ORNL develops error correction system to enhance 3D printing of large composite parts

A test object is 3D printed using a new system to monitor for errors and correct them automatically while manufacturing large items made from composite materials. Source | Carlos Jones/ORNL, U.S. Dept. of Energy

Researchers at the U.S. Department of Energy’s (DOE) Oak Ridge National Laboratory (ORNL, Tenn., U.S.) have created a novel tool that can catch and correct potential mistakes in real time while 3D printing large composite parts. The automated system could help U.S. manufacturers produce large, custom parts with fewer defects, potentially reducing waste, lowering costs and strengthening domestic competitiveness in additive manufacturing (AM), which includes 3D printing.

Large-scale 3D printing (or LFAM) directs heated plastic — often reinforced with either continuous or discontinuous fiber — through a robotic nozzle, arranging layers to form parts such as walls for the building industry or aircraft wings and car bumpers for the transportation sector. There are many printing variables that control whether layers are hot enough to stick together yet firm enough to hold their shape, a manufacturing balancing act that requires constant supervision. 

Featured Content

Episode 55: Dave Noblin, Greene Tweed READ MORE

Report: Composites and Carbon Fiber Use in Hydrogen Storage READ MORE

Automated Composites Knowledge Center READ MORE

ORNL researchers created a controller that supervises automatically, freeing workers to focus on more complex tasks. The controller system is equipped with sensors tracking the position of the robotic nozzle, the printing speed and the temperature of the plastic being dispensed. The team augmented the sensor suite with low-cost thermal cameras mounted around the printing nozzle. These are used to monitor the temperature of the deposited plastic as it cools.  

Computer vision, a type of AI that allows a machine to interpret images, enables the ORNL-developed controller to identify the location and temperature of hot material within a live-streamed thermal image. If the controller spots a deviation from the target temperature, it adjusts the speed of the 3D printing process so each layer cools to the target temperature before the next one is added. This ensures the proper shape and binding between layers, reducing failed prints and wasted material. 

A ring of tiny thermal cameras points at the robotic nozzle depositing plastic composite during testing of the ORNL platform for correcting 3D printing errors using automation. 

“It controls the process almost like a human would: by observing and nudging the setting until it reaches the desired outcome,” says Kris Villez, the project’s lead researcher, who partnered with University of Tennessee graduate student Chris O’Brien.

To test this, the team first calibrated the control system and adjusted the tight crown of six thermal cameras on the robotic nozzle. The assembly resembles a column of metal tubes laced with colorful wires suspended inside a printer the size of a boxcar. Researchers prepared to monitor how reducing print speed would affect the temperature of the layers. As each layer was dispensed, the print bed — serving as the floor — lowered slightly to make room for each new layer.

The machine printed a hexagon bigger than a truck tire to demonstrate the controller’s performance on a full-scale part. The job started with a low print speed to challenge the new controller. This resulted in material that was about 30% too cool when the next layer was applied. Detecting this, the controller automatically increased the print speed to maintain the best temperature for layers to fuse correctly, demonstrating real-time correction in action.

O’Brien says the tool can detect and correct temperature differences down to just few degrees, which is critical since temperature variations are a common cause of ruined parts. Moreover, unlike some monitoring systems, ORNL’s controller does not need retraining for every new design, saving time and computing power while increasing flexibility. It is designed to work with any large-area composite printer, any type of plastic and any shape.

This model also used machine learning to create a virtual replica of the physical printing process (i.e., a digital twin) to ensure risk-free experiments with new shapes and materials.

This project built on a previous ORNL study with Purdue University and the University of Maine (UMaine) that showed the benefits of combining thermal images with a statistical model to improve fault detection in large-scale 3D printing. More recently, researchers at the University of Tennessee-Knoxville and ORNL proved this approach is capable of reliably catching print speeds as little as 15% different from the programmed settings. 

While the earlier project automated fault recognition, the new system goes further by instantly correcting errors.

“There is a vast opportunity space to make these machines more intelligent and more responsive,” Villez says. “In the end, we’d love this to work like baking bread: You set the oven temperature, put in your dough and return when the timer goes off to see if it’s done. You don’t have to monitor the oven temperature in real time throughout the baking.”

Other researchers who contributed to the project include ORNL’s Katie Copenhaver and Alex Roschli, with funding from DOE’s Advanced Materials and Manufacturing Technologies Office. 

For related content, read “LED technology improves the tensile strength of Z-axis interlayers in composite 3D printing by 30%.”