 I'm going to try to sketch out what makes for a good CAD model here, along a few different axes, but it should be plain that I'm just scratching the surface. It's a multi-dimensional subject with many layers, so please don't malign me if I have to trim something critical for time. I'll circle back and square things up in a future video. Hi there, this episode is going to be a little different than my usual fair, and probably not very interesting if you don't work with CAD software like I do, but I've been wanting to make this video for a long time. Originally, I was going to make it the first episode of a whole self-contained series on CAD, but I've decided, screw it, let's just do it. Computer assisted drafting, or CAD, is one of the most useful tools for folks who want to make stuff. Instead of trying to hold complex 3D shapes in your head, then dissecting them into 2D drawings, finding some way to make those shapes, attempting to put them together and only then realizing that they're absolute garbage for whatever you wanted to do in the first place, now you can build a model on a computer and realize that your idea is garbage very quickly and efficiently. If you're lucky, sometimes you can update the model in some way that makes it not garbage before you actually build the thing. That capacity for rapidly visualizing and effortlessly changing a design as it gets fleshed out is why people who can make good CAD are highly sought after. Of course, that's not where most of us start. Lots of folks, myself included, are thrown into the deep end when they start using CAD, struggling valiantly to make an exceptionally complicated and frequently obtuse bit of software do anything that's remotely useful for an urgent problem they're trying to solve. Understandably, a situation with lots of pressure and little guidance often results in really bad, fragile, hacked-together models, as well as a set of bad habits that can make modeling even harder. Without a grizzled CAD veteran looking over your shoulder pointing out how you might do things differently, you might not even realize that you're wasting hours of your life repairing broken models that could be better spent, I don't know, learning how to knit if you approached modeling a little differently. What I'm hoping to sketch here is a general theory of good parametric CAD practice that should apply regardless of what program you end up using. A set of high-level concepts that might give you some sense of what you're aiming for. Obviously, if you're solving into your keyboard trying to figure out how to make literally anything, and believe me, I've been there, these ideas are going to be aspirational at best. But maybe you'll have a little more patience for learning the ins and outs of notoriously unfriendly CAD software if you have a couple rules of thumb guiding how you go about modeling. Some general advice so that maybe you don't have to fight the software and yourself. If you're a coder, you may hear some things that rhyme with good programming practice. After all, CAD is, fundamentally, feeding a computer a list of sequential instructions to achieve some output. At the end of the day, CAD is just programming shapes. First, let's talk about dependencies. Every edge and surface of a part represents a series of calculations. For example, to make this whole, the software has to figure out where the top surface is, where the bottom surface is, where the whole center is, what shape the whole is, how that 2D shape traces through space to create an enclosed volume, and which side of that volume should count as the inside or outside of the part. Now, there are a theoretically infinite number of ways we can tell the computer each of those values. We could add dimensions for it from the origin of the part. We could create dimensions off the sides of the part. We could create dimensions from some other feature. We could create some dimension sketch and reference that. We could create a sketch that references a sketch and reference that. We could... you get the picture. The choices we make about how to define all the calculations that create the whole say a lot about how we're thinking about it, which in turn will make our model work in a specific way, making certain changes easier and others harder. Do we care most about the whole being in a certain location relative to the origin, or a certain distance from a specific edge? Is it built to always be at a right angle relative to the top surface, or did we include some extra moving parts when we built it, just in case we need it to be diagonal someday? Importantly, the more stuff we involve in the calculation of where the whole goes, the more steps the computer needs to figure out where to put it, and the more fragile that calculation becomes. If we make the whole depend on a sketch that's built on a sketch that's built on a sketch that's built on a sketch, not only is each one going to need to be calculated by the computer, which can get slow, if just one link in the chain breaks, the whole stops working and so does everything that depends on it. Being deliberate about how you define features and leaning towards a flatter hierarchy where it makes sense can remove some of those dependencies and make your model lighter and more resilient to changes. If you have to choose between drawing on a default plane that's