This is the first post in our new series Made to Make, a series of unashamedly open, unabashedly transparent posts about what we’ve learned while working to prove to ourselves that we can make a great product (that’s also affordable) right here in America. The posts in this series cover the full history of our product and processes, from made to make—how we’ve made our aligners in the past all the way up to how we make them today.
Dental models and ClearCorrect go way back. From day one, they’ve been an integral part of our product and our process. Every aligner we make is formed on a dental model, and until recently, we included those models in every phase of aligners we shipped.
This January, we announced we’d no longer be including models with our aligners. It was a big change for us and not one we made lightly (we actually looked for ways around it for over a year). But as I hope you’ll learn through reading this post, when it comes to making a better product, we don't shy away from change. We do our best to embrace it.
There's probably nothing we’ve changed more in our process than our dental models. Their story is one of constant iteration, improvement upon improvement upon improvement. It's a great story that's taught us a lot about what it takes to make a better product. It begins over seven years ago, and like any great story worth its salt, it all starts in a garage.
All we needed to know about manufacturing models we learned in Paul’s garage.
You may or may not know Paul Dinh. He’s a man of many hats here at ClearCorrect. One of his lesser-known hats is that of ClearCorrect co-founder and inventor of our original manual manufacturing process.
Before ClearCorrect, Paul worked in his own lab as a certified dental lab technician. He made all sorts of orthodontic appliances, but clear aligners had become his specialty and his passion. With a little resourcefulness, a good deal of ingenuity, and decades of prior art on the subject for reference, Paul had devised a clever process for making aligners—entirely by hand—that could effectively and accurately treat both anterior and posterior teeth, both simple and advanced cases. This wasn't just any old manual process.
One day in late 2006, Paul walked me and Dr. Pumphrey (ClearCorrect founder and my dad; we call him “Dr. P”) through the basic steps of his process in a makeshift lab he’d set up in his garage:
- Create a stone master model for each arch from PVS impressions.
- Section the teeth on the master models with a saw and then float them in wax.
- Plan out the prescribed treatment in steps of movement. Each treatment step would be an aligner.
- Move the teeth on each master model one treatment step.
- Duplicate the adjusted master models to create stone casts.
- Thermoform the casts.
- Trim the thermoforms down to aligners.
- Repeat the process from step 4 for the next treatment step.
Now, there’s a lot I could write about what I saw that day, but pertinent to this post is step 5, creating the casts, the models we’d later start including in every box of aligners we shipped.
Paul used a molding process to create his casts similar to the traditional model duplication process practiced in most dental labs: First, he soaked the master models in water until they were fully saturated (wet stone was easier to demold later in the process). Then he made molds of the masters using impression trays and alginate. Once set, he demolded the masters from the alginate casts and poured up the molds with dental stone, creating duplicate casts. Finally, he cleaned the casts and readied them for thermoforming.
It took Paul about an hour per treatment step to create casts using the process he showed us in his garage. It was a time-consuming and labor-intensive job, but it worked, and it taught us everything we needed to know about making models—making anything, really:
- Use any process that works to start.
- Study it. Learn from it.
- Use what you learn to improve the process when needed.
Though inefficient, for the most part, Paul’s process worked. It worked for our first case, our first hundred cases, our first thousand cases. It worked and worked and worked.
Then one day, it didn’t.
Don’t reinvent the wheel—just make it plastic!
Like the company itself, most things we do are born out of necessity. Shipping models is one of those things. It’s something we started after Dr. P had a patient lose an aligner for the first time and we realized we had a problem.
As effective as it was, our original manual process definitely had limitations. A big one was that we couldn’t easily reproduce aligners for treatment steps we’d already made. Once we had moved the teeth on the masters, we had no way of accurately moving them back.
The only place a previous step possibly existed was in the duplicate casts from step 5 of Paul’s process, but that was never guaranteed. See, the problem was in the perishable materials we used to produce our casts: alginate for molding and dental stone for casting. Both worked well for immediate use, but neither held up well over time. The alginate molds would eventually dry out, and the stone casts, fragile as they were, would often break in the thermoforming and demolding process.
With orders for replacements stacking up—lost or broken aligners being an unavoidable certainty we learned—it was time to improve our process. We decided rather than changing everything, we’d focused in on step 5. Since the problem was perishable materials, we’d switch to non-perishable materials. Specifically, we switched from alginate to silicone rubber and from dental stone to a polyurethane plastic.
