PMI

Two weeks ago, Spatial hosted a booth at the CONTROL Exhibition in Stuttgart, Germany.  I hate to follow John's recent post with another one about a trade show, but this one is worth discussing - let's just call it "Interesting Shows - part 2."

All that these devices do is measure points in space  

Of course they do that with tremendous variety, which is how they differentiate themselves from each other. Differentiation can be on the accuracy of measurement, point gathering speed, physical access (e.g. you can't put the wing of an airplane in a bridge machine, so you use a hand held device), and much more.  But the one thing they have in common is that they're still all trying to do one basic thing - give you three very, very accurate coordinates, many, many times over.  

Obviously, Spatial is very involved in this part of the measurement market, particularly as more and more systems are moving to measuring and comparing to 3D parts rather than 2D drawings.  One thing in particular struck me throughout the show - almost every discussion I had turned to the subject of PMI (or GD&T) at some point.  There was a time not so long ago when using PMI in CMM applications was a new idea.  When we first added PMI to our 3D InterOp product line, we had many customers excited about it, but mostly in principle. Very few were actually doing anything with it.  Today the discussion is totally different.  We're seeing applications do everything from drive automatic test plan creation to automatic post-process comparison between the gathered points and the tolerances originally specified by the designer.  

I recently found a really interesting technical article describing the difference between semantic and visual PMI.

First some background . . .

For the first few years that Spatial 3D InterOp offered PMI, there was one topic that really confused me: semantic PMI.  What did the term semantic mean when applied to dimensioning?  Actually my lack of understanding went deeper than that, what was the big deal about dimensioning and PMI anyway?  (I'm no ME) -- I had to do some catch up to understand what on earth a geomtol was and why it was important.  Prior to then, I thought PMI was just +/- some tolerance on the length of an edge, right?

So I learned that it is more complicated than that and that there wasn't even a standard way of representing PMI.  In fact, not only wasn't there a standard, but there wasn't even a common ideology on the structure.  There have historically been two competing ideologies: semantic and graphical.  Spatial started offering semantic initially to meet the automation needs of our CAM and measurement customers, while graphical PMI was more popular at the time.  In more recent years, these two ideologies are starting to merge together, as we'll show in upcoming releases (sorry for the marketing, hard to talk about this topic without discussing our product line).

So about the article . . .

As Fischer nicely explains, "semantic data captures the meaning" whereas graphical is presented "for human consumption."  In computer science terms, you get a class structure in memory which represents the PMI , providing access to its specifics, such as geometric tolerance type or magnitude through a class method or property - this enables automatic creation of machine paths and test plans.  The article also discusses some of the inherent difficulties with semantic PMI, which we struggle with too, by the way, stemming from the lack of a common standard.  

 An example of this that we've seen is in ProE/Creo, which allows you to put tolerances on driving dimensions.  Driving dimensions are the various dimensions defining the features which ultimately result in the final solid, but they may not necessarily be dimensions that are meaningful to the final solid.  See the example below in which I've created a geomtol between the solid and a construction plane.  Ok, this is a very simple example of a feature, but the significant point is still illustrated: unless you understand the relationship between that plane and the solid (i.e. the feature, i.e. ProE's "secret sauce"), that dimension and geomtol are meaningless.  This is an inherently different style of tolerancing than what is used in either UG/NX or V5, which makes standardizing the data between them difficult.

Anyway, to make a long story short, in my quest to understand a topic that confused me, I found out that there is general confusion and inconsistency on the topic . . . but people are working on it.