Engineers were sold a lie decades ago when we were told Parametric Modeling would make CAD robust, easy to manage and easily changeable. It turns out the total opposite was true. I’d like to convince you to stop using it as a default. It is, ironically, bad engineering practice for two reasons:
- It leads to easily breaking models without any debugging as to why the model broke
- It’s often unsafe and bad engineering practice. This is a hefty and dramatic accusation, but I provide an example below
Before starting, I want to point out that even getting engineers to even agree on what the term means is a problem in and of itself. Here’s the most basic definition I can find: geometry and variables you define and can change at any point that impact the definition of a CAD model.
Modeling History does not equal Parametric Modeling. You can have history based modeling (what many engineers mistake for Parametric Modeling) without parameterization — Shapr3D is a perfect example of this. It lets you define a model with a history so that you can make edits in the past, but doesn’t force you to define parameters that may break things later.
The first problem of Parametric Modeling: It leads to easily breaking models without any debugging as to why the model broke. Maybe you have a fancy spreadsheet with a bunch of variables (thickness, part width, counts, etc.) defined. Maybe you use the parameter tool inside of the model itself. It doesn’t matter — at the end of the day, you’re defining parameters abstractly like variables in a program. These variables sit totally disconnected from how the model is built until you link it to a dimension or other variable.
This is where the horror happens — you build your model, and everything works. You later make one change (say, the number of holes in a plate from 6 to 8), and the model updates to propagate it, and suddenly everything breaks. Why?
Each CAD package has its own implementation of topological naming. This topological naming decides the naming of faces, bodies and other geometry. Those names are decided programmatically, and are hidden from the user.
This is where things go wrong. An innocuous change in Catia (two extra holes in a plate) has totally broken your model, because it turns out parametric modeling was not designed into the modeling standard from the get go thirty years ago. Hole number 143 ten sketches away is now hole number 145 and everything referencing it just broke.
I want to be clear here by making a simple statement — there is no CAD package on the planet that has totally addressed this issue. Some (Siemens NX and Shapr3D) are much better than others (Solidworks) but the problem lingers because of the fundamental way these applications are built (Sketches, features & Parameters).
That’s just from the CAD side of things. Let me explain why it’s even more terrifying on the engineering side.
It’s unsafe and bad engineering practice in the worst case - I will use a simple example to illustrate the issue here. Let’s say we’re designing a plate that mounts to another part with four holes. This plate carries shear and moment and resolves loading through the four bolts.
Assume we are good engineers and define this bolt pattern upstream in an ICD, spreadsheet, or parameter list somewhere.
An engineer who does not own this part, but owns the mating part decides to change the bolt pattern because it better accommodates his design. I won’t get into the million ways that miscommunication, bad process or time crunch can result in the downstream engineer not realizing that their part changed, but I can outline how it can happen in CAD without them being any wiser.
The model updates this plate, and results in the following:
The downstream engineer has done nothing incorrectly — they followed every business process, ran the stress analysis months ago, documented it in confluence, and has checked every box possible. Yet, when they go to release a revision of this part their edge distance has halved and their FOS has reduced by a factor of 2 in the shear out failure mode.
You could argue that there are processes upstream that should catch this, but I would say that organizations that have those are rarer than organizations that do. Additionally — this change happened without the intent of the engineer, which means should they revise the part for any other innocuous reason (material change, SNCRs, etc.) the model will update without the engineer ever realizing unless they’re incredibly eagle eyed and thorough (Coincidentally, this engineer was fired before the next revision of the part was ever released because they were so eagle eyed, thorough, and importantly, slow).
The good news is this problem isn’t as terrifying as it could be. Software such as Risa3D that is used to design the buildings we live and work in recognized this problem from the onset and have specifically followed the way engineering workflows operate to avoid parameterization changing a model without the designer knowing.
With all of that said - parametric modeling is a powerful tool that has its place - but its place is not the default way to think about modeling. It’s useful as a tool to create parameterized assemblies that you know will have many configurations. That doesn’t mean it’s the most robust, easy and interoperable way to share files; that distinction is held by direct modeling with BREPs as ground truth (not parametric based history models).
Ultimately, what matters is the BREP that is generated at the end (the only standard that matters) as it is what ends up generating G-code, which generates the part. It doesn’t matter how you got to that BREP.
A new argument for my thinking around Direct Modeling as the main and effective way to work on complex assemblies in the real world was understanding that if you are to work inter-department, inter-company, or inter-whatever, you’re able to work on my assemblies and I yours if we agree to utilize direct modeling as the method for geometry creation and modification. Any other way doesn’t work, as we (the planet) all have to agree to only using Solidworks on the current version otherwise. We also know that this is impossible.
