Generative design software isn’t as popular as it could be. In today’s green-aware environment, minimizing resources in manufacturing ought to be a no-brainer, a task at which generative design excels. And yet.
There are reasons for the reluctance to deploy generative design.
- It has been expensive; Autodesk at one time charged US$8,000/year/seat (source), though now it’s less
- It’s an extra step in the manufacturing process: design -> generative optimization -> simulation -> machining
- The organic output typically is meshes or NURBS, which traditional MCAD programs find difficult to edit
- Its use of bone-growth algorithms (source) result in knobby-looking, spindly shapes filled with lattice holes that are easy to 3D-print but more difficult to produce with regular CNC machine tools, which are much cheaper to run and more commonly found
The ultimate aim is for generative design to automate design and manufacturing, to the point that some CAD vendors have taken to calling theirs “artificial intelligence.”
The idea: once designers specify constraints, the generative design software output optimized parts, which are manufactured by 3D printers -- all automatically. Constraints include the volume in which the parts must fit, connection points with other parts, allowable stresses, and acceptable materials.
Generative Design At 40
At age 40 (source), legacy generative design software needs new thinking. I’ve come across two companies, InfinitForm and Hyperganic, doing the re-thinking, by making the claim that designing for additive manufacturing (3D printing) is too difficult (source) and too expensive (source) -- as compared with subtractive manufacturing, such as using CNC machines. InfinitForm figures that generative design software could eventually do 99% of the work, from design to manufacturing.
Each firm has written its own generative design software, and works towards algorithmic design but with opposite results. For InfinitForm, the solution is to have generative software output prismatic shapes, which are suitable for subtractive manufacturing. For Hyperganic, the solution is to drive down the cost of design.
InfinitForm: Generative Design for CNC
After getting a PhD in topology optimization, Michael Bogomolny founded (and then sold to Carbon 3D) his first generative design firm. In CogniCAD, he had found a way to more accurately emulate microstructures of artificial bones by having lattice-grain bias in the direction of force.
In an interview with me (source), Bogomolny said that after leaving CongiCAD, he wondered if it might be possible to get prismatic parts out of generative design software. Prismatic parts, with their flat sides and round arcs, would be editable in 3D MCAD software and easier to machine with CNC tools than the organic shapes usually produced by generative design.
He targeted CNC machines, because they are cheaper to run than 3D printers, and once set up, are ideal for mass production. He notes that prismatic shapes benefit from being more accurate to measure with coordinate measuring machines than organic shapes. Working with solid materials, like aluminum ingot, means you know what’s inside the material, which is not necessarily the case with 3D-printed materials, he argues, and so 3D-printed ones are not used for critical parts, such as in aerospace.
In May, 2024, his company came out of stealth mode with a closed beta. The software figures out the best compromise between least material and greatest manufacturability, and then adjusts surfaces to be prismatic: flat faces with round arcs. Fillets assume the radius of cutting tools, while narrow webs and slots are made wide enough to be machined. No gnarly, bone-like structures.
I asked Bogomolny how he gets those flat sides out of generative design. “It is our secret sauce,” he replied. My guess is that InfinitForm runs something like a civil engineer’s cut-and-fill operation for road construction: removing material from one area to add it to another -- all the while repeatedly checking that the parts still meet the constraints.
Being prismatic, parts can be brought into 3D MCAD programs to adjust the design, such as unnecessary features InfinitForm might have left in. For example, in a demo I watched, there were some shallow dimples in the final result, which would need to be edited out.
InfinitForm makes the claim that it reduces the D2M (design to manufacturing) process from weeks to minutes. A part typically takes 4-20 minutes to process, depending on the complexity of the part and whether both organic and prismatic outputs are wanted.
To start off, the kernel was OpenCASCADE, but the program is now being ported to Parasolid, allowing InfinitForm to run inside of MCAD programs like Solidworks. For now, it is based on CUDA with Amazon’s cloud, using nVidia’s massively-parallel GPUs [graphical processing unit] for speed. The plan is to eventually let it run on desktops, also on a GPU.
In 2024, InfinitForm raised an undisclosed amount in seed funding from Schematic Ventures. He is not planning to sell the company, but wants to grow it into a large one. As the software is still in closed beta, no pricing is set, but the plan is to ship in Q4 this year. https://infinitform.com/
Hyperganic: Cheaper 3D Printing
Lin Kayser, co-founder of Hyperganic, agrees with Bogomolny that 3D printers are too costly. Manufacturers hand-make their products and they “are sold like luxury cars, one by one,” charging “outrageous” amounts for 3D printing materials, he said (source).
He, however, disagrees with Bogomolny over how parts should be manufactured. Kayser feels that CNC machines are too limiting, as they can be set up to mass produce only one kind of part at a time; to mass-produce the next part requires expensive change-overs. By their nature, 3D printers effortlessly switch between producing different parts. So, his solution is to greatly drop of the price of 3D printers by mass producing them with 3D printer farms.
The problem, however, is that the price-drop is only an idea: Kayser doesn’t own a 3D printer company. So, his former firm (he is no longer with Hyperganic) is tackling the other side. HyDesign’s expertise is generating the 3D parts quickly, and then exporting a file that is not too large but of good quality to the 3D printer. The kernel in this generative design software has some Assembly-code optimizations for speed, and it runs on the cloud.
You can define lattices in Hydesign using the TPMS (triply-periodic minimal surface) feature, in which algorithms determine if there ought to be material or space in each point of the part, or else you can access the UltraSim lattice library from BASF’s Forward AM to explore materials and their properties (source).
Hyperganic wants to become a one-stop solution from engineering to manufacturing, and so it acquired DirectFEM (finite element analysis), which deploys a quasi-mesh-less method to work directly with Hydesign’s voxel data model (source). No meshing needed -- a bottleneck in traditional FEM; users can evaluate algorithmically-engineered objects quickly.
Currently, Hydesign imports STL (from MCAD) and OBJ files (from industrial design), as well as places PNG raster images to define features. The roadmap plans import of native CAD formats in the future. Product manager Weijie Zhao told me that there is no export to native CAD formats at this time, because there is no practical mathematical approach to generate a CAD boundary representation (b-rep) from an arbitrary voxel geometry representation, echoing InfinitForm’s Bogomolny’s point. You can, however, export models to the 3MF format, which is optimized for additive manufacturing.
In 2021, Hyperganic received $7.8 million from HV Capital and other investors, in addition to earlier undisclosed funding. Pricing for the software starts at e25/month (about CAD$37), approximately a sixth of Autodesk’s fee for accessing generative design. A free tryout is available at https://www.hyperganic.com.
Conclusion
Both of these companies are new, and are building out their algorithmic engineering prowess. Manufacturers look for ways to cut costs, and, sadly, one way may be to replace designers with these D2M software programs. My hunch is, however, that the 99% replacement won’t ever happen.
[This article first appeared in Design Engineering magazine and is reposted with permission.]
Comments