// // Leave a Comment

Machine Learning Making Real-Time Fixes to 3D-Printed Metal Parts

Machine Learning Making Real-Time Fixes to 3D-Printed Metal Parts


Using Real-time video and neural networks with advanced algorithms - fixes to metal 3D parts can be applied 'Before they are even built'.


Over the years 3D printing provided the industrialists, SME's and the gadget enthusiast with many false dawns. Limited materials and reliability issues coupled with software not developing at an exponential enough rate to complement the visionary ideas all contributed to the venture capital and general investment being allocated elsewhere and 3D printing 'Addictive technology' fading from the public consciousness.


However, over the past five years, the average SME and tech enthusiast at home has had to be content with the advancement of digital print and cloud-based software. The transformation that the printing industry is currently undergoing from traditional to digital print has seen significant improvements in toner cartridges, printer inks and 3-in-1 device technologies.


However, the 3D printer scientists and developers have been making advancements that are light years ahead of anything we have seen in the average office building during that time.

While printer ink and printer cartridges and the like have been marketed to a point of saturation 3D printing has been quietly going about its business moving us closer to the future of manufacturing and piling even more pressure on the other side of the printing industry.

Cutting-edge solutions that are making industries take note are opening the doors for the creative minds to dream big once again of the possibilities where this technology can take us, but with a direction and purpose previously unseen.

Production – First Time – Every Time – A Manufacturer’s Dream 




Engineers have made a breakthrough analysing the physics of metal 3D printing using imaging techniques with sensors to increase the quality in metal 3D printed parts whereby they print first time, every time.




Now integration of machine learning can process data during the build in real-time and detect in milliseconds if the build is of high or substandard quality.

Also, they have developed (CNNs) Convolutional Neural Networks which act as an algorithm used in processing images and videos to predict the outcome of the build within ten milliseconds or less.

This is a considerable advancement as the previous analysis was conducted post print and proved costly if the part was inadequate. Besides, print times for metal components could take days or even weeks to print so with the introductory of CNNs it can push the industry forward in a quantum leap.

It will be able to detect an abnormality to adjust and fix it in real-time to produce perfect results every time, making this a significant breakthrough and further enhancing 3D printings reputation as the solution for manufacturers big and small.

The process that has been developed took over 2000 video clips of various conditions of laser tracks that were melted to see variations of speed and power.

A 3D height map tool scanned part surfaces, gathering information that trains the algorithm to analyse the convulsions (each section of the video frame). A process way too time consuming for humans.

So, the algorithms by using the same theory and model are then able to predict the build's width and what would be standard deviations. The algorithms provide data to analyse to determine the quality and whether the build will be acceptable. Statistically, the neural networks detection figures were 93% accurate.

Researchers have spent years collecting all kind of real-time data relating to the 3D printing process of laser powder bed-fusion metal including:

  •  Video
  • Optical Tomography
  • Acoustic Data

It became clear that it wasn't feasible to analyse the data manually and the idea of trying to incorporate neural networks could simplify the method and speed up production.

It was a case of connecting the dots, a little like the human brain uses vision to navigate, the same applies in navigating through the 3D printing process where the algorithms uses that as sensory data.

This could be a tip-of-the-iceberg scenario at theoretically the neural networks could be used in other 3D system using and following the same formula.

There is no reason why, with standard machine learning techniques collecting video, optimal topography and acoustic data, scanned with the height map couldn't generate the same requirement of information to be instantly acted upon.

The only perceived gap is there are voids in specific builds that can't be detected using a height map, but ex situ x-ray radiography could be a solution in measurement. The next phase is to look to use other forms of sensory technology other than video.

Conclusion

This is yet another breakthrough in the world of 3D printing, occurring on a regular basis. With the medical, aerospace, aviation, dentistry and fashion industries already well past research and development stage with 3D printing and with billions of dollars now readily flowing into all kinds of industries and a gathering attention from venture capitalists, the industry is once again on the move. 3D printing is answering the call for higher output and faster speed with less material usage and waste. It seems that finally 3D printing is losing its 'Hype' tag and will transform the world we live in as it has promised to do so for some time.


About Author

Heidi Kovic is a tech blogger with a keen interest in business affairs, she likes to travel and is a big into global cuisines. She is associated with Printzone who are leading stockists of printer cartridges and ink cartridges in Australia.

0 comments:

Post a Comment


Found Spell Error , Need to add more content use this form to Suggest Edit