Opinion
Generative design optimization has a big problem, in my opinion. It’s been around for a long time (since 1994), and it seems to be just not that popular.
Michael Bogomolny formed a new company whose InfinitForm software outputs prismatic generative designs that can be cut by regular machine tools, which are much more common, much faster, and much cheaper than 3D printing.
I interviewed Mr Bogomolny last week. Here is our chit-chat, edited for clarity and brevity.
Ralph Grabowski: I want to talk about your current company, InfinitForm, but also get a little bit of background on CogniCAD, your first company. Where were you before CogniCAD?
Michael Bogomolny: Before CogniCAD, I did my PhD in the field of topology optimization, and then I worked in the defense and automotive industries. I worked for Altair for close to three years as a software developer, and then I founded ParaMaters and our CogniCAD software. [In 2016, ParaMaters raised an undisclosed amount from Avi Reichental.]
Q: Why did you sell CogniCAD?
A: Because it was a pretty good offer from Carbon, the 3D printing company. CogniCAD was designed for additive manufacturing, and it made sense to align materials, Carbon machines, and our software.
Right now, the software is disconnected from the printing machines, and materials are supplied by other suppliers, so there always is disconnect. If they communicate well, then there's a lot of capabilities you can build around.
Q: Do you have a price for your sale?
A: That's not disclosed.
Q: Let's go on to InfinitForm. Your site has a graphic showing CAM being brought into CAD. So how does InfinitForm work inside a CAD program?
A: Maybe we should take a side step back, to explain the state of design-to-manufacturing process, and generative design software in general.
Generative design can be promising to automate designs for manufacturing. The reality is that the way generative technology works, it produces organic shapes. [Generative design software is based on bone growth algorithms.] With organic shapes, there are multiple problems:
-
First, it's difficult to machine or even to cast those organic designs. You can 3D print them, but it's difficult to do to manufacture using traditional [subtractive] manufacturing technology.
-
Secondly, the representation of these organic geometries is either meshes or NURBS [non-uniform rational B-splines]. For engineers who use traditional CAD systems like Solidworks, Catia, and Siemens, it's difficult to edit organic designs represented as meshes or NURBS. There are changes to designs and in versioning, and it's difficult to make changes; you have to basically start from scratch.
-
Thirdly, how do you do quality-control for [3D-printed] parts [such as with CCMs, coordinate measuring machines]? There's no dimensions really to measure. What do you measure exactly on organic shapes? It's very difficult. So the QA [quality assurance] process is not straightforward with organic parts.
The difficulty today is that generative design works well for additive printing, but not for traditional [CNC] manufacturing.
Design-for-machining has been done manually:
- You have to draft something in CAD.
- Assuming that the design engineer is experienced enough to apply machining constraints, it then goes through CAE [computer-aided engineering] simulation. There can be iterations between CAD and CAE.
- Then it goes to design reviews.
- If the design engineer is not experienced enough, after the design review the part can come back to CAD engineer to make changes, make it manufacturable, and back and forth...
- ...until it goes for quotation.
- When part goes from quotation to the machinist, he doesn't know what the design intent is. The engineer has to find the setups, how to set up the part to machine it in most cost effective way, if it's mass production.
To summarize, on the one hand, we have a fully manual process of design-to-manufacturing. On the other hand, we have generative design, which is intended to automate design-to-manufacturing, but is only additive-oriented and doesn't exist in a workflow.
So what InfinitForm does is to generate machinable designs in a prismatic CAD format. We generate designs that are constrained with arcs and lines, and it outputs an editable CAD model. It can be machined with all the machining constraints intact.
This is our biggest difference from traditional generative design software. And that's a big step, a huge step in the process.
Q: As I was doing research on your company, I realized that you have a significant breakthrough here. How are you able to get flat sides from generative design?
A: That is our secret sauce. It’s exactly what we do.
Instead of outputting an organic piece, we have now suddenly something that looks like humans designed it [from lines and arcs]. You take it into CAD and make changes to the model, without any problems.
InfinitForm takes you 90% towards the final design. We're not replacing engineers here. It's an assistant design copilot, which helps engineers to optimize as close as possible to the final design.
Q: I believe your software runs on the cloud, but you also have plans for a desktop version?
A: Indeed. We launched our beta on the cloud. We will store local versions. Our longer term strategy is to be integrated into existing CAD programs, so that users can run InfinitForm inside their CAD software as a plugin.
Q: How will it be integrated, through an API or a file link?
A: We're experimenting right now, probably via APIs.
Q: You're currently using OpenCascade for the solid modeling kernel, but you might be moving to Parasolid?