As a solution for dried out molds and fragile casts, it worked remarkably well. We could easily replace any aligner we’d previously made by simply re-thermoforming the plastic cast. If the plastic cast had somehow been damaged, we could easily remake it with the silicone mold. It worked so well, actually, that we were soon faced with a new problem: what to do with the thousands of silicone molds and plastic casts that were overtaking our then tiny lab.
Molds and casts were everywhere. They filled our shelves. They lined the walls. They were crammed into every spare nook and cranny we had. And just when the mounds of casts and molds had piled so high they were ready to topple, we got an idea. Maybe there was a way we could reduce our storage burden while at the same time adding some useful functionality to our product. What if we sent the casts to the doctors? We could ship them with the aligners. Our doctors could use them as study models for reference or as thermoforming molds to make replacements themselves if they wanted. And if our doctors chose not to make their own replacements, we’d still have the molds that we could use to make the replacements for them. It would give more control to our doctors and cut our storage needs in half.
Best. Idea. Ever. We thought so back then, anyway, and we started including plastic models in every phase we shipped.
Having made models the new co-star of our product, we next set out to make them actually look and play the part. As any of our veteran doctors can tell you, our early models were, in a word, crude. If they were ever going to become anything to brag about, we’d have to do a whole lot better than was possible with a
manual handcrafted process.
Dōmo arigatō, Mr. Roboto.
We affectionately called them "Japanese robots." Technically, they were tabletop CNC milling machines made in Japan, but in our eyes, they were as high-tech as R2-D2. It was mid-2008 and they represented a significant milestone for ClearCorrect, the start of a whole new era: mechanized manufacturing.
We had eight of the Japanese mills set up in the back room of our newly acquired 2200 sq. ft. facility. We’d moved into our new expanded digs earlier that year ready to go all in on computer-aided manufacturing (CAM), what we were certain was the future of ClearCorrect model fabrication.
It seems like as soon as we could get the machines out of their crates, we had them running non-stop. Twenty-four hours a day, seven days a week, they hummed away, milling 3” x 3” x 1” polyurethane blocks into dental models. One mill could produce a model in about an hour. With all eight going, we were kicking out an entire phase (eight models) every hour.
One phase an hour! Compared to the days it could take with our manual process, one phase an hour wasn’t just fast, it was light speed. It meant better models faster than we’d ever imagined. It meant faster service for our doctors than ever before. It was amazing. It seemed like overnight, we’d become a high-tech company, and we could barely contain our excitement about it. I recall telling more than one person to “picture it: we’ll fill manufacturing bays the size of football fields with these things.” These Japanese robots were going to take us places.
Oh, how quickly things can change at a fast-paced, high-growth startup.
After running for just 6 months, we learned something about our little friends: As much as they reminded us of R2-D2, they were nowhere near as resilient. Where R2 could take a laser blast and still get the job done, our little guys, it seemed, couldn’t handle the slightest bump of the table without going out of whack. They weren’t made to run as hard as we were running them, and only 6 months in—just 6 months!—the wear and tear was beginning to show.
CAM or “milling,” as we simply referred to it, was well-suited for our needs at the time even if our Japanese robots were letting us down. As a company that was largely bootstrapped, we needed a flexible and affordable solution for making models. Milling was exactly that. Unlike other technologies we’d looked into back then, milling was so incredibly open that we had choice—choice in machines, choice in materials, choice in software. (We're big fans of choice.) The wide variety of choice meant we could swap out pieces of the process until we got the best possible combination. And so that’s just what we did.
In 2009, we swapped out our eight delicate Japanese robots for two industrial-grade, American-made (German-named), 5-axis, high-speed milling behemoths. Each weighed about 4 tons and took up so much space that we had to rent the suite next door just to house them. Designed to mill steel with impressive speed and accuracy, they ate through the plastic blocks we fed them twice as fast as their Japanese predecessors, taking our time per model down to just 30 minutes.
After months of them proving themselves steadfast and reliable, we ordered the first two behemoths some friends. And then some more. And then more again. Eventually, we had 18 of the giants crammed into the neighboring suite where they sat running around the clock, formidable and imposing, and insatiably hungry for our particular flavor of 3” x 3” x 1” blocks. From rickety robots to greedy giants—the transition was an awesome thing to see.
Our new industrial mills offered unprecedented (and seemingly endless) speed and power. No matter how hard we pushed them or how fast we ran them—we eventually got the time down to just 12 minutes per model—they never tired.
As you can probably imagine, the speed and power was intoxicating. It was so intoxicating, in fact, that it wasn’t long before we were drunk on it. Too bad drunk people do stupid things.