never going to stop working, or drawing on a surface that's built on a feature, that's built on a sketch, that's built on a default plane, all things being equal, you should lean towards the approach with fewer middlemen. Obviously, it's possible to go overboard. If you build every feature with no dependencies whatsoever, they'll always work, but if something really does depend on something else, you'll have to update it manually every single time as controlling feature changes, which is both laborious and prone to error. Some things should move together for the model to make sense, but avoiding needlessly precarious or nested references is, generally speaking, a good goal. Next, let's talk about feature order. Whatever CAD program you're using, you should have ways to modify the sequence of steps the program takes to make the desired shape, and that sequence matters a lot when deciding how the model should work. Take this thingamajig. I've built it as a half model, which lets me keep things symmetrical so I don't accidentally add a feature to one side and forget to add it on the other side or something. Unfortunately, this text I've put on here is mirrored too, so it's just an ugly mess right now, but if I drag the feature that creates that text below the mirror feature, now everything's fine because the mirroring happens before the text. Meanwhile, this whole is looking kinda lonely. I could add a new feature to give it a matching partner on the other side, or I could edit the feature to add a second hole, or, boop, just move it up here. Now it has a friend. Feature order can make your model easier or harder to work with and update. Details like rounded corners or decorative patterns can be fragile, so putting them near the end of the list gives the model a good chance of continuing to mostly work, even if they break and disrupt everything that comes after them. Rolling back and forth through the feature tree to add new features where they make the most sense can prevent annoying CAD stuff, like remaking the same hole over and over because it keeps getting covered by new features or accidentally creating references to unimportant details that cause errors when those details get deleted or changed. Finally, let's talk about readability. This entire part is driven by this one enormously complicated sketch. I can modify one dimension and voila, everything updates automatically, but man, I do not envy whoever has to figure out how to make a simple change to this thing, picking around all those sketch segments to find the one that will do what they need it to do without screwing up the whole damn thing. It's like I've tried to cram an airplane cockpit into a shoebox, maybe controllable in theory, but hopelessly eligible and liable to fail disastrously if someone sneezes while they're working with it. Making a model readable is tricky and requires a fair amount of proficiency with the software to know what features will communicate what you're trying to say about the part most effectively. If you think about writing instructions or a recipe, there's a balance between giving too much information to process all at once and belaboring every single step in agonizingly boring detail. Similarly, trying to cram everything that defines a shape into one incomprehensible sketch makes a model hard to work with. And of course, taking 50 steps to describe something trivial isn't really easier. Ideally, a model should speak to whoever's working with it in a straightforward and intuitive way. So anyone who picks it up, including yourself, six months in the future when you've forgotten what the hell you were doing will have a good idea of where to look if they want to change something about it. Now, there are lots of little incidental tricks and idiosyncrasies that can make your models better without much work. Building the default reference planes in the geometric center of the part instead of somewhere random. Patterning bodies or other dumb geometry instead of forcing the computer to recalculate a feature over and over again. Fully defining and dimensioning sketches without creating a bunch of unnecessary geometric relationships. Mating parts with reference geometry instead of surfaces. Avoiding splines unless an industrial designer has a gun to your head. These are all useful tricks and you'll inevitably discover more of them as you work in CAD and watch models explode. But they often require a nuanced understanding of your specific software and how it works which is more than I can give you in a 10 minute video. Quilts, bodies, there's no telling what nonsense you'll have to translate into the language of your particular program. Still, if you keep your sight set on the angle of a simple but not too simple adaptable readable model. If you think about how to build in the dependencies and calculations that best represent what you want it to be and no more than it needs, you'll be doing a lot better than most folks who get into CAD who think that so long as it looks right, it's fine. What are the most useful tricks that you've learned for working with CAD? Did you watch this entire video despite the fact that you don't actually use CAD at all? Please leave a comment below and let me know what you think. Thank you very much for watching. Don't forget to blow up, subscribe, blah, share and don't stop thunking.