We’ve got so much happening here at Adafruit that it’s not always easy to keep up! Don’t fret, we’ve got you covered. Each week we’ll be posting a handy round-up of what we’ve been up to, ranging from learn guides to blog articles, videos, and more.
Are your backyard games of cornhole or horseshoes getting pretty serious? It may be time to build a professional-looking scoreboard!
Use the Matrix Portal and LED matrix display along with Adafruit IO to score a friendly competition between two players or teams! You can use your smart phone’s web browser to adjust the scores in real-time. All coded in CircuitPython.
We hear a lot about engineering hardware and software and other accompanying technologies for 3D printing, so the idea of going in reverse may raise an eyebrow or two; however, scientists from the NYU Tandon School of Engineering are using machine learning and reverse engineering to test vulnerability in 3D printing toolpaths.
Security in 3D printing has been an ongoing concern for years now, and the focus of numerous different research studies. On a more topical level, there are worries about criminal factions using the technology for evil purposes like fabricating skimmers, making guns for nefarious purposes, and even 3D printing packaging for illicit drugs. On a much deeper, more analytical level, there is vulnerability to cyberterrorism, whether in tampering with critical parts for aerospace applications, creating product defects and causing safety issues and liability, or even interfering in military operations.
The researchers, led by Nikhil Gupta, a professor in the Department of Mechanical and Aerospace Engineering, enlighten the public on worries that most 3D printing users would never consider: the potential for stolen trade secrets through analysis of layered materials. Gupta and his researchers have been tackling this issue for years now too, examining risks throughout the online world, but with an emphasis on the potential for cyberterrorism in 3D printed parts.
For 3D printed parts to offer functionality and high performance, many factors are “fine-tuned,” and this is what an interloper could uncover in analyzing toolpaths contained in CAD files; in fact, the researchers consider much of that data to be easily copied and stolen.
“A dimensional accuracy with only 0.33% difference is achieved for the reverse engineered model,” stated the researchers.
Also working on the project were NYU Tandon grad students Kaushik Yanamandra, Guan Lin Chen, Xianbo Xu, and Gary Mac, demonstrating that fiber orientation can be intercepted with micro-CT scanned images. Loss of trade secrets means stolen intellectual property in most cases, along with what could be substantial investments in research and development costs too.
While spying via 3D printing presents obvious gray area regarding legality, theft of intellectual property is often taken much less seriously outside of the US—with countries like China being known for their irreverence toward IP law.
“Machine learning methods are being used in design of complex parts but, as the study shows, they can be a double-edged sword, making reverse engineering also easier,” said Gupta. “The security concerns should also be a consideration during the design process and unclonable toolpaths should be developed in the future research.”
MX3D’s steel bridges are an inspiring sight to see, but, even if bridges are what the Dutch firm is known for, they are not the only thing the firm is capable of making. The company now has released a new 3D printed robot arm component made with its metal AM system, which relies on an industrial robotic arm of its own.
Made together with industrial automation company ABB and software simulation firm Altair, the new arm has been optimized by the Altair team working in conjunction with MX3D. Altair’s generative algorithms were not only used to cut part weight in half, but also to improve toolpath planning on the printer to increase the print speed. The total print time was four days and connecting surfaces were finished on a three-axis milling machine. The part has now been installed and is in use on an industrial robot.
Are we seeing larger-scale 3D printing coming into its own? Firms are bridging the gap between the virtual and real-world through connecting data to optimized toolpaths, designs, and parts. Driven by resolution limitations, difficulties of working with industrial robots (lack of memory, proprietary syntax), and a strict regulatory environment large scale firms are turning to software to solve their problems.
We’re seeing a remarkable difference between the “house printing” companies—who seem, on the whole, to be rather optimistic and cavalier about their endeavors to print buildings—and the large scale part printing cohort of enterprises. The latter, which includes MX3D, seems much more in tune with regulatory requirements, certification, and software than the former. Perhaps, because you can’t really sell a bridge ex-works, while a demo house doesn’t have any regulatory requirements, so the parts builders have been put onto a more difficult digital path.
But, through controlling toolpaths, FEA, weight reduction, and using this as a tool to try to get parts built correctly, companies have been forced to deal with these things early on in their machine and process design stages. This, in turn, has led to them being better placed to build actual parts for the actual world. Meanwhile, the “housebuilders” are building much larger more media-savvy structures that have yet to be subject to many thoughts on how they will be built safely.
