At software developer Vixiv, the artificial intelligence (AI) has taken over.
Vixiv's software can save on the need to use iterative calculation to create lattices. The price: Vixiv has created and tested about 20,000 lattice samples to develop its AI model for lattice design. In this photo taken at the company's Cincinnati facility, CEO Aaron Chow is seen with just some of the samples tested so far. More recently printed lattice samples awaiting testing are seen in the image at the top of the page. Source (all photos except where noted): Additive Manufacturing Media.
Team members here used to manually design 3D printed lattice geometries for physical testing. Data from this testing have fed an AI model for designing with lattices. Now, after over a year of this work, the model is sophisticated enough that the AI knows where its confidence is lacking. Thus, the AI now designs lattice geometries to be evaluated. The new geometries are a surprise to the team every week. They receive these lattice models and print and test them as instructed, to continue to make the model better.
Why physical testing of lattices and why AI? Because lightweighting should not be such a heavy lift. That is Vixiv’s mission, and the need it aims to meet. More specifically, lightweighting via 3D printed lattice designs should be a heavy lift only one time, only for a source such as Vixiv that develops the data model, which can then make latticing easy thereafter. The startup, founded in Cincinnati, Ohio, is now introducing version 1 of its software tool based on this work. For now, the tool is for 3D printing in PA12 nylon through HP’s Multi Jet Fusion platform, but other AM platforms and other materials will follow. (Next up: lattices for laser powder bed fusion of titanium.)
The software aims to make lightweighting easy to apply. The user inputs a part model and characterizes the applied force. The software responds with design options for latticing the part's volume. Source: Vixiv.
The software is an alternative to lattice modeling via generative design, meaning it is an alternative to iterative calculation. Vixiv does not use iterative calculation, but instead draws on an AI model that already “knows” the equivalent of every lattice model and which lattice forms go best with any possible part shape and load condition.
“Think of it as a Lego library,” says CEO Aaron Chow. “We have tested every lattice Lego block to know how they work, and how they interact with one another.” The software can then select or interpolate blocks out of a wealth of known solutions and put them together in a few minutes.
The intended user is the part designer who could benefit from lightweighting but lacks a practical way to apply it.
“Let’s say you have heard additive manufacturing can help make your part lighter — what do you do then?” says Chow. Developing several solutions for evaluation using a generative design software tool could take weeks, and the ultimate solution would apply only to that one part. By contrast, Vixiv’s AI-driven solution outputs several workable lattice strategies for any part geometry and load case that is input, and it delivers this result in a moment. “For any new lightweighting need, we get you to an actionable part quickly.”
The result, the Vixiv team believes, will be to take a promise of additive manufacturing that is now broadly recognized — lightweighting — and make this promise easy to realize as well.
Weight is not the only meaningful consideration. Different lattice choices will work for a given problem, so the software characterizes the possibilities. Some choices produce a lighter part at the expense of a more intricate part that might be harder to postprocess. Source: Vixiv.
Lattice Tradeoffs: It’s Never Just the Weight
For every model and load application input to the software, multiple solutions are offered using various lattice possibilities. Lightweighting is rarely ever the only consideration, Chow notes, so the solutions are presented with the tradeoffs apparent. Potential solutions are all scored according to lattice cell type; cell size; beam thickness; stiffness; total number of cells in the solution; total mass; and the mass reduction percentage compared to a solid form. It might be the model delivering the lowest mass on the list is the best choice, but perhaps not.
The same part with different lattice designs. The chance to easily see different lattice choices, and toggle between them, is another meaningful benefit of the software. Source: Vixiv.
For example, the number of cells suggests the likely postprocessing difficulty. “A greater number means cells are smaller and harder to clean up,” says Chow. And even further downstream from 3D printing is the question of commercial appeal. The capability to visualize each lattice solution within the model, particularly for parts in which the lattices remain visible, is a vital resource here. For lattices that might be odd or surprising in appearance, Chow says, “Engineers can show it to the marketers and say, ‘Are you ok if it looks like this?’”
That insight from the modeling hints at another important power of the software: The CAD capability offers (and needs) something more than stacking lattice blocks. Because a user’s part model might be any shape, the Vixiv software is able to calculate where lattice blocks can fit within the volume, where solid regions are needed instead, and how these solid regions affect the overall performance of the lattices and mass of the part.
“And the user can print out any design to validate the software’s predictions about it are true,” he says.
But what if the user lacks this capacity — is owning AM a prerequisite? Chow says even this limitation does not have to preclude lightweighting. A forthcoming software feature is a portal for ordering prints from Vixiv directly. Parts can be made in the very same lab where the lattice samples were made to develop the AI model.
Lattice Samples: Life After the 1,000-Piece Builds
In Vixiv’s lab, about 20,000 polymer lattice samples have been 3D printed so far via Multi Jet Fusion. All of these have been subjected to physical testing, generally crushing them to failure. This physical work, and the data collection that go with it, are the investment Vixiv has made in educating its AI model, so that generative design, with all the time and computational power it requires, is not needed to solve latticing problems. In short, the reason the software provides a solution in seconds is because of the months the team devoted to this work.
Lab manager Michael Van Dorselaer produces lattice samples for ongoing testing and further development of the model. The work to date has been in polymer. Next step is adding metal lattices made through laser powder bed fusion. He has begun making metal samples as well.
Michael Van Dorselaer is the lab manager, overseeing all the AM equipment, including an Xact Metal laser powder bed fusion machine now making samples to expand the AI’s knowledge into metal lattices. He is the one who responds to each new order for new lattice samples.
CSO Zachary Beller performs physical testing of the specimens, generally to failure. Certain lattice designs "fail in interesting ways," he says. This one sheared cleanly so that the formerly cube-shaped lattice specimen left two pyramids behind.
“If I can get the printer stared by noon every day, then things will go well,” he says. Build cycles are 16 to 20 hours for a set of polymer samples. Postprocessing steps include unpacking, sand blasting and washing with glass beads, all of which can be performed within the following workday if the build begins early enough. The samples are then sent for stiffness and load capacity testing, often conducted by Chief Science Officer Zachary Beller, usually resulting in the destruction of the sample.
That metal parts are now being integrated into the mix and the model is fitting in various ways. The model’s expansion of material types and AM methods is necessary for greater use, of course, and meanwhile, the Multi Jet Fusion 3D printing of samples has gotten easier.
“We have cranked through the 1,000-piece builds of small samples. Now, the builds involve larger cells, with 15 or 20 pieces per build,” Van Dorselaer says.
The design of each of these cells is now determined according to the AI model’s evaluation of its own confidence. That is, Van Dorselaer is the one living the surrealness of receiving his work instructions from AI.
“The computer tells me what to make,” he says. “Then, we make it and we crush it.”
The current Vixiv team. A forthcoming feature: the ability to order 3D printed parts from the same group and facility that ran the lattice testing. Left to right: Zhenyu Dai, Zach Beller, Aaron Chow, Michael Van Dorselaer, Massimo Vancheri. Source: Vixiv.
