I have seen that newcomers to MOI have problems with complex geometry. It is natural that one will have trouble while learning. When I started, my fear was that I would encounter some configuration that I could not resolve with the tools that Michael has given us. After thousands of hours of working with MOI I know that I can always get what I want -- but it may take more work than I would like to spend. I accept the time involved because the results are spectacular, and I don't know any other way to get it done. The model that I'm including here is an example of what I'm doing. A lot of the models that I see are just intended to look like something; they don't have to be made and function. This is a cylinder which will be cast, machined, and part of an engine that will be used in a motorcycle. Dimensional accuracy is critical. Everything was done in MOI -- I didn't have to export anything to another program to solve some problem that I couldn't solve in MOI -- as other users have sometimes suggested is the only way out. People who have trouble with 'fillets' may be shocked at what they see here. I actually think that the ability to blend one surface into another is a strong point of MOI. You can judge for yourself. The file is too large to post directly (even though I deleted much!) but try this link:

Cyl Less Manifolds_3dm.zip

Great job Tim, that is an awesome model! Thanks for posting it.

- Michael

Super, very trustworthy model.
Mik

Looks so good!

So lot of "manual filleting" by trimming and blending right.

I'm also modeling something similar for a technical illustration I'm working on. Haven't worked on the engine block yet. Not sure how difficult that will be.


Nice work!

Cool result! Bravo!

But what is the general concept of this modeling?
You say just "blend one surface into another" but how..in delicate junctions between crossing parts...
That is the only secret ? ;)

Have you a concret workflow on a difficult part ?...

Actually, the word 'fillet' applies to a special case. When two surfaces come together with a sharp edge, you can put a fillet on that. In my work, I don't do that much. This model was built from the inside out, the "things" on the inside are determined by their functions. But these things are only the inside surfaces. What I have to do is add wall thickness. But the walls overlap and the surface of one doesn't neatly fit the surface of the thing next to it. So what I have to do is to build a bridge between the surfaces, which is quite a different sort of a problem from a 'fillet'. There is no way to write an algorithm to handle problems like this, for every problem is unique. What I start with -- several surfaces that don't seem to relate in any way to one another -- looks like a mess, and I don't have any clear vision of what this should look like. But I have techniques for approaching these problems. Obviously, they work; but as I said, the time required can be great. I think it would be helpful if I could put together a video of the general rules that I follow, but that probably won't happen, as I don't know how to do that and I don't have time for it. But I can give you some clues. You can start by drawing on the surfaces in 3D mode. Slowly, the lines that I draw begin to show me how I must trim the surfaces. The first step, then, is to 'see' what needs to be done. Then, these lines that I drew are not really on the surfaces; the points selected are on the surface, but those points are far apart and the line is not on the surface between the points. Also, the lines are not fair. So I use them only as a guide. I then draw a 2D curve that more or less corresponds to the rough curve, but it is fair and precise in certain ways; you can do this in 2D, but it is impossible to do this in 3D, directly. I project those 2D curves onto the surfaces and then I have curves that fit the surfaces, are fair, and I use those projected curves to trim the surfaces. Now I have well-formed gaps between the surfaces. I use 'Blend' to connect them, but this is only the starting point. 'Blend' is beautiful at connecting surfaces, but it only works with two, and only between two edges. This usually results in a bunch of 'fillets' that don't fit together very well. That's okay. Using 'Construct', 'Curve', Iso, I make cross-sections on those blends; as many as I think I need. (Sometimes, 'Iso' won't give me what I want, so I 'Project' a line onto the blend.) Now, I delete all those blends, except in certain cases where one is good. The next step is to use 'Network'. This is another great tool, but it doesn't take into account surface contours. What 'Network' does is use four edges, where 'Blend' only uses two. So the trick is to use 'Blend' and 'Network' together; each makes up for the limitation of the other. So my method depends in large part on 'Project', 'Blend', and 'Network'. There are many tricks that one picks up and these are necessary to deal with the infinite variations of complex geometry that come up, but those three tools are the keys to the kingdom.

Great sample model!

Thanks for the infos!

So the 3 keys are Iso - NetWork - Blend! ;)
in waiting an hypothetic video! :)

An idea of time for modeling the object above ? (with technic assumed)
And have you try to transform it in a "solid" in Moi ?