How dense can you be?
Stan Lee should have made it, “With great power comes great stupidity.” That’s how I felt, anyway, as I did my best to listen to the fireman explain what had happened. It was 2 AM Saturday morning, April 10th, 2010. I’ll never forget it. Water puddled around my feet, dangling wires and fluorescent lamps swung around my head, the noxious stench of melted plastic assaulted my senses, and every word the fireman spoke sent me reeling.
“The fire was contained to this machine.” The fireman pointed to the charred bones of a former giant.
“The heat from the fire was extreme. It took out the ceiling.” He pointed to a mess of twisted wiring and framing that used to be a dropped ceiling. “And it damaged these other machines over here.” The safety glass in the neighboring machines several feet away had gotten so hot it had melted.
“The real damage came from the smoke, though.” He pointed to the soot covering everything, then walked me out the back door and pointed to the long line of opened doors running down the full length of the building. “It’s in every suite. They’re airing them all out now.”
“Looks like your guy tried to extinguish the fire himself.” He pointed to what looked like 3 or 4 exhausted fire extinguishers scattered across the floor as he walked me back into the building. “We had to flood the machine to stop the fire, so it’s no surprise he couldn’t beat it with those. It was burning some serious fuel.”
The “fuel” was the polyurethane shavings that had accumulated in the bottom of the machine as model after model was milled.
Stupid move #1: We never tested whether or not the polyurethane material we used was flammable. We assumed since it worked great for molding our models, it would work equally well for molding the blocks we milled into models. Later controlled testing showed the stuff to be as combustable as sawdust laced with gasoline. The smallest spark could ignite it.
Stupid move #2: At some point, to save time between runs, we started allowing the machines to fill with shavings rather than emptying them as frequently as we should have. With what was essentially 18 giant barrels of fuel crammed into a tiny space, we’re lucky the fire wasn’t much worse than it was.
Footage from a security camera we had installed in the manufacturing area just days before the fire provided the details the fireman couldn’t. In disbelief, I watched the whole scene unfold on a tiny TV in our HR office. The timestamp up in the corner of the screen told me it was just after midnight. A lone night operator was tending a resting giant while he bobbed his head to whatever tunes he was listening to through the white earbuds he had jammed in his ears. Behind him, seemingly content, a giant was feeding on one of its first 3” x 3” x 1” blocks of the day.
Suddenly, there was a small flash. Through the safety glass of the feeding giant, I saw sparks fly. Seconds later, another flash. More sparks. Lone Night Operator, attentive as he was to his resting giant, was totally unaware of the giant just behind him, the one getting rather upset, the one about to ruin his night.
Flash! More sparks and then a soft glow.
Astonished by what I was seeing, I started calling out to the screen like I was watching some bad horror flick where the college co-ed doesn’t see the serial killer standing right behind her, “Turn around! Look behind you! Can’t you hear that?! Take your damn headphones off! Behind you! TURN AROUND!”
Stupid move #3: We failed to tell Lone Night Operator that because the machines could possibly get bad G-code—the programs that told the machines how to move—he needed to be able to hear them running so he could stop them if he heard anything weird…like them crashing into themselves. Happy giants sound different than grumpy giants, but with headphones on, they all sound like Snoop Dogg.
The soft glow grew. Flames started licking the safety glass. Sparks were still flying. Lone Night Operator was still dutifully focused on his resting giant, still jamming away and tuning out the world.
Black smoke started to billow from the burning giant. I watched it waft across the aisle and engulf Lone Night Operator. A bit confused, a bit alarmed, he turned and, spotting the source of the smoke, sprang into action. FINALLY! He ran off-screen and returned moments later with a fire extinguisher. He approached the burning giant and hit its emergency stop. Then, well-intentioned as could be, he hit the auto-door button, flinging open the doors of the machine and...feeding the suffocating flames with oxygen. Whoosh! What had been a relatively docile fire became a roaring, raging inferno.
Knocked back by the flare-up, Lone Night Operator took a moment to re-muster his courage. He then pointed the extinguisher at the giant’s fiery mouth and attacked the flames. Nothing happened. Puzzled, he looked down at the extinguisher, fiddled with what looked like the handle, and attacked again. Still nothing. Apparently deciding that particular extinguisher was busted, he tossed it aside, ran off-screen, and returned with another extinguisher. He attacked again. Still nothing. Tossed it. Got another one. Attacked. Nothing. Another one. Attacked again. Nothing again. He looked at the handle of the latest extinguisher, fiddled with it longer this time, attacked again, and—success! He did his best to aim the frosty blast through the open doors of the machine, though he might as well have been trying to blow out the flames with his lungs for all the good it did. He dropped the exhausted extinguisher, and just before the screen went black from the thick smoke pouring out of the machine, I saw him pull his cell phone out of his pocket to presumably call 911 as he ran out the back of the building.
