At the recent AI Accelerator Summit in Boston, Additive Leader at GE Global Research Brent Brunell addressed the growing need for AI-enabled applications within the context of industrial metal 3D printing. With an audience made up of some of the leading hardware and chip innovators in the tech industry, Brunell explained that in order for metal AM to be broadly industrialized, the sector needs faster computing capabilities and more computational power.
This reliance is related in large part to the simulation process, which is a critical step in ensuring that a given part will be successfully printed using a set of parameters. As Brunell explained in a recent post on Linkedin, the computational power required for a single AM simulation is substantial.
He says: “just consider that printing one small metal fuel nozzle part that you can hold in your hand can generate 36 terabytes of data. That’s three times total amount of data Twitter generates in a day! And this data is not trivial.”
In order to scale the process up to industrial levels, there is therefore a serious need for more powerful, AI-driven computing capabilities. This need is being addressed at the GE Research Lab, where Brunell says experts are working to integrate new applications of AI into industrial AM machines and processes.
In the video below, Brunell demonstrates not only the real-time speed of the DMLM process but also the simulation of the laser path in the metal powder. Slowed down by 1000x, you can really see the critical information that the simulation provides, from the melt flow, to the hardening of the metal, to pores and depressions affecting the layer below.
By watching the video, he says: “you’ll get an appreciation too for just how fast the printing process is happening, as our AI algorithms account for all the factors we want to measure to ensure we get the first part right every time.”
At this stage, GE has the AI capabilities to measure the correct parameters of a part build. Going forward, it hopes to advance computing speeds and computational power to process builds in real time. Looking at numbers, Brunell says that a typical speed today can reach up to 2 kilohertz. In future, and in order to exploit the maximum benefits of AI for additive manufacturing, speeds up to 20 kilohertz will be needed.
“The good news is that we have full ecosystem of established hardware players and emerging venture startups working hard to ramp up speeds,” he concludes. “We connected with several of them in Boston and look forward to what can be accomplished together in the days, weeks and months ahead.”