Always with the incredible free SimLAb Composer lite ;) 1 second render :)

"So the 3 keys are Iso - NetWork - Blend! "

The three keys are 'Project', 'Blend', 'Network'. The first problem is to get fair curves on the surface. In 3D mode, you can 'draw' on a surface, so long as you are in 'Object' mode and have 'On Surface' selected. Then, every point that you pick will be on the surface; but you will notice that if the picked points are very far apart (and the surface has much curvature) the line seems to disappear. This is because the curve is finding it's own way between points and is not on the surface. (BTW, for those unfamiliar with the term, 'Fair' refers to -- I will call it -- constant curvature. It is very different from 'smooth'. The term comes from the building of ships and yachts. Take a fiberglass hull and paint a line on it, from stem to stern, which looks straight from the side. Now go to one end and sight down the line. If the hull is not 'fair', it will look wavy. The surface may be quite "smooth", but if any line along the hull looks wavy, the surface is not 'fair'.) When you draw a curve this way, it will not be fair. What you need is a fair curve, on the surface. To get that, draw a curve on a plane (2D) and use as few points as you can get away with. You can easily move those points around to make the curve fair. The curve that you drew on the surface will only be a rough guide. When you 'Project' that 2D curve onto a complex, 3D surface, you will get a fair curve that is on the surface (because the projection will use as many points as necessary to make it on the surface). A very neat function of 'Project' is that after the projection, you can move the points on the 2D plane, and the curve on the 3D surface will adjust to the change! With this trick, you can fine-tune the projected curve and retain the fairness. (There is one case where this doesn't work: If you project onto a cylinder or anything like that, you will get two projections, on on each side. In this case, you can't adjust the curve after the projection. I guess that manipulating two curves at once is too difficult.) Once you have the curve that you want on the surface, do 'Select Points' and look at it. If there are a huge number of points, this may be telling you that the surface is not very fair and so many points are required to make the curve conform to the surface. You might 'Rebuild' the curve with fewer points, but if you reduce the number too much, you will have trouble trimming the surface. And while you have the points showing, zoom in on the ends and move the end-points to the intersection of adjacent surfaces, or to the edge of the surface to be trimmed. This is an important step, as it will eliminate a lot spurious end-points which can give you much trouble when do other operations. One more point on using 'Project': You may assume that you have to draw your 2D curve in one of the orthogonal views. But in many cases, that will not give you the result that you want. You need to project more or less perpendicular to the surface. So you will often have to make a plane that is more or less perpendicular and draw on that. (I thought that I would have to reset the 'C Plane' to this new plane, but usually 'Project' is smart enough to know what you are doing and will just do it. If it doesn't want to project, then reset the 'C Plane'; but 95% of the time, that's not necessary.) I've explained this at length because you have to get this right before you use 'Blend' and 'Network'. 'Iso' is certainly helpful in making the cross-sections that 'Network' needs, but I would call it a secondary, not primary function. Play around with this and see how it works for you. If you have trouble, come back with a question.

Sorry, I forgot your second question: How long does it take to make a model like this cylinder? I don't really know. I had only done one other model with MOI (or NURBS) before I started this cylinder, and I didn't know what I was doing. (I do have a lot of 3D experience, going back to the early 1980s, but that was with polygons, and NURBS is very different.) Learning is a slow process! A lot of that early work was poor and I had to go back and fix many things. There is still much on this model that can be better, because I accepted things that weren't really right, but were close enough that the errors would never be seen at the resolution of 3D printing. I want to say something about design. Companies like 'SolidWorks' teach people to do everything in 3D and then make 2D drawings from the 3D model. Except in simple cases, I think this is insane. I do all the design work in 2D (using 'Q-CAD'). Then I can import DXF drawings into MOI and build the model from these. It is very easy to make small changes in 2D; e.g. moving a screw location. Doing the same thing in 3D is very time consuming. I haven't tried to make a solid of this, but you can try it. I rather think that it is too complex.

Many thanks for the very detailed answer(s)!

So that is your "first" try of complex Moi object ? Very convincing result! Bravo!

You don't speak about the projection "nearest point" option - does this will be useful or not in your process ?

When I speak about "3D solid" result it's just when you "Join" all surfaces Moi makes a unic "Solid" waterproof (if surfaces are well build)
Indispensable for 3D Print or future boolean operations between volumes!
So maybe that takes many or few time or not - or maybe that is not possible because some problem are always present.

For your object that is one minute - and result is not a solid but stay "Linked surfaces"
So maybe some little adjustments are necessary for have a 3D printing object...
But like is it's yet pretty cool for just render images or maybe animation.

Do you know the native ExplodeMove function ?

Press Tab and write ExplodeMove (with or without Capital(s) anywhere but without space) or by shortcut making (on the 2nd column)

That can be help...to understand the surfaces! ;)

I'm afraid that my explanation yesterday may have done more to confuse than inform. I will address this in the following post. Now, I will tell you why this model can't be turned into a 'solid'. I made this model for display only; for example, to render as you did. This model will be changed for 3D printing and casting. For example, if you look, you will see that the screw holes have no bottoms! Obviously, that won't make a solid. I should have told you this; I'm sorry that you wasted time on it. Also, on the model to be printed, I have to add material to all the surfaces that will be machined after casting. This model is just to look at.

