As additive manufacturing gains acceptance as a production tool, the size of a typical 3D print slice data file is predicted to grow cubically. Handling all of these data could become an issue for the additive industry. Harshil Goel, CEO and Founder of Dyndrite discusses potential solutions to this problem. Harshil is an inventor, engineer, and mathematician with degrees in Pure Mathematics and Mechanical Engineering from UC Berkeley.
Several aspects are culminating together and amplifying each other that lead to this data handling challenge. “It’s sort of a Moore’s law for manufacturing and that if you look at manufacturing printer bed sizes over the last few decades, they are growing in size,” notes Goel.
Average build sizes were less than and up to 250 to 400 millimeters cubed. Now several machines are over one meter cubed, so the volume has grown cubically. Because of that, the amount of data that goes into an additive system for a build is growing.
This situation is compounded by the fact that the resolution of the machines is also increasing. And the amount of metadata you need to supply to the printer to properly print the files is also increasing. “It’s just sort of this cascading phenomenon that’s causing what I’ve dubbed a data explosion issue. These factors are interrelated, and they’re all stemming from printers getting more robust as they shift from prototyping into production, and basically, the software really hasn’t kept up with it.”
One drawback of the data deluge is that it slows the additive machines’ speed when making parts. “The CEO of SLM said at one point that their machine could literally print faster than anyone could prepare files for it. What we’re realizing is unlimited complexity and on-demand with high mix, high volume designs are actually compute limited. It’s not hardware limited anymore.”
Now, this is not a new profound problem. The makers of dot matrix printers from years ago developed a solution. “They enabled a software program known as postscript, which is an interpreted language for taking three-dimensional data and sending it to the machine. And the key to postscript compared to what was before was that it computed the data that the printer needed on the fly. And because it could do that, it could compute things at the resolution of the machine.
“So, if you’re printing a really big banner or a book, it would always compute it at the max resolution of the machine. And if you take that analogy and bring that into 3D land, you realize we’re literally talking about the same problem. We have printers, they’re now 3D instead of 2D. And we have high resolution, effectively perfect data, namely CAD data, instead of fonts. And why not then compute everything at the resolution of the machine, on the fly, with a Digital Front End (DFE). So ironically, we don’t actually need a fundamentally new solution, all the pieces and components of what we need to make that solution already exist. It’s just recognizing and learning from the past and putting them together again in the right way.”
A Digital Front End
A Digital Front End is a compact solution of software or firmware. “It is a computer, it’s GPU’s, and it’s even potentially controllers.” It’s a high-performance computer with specialized hardware to run 3D printing devices. It can be included in these machines or it can live next to them. And it will serve as the number-crunching unit that feeds data to the machine.
“For example, in a server with an optical interconnect, there are different network topologies. But the key idea is compute is necessary for modern manufacturing and where the compute is located is going to be important. This harkens back to the postscript comment I made earlier. Adobe’s postscript system required all the printers to have a Motorola 68000 chip in it. So, this concept is not a new thing. The modern-day version of this is DFE with an Nvidia chip in it instead.
Key features of a DFE
A DFE is sending data but it’s also getting data out of the printer and using it. As 3D printing machines become more sophisticated with sensors, cameras, infrared, and other components, you might want to use that data for process monitoring and control. To compute that volume of data and then use it to tweak or modify the print while it’s happening, you need fast and sophisticated software and you need hardware to compute it with.
A DFE will become important for automation. “CAD data is a lower fidelity way of dealing with complex designs because you don’t need all the large mesh files. Also, the meshes, of course, are pathological in nature because you can only manually fix them, you can’t automatically fix them, which to me immediately disqualifies them from being a production process. But with the CAD data, you might want to compute it for specialized operations or automation. So maybe you have machines with special controllers that enable you to print splines and you want to use the CAD data natively on the printer to improve the fidelity of your manufacturing process.
“Or maybe the CAD data comes with specialized color information that tells the machine where to do things on different surfaces. And so you can actually delay the compute to when you need it as you need it and do it on the machine. For example, the CAD data might say, these surfaces are special. They need to have better surface qualities or finish because they’re going to make something where this thing is a label pad. And you just need to make sure that the thing that’s making it makes something that’s readable. There are different ways to do that. And by upscaling the compute and improving the fidelity of the data coming in, namely the CAD data, not the mesh file, you can dramatically, in my opinion, transform the workflow and basically help automate it and make it production ready.”
The DFE eliminates the need for other 3D printing programs such as 3MF and .STL. Because printing in a production environment will use CAD data. So there’s no reason to use 3MF or mesh files.
The additive industry has reached a point where further advances are delayed because of software.
“We’re at the unique point where the manufacturing hardware has actually surpassed the software, meaning people are literally designing things they can’t manufacture. And it’s because of the data problem. Whether they’re doing it in traditional CAD or topology optimization or whatever, the thing that we’re working to solve is the compute problem. If you develop a fantastic, crazy design, you will usually have to make it simpler to actually print it because the amount of time it takes to compute.
The state of AM with regard to manufacturing
Goel sees two sides to the state of manufacturing with AM. Additive manufacturing is already used in a production environment in a number of use cases. “We’re talking not onesy, twosy things. We’re talking about thousands and tens of thousands and hundreds of thousands of components being printed, for example, for the dental market. It’s also being done in the automotive market as well. So if you look around, there are actually quite a few things that are being done in that vein. So what this Digital Front End does though, in my opinion, is it opens the possibilities of where else production AM could be applied.
“Basically, it’s reducing the operational expenditure cost of doing additive manufacturing, which I think is little understood and not really considered a concept. Everyone is focused on how many million dollars or a thousand dollars machine costs. And they’re not considering the cost of time. Let’s say you load a file and it takes 15 minutes every day to load that specific file. You do that four times a day with 10 people in your company, a $100 to $200 burn rate. You’ve now wasted $20,000 a week loading in files. That adds up quickly. So having a system that is compute optimized, that is pushing the automation and then pushing honestly open standards, makes it easier for more people to qualify materials quickly. I think is hopefully going to accelerate the adoption of production, added manufacturing and more use cases.
Production versus prototyping
The question of when will additive manufacturing be manufacturing continues to be a concern for potential users.
“I feel it’s a marketing problem,” notes Goel. “For example, one common thing AM is known for in the industry is for the production of a lot of aerospace components. Especially in the rocket industry. Aerospace companies are spending millions of dollars producing things. And even though they have high double-digit percentage failure rates on the prints, they’re still spending the money because it’s worth it. So of course our goal is to reduce the failure rates so that it’s even more worth it. From a cost and money standpoint. In the consumer sector, dental applications and electronics applications, again, people are producing thousands and thousands of parts. I think what will happen in the next two to five years is people are starting to really figure out additive’s place.
“And we’ll start to use it, not in necessarily “let’s use it for everything kind of way,” but we’ll really hone in on where it fits, how it’s used and how it needs to be used. For example, maybe you don’t actually use the additive parts as the final component. Maybe you’re using a printed part as a sacrificial component, as a part of a production process. Not everything has to be black and white in that sense. And so I do believe though that additive is going to find its place, I think enabled by software. I think that’s the real bottleneck here over the next two to five years.”