In 3D printing for construction, it would seem that the earlier on your business model encounters regulatory opposition, the earlier you will design safety, reliability, and repeatability into your process. Logical perhaps, but not something considered so far by the industry at large. One will expect however that the “go big or go home” crowd will seem to be ahead initially, but then take much longer to develop process control once they start building parts that will go on the open market and touch the realities of such arcane and frightening things, such as the law.
Whereas houses may be the best clickbait, there are myriad of other parts that can be built with robot arm construction systems through 3D printing. Generally, we can see that our market does nanoprinting on the submicron and micron-scale (femtoprint, nScrypt), microprinting on the mm to micron scale (3D Micro Print), regular 3D printing which starts from several mm parts to around 50 cm parts (RepRap, Ultimaker), medium format printing which is for parts of up to one cubic meter (BigRep, Builder), large format 3D printing for parts from one cubic meter to around ten cubic meters (CEAD, BAAM) and macro 3D printing which is parts that are larger than 10 cubic meters (3D Printhuset).
At each and every scale we can see a strange thing happening. Scale drives accuracy which drives value which, in turn, determines go-to market and that determines the level of quality leveled at the part. This is super logical in the sense that small things often have to be precise in order to exactly fit small assemblies, which in turn are likely to be a part of something complex that needs high tolerance—a watch, for example.
At the same time, if you can make things that are 1 mm x 1 mm or less, then a stent is something that you can do and you won’t think of car bumpers. Of the total set of things sold in the 1mm x 1mm x 1mm range, often a disproportionate number of these things actually have high value due to their precision manufacturing requirements.
This is, again, logical but could go against the conventional wisdom that more material equals more expensive production cost or the “rule of most things” that stipulates that larger things are typically bigger. In the mid-ranges, there also seems to be an ongoing effect whereby, if the things that you print are likely to be the same size as inexpensive manufactured goods but are more difficult to make, larger and smaller things can vary more widely in price. Production difficulty, in large or small structures, drives price and applications, as well. I’m not saying that size is solely deterministic, but we are seeing effects here.
On the micro- and nanoscale, quality systems are adopted rapidly by participants due to their adjacency to the medical business. If medical is the most profitable thing you can do and just about the only thing you can do, you’re going to end up having a cleanroom. Meanwhile, it took a long time for a lot of service bureaus to turn to ISO, and desktop machines are currently still sold with a warranty that scarcely lasts past the UPS carrier’s hands. Now increasingly, quality systems and certifications are being adopted by desktop companies and service bureaus. In larger-scale things, we’re seeing medium format start to look at quality now.
Many of us are familiar with the innovator’s dilemma, whereby a large volume good enough product displaces a better more expensive earlier one. Could we in 3D printing see a similar effect where higher quality systems engineered for smaller sizes could displace established entrants with larger sized parts? If Prusa and Ultimaker were good at precision in the 10-cm range, wouldn’t it be fairly easy for them to scale their systems on the back of their existing installed base?
Crucially, they wouldn’t have to adapt all systems completely, but just make some components stronger to reach the next size of medium-format machines. If they jumped to the Cincinnati BAAM category, of course then they’d have to completely re-engineer everything, but the adjacent category would be simple for them to do. But, for them to work at the microscale would mean a lot of adjustments to their current design and manufacturing of hardware components as well as working in a higher quality standards way.
This leap would be daunting, especially since the volume of products made with the smaller category would be less than with their own. Furthermore, they could expect to sell less material and fewer machines in the smaller size category, but more material and fewer machines in the one-size larger category. Especially consumables driven firms or companies such as polymer firms will benefit from more parts, faster print speeds and larger sized parts. The sum total of these effects could indicate pressure on firms to move into larger scaled manufacturing all the time, but ignore smaller scales.
If we look at MX3D for example, we may think of its bridges which it may sell in the hundreds if it got them right and could certify them. But, MX3D also can sell many more smaller components at larger volumes as well. Its Takenaka connector for example needs precision, but this component could sell in its thousands. Bike frames need to fit with precision components, such as derailleurs, and the precision and volume required for these components can drive its other businesses. Operational advantages gained here could be used to earn margin on larger components, such as bridges, that few can make. It seems blindingly obvious if we compare it to bicycle companies moving to passenger cars and then sometimes to vans and sometimes to trucks. This development seems to be a very similar one.
