated polygon only affects the polygons it is connected to, so on the fly analysis of the design can be performed without the need for a separate analyst.

It's kind of appropriate that a Chinese guy and an Indian guy are the ones coming up with ways to eliminate jobs.

Re:Making a 2 stage process into a 1 stage process

[riprjak]

You can only do away with the analyst if the CAD operator is competent to correctly interpret the results.

The problem with simulation is just that, it is *not* real world performance. Gaining accurate and useful data from FEA or CFD requires competent, skilled and highly trained operators (yes, all 3). Otherwise all you get are pretty coloured pictures.

In line analysis tools built into CAD for near real time analysis of models is nothing new. Its been around for the better part of a decade.

The difference here seems to be swapping your traditional mesh for an interlinked series of geometric primatives; essentially trading one form of approximation for another.

More power to them if this is indeed faster and equally accurate; but I doubt it will be, most of the difficulty these days is not mesh generation (damn near automatic) or Finite Element mathematics (sparse matrix solvers on modern CPUs make a meal of such calculations)... the time comes from the hideously complex multi-physics equations themselves; and these must remain common for equivalent accuracy.

Either way, the press release was typical oversold hype, but the tool SEEMS like it could be useful in low end design descisions.

err!
jak.

Re:Making a 2 stage process into a 1 stage process

[QuantumG]

Because that would require IT to actually do something.

hardware design

[doubleshot]

The title makes reference to hardware design, in most of us in the technological world this seems like it might be a new PSpice or the likes. However this is a mechanical engineering software program, not a hardware design program. Agh, I got all excited for a second.

It's NURBS for stress analysis. Here's the paper.

[Animats]

Here's the technical paper: Constructive solid analysis: a hierarchical, geometry-based meshless analysis procedure for integrated design and analysis. This extends finite element analysis from rectangular cells to elements defined by NURBS surfaces. Difficult mathematically, but if you set up the problem that way, fewer cells and less number-crunching is required. Very nice.

This may have applications for soft-body physics in games.

Re:Nothing new...maybe

[Overzeetop]

The press release sounds like it was written by somoeone who has never worked in an integrated production environment. Pro/E has had the end-to-end capability for close to a decade (though at $50k a seat). I didn't like (or didn't trust) the meshers in the 90s, and solid elements also didn't always behave properly, especially at boundries, so I did most of the work by hand. Parts of the writeup seem to suggest that the FEM created would contain boundries that would remain static, and the re-meshing would only occur in defined areas, reducing the amount of time to create and (ideally) invert the matrix on the next iteration (sorry it been a long time since I did FEM, excuse the glossyness of that statement). If thats the case, you're really only talking savings in computer time, which for common manufactured parts is probably down in the noise, costwise - who cares if you have to solve a 1e4DOF model if you can do it in a few seconds? The abstract of the actual paper, linked in another post, seems to imply that they're not using FEM to solve the optimization, which would be revolutionary.

Re:Nothing new

[EndingPop]

This particular paper isn't new, and neither is design optimization. NX, Pro/E, ABAQUS, ANSYS, etc. are all just analysis tools. True design optimization, that is, the ability to find non-intuitive solutions to problems with literally hundreds of variables is a burgeoning market. This article is surprisingly sparse with details, but topology optimization is powerful in that it can bypass the mesh generation because it uses the same mesh with every analysis. It does several analyses (sometimes hundreds) and each time it changes the "density" of each element until you reach a predetermined weight reduction while maximizing stiffness. This is actually only a tool for preliminary design, since it only deals with strain-energy; it never even looks at stress. In fact, the solution is independent of the actual magnitude of the loads applied, only the location of the BCs.

I think I'm already doing this

[JoshWurzel]

I'm a mechanical engineer who uses Pro/Engineer at work. I design geometry and then use built-in software (Pro/Mechanica) to analyze the parts every day. How is this new?

At first it sounded like it was simultaneously computing something about parts as you design, but this just isn't possible or even useful as far as I can tell. The whole point of having a separate step is that you can define different loading scenarious. "What is the strength of this part?" is a meaningless question. What you want to know is "how much force will it take to displace the end of this beam by 2 inches?". There just isn't anything to compute on the fly.

So again, what is it that this is doing? The integration of FEA modules in CAD is already pretty seamless. It saves the analysis features as part features.

Reading the article, I see some interesting quotes:

"The way it is now, the same CAD software used to make the shape of the part can't be used to analyze the mesh" - flat out wrong. Pro/E, ANSYS, and NX all have integrated FEA modules. Its a separate module, true, but you just push a button in the GUI to change your mode from modelling to analysis.

"After the designer designs the object, it is thrown over to the analyst, and the analyst says, 'OK, I think, based on my analysis, that your design has to be modified this way,' and then throws it back to the designer, who makes the modification" - sorry, but this is not the way things should be designed. The company I work for is small, but in general there are only a few designers who have *extensive* knowledge of manufacturing capabilites and extensive experience. This enables them to make good judgements when designing parts. When they do make a mistake, the engineers have sufficient CAD skills to make the change themselves. There is very little "back and forth". Now, if the analysis indicates the design is completely non-viable (i.e. "Not strong enough now and never will be!") then the designer goes back to the drawing board. But that's an entire re-design and not nearly as similar to phone-tag as the article makes it sound.

And don't even get me started on the fact that its written in Java...it may be a fine language for web apps, but I've used FEMLAB and if someone else writes a CAD package in Java I may well be forced to hit them in the face with a trout.