3D printing titanium: Learning to learn from success
Dear reader —
This report is an update to my experiences in metal 3D printing; it describes a good chunk of what I’ve been working on over the past four months. While I’d like to say that it stands on its own, I think there’s some context — especially if you’re just beginning to explore metal 3D printing — that can be gained from reading my earlier posts (starting with my long “Bin of broken dreams post”) first.
For background: I believe that functional, engineered consumer products made by additive manufacturing are an inevitability. In order to prepare myself for that inevitability, I’ve been developing a line of bicycle parts made by metal powder bed fusion, a process that’s used extensively in aerospace, medical, and tooling applications. My last post described the difficulty and constraints I’ve experienced in part geometry and build orientation. Since then I’ve turned a corner, and today I can say that I’ve successfully designed, built, and tested a functional product — which at last puts me in the position of needing to learn from success.
As I’ve documented the process — and frustrations — of developing metal 3D printed parts, I’ve been pleased and surprised at the number of people who’ve reached out to me to commiserate (n.b., if you’re reading this and want to do so yourself, please drop me a line). Without exception, they have expressed solidarity. “We share all of your frustrations,” one person said. “I have been through the same pain as you,” said another.
One of these people was Tom De Bruyne, General Manager at Layerwise. Layerwise is a Belgian company which was started out of the Catholic University of Leuven (one of the premier centers of additive manufacturing research); it was acquired by 3DSystems in late 2014. They’re famous for being one of the few service providers who built their own laser metal powder bed fusion machines, and have a ton of experience making 3D printed parts at both prototype and production scale. We struck up a conversation, and soon agreed to work together.
While popular opinion would have you think that quantity is a non-factor with 3D printing, the realities of running a service bureau are much to the contrary. To job shops, quantity is a critical factor; if a part will be produced at large volumes, every detail of its design and manufacturing life cycle must be examined. If, on the other hand, you’re printing a tool — or a prototype of a part that will be manufactured conventionally — then most shops will focus on getting the first print right without modifying its underlying geometry.
My project falls somewhere in the middle: while my design is certainly imperfect, there are many aspects of it which are very close. Moreover, it poses challenges (most notably its opposing cylinders, oriented 90° apart, and also its thin-wall construction and bolt boss) that will exist throughout any redesign, and solving them now will only improve my ability to deal with them in future iterations.
At the current juncture, the key questions to test are:
- Can we reliably build my current design with minimal post processing?
- Does my current design meet the necessary performance standards (strength, security, etc.) for bicycle seatmast toppers?
In practical terms, the first question boils down to whether we can build a part that can be installed on a bicycle. This means two things: maintaining inner diameters which are round and dimensionally accurate to within +/-.006", and having a bolt boss on the long cylinder which, when tightened, is capable of securing the part to a bicycle’s seatmast.
My last prototype was built on its side, and the long cylinder’s aft wall distorted as it was sintered. We added a series of solid ribs in order to counteract the built-in stress in that wall, but it ended up like a game of whack-a-mole: each reinforcement just moved the stress somewhere else. A new approach was needed.
Unlike most of the US job shops I’ve spoken to, Layerwise bundles together many low-volume customer orders into each build; my part was printed alongside a handful of other titanium parts in one of the 15–20 machines that Layerwise designed and built themselves. This means that Layerwise is able to process a large value of parts at once, even if they’re only building prototype quantities of each design.
This poses a significant challenge: if one part fails, it can jeopardize many customers’ orders. To mitigate this risk, Layerwise constantly monitors a slew of process signatures, and can adjust machine parameters on-the-fly if they detect problems. They’re also working on a layer-based deposition control & verification system (the details of this are secretive, but it sounds similar to the optical tomography systems that other machine manufacturers are working on today). Still, each part is evaluated carefully beforehand in order to anticipate and avoid failures. Especially for short-run prototypes, it’s usually better to over engineer the support structures (and ensure a successful build) than skimp out and need to rebuild the part later.
