3D printer with machine vision and artificial intelligence
Oak Ridge National Laboratory is studying methods for making parts of a nuclear reactor that would become the world's first completely printed on printers
Researchers at the U.S. Department of Energy's Oak Ridge National Laboratory have applied artificial intelligence methods to evaluate the quality of powder 3D printers. The program, called Peregrine, is able to assess print quality in real time, without using expensive control equipment. Standard video cameras are enough for it - cameras with a resolution from 4 to 20 megapixels were used in the experiments.
Powder printers work by layer-by-layer fusing or bonding the powder laid out on the work surface - usually metal. Uneven distribution of powder or glue, pores, uneven heating and other factors lead to defects in the finished part. Peregrine analyzes the camera image using a convolutional neural network. The algorithm takes into account the peculiarities of edges, lines, corners, and texture of each layer.
Artificial intelligence learns to assess the quality of 3D printers
The developers believe that the big advantage of Peregrine is its ability to work with any powder printer. The program has been tested on a variety of printers, including those used at Oak Ridge National Laboratory to make parts of a nuclear reactor, which is expected to be the world's first reactor printed entirely on 3D printers.
This is the power of integrating reinforcement learning and other advanced AI methods with 3-D printing to create parts that are of controlled quality and whose production is governed by the performance of the part and not just its appearance. This advance is critical for the continued development and proliferation of 3-D printing into medical and other regulated industries. - said Vadim Pinskiy.
Commercially available 3D printers typically offer only high-speed, high-precision, or high-quality printing. Rarely do they offer all three simultaneously, which limits their suitability as a production tool. Today, 3D printing is used primarily for prototyping and small-scale production of specialized parts.
Solving this problem, Inkbit, an American startup hailing from the Massachusetts Institute of Technology, is working to bring the full benefits of 3D printing to many products that have never been printed before - and aims to do so on a scale that will radically disrupt manufacturing processes across industries.
The company achieves this by combining its multifunctional inkjet 3D printer with machine vision and machine learning systems. The machine vision system comprehensively scans each layer of an object during printing to correct errors in real time, while the machine learning system uses this information to predict the deforming behavior of materials and make more accurate final products.
The company says it can print more flexible materials much more accurately than other printers. If an object, including a computer chip or other electronic component, is placed on the printing area, the machine can accurately print materials around it. And when the object is finished, the machine saves a digital copy that can be used for quality control.
Today's manufacturing processes use a variety of materials, some of which are difficult to 3D print, resulting in uneven distribution and printing process failures such as clogging. They also tend to shrink or round at the edges over time. The printer developed at Inkbit is capable of printing a record 10 materials at a time using machine vision technology. At first, the engineers used a simple 3D scanner to track the progress of the work. For Inkbit's first printer, they decided to significantly improve the "eyes" of their machine and use an optical coherence tomography (OCT) scanner, which uses long waves of light to see through the surface of materials and scan layers of material at a resolution no greater than a fraction of the width of a human hair.
Since OCT scanners have traditionally been used by ophthalmologists, the only devices available were too slow to scan every layer of the 3D printed part. So the Inkbit team developed their OCT scanner, which they say is 100 times faster than anything else on the market today.
When a layer is printed and scanned, the company's proprietary machine vision and machine learning systems automatically correct for any errors in real time and proactively compensate for deformation and shrinkage of impermanent material. These processes further expand the range of materials the company can print with, removing the rollers and scrapers used by some other printers for accuracy, which tend to jam when used with difficult materials.
The system is designed to allow users to prototype and make new objects on the same machine. The current Inkbit industrial printer has 16 printheads and a print unit large enough to produce hundreds of thousands of fist-sized items (or fewer larger items) each year. The machine's non-contact inkjet design means that increasing the size of subsequent iterations will be as easy as expanding the printing plate.
The company currently has only one production-grade printer in operation. But it will begin selling products printed with it in the near future, starting with a pilot project with Johnson and Johnson, and then bringing its own printers to market as well.
Vadim Pinskiy is the vice president of research and development at Nanotronics, where he oversees product development, short-term R&D, and long-term development of AI platforms. Vadim completed his doctorate work in neuroscience, focused on mouse neuroanatomy using high throughput whole slide imaging and advanced tracing techniques. Previously, he earned his master’s in biomedical engineering from Cornell University and his bachelor’s and master’s in electrical and biomedical from the Stevens Institute of Technology. Vadim is interested in applying advanced AI methods and systems to solving practical problems in biological and product manufacturing.