Additive manufacturing lets you create all sort of things, from houses to airplane components and hip replacements. But it’s not perfect. Sometimes things go wrong during the printing process and you end up with a faulty product. These errors can be caused by many factors including the material, the temperature, or the speed of the printer. Humans can monitor a printer, but not many at the same time and not for all of the hours or even days it can take to print a part. Humans also can’t intervene quickly enough to repair an error if it occurs.

That’s why we created a new system that uses artificial intelligence (AI) to learn how to adjust 3D printing parameters to prevent and correct errors. In this recently published paper, we used two sources of data: the metadata from the printer, which tells us the printing conditions; and the video of the printing process, which shows us what’s happening on the print bed. We trained AI models that can predict how much material should flow out of the nozzle at any time to achieve a good print. We also built a fast control loop that can use these predictions to adjust the material flow on the fly and fix errors as they occur (fig. 1). We found that our system could learn the optimal parameters for a new material quickly, and could correct errors on new shapes effectively. Our system could also handle complex materials that change properties depending on how they are printed.

CAXTON - The Collaborative Autonomous Extrusion Network

Our method is based on a network of 3D printers that work together and learn from each other. We call it CAXTON: the collaborative autonomous extrusion network. Each printer in CAXTON can print non-stop and collect data, without human intervention. The data are images of the printing process captured by the webcams. CAXTON uses a special type of AI system called a deep convolutional neural network to analyze the images. It can compare the images with the ideal printing settings and find out if there are any errors. For example, it can detect if the material is too hot or too cold, too fast or too slow, or too thick or too thin. It can also tell how to correct these errors by adjusting the printing settings accordingly.

CAXTON can do this for multiple errors and settings at the same time and in real-time. It can also learn how different settings affect each other and find the best combination for each situation. It can even discover how to print with new materials that it has never seen before. CAXTON is different from other AI systems that are used for 3D printing monitoring. Most of them only detect errors after they happen and rely on human input to label them. CAXTON can prevent errors before they happen and label them automatically. This makes CAXTON more efficient and accurate.

‍

We believe that CAXTON can be applied to other types of 3D printing and other manufacturing processes that use similar data. AI can help improve 3D printing quality and reliability, and make it more accessible, efficient and versatile for everyone.

Acknowledgements

This work has been funded by the Engineering and Physical Sciences Research Council (EPSRC) Ph.D. Studentship EP/N509620/1 to Douglas Brion, Royal Society award RGS/R2/192433 to Sebastian Pattinson, Academy of Medical Sciences award SBF005/1014 to Sebastian Pattinson, Engineering and Physical Sciences Research Council award EP/V062123/1 to Sebastian Pattinson, and An Isaac Newton Trust award to Sebastian Pattinson. For the purpose of open access, the author has applied a Creative Commons Attribution (CC-BY) license to any Author Accepted Manuscript version arising.