Layerwise is secretive about their custom-built machines, but they did tell me a few details. They each contain a single laser, and are built around roughly the same form factor (275mm x 275mm x 420mm) as most of the industrial machines on the market today. I also understand that Layerwise uses a recoater blade (unlike 3DSystems’ ProX line, which uses a roller), though I couldn’t confirm whether it’s a hard material (like the high speed steel blades that most EOS machines use) or a soft one (like the polymer wipers often used on Concept Laser machines). They’re able to print in about 15 materials, with the majority of their work being done in titanium. They monitor temperature, pressure, and oxygen content on-the-fly, and are working on additional variables — including full melt pool analysis.
Layerwise has in-house wire EDM, and has a daughter company that does 5-axis CNC machining. But most prototype parts, including mine, are finished by hand. When validating a part for production, Layerwise tests the full manufacturing lifecycle, building multiple full batches of parts and sending them out to be post processed. Once they’ve validated the process, they will in some cases hand off production to other 3DSystems divisions.
I worked with Martijn Vanloffelt, a project engineer at Layerwise, to prepare my part to be built. He used a few key tricks:
In order to maintain roundness in the saddle clamp cylinder (the shorter of the two cylinders, which was was going to be oriented more or less parallel to the build plate), Layerwise reinforced the inner diameter with three serrated discs.
They also oriented the part slightly off-axis in both the X and Y axes. This brings me to a point that’s worth highlighting: In metal powder bed fusion, a part’s orientation has a number of effects. First of all, any surface with an angle of less than about 45° (depending on material) must be supported. As a result, it’s generally better to orient a part so that all overhangs are as steep as possible.
But in addition, one must consider the angle between the part and the recoater blade. If the part lifts up at any point during the build, the recoater blade will strike it. The longer the area of contact is, the worse the result will be. Some machine manufacturers offer alternative recoaters to lower this risk (3DSystems ProX uses a roller; EOS offers a carbon fiber brush; Arcam uses a metal comb; and both Concept Laser and SLM offer soft polymer blades), but most use a piece of high speed steel (or, in the case of older EOS machines, a ceramic blade) to spread powder across the build platform. Regardless, it is usually better to orient parts slightly off axis in the XY plane, so that the blade doesn’t contact the part’s walls all at once.
Orienting parts off axis can also help improve surface finish. When a cylinder’s axis is aligned in the XY plane, the top face will exhibit an undesirable stepped appearance; my earlier prototypes all had this feature. When a part is oriented off axis, the surface finish is generally more consistent.
I should note that none of these techniques is guaranteed to work in all cases. Layerwise has a lot of experience building a wide variety of geometries, and has developed a sense of how to anticipate and deal with issues as they come up. I got the impression that the techniques they used on my part are things they’ve used in the past, but each design is different — and even a tried-and-true method of dealing with one design isn’t guaranteed to work well on another.
The Layerwise team also put a lot of care into generating mesh supports. Like most of the additive metal industry, Layerwise uses Materialise Magics for their final build prep, and they’ve developed expertise in exploiting the software in creative ways. I’m not able to share a detailed description of the supports Layerwise created for my part, but I can say (and anyone in the industry could confirm) that they were needed in four areas:
- Underneath the part to anchor it to the build plate.
- Inside the saddle clamp cylinder.
- Inside the window in the center funnel area.
- Inside the seatmast clamp bolt boss.
Layerwise took great care to orient the part such that it didn’t require support structures inside the hidden voids in its center section. This is something that designers and project engineers alike need to think about as a part heads into production. Not only can powder bed fusion not make fully sealed voids (if you printed a sealed sphere, the entire center would be full of trapped, unmelted powder at the end of the process), but many geometries will require support structures in areas where they’re difficult or impossible to remove. For instance, a Klein bottle could be printed in metal — but no matter how you oriented it, there would likely always be support structures stuck inside its fat end. Because of the angles in my part, it was possible to avoid this — but a different design might not fare as well.
The first part Martijn printed was a big step forward: The build completed successfully without collapsing. However, a new problem arose. The windows on the seatmast clamp area caused the two “leaves” of that cylinder to twist as they were built. By the time the window closed back up, they had become misaligned, and a witness was clearly visible where they joined back together. The part had a clear flaw — and it wouldn’t be acceptable for production.
