Voxel Developing AI Alternative to Generative Design

One of the challenges with generative design is the computational power required. Additive manufacturing offers a means of realizing intricate geometries that are tailored and optimized to the set of performance conditions required for a specific need, but in applications optimized for a larger set of conditions, accessing and employing the cloud-based processing capacity necessary for iterative computations loops becomes a meaningful share of the cost of the capability.

Instead, what if there was an expert system that already “knows” the right or nearly right geometry for a given problem, so that no generative design and therefore significantly less computation is needed?

That aim is the promise that Ohio-based startup Voxel is working toward. The company recently moved into a new headquarters in Cincinnati where it will perform the 3D printing and physical testing to develop this system.

Artificial intelligence (AI) replaces iterative computation in Voxel’s system. Co-founder and chief technology officer Zachary Beller says the system uses “an AI-enabled backend” to “relate physical performance to geometric parameters” in order to suggest a form that is “better than 90 percent of the way there” to being fully optimized for the input aims and requirements of the part. And all this comes without computation, without generation.

He offers the analogy of an I-beam. Beller says, “Engineers in construction don’t design I-beams. They specify them based on known performance of these shapes.”

The way I-beams are known, the Voxel AI will “know” the performance of an uncountable number of complex, organic, unnamed geometric forms.

Building this tool means feeding the AI. That is a key purpose of the new facility. The Voxel team here is 3D printing small cubic unit cells in many different geometries (see photo) so that these geometric forms can be tested and the data can be applied to building the neural network AI system. Unlike OpenAI and its tool ChatGPT, Beller says Voxel will own all the data on which its AI is built.

Funding for the work comes from various sources. The company has venture capital investment. In addition, it performs AM consulting and prototyping work on its HP and Xact Metal machines using the design understanding it has developed so far.

For the system it is developing, there is a long way to go, but work is straightforward. “For now, we are solving the lightweighting problem in polymer,” Beller says. Unit cell forms are being tested with this purpose in mind. That is, the geometry samples are being evaluated for stress and strain, one cell and one linear axis at a time.