A: Indeed, we’re moving to Parasolid. I would say Parasolid, among all the kernels, is probably the most stable and most widely adapted. There are advantages to using Parasolid when integrating with other platforms. Yeah, we know that some platforms use ACIS.
Q: I can see the attraction of starting with OpenCascade, because it is free, but isn't switching to a new kernel a big job?
A: I thought it was going to be a few weeks to get the commands to work, but it involves some investment. OpenCascade does the job; we just feel that we need to be more stable, because users don't know what's going on the back end. They want to see a robust model.
Q: [InfitinitForm gets is speed from running on GPUs.] What kinds of GPUs are you running on? Is it nVidia or does it matter?
A: The GPU [graphics processing unit], in general, doesn't matter. Right now, InfinitForm runs on CUDA [nVidia’s compute unified device architecture] (source).
The speed to generate an optimized part depends on its complexity. It can take eight minutes or twenty minutes for the entire process of pre-processing, optimization, and then post-processing for both organic and prismatic shapes. It’s takes half as long when we output only the organic part without the prismatic part; then it can take four or five minutes. If you want to go to higher resolution, of course it will take a longer time.
Q: I would think that a prismatic shape would require less resolution than an organic one.
A: That's part of the secret. But you also need enough feature-size controls. For CNC machining, you don't want thin members, you need reasonable thickness. So, that's why we don't need high resolution.
Q: Do potential customers believe your numbers, your speed?
A: Yes, they use it and they see the speed. This is just first version of InfinitForm. Not all functions are shipping to customers yet; we will release them gradually.
Of course, we plan to speed it up significantly. Our solvers are very scalable. So, it will run much faster as we allocate more hardware. Right now, we're using the smallest instance of a GPU instance on AWS, to minimize the cost.
Q: Your Web site still requires people to apply for Early Access. So that means you're still in beta and you want to select who you take on as a beta customer?
A: We already have beta customers who are using the platform. We are going to expand the beta to get more and more partners.
Q: Do you have a date for general availability?
A: Probably Q4 later this year.
Q: I'm wondering if you know of anybody else doing what you're doing?
A: Not that we know. We are the only platform that's producing prismatic outcomes automatically.
Q: How easy would it be for a competitor to copy you?
A: I'm not saying we are the smartest in the world, but it's pretty challenging problem to solve.
Q: How long have you been working on it?
A: Since November, 2023. So eight-nine months. It's a team of very skillful people. We are nine people, six with PhD degrees, with probably one of the best schools in this field. [At the end of 2023, InfinitForm raised an undisclosed amount from Schematic Ventures.] (source)
Q: How did you get the idea for InfinitForm?
A: It came naturally to me when we left Carbon. I realized we had developed an advanced technology that could automate design, but one that was only for the additive market, and the additive market is not growing.
There is a huge traditional manufacturing market [of subtractive manufacturing] out there, which has no automation. We started to dig deeper, and understood that it's a very challenging problem, because to create organic designs you nearly don't place any constraints. You apply topology optimization and there's geometry processing, which also is not trivial.
But to constrain design exactly to these prismatic shapes, and to be able to create designs that are created like humans, is an extremely difficult problem.
Q: And I suppose ultimately you would want to sell your company.
A: We want to build big company this time. We have something big here.
Q: So obviously there's more features you want to add over time. But could this idea of prismatic generative modeling branch out to other areas?
A: Indeed. So we started with CNC machining, and we are expanding to die casting and injection molding.
Q: I recognize what you're saying that additive machining is just not that big. In fact, the only place I've seen it at a large scale was in the basement of the Sagrada Familia in Barcelona, where they use many 3D printers to produce parts to see if they'll fit in the cathedral. (source)
A: If we talk about seeing those Gaudi buildings, they are organic -- gosh, how did they make it a hundred years ago manually. How much effort is that? It's crazy!
Q: I think that with our computers and their Undo buttons, we just don't grasp that. Gaudi’s work was all handmade. There's no extrusion 3D printer doing that.
There's like 7,000 small machine shops in Germany that service larger companies. They're not going to be replacing their lathes and drills with 3D printers. So that's your market opportunity.
A: Also, if you look at the cost of the part I’m holding, mass producing it with CNC machining is going to be $100 per bracket. If you 3D print it, it's probably going to be ten times more expensive.
The quality of material in additive manufacturing is difficult to control versus CNC machining. Especially in the aerospace industry, you can't 3D-print metal parts for critical components of aircraft. They only implement 3D metal parts that are for non-critical components.
I think that's where we will have the ability to gain better adoption in the industry with generative design tools than what we've seen in the past.
Q: InfinitForm might solve the problem that organic generative design is not that popular.
A: Independently, the prismatic output can be done without organic. Right now for the beta, we create both, so that users can see and compare.
Comments