Here is a picture that will help explain one of my methods. The job here is already finished, so you will have to imagine that the fillet in question is not there, and the red curve is just a rough idea drawn on the surface. Because of the angle, there is no way to project onto this correctly from any of the orthogonal views. So after I've drawn the rough curve, I adjust my viewing angle to that which will work best for the projection. I then draw a line that corresponds to that angle (that's the green line almost hidden among the selected lines). I move that line away from the surface and then I make the plane perpendicular to that line. Next, I copy that line to several points along the rough curve. The lines selected are those lines, which I have extended to fall on the plane. Now I can draw the blue curve between those points. After drawing that curve, I delete the selected lines -- they are no longer necessary. Already, this method washes out a lot of the irregularities of the rough curve, but if necessary, I move the curve's points on the plane until I have a fair curve. Now I can project that curve onto the surface. As I said, even after it is projected, if it doesn't look quite right, I can move the points on the plane again and the projected curve will automatically follow. This is very neat. When I'm happy with the projected curve, I 'Show Points' and move the endpoints to be certain that they are precisely where they need to be. I then use that curve to cut the surface. I'm afraid that this sounds awfully complicated, but when you've done it a hundred times, it goes smoothly. One thing that this method permits is a gap between surfaces that can vary in any way that you choose. It doesn't have to be straight or tapered. It can be narrow at the ends and wide in the middle, or whatever suits the model. I want to mention something else. When you use 'Blend', the bulge factor should be different depending on the angle between the surfaces. But what do you do when that angle changes over the length of the 'fillet'? Simple: At the wide-angle end you will want a bulge factor of maybe 0.8. At the end with the more acute angle, you will want a bulge factor of maybe 1.2. So first, blend the edges at 0.8 and make a cross-section (usually with 'Iso', or just turn the end of the blend into a curve). Delete that blend. Do another with the bulge factor at 1.2. Make a cross-section at the acute end. Delete that blend. If you want to be cute, do another blend at 1.0, and make a cross-section in the middle. Delete that blend. 'Blend' forces you to chop up a fillet wherever there is a single edge. You can now 'Merge' all of the edges that you trimmed to accommodate the 'Blend' tool, and 'join' all the edges so that you then have only two long curves. Select those and all the cross-sections and do 'Network'. Do you get this? You can make fillets that vary in width and vary in bulge-factor and that cover several surfaces! In other words, you have complete control over what you get. Yes, there is some work involved, but there is no way to automate this process and retain this control. And now you see why I say that doing 'fillets' in MOI is a strength, not a weakness.

Here is another model that you can play with.

Hi Tim,

Thanks for explaining!

<<that you wasted time on it

No problem... I just load it --- Join it (no solid) --- ExplodeMove it and render! ;)
All that take 2 minutes! ;)

I will study your last explanation and model...as soon as i have 5 free minutes!

Stranger things indeed! :)

Here is another example. You can see that there are several 'parts' which do not come together in any way. There forms are dictated by their functions. But all these 'parts' have to come together to make one part. As I said, the problem here is nothing like a 'fillet', although the end result often looks like a 'fillet'. This sort of a problem comes up in my work more often than the 'fillet' problem. And when confronted with something like this, the question is: How do I make these 'parts' conform to one another. Let me be perfectly frank and say that I can very seldom see how this should work. So I need a method, like an algorithm that let's me me begin in the dark and slowly come to the light. In this example, I began by making the surface in the center, which goes from a flat plane on top to an ellipse at the bottom. I used 'Blend' to make that -- but first I had to extrude the ellipse at a certain angle that would give me the correct form. With that surface done, as you see here, I went about "drawing" lines on the surfaces of the other parts. The blue lines shown here are not the first or the last that I drew. I draw some lines, and then rotate the view around and look at the whole scene from different angles. My first effort is usually quite poor, but knowing what won't work is information leading to the solution. I keep refining these rough lines until I can 'see' how the connecting surfaces must be. When I finally know what I'm doing, I apply the method that I previously explained, to go from these roughly drawn lines to fair curves that lie on the surfaces. With these I trim the surfaces. Then I use 'Blend' and 'Network' to fill in the gaps. 'Network' is a fantastic tool, but you have to provide enough information for it to work without deformities. 'Blend' is the primary tool that you use to set up the network of curves (the information) that 'Network' needs. When you master this, the results are wonderful.

I gave you a problem which many people might find impossible to resolve. So here is how it turns out.

And natural boolean between volumes don't can give some result ?

Does all that is real pieces or imaginative stuff ?