If this holds true, then for MX3D, the future could be in making many medium-sized parts for a larger scale future. In Dutch we have an expression, “wie het kleine niet eert, is het grote niet weed”, which means, “he who does not honor the small things does not deserve the large.” For 3D printing, this expression may hold very true indeed.
Focusing on 3D-printed materials for construction, Bagheri and Cremona assess the potential for machine learning. Experimenting with geopolymer samples and different compositions, the authors evaluated target variables in machine learning. They began by looking at the compressive strength of geopolymer binders and the elements involved, to include:
Features of raw materials
Chemical composition of the aluminosilicate resources
Formulation of the alkaline activator
Alkaline ions in the activator
Fraction of silicate to hydroxide compounds in the activator
Water to binder ratio
Formulation of aggregates
Upon 3D printing, factors grow to include:
Shape of prints
Rates of extrusion
Preparation and formulation of materials
“Given an innumerable number of independent variables, the prediction of the compressive strength of printed geopolymer samples without the use of a machine will generate a high level of error,” stated the researchers. “For instance, one can predict the strength of samples that are classified into four categories with 75% error. However, the use of machine learning would reduce this error significantly as can be seen further in this work.”
Current data offers benefits to researchers as they are able to learn more through printing variables and changing parameters:
“Among the mentioned effective parameters, the content of the fly ash, the content of the ground granulated blast furnace slag (GGBFS), as well as the ratio of boron ions, silicon ions, and sodium ions in the alkaline solution have the most significant impact on the compressive strength,” stated the researchers.
A small 3D printer was used to fabricate samples for the study, consisting of a piston-operated extruder. The researchers used vibration to make sure the mix was compacted, with resulting sample dimensions of 250x30x30.
Statistical summary of the input data
Target data classes
Slag was found in the geopolymer mix, and also displayed better compressive strength; conversely, samples with more sodium showed decreased compressive strength.
DT flowchart of the ctree function
Increased boron raised sodium ions, while lessening compressive strength—with the same shown in terms of lesser slag content too. Silicate is also a critical ingredient for strength development and cross-linking.
Confusion matrix of ctree function based on actual values
Confusion matrix of ctree function based on predicted values
Ultimately, Bagheri and Cremona discovered the true prediction value to be 63 percent.
Confusion matrix of rpart function based on observations
Confusion matrix of rpart function based on predictions
“The predictions could be compared in two efficient ways. First, the simplicity of the model could be assessed based on the predictions rules and comprising the number of parameters. Accordingly, rpart function is far more uncomplicated with only two parameters for 50% of the predictions and three parameters for another half,” concluded the researchers.
“Whereas, ctree function used four factors for 74% of the predictions and two factors for only 26% of the predictions. Secondly, the cumulative accuracy of each prediction function was used as a comparing criterion. The cumulative accuracy factor was obtained by multiplying the number of predictions in each category and the appropriate positive predictive value.
Acquiring 70% cumulative accuracy for rpart function with respect to 63% for that of ctree function evidenced similar but slightly better performance for rpart function to predict the compressive strength of 3D-printed boron-based geopolymer samples. Moreover, the importance of the percentage of slag and the ratio of boron ions can be seen in the decision trees created by ctree and rpart functions, respectively.”
3D printing in construction continues to be of growing interest, with the potential for homes, offices, and even entire villages to be built with a variety of different printers and materials. What do you think of this news? Let us know your thoughts! Join the discussion of this and other 3D printing topics at 3DPrintBoard.com.
Comparison of the results: from laboratory test to machine output
Naftaliev first came to our attention when he published work showing us how to take a 2D image and turn it into a possibly 3D-printed file. Creating 3D content, moreover 3D printable 3D content is difficult. CAD is still too complex for most people and 3D scanning works, but is finicky. Everyone can either draw or take photos that lead to digital 2D content, however. If we could easily take 2D data and make it 3D printable, one could much more easily make their own 3D-printed products, let consumers mass-customize things or make custom-fit things like shoe soles.
So, Naftaliev’s work on the cutting edge of AI and 3D printing is, to me, of potential crucial importance to the future of 3D printing. Eventually raw computing power, improving cameras and better software could help us get to a stage where all of our phones are 3D scanners that can be used to create 3D content easily. Until then, and also subsequently, AI and machine learning could let us take much more content and make it 3D.
