A new 3D printer uses machine learning and machine vision to check objects during the printing process for errors. By correcting these errors in real-time, the technology helps reduce the number of failed prints. Machine learning is also used to draw lessons from detected problems, which helps improve the printing process down the line.
The 3D printer called 'Snapper' has been developed by Inkbit, a startup founded professor of electrical engineering and computer science Wojciech Matusik and former MIT students Javier Ramos, Wenshou Wang, Kiril Vidimče and David Marini. Some of them were previously involved in the development of a 3D printer that could print ten materials simultaneously by using machine vision technology.
Giving 3D printer eyes and brain
With support from the Deshpande Center, a part of MIT that helps students commercialise developments, this idea was developed further. This led to a 3D printer that can monitor the progress of the printing process using a simple 3D scanner. Following this development, Inkbit was founded. "The company was born from the idea of adding eyes and brains to a 3D printer," explains David Marini, co-founder and CEO of Inkbit.
Under the Inkbit banner, the eyes of the 3D printer have been further developed. The printer uses optical coherence tomography (OCT), a technology traditionally used in ophthalmology. However, scanning each print layer requires rapid scanning, something that was not possible with off-the-shelf OCT scanners. Inkbit therefore created its own OCT scanner that works 100 times faster than commercially available OCT scanners.
Checking every layer in real-time
This scanner is deployed to check each layer for errors in real-time during the printing process. If problems are detected, machine learning algorithms adjust the printing process in real-time to correct it. For example, a material may shrink during cooling, which can lead to distortions. The system detects these deformations and automatically compensates in subsequent layers.
The use of machine learning and machine vision in 3D printing offers a variety of possibilities, Inkbit claims. For instance, the company claims to be able to print flexible material more accurately than other 3D printers. Flexible material tends to shrink more than harder materials, something Inkbit's technology can compensate for in real-time. However, the system also allows accurate printing around an existing component, integrating that object into the print. Think of a chip or other electronic component.
Transformation
"Everyone knows that the benefits of 3D printing are huge," Marini said. "However, many people run into problems when embracing it. The technology is simply not ready yet. Our machine is the first that can learn the properties of a material and adapt its behaviour accordingly. I believe this will be transformational, as it allows anyone to turn an idea into a usable product very quickly. This will create business opportunities."
Inkbit currently has one operational 3D printer. This printer is equipped with 16 print heads, allowing it to print objects composed of multiple materials. The print head, thanks to its size, can print fast enough to create hundreds of thousands of fist-sized objects per year, according to Inkbit. The company points out that the print head can be easily enlarged, making it possible to produce 3D printers of a larger size.
The company will start selling the first products printed with its 3D printer later this year. A pilot project with pharmaceutical company Johnson and Johnson will initially be set up for this purpose. Inkbit expects to launch its 3D printers in 2020.
GE is working on similar technology
Inkbit is not the first party working on technology to detect errors during the printing process in real-time. GE announced in late 2017 that it was working on such a technology based on high-resolution cameras, machine learning and artificial intelligence. In the future, GE's technology, like Inkbit's 3D printer, should be able to correct detected problems in real-time.
"By integrating edge computing, we have created the 'digital eyes' to track every layer of every print," Randy Rausch a senior engineer in embedded computing at GE Research, explains on GE's website. With edge computing, collected data is not sent to the cloud for processing, but processed directly locally on the 3D printer. "We want the manufacturer to know in real-time whether a print is good or needs to be destroyed," he says.
Understanding print layer quality
Machine learning algorithms are used to perform analysis on scans. This gives the operator - and eventually the 3D printer itself - instant insight into the quality of each print layer. The technology is being tested at GE Research in New York. In doing so, Rausch and his team deliberately adjust the printing process to introduce defects in a print layer to test whether these defects are detected. The holy grail, according to Rausch, is being able to manage print process with the speed and accuracy needed to prevent or fix problems in real-time.
Author: Wouter Hoeffnagel
