Machine Learning for Improved Autonomy
The US Office of Naval Research (ONR) has awarded a two-year, $5.8 million contract to a team led by Lockheed Martin’s Advanced Technology Center to develop software models and sensor modifications for specialised multi-axis robots that use laser beams to deposit materials in the additive manufacturing process, otherwise known as 3D printing.
Today, 3-D printing generates parts used in ships, planes, vehicles and spacecraft, but it also requires a lot of babysitting. High-value and intricate parts sometimes require constant monitoring by expert specialists to get them right. Furthermore, if any one section of a part is below par, it can render the whole part unusable. Which is why ONR and Lockheed Martin are exploring how to apply artificial intelligence to train robots to independently oversee—and optimise—3D printing of complex parts, in order to build better components.
Researchers will apply machine learning techniques to additive manufacturing so variables can be monitored and controlled by the robot during fabrication. Currently, technicians spend many hours per build testing quality after fabrication, but that is not the only waste in developing a complex part. It is common practice to build each part compensating for the weakest section for a part and allowing more margin and mass in the rest of the structure. Lockheed Martin’s research will help machines make decisions about how to optimise structures based on previously verified analysis.
That verified analysis and integration into a 3D printing robotic system is core to this new contract. Lockheed Martin, along with its seven industry, national lab and university partners, will vet common types of microstructures used in an additive build. Although invisible from the outside, a part could have slightly different microstructures on the inside. The team will measure the performance attributes of the machine parameters and these microstructures, aligning them to material properties before integrating this knowledge into a working system. With this complete dataset, machines will be able to make decisions about how to print a part that ensures good performance. The team is starting with the most common titanium alloy, Ti-6AI-4V.
“We will research ways machines can observe, learn and make decisions by themselves to make better parts that are more consistent, which is crucial as 3D printed parts become more and more common,” explained Lockheed Martin’s project manager, Brian Griffiths. “Machines should monitor and make adjustments on their own during printing, to ensure that they create the right material properties during production.”
“When you can trust a robotic system to make a quality part, that opens the door to who can build usable parts and where you build them,” added Lockheed Martin Fellow for additive manufacturing, Zach Loftus. “Think about sustainment and how a maintainer can print a replacement part at sea, or a mechanic print a replacement part for a truck deep in the desert. This takes 3D printing to the next, big step of deployment.”