Machine learning and AI, however, are kind of like a magical sauce that is supposed to make everything better for everyone all the time. I remember when 3D printing was seen in the same light. I personally tried to be a realistic, enlightening, but not dazzlingly optimistic guide for people through these hype times. For AI and the intersection of machine learning and 3D printing, Naftaliev is this person for me. Not for me alone, however.
“image processing, AI, 3d modeling, technological advancements . But, more importantly, it is a step to try to democratize access to information to anyone around the world. Academic research and papers can be very hard to figure out even for people who are working in the subject. Reading just one paper and truly understanding what is going on can take several good days of work and sometimes requires access to people who have knowledge in the field. And, what’s more, a lot of the research does not come with an open source where you can try to test things out yourself (it is extremely hard to replicate the code and results of a paper, if not impossible because of access to training data and computational resources). The research that does come with code many times is still hard to figure out, sometimes there are bugs or things in the code that do not align with the research. I want to help make all of this more accessible to people everywhere.”
He feels that “if I or anyone else has put the time and effort to understand some new research that is out, why not share it with others.” He does each live lecture twice, once for the east side of the planet, once for the west. He then offers these lectures for download. The next lecture deals with generating art using neural networks.
In the future he hopes “to get the authors of the most important researches in our field to come and present their papers, code and the latest advancements – live, online, for anyone who is interested in learning more.” The lectures are clear and super interesting but not necessarily for casual viewing, so paying attention helps. Naftaliev means for them to be for,
“People anywhere in the world with technical orientation who are interested in machine learning, or are already full-fledged practitionersresearchers who want to expand their understanding of sub-topics in this sphere. We are also touching the boundary of digital art, so people from the digital arts that want to see what the latest technological research in the field can do and how they can use it for their art.”
In terms of background,
“Mathematical and programming background is a plus. We do explain basic concepts in machine learning if we see that the audience is not fully familiar with them. Every participant who signs in to listen to a lecture fills out a small bio about himself so we know how much intro material we need to explain and how deep we can dive.
Naftaliev hopes that you can learn,
“Which papers and open source findings are interesting and relevant, the current state of the art results and how to replicate them, current technological limitations in industry and academia, expend your horizons about what is possible to achieve with AI in everything to do with image processing, 3D modeling, signal processing and more. I am also experimenting with allowing people to get to know each other and form connections around the world by sharing these common interests.”
You can find the YouTube channel here. Below you can see how you can go from 2D to 3D using neural nets.
I think that this is fascinating and, with Naftaliev’s help, you can be transported to the cutting edge of making 2D 3D and understand more about machine learning. I really think that this is an emerging frontier for our industry and am very grateful that Naftaliev will be giving this series of lectures. Subscribe here.
We’ve got so much happening here at Adafruit that it’s not always easy to keep up! Don’t fret, we’ve got you covered. Each week we’ll be posting a handy round-up of what we’ve been up to, ranging from learn guides to blog articles, videos, and more.
A Poster of CAD Design Tips for #3DPrinting #CAD @BillieRubenMake
“Maker of many kinds, serial skill-collector, and STEM-student,” Billie Ruben, has created this very useful little poster of do’s, don’ts, and other design considerations for creating CAD files for 3D printing.
You can download a large version of the poster, along with two other 3DP tips posters she’s created here. While on the Imgur link, don’t forget to check out Billie’s other two posters with more tips on what modeling programs to use for different types of 3DP and the ins and outs of 3D printer bed leveling. Read more!
Museums today are expanding public access to their art collections, not just inside the walls of the museum but also outside. Digital initiatives are bringing artwork, once relegated within the confines of the museum, to a 21st century global audience. These modern museums have essentially become the new content providers. Much like the modern companies Netflix and Pandora that provide video and music content, museums are becoming their own content providers with their collection of paintings, photographs, jewelry and other media. Digitizing these collections and making it publicly available brings this material to a global audience.
One such museum that is at the forefront of this endeavor is the Cleveland Museum of Art.
This project accesses the CMA API using CircuitPython running on the PyPortal. It uses the API to pick a random item from the collection. It converts and resizes the JPEG image from the collection to a BMP image sized to the PyPortal using Adafruit IO image conversion service. Finally, the converted image is downloaded to the PyPortal for display. A new feature of the PyPortal allows the correct scaling of portrait images, which is a great feature for this project. Learn more!
