Greg Huff, assistant professor of electrical engineering, and Dianna Sessions, PhD at the University of Pennsylvania, have created the world’s first real-time 3D printing error diagnostic methodology.

The authors of the new work decided to create the first methodology for diagnosing printing errors using machine learning. According to the researchers, this could make printing efficient in terms of time, cost, and computational throughput.

Until recently, the only way to test a 3D printing process for errors was to measure and test the device, or use built-in simulations. Both methods are expensive and computationally inefficient.

Previously, in a previous project, the researchers attached cameras to printer heads and triggered a snapshot when something was being printed. As a result, they created a dataset that can be combined with an algorithm to classify types of printing errors.

We use this information to predict electromagnetic performance, so there is no need to run simulations during production. If we have an image, we can tell if a certain element will be a problem. Therefore, we trained the neural network.

Dianna Sessions, PhD at the University of Pennsylvania
If the framework is used when printing, then errors can be detected. Therefore, the influence of errors on electromagnetic characteristics can be determined in real time.