CAD is a complex cognitive skill, comprising a large set of interrelated constituent skills with different characteristics and different learning processes underlying their acquisition.
One of the the most effective ways of making CAD more usable is to reduce the number of constituent skills which it comprises.
I’ve never seen any even reasonably complete listing of the constituent skills required for CAD. It might be interesting to try and put together such a list, but, for the moment, let’s look at just one of the important constitutuent skills:
This is not a trivial skill. Even when working with 2D AutoCAD drawings, it can be a challenge to make changes without knowing ahead of time how the drawings are structured. When it comes to history-based 3D (what we’ve commonly, if not a little dismissively, come to call parametric feature-based solid modeling), the problem sometimes becomes intractable.
There is plenty of research showing that editing history-based models is a big problem for CAD users. This is primarily because the task requires not just the skill of deconstructing model geometry (e.g.,figuring out how the geometry should be changed), but also the skill of deconstructing the history of how that geometry was originally created.
The history trees of typical models can have from dozens to hundreds of entries. In order to effectively edit one of these models, you need to dig through all (or many) of these entries, to find their dependencies—which are often unobvious. The process is no easier than trying to read through the source code of a complex computer program, to figure out how it works.
The challenge is to find a way of modifying CAD models without needing to deconstruct their history trees. Work on this has been ongoing in academia for about 20 years. In the commerical CAD industry it’s taken a bit longer to get right.
Direct (or explicit) modeling CAD systems have been around far longer than history-based systems. Ivan Sutherland’s 1963 Sketchpad was an incredibly intelligent CAD system (don’t miss watching this discussion of SketchPad by Alan Kay.), most commercial CAD systems developed from that time until the late 1980s were direct modeling systems, in which you directly edited the geometry of the model. Though Pro/E ushered in the era of history-based modeling (or, rather parametric feature-based solid modeling), it did not kill the direct-modeling business. Direct modeling CAD programs such as CADKEY, Autodesk Mechanical Desktop, ME/30 (from HP and then CoCreate) and many others continued selling in significant, if not dwindling, quantities.
IronCAD and CoCreate started to introduce intelligent editing capabilities to their direct modeling CAD programs in the mid to late 1980s, but it wasn’t until a few years ago that the game really changed, with a number of CAD programs adding feature-inference on top of direct modeling.
These products, from companies such as Siemens PLM, SpaceClaim, Kubotek USA, PTC and IronCAD are now commonly called “direct modeling” CAD programs. (I’ve pointed out that direct modeling has been around since the mid-1960s far too many times in the past, so I’ll just go with the flow for now, and use the same term everyone else does.)
What makes today’s direct-modeling CAD programs significant is their usability. With one of these programs, a CAD user doesn’t need to learn the skill of deconstructing model history to be both effective and efficient.
I’ve seen a lot of discussion about whether direct modeling or history-based modeling is “better.” That’s not a discussion I really want to get into yet. It’s reasonable to mention that major aerospace and automotive companies use direct modeling software for growing number of applications. PTC, which sells both direct and history-based tools, has major customers using both types on different product development programs, apparently with great success.
What’s really interesting to me is the potential of comparing the effectiveness and efficiency direct modeling versus history-based tools. While there’s a lot of anecdotal information about this floating around, to my knowledge, there are no carefully constructed research studies available.
If you dig into Google Scholar to look for academic articles on learning CAD, you’ll find one name comes up more than any other: Professor Ramsey F. Hamade, of the American University of Beirut. Dr. Hamade’s research on CAD learning is published in a number of academic and technical journals, is cited by nearly all researchers in the field, and makes for really interesting reading.
I exchanged email with Dr. Hamade recently. Here’s what he had to say on the subject:
[Direct modeling] comes across as more natural and less restrictive. Therefore, I would tend to think that such modeling should be faster and less complex perhaps resulting in shifting the learning components, both declarative and procedural to faster and ‘simpler’, respectively. Unfortunately, I have not had the opportunity to perform experiments on Creo (or the like) in order to evaluate whether these ‘logical’ expectations will hold water. I teach the CAD course (where I collect data) in the Spring so it may be a while before we can make a determination.
Dr. Hamade’s research to date supports the notion that CAD is a complex cognitive skill, and points to significant differences in usability between different systems. It’ll be interesting to see what he finds when he gets a chance to formally compare the learning processes for direct modeling versus history-based modeling systems.
Note on the sphere image: It’s not a sphere. It’s a filleted cube. Here’s a challenge for you: Make a 3D model, with as many convoluted features as possible, that looks just like a sphere and has a class-A surface (G2, I think. G3 continuity wouldn’t apply to a fixed radius curve.)