I’ve been thinking about recent developments in AI technology and how they might dramatically change how we make parts.
Today, the process required to make proper parts is quite lengthy and requires the use of multiple tools — and that’s even if an individual is designing and printing parts. It gets far more complex when there are groups, subcontractors, and more involved.
Let’s look at the simple case.
- A part is required, and it must perform certain functions. These are defined as the first step.
- A CAD tool is used to transform the requirements into a (hopefully) useful design.
- Sometimes, the design is tested virtually using a simulation to learn about the stresses the part must endure.
- The design is prepared to print using slicing software, with (hopefully) the optimal print parameters.
- (Hopefully) the resulting part is successfully printed, and it can be physically tested.
- Usually, the part is not correct for one reason or another (wrong geometry, concept incorrect, slicing not optimal, wrong materials, flawed print job, etc.).
- Start over at the top and repeat the steps until the best part is obtained or someone declares “it’s good enough, let’s go for lunch.”
This is more or less the workflow that is currently used. Multiple tools are used: CAD, simulation, slicing, etc., and those all require training, sometimes significant.
You’ll notice that the process involves a lot of iteration and hope.
Some of the trouble has been reduced already through the use of AI technology. For example, print monitoring can now be mostly done using AI vision tools, freeing the operator from watching the machine.
But what if we took this a lot further? What might the concept of part workflow look like in a powerful AI future?
I expect AI automation will gradually increase in several areas:
3D printer operation: Future 3D printers should have even more advanced AI that would automate even more steps for operators, perhaps even some of the material mounting effort required today. We are already on this path with many of today’s 3D printers.
Slicing: A huge amount of intelligence could be injected into future slicing software. I could see the following:
- Automated placement and orientation
- Automated simulation to devise optimal toolpaths for strength
- Automated material type selection
- Automated job dispatch
- And more
Almost anything today’s operators “know” about how to set up jobs could in theory become part of the slicing software and be performed automatically. You’d just need to drop in the file and indicate how many you need. This could significantly reduce iteration.
Design generation: This is a growing area, and we’re just at the beginning of it. Today, precision 3D models are prepared by experts using complex CAD software tools. Eventually — and this might be a few years away yet — it may be possible to generate the same precise designs using prompts. I can see a future design “talking” with an AI CAD tool to make modifications, much like we do today to modify 2D images. Such a tool could also make use of simulation knowledge, and be far more likely to design the right part immediately.
There’s another step that can occur here: what if an AI agent was able to understand these steps and take them on by itself? In that scenario, the users would essentially describe what they want, and the agent would then perform the rest of the steps, all the way to producing the part.
There would be little need to understand how a CAD program works, nor how to slice a job, nor very much on how to operate the 3D printer.
That might sound ridiculous today, but we can already see steps being taken in that direction. The AI technology advancement curve is becoming very steep, and it would not surprise me to see tools with these capabilities emerging in a short while.
