Annie Wang was a relative latecomer to additive manufacturing (AM), having not been exposed to it until she was pursuing her master’s at the Wharton School of Business. It was during this time that she met her future business partner, Zach Simkin.
The pair quickly saw the potential of the technology and set out to learn as much about AM as possible, which led to their co-founding of Senvol LLC, where Wang serves as president. Their mission is clear—using data to help other companies effectively implement additive technologies. While this is easier said than done, there’s an added degree of difficulty—avoiding falling off the data cliff.
We’ll come back to that cliffhanger, but first let’s start with more about Wang and Simkin’s auspicious meeting and introduction to AM.
“It’s a funny story,” Wang confides. “Zach and I were both working on our MBAs and were in a class called Innovation. Our professor took out a MakerBot and said to the class, ‘This is a 3D printer.’” She laughs. “It was the first time we’d seen one or even heard about the technology. And since the class was about innovation, he then challenged everyone there to develop ideas of what they would make with it, given the chance.”
Their fellow students’ ideas ranged from jewelry and glasses to puzzles and games. Wang and Simkin were the only two to offer ones with a business-to-business, industrially focused angle: hers was spare parts, his was munitions.
“That’s how Zach and I first learned about 3D printing, and because we’re both very entrepreneurial, we decided to start our own company,” Wang says. “After graduation, we moved to New York City, where both of our spouses had jobs.” (For more on Simkin’s story, see “Driven by Data” in the October 2022 issue of Smart Manufacturing.)
Wang doesn’t recall what grade she received for that class. But given her track record, it likely starts with an “A.” Since then, she’s been an integral part of the team that has developed numerous data-centric tools for the AM industry. The first was the Senvol Database, which is a searchable (and free) online repository of AM machines and materials.
“I remember asking myself questions like, ‘If I wanted to find a 3D printer with a build box size of X, Y, Z that could print a specific titanium alloy, how would I do that?’ The answer back then was to spend hours sifting through spec sheets. Now it’s a basic query,” Wang notes.
To say that the Senvol Database has grown since its initial release is a gross understatement. While she doesn’t have the exact figures, Wang concedes it “wasn’t more than a few hundred records in all” in the early days. Today the website lists more than 1,900 machines and 4,100 materials—and climbing.
Those are big numbers, but it’s important to note that many of these have been discontinued and are flagged as such in the Senvol Database. Why keep old records? Simple, Wang explains, “As with any manufacturing technology, some people continue to use old machines and old materials. We support them by keeping the data they need to do their job.”
More machines and materials also mean more companies. And the industry continues to transform, Wang says, noting the first RAPID + TCT she attended in 2013 was a fraction of its current size.
“We’ve seen companies come and go—some failing, others getting eaten up by larger firms—but the trend continues to be one of tremendous growth and proliferation,” she says. “It’s a very dynamic industry, and I don’t see that changing anytime soon.”
With this dynamism comes data—lots and lots of data. Senvol has made it its mission to not only make sense of it all, but leverage it for the greater good. The company has built on its success with an application programming interface that allows customers to integrate their own software with the Senvol Database.
Senvol has also begun offering indexes—AM data sets that “would otherwise take months to develop” as well as standard operating procedures (Senvol SOP) that help ensure customers generate “pedigreed data” from their additive endeavors. Yet it’s Senvol’s latest product that has Wang most excited. Not surprisingly, it also revolves around data.
“Think about laser-powder-bed fusion,” Wang says. “Early on, people used all kinds of metal feedstocks, some of which worked well, while others didn’t. It was basically a shotgun approach.
“Since then, they’ve become more sophisticated,” she continues. “The industry realized that, of the dozens and dozens of parameters—many related to the feedstock, others to the machine itself—some play a bigger role in build success. The trick is determining which ones are most important.”
Whittling down these many variables—particle size and morphology, chemical composition, flow rates and layer thickness, as well as laser speed and power—is a tedious and costly exercise, after which the machine owner is left with a proprietary recipe that works for a specific application, but may or may not be optimal. The solution to this inefficiency, according to Wang, is machine learning (ML).
Conceptually, she notes, ML has been around for a long time, and the mathematics behind it actually precedes the birth of AM. But as with finite element analysis and other forms of engineering simulation software, the growth of ML and artificial intelligence as a whole, has long been hampered by a lack of computer power. That’s no longer the case.
“Back then, it would have taken months or even years to run some of the algorithms we and others are now using,” Wang says. “That’s not to say machine learning’s always fast, but it’s definitely more practical than it once was.”
As a result, there’s no longer a need for the trial-and-error approach to AM optimization. According to the Senvol ML product page, users can predict mechanical performance from a given set of process parameters and determine in advance what parameter values they should use to achieve specific performance goals. The software also “learns” as it goes, applying its knowledge to new or untested scenarios while guiding users with recommendations as to what data points they should be keeping an eye on for additional improvements.
“ML will reduce the amount of testing and tribal knowledge needed to optimize the AM process,” Wang asserts.
Due to strict confidentiality agreements, Wang is unable to share any recent ML success stories. However, she’s quick to add: “I know it saves them significant time and money.”
This last point illustrates a potential roadblock to AM proliferation: When every company is continually reinventing its own wheel and keeping the results for itself, the industry as a whole suffers.
However, Wang points out that there is some collaboration and information, albeit in an informal sense. This includes standards development committees, such as those sponsored by SAE and ASTM that Wang has served on. “It might not be in great detail,” she says, “but people talk about their experiences with certain machines or certain materials, and—as members of these groups—are encouraged to share the results whenever possible.”
There’s also the Additive Manufacturing User Group (AMUG) and America Makes, an organization that Senvol has worked closely with since its inception. Both promote knowledge sharing and education, as does SME’s Additive Manufacturing Technical Leadership Committee, which Wang recently joined.
These groups play a crucial role in the standardization, promotion and further development of additive technologies. But as Wang points out, so do many of the companies listed in her database.
“As AM has grown, we’ve seen a number of large suppliers like BASF and Carpenter Technologies take an increased interest in the industry and focus more of their efforts on making metals and polymers specifically for 3D printing. Similarly, HP and other equipment manufacturers are taking an active role in its development.”
The U.S. Army shows similar interest. “That’s been a very interesting project,” Wang explains. “In a nutshell, we are attempting to use machine learning and process parameter optimization to achieve identical mechanical properties in parts printed on different machines located at different sites or locations.”
It’s more difficult than it might sound. Wang tosses out a hypothetical situation where 300 watts of laser power delivers the perfect part, but a few watts above or below this target value can cause a build failure. And because all systems contain some small amount of error, two supposedly identical 300-watt machines might not produce the same results. It’s these types of inconsistencies that Senvol is working to alleviate.
“You need a robust processing window, one that accommodates these margins of error and doesn’t have what I call cliffs on either side of the ideal process parameter point,” Wang says. “That’s the great thing about machine learning, in that it actually maps out the entire processing space so you can understand and quantify that window and then make parameter adjustments to compensate; you’re avoiding the cliffs, if you will.
“Doing so gives you the desired results much more consistently, and without all the testing and fine-tuning that’s normally needed,” she concludes.