Stupid move #4: The extent of our firefighter training was putting fire extinguishers on the walls with signs pointing to them. We never trained Lone Night Operator how to use those fire extinguishers. Never told him that in order to make the extinguishers work, he had to first remove the safety pins.
Stupid. Stupid. Stupid. It’s all I could think for weeks after the fire. We should have known better—I should have known better. I couldn’t blame Lone Night Operator. It fell on all of us. Especially me. We had gotten careless, and it was unacceptable. It was time to improve things again.
We started by posting a safety officer who was also a volunteer firefighter in his off time. He made sure we had the right type of extinguishers and that we actually knew how to use them.
We established better safety protocols for manufacturing, which included losing the headphones and emptying the machines more often.
We built fail-safes into our G-code so it was impossible for a machine to crash, even if the code was bad. And we upgraded the settings directly on our machines so at the slightest hint of a crash—if the fail-safes, um, failed—the machines would hit their own emergency stops automatically.
Finally, we tackled the material. This one took months. It turns out, apart from being extremely flammable and perhaps a pain to mold ourselves, our polyurethane blocks had a lot going for them. Most importantly, they were the perfect density. Not so dense that they broke our cutting tools when we ran at high speeds, but not so soft that they collapsed under the pressure of being thermoformed. Matching that density ended up being incredibly challenging. We tested dozens of materials and finally settled on the closest match we could find: high-density polyethylene (HDPE).
HDPE also had a lot going for it. It met our density requirements, it was low-cost, it was widely available, it came pre-cut in 3” x 3” x 1” blocks, it came in a nice bright white color (which looked pretty slick in our bright green boxes), and of course, it was completely non-flammable. It had one drawback we had to live with, though: It required deburring.
Deburring is the process of removing burrs, the cuttings that don’t fully separate from the model during milling. Our HDPE models came off the mills covered with the things. A quick pass with a stiff-bristled brush took care of most of the burrs easy enough, but it meant adding a process between milling and thermoforming where there hadn’t been one before. Initially, we could deal with it. Eventually, however, as our case volume grew, the amount of labor and time required to deburr tens of thousands of models simply became unsustainable.
As great (and great looking) as HDPE was, we all knew it would never work as a permanent solution; deburring just one of those bright white models would tell you that much. But what no one knew was just how radically we’d change our process to find a better one, or how far we’d go to keep it secret—top secret.
Quiet, please. Skunks at work.
If you walked by the building, you’d never have guessed what was hidden inside. Not this building. Bars covered the windows. Scrap metal and old tires littered the parking lot. And the neighborhood...well, let’s just say it would have given Mister Rogers conniptions. No, if you saw this place, you’d never have guessed that just inside, just behind the facade of filth and grime, there was hundreds of thousands of dollars worth of high-tech equipment silently running, silently redefining ClearCorrect’s future.
It was early 2011, and our R&D team had found the place at my request. They had been working on something so radical, so disruptive to our existing process—to our existing thinking—that I’d asked them to keep quiet about it until they knew it would work for sure. No use getting everyone worked up over a bunch of nothing. So they’d found this rundown, out-of-the-way place and set up what amounted to ClearCorrect’s very own skunkworks division. Their one and only objective: figure out how to make 3D printing work for ClearCorrect.
With its increased coverage in the news lately, it would be easy to think 3D printing is a new technology, but it’s actually been around, in one form or another, since the early ‘80s. Of all its advances since then, the most exciting to me is how accessible it’s become.
Until recently, 3D printing as a manufacturing process was simply too expensive for us to implement at our scale, regardless of how incredible it was. Though we’d considered several forms of 3D printing early on in our history, it had always been cost-prohibitive. It generally meant buying really expensive machines and using really expensive, really proprietary material. Compared to our entirely open milling process, where we had choice in machine and material and could produce sufficient results, it just couldn’t compete.
But by 2011, that had all changed. For one, we’d grown quite a bit by then and had the resources to support the transition from milling to 3D printing, but additionally, an Israeli company named Objet (today they’re known as Stratasys) was changing the 3D-printing game so smaller players like ClearCorrect could at least afford a seat at the table.