Machine learning is only about as complicated as cereal and marshmallows in this guide.
This project from Google uses a laptop’s built-in camera to identify various cereal and marshmallows. The computer then sorts them based on a model you train. A Circuit Playground Express communicates with the computer to decide when to sort which marshmallow/cereal via a micro servo.
Codecademy, an online interactive learning platform used by more than 45 million people, has teamed up with the leading manufacturer in STEAM electronics, Adafruit Industries, to create a coding course, “Learn Hardware Programming with CircuitPython”. Starting today, the course is available in the Codecademy catalog.
In this project we’re building a rig for installing heat set inserts. Use 3D printed parts and hardware to build a solder rig with smooth linear roller action! Make perfectly straight inserts with precision using a tip for installing inserts.
Make a light up Master Sword from the Legend of Zelda! This build has motion activated sound effects and LED animations!
The Adafruit Feather and Prop-Maker FeatherWing has you need to add lights and sounds to your projects. NeoPixel LEDs are fitted inside the blade.
This uses the built-in accelerometer and audio amp. When you swing it around, it’ll make different sound effects. On heavy hits it makes flashes and fade the colors of the LEDs.
It has pulsing animation and an idle sound effect that loops in the background. You can make this fit your project by customizing the colors or adding different sound effects. You can recharge the battery or even add new sounds with the USB port. The sword mounts on a computer, just like a USB Drive!
The Raspberry Pi 4 Model B is the newest Raspberry Pi computer made, and the Pi Foundation knows you can always make a good thing better! And what could make the Pi 4 better than the 3? How about a faster processor, USB 3.0 ports, and updated Gigabit Ethernet chip with PoE capability? Good guess – that’s exactly what they did!
The Raspberry Pi 4 is the latest product in the Raspberry Pi range, boasting an updated 64-bit quad core processor running at 1.4GHz with built-in metal heatsink, USB 3 ports, dual-band 2.4GHz and 5GHz wireless LAN, faster (300 mbps) Ethernet, and PoE capability via a separate PoE HAT.
Stratasys and DSM Venturing (venture capital arm of Royal DSM) lead the way in yet more financing for startups, acting as the major sources of funding support of $12 million total, in an equity round for Inkbit—the 2017 spinout of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. Headquartered in Medford, MA, Inkbit’s technology is based on research led by Professor Wojciech Matusik which has evolved into a vision-based, AI additive manufacturing platform.
“As pioneers of jetting-based additive manufacturing solutions, we are excited to help Inkbit bring their technology to the factory floor. Vision-based feedback control and artificial intelligence will take additive manufacturing to a whole new level and will help to enable its widespread use for production,” said Ronen Lebi, Vice President of Corporate Development at Stratasys.
Up until this point, a lack of technology in the form of inkjet 3D printers has held back many innovators seeking superior performance—and this means improving reliability, accuracy, and the need for planarization. Known as the company that created the first 3D printer offering vision-based feedback control and artificial intelligence, Inkbit technology is unique due to the inclusion of machine vision and machine learning.
The first 3D printer created by Inkbit. (Image: Inkbit)
“We are excited to partner with such an extraordinary team of industry-leading players and impressed by their entrepreneurial spirit and commitment to innovation,” said Davide Marini, Inkbit cofounder and CEO.
“The composition of this syndicate was chosen to maximize the speed of development and commercialization of our platform, with each investor bringing to us their unique expertise in equipment manufacturing, high-performance materials and applications in robotics, medical devices and life sciences tools. Our value proposition to customers is simple: we are adding a layer of machine vision and machine learning to material jetting, increasing its accuracy, reliability and enabling its use with production-grade materials.”
With this latest infusion of cash, Inkbit plans to industrialize their AM system further by:
Integrating multi-material and volume manufacturing requirements
Expanding the set of materials for medical, life sciences, and robotics applications
Installing the first units for customers
Currently the Inkbit team is working with customers like Johnson & Johnson in an early access program, with a release date of 2021 set for select customers.
“Materials always play a major role in industrializing breakthrough technologies and in additive manufacturing they become absolutely critical. We are delighted to have Inkbit in our investment portfolio and look forward to helping them develop the best materials for customers world-wide,” said Pieter Wolters, Managing Director of DSM Venturing.