Now, no process is perfect, not even 3D printing. All processes have their own advantages and disadvantages. While Objet had made 3D printing more accessible for us, we still needed to figure out how to make it work better than what we were already doing. And frankly, that wasn’t an easy thing to do. We had hundreds of thousands of hours of experience milling hundreds of thousands of models. We knew how to turn a block of plastic into a dental model. We’d gotten pretty good at it. If we were going to make the switch to 3D printing, we had to get that good with the technology in a fraction of the time.
And so we found ourselves in that grubby hole-in-the-wall of a lab. Our R&D team (with some friends from Objet) basically lived there. They worked around the clock, fiddling with settings, learning the machines inside and out, working out workflows, proving out and abandoning theories, and essentially condensing months of trial and error down to a few weeks. When they were done, we had a process we could start with. It certainly wasn’t perfect and it wasn’t superior to our milling process in every way, but it was better in enough meaningful ways that we decided to pull the trigger on a full-blown transition from milling to 3D printing.
Over the following months, we swapped out mills for printers until every model we made was 3D-printed. As our case volume continued to grow, we added more and more printers, learning more and more about them along the way. When we made the move to Round Rock, we built out our space as a state-of-the-art 3D-printing facility, including everything we needed to maximize our use of the technology. And today, we’re one of the largest users of the technology in the world, making more models more accurately than ever before and more efficiently as well—our average time per model is just 6 minutes these days.
So if our models are so great now, why’d we decide to stop including them in our phases? Well, we do a lot more than make models, of course. Models are just a byproduct of the aligners we make. Similar to how we’ve worked to advance our model fabrication process over the years, we’ve also worked to improve our other processes. While we’ve tried our hardest to contain those improvements to each process, sometimes there’s just no way around changes to one process impacting another. In this case, it's improvements to a robot named Abby and her temperamental laser that get the blame.
Should we cut with a laser or focus like one?
As you might know, a while ago we started trimming our aligners with a robotic laser-trimming system. It's one of the most advanced systems we have on our production floor. At the core of the system is Abby, a bright orange ABB robot that we love. (She's an actual robot; nothing like those Japanese knockoffs we used to have.) All day and all night, aided by her human counterparts, Abby picks up thermoformed model after thermoformed model and positions it with robotic precision in front of a CO2 laser as she contorts to trace unique gumline after unique gumline. Seeing as no two models are ever the same, we rely on a bit of software wizardry to teach Abby on the fly exactly how she needs to move every time she picks up a different model. (It's pretty cool.)
Abby does a great job for the most part, and rarely complains. Her laser, on the other hand, likes to be moody. I don't know…maybe it's because we never named it.
Anyway, if you've done a case with us in the last year or two, you've probably noticed some slight surface scarring on the models we've included with your aligners. It's usually just below the gumline and often most noticeable across the front of the models. That scarring comes from Abby and her laser. If your models don't have that sort of scarring, it's likely Abby never touched them.
Now, as advanced as Abby and her laser are, they can be better. Particularly her laser. Just like we did with our model fabrication process, we started with what worked and then improved it over time, bit by bit. But then we hit an impasse.
The dilemma: If we made the system cut any better, whether through increasing the cutting power of the laser or swapping the laser out for a different system altogether, that slight surface scarring on the models became deep structural wounds; the integrity of the models was completely destroyed. If we preserved the models so we could continue shipping them, our aligners—the actual product—wouldn't be as good as they could be.
We had to decide what was most important. Would we focus our efforts on making a better overall product at the cost of a feature many of our doctors had come to love, or would we maintain the status quo? After struggling with the issue for over a year, we finally decided what we’d honestly known all along: shipping models had to go if we were going to make the best aligner we could make.
Today, we still count on Abby to do her part. We've swapped out her laser for a humble mill. (We’ve named him Milton, by the way. Figured we wouldn’t chance the whole no-name thing.) He's not as flashy (or flakey) as the laser, but he's solid and dependable, and he's helping Abby make the cleanest cuts we've ever seen. As expected, he’s a bit rough on our models—he destroys them, in fact—but don’t let that fool you. Our models are still better than they’ve ever been. Now they simply serve a different function, a focused function. And our aligners are all the better for it.
As we continue to improve our product and our processes over the years to come, who knows what’ll happen to dental models at ClearCorrect. One thing I know for sure, though: the lessons they’ve taught us will hold true whatever we do. We’ll start with what works. We’ll study it. Learn from it. And when the time comes to improve it, which it inevitably will, we’ll be ready, ready to make the only kind of product we were ever made to make—a better one.