In the next build, Martijn added a curved, perforated disc to each of the seat mast cylinder’s windows, keeping them aligned as they grew. The part that resulted was a full success, printing with clean, smooth surfaces and good near net dimensions.
Considering how much support material needed to be added back into the seatmast clamp area just to get it to build properly, I’m struck again with how inefficient my design is. The windows in the sides of the part are meant to reduce both weight and cost, but a bunch of energy is put into supporting them — and then cleaning the temporary supports out again. Instead of windows, I could just as well have replaced the walls with a lattice structure that would both decrease mass and be self-supporting during the build process, bringing the part’s cost down.
This hammers home a point that has plagued my design process: Without knowing — and, optimally, having input into — how a part is going to be oriented and supported during its build, designers are doomed to creating inefficient designs. Designing for manufacturing requires an intimate knowledge of the manufacturing process, including direct access to detailed information about how the part will be oriented and supported. But in most designer/service provider relationships today, that information comes well after many of the important design decisions have been made — if it comes at all. As a result, it often takes a large investment (both in time and money) just to prove whether additive can possibly be used to create the part at hand — and once that’s been proven, many additional iterations are sure to be needed.
This is a key problem in today’s additive manufacturing supply chain: while parts are usually designed in a solid modeling environment (often Autodesk Inventor or Solidworks, each of which cost between $5–10,000), builds are oriented and supported in Materialise Magics — which costs an additional ~$30,000. As a result, independent designers are stuck with a disjointed process, which requires costly iterations and lots of communication with the service bureau who’s preparing the part to be built.
Regardless: At this point in the process, it didn’t make sense to redesign the seatmast clamp area to reduce supports. Martijn’s build had a very high likelihood of completing successfully, and it was time to put it to the test.
After printing it, Layerwise did a bunch of post processing before sending the part to me:
- First, the entire build plate was stress relieved. Layerwise’s stress relief process is proprietary, but a typical process would involve putting the build plate in a furnace and bringing it to 600°C over a period of an hour, then holding it there for three hours before turning it off. In theory the furnace is either argon purged or vacuumed, but in practice it may contain small amounts of oxygen too. Layerwise says that the vast majority of the stress relief that they do is performed in a vacuum, but argon is typically used on prototype parts.
- Then the parts were removed from the build plate. Like most shops I’ve spoken to, Layerwise uses wire EDM — though bandsaws are also common.
- At this point, each customer’s part is separated and processed on its own. Supports are removed by a totally unsexy manual process, often involving wrenches, picks, and mallets.
- Where support structures have been removed additional cleanup is usually necessary. On prototype parts, Layerwise makes extensive use of rotary grinding bits.
- The inner diameters of my part were both ground to their final size. Layerwise told me that this process was 100% manual, and I was blown away at the precision and consistency of the surface finish.
- Then, any remaining features were micro shot peened with a nonabrasive ceramic medium.
At this point, Layerwise sent me the part. Still to be done, however, was to tap the female threads in the seatmast bolt boss.
Herein lies an important point: metal 3d printing does not, in general, produce usable mechanical features like threading. In conventional manufacturing, threading is often just another step on the same machine: mills and lathes can both easily create female threads. But with additive, threading almost always requires secondary processing. As a result, the design files that are loaded into Magics only contain plain-bore thru holes; any threading specifications must be documented (and manufactured) separately.
So, the part that I received simply had a 4.2mm hole in it; it was up to me to cut the M5 female threads. “No problem,” I thought. I’ve got a tap handle right at my desk, and am more than comfortable using it.
At this point, I became painfully aware of what’s called alpha case. Alpha case is a very hard, brittle layer of oxygen rich titanium in a part’s surface (for an interesting study on alpha case depth, see this); it’s the result of the titanium having been processed at high temperatures in environments where oxygen is present. And as I tapped the hole in the first part that Layerwise printed me, I realized that it’s very, very difficult to cut.
In order to make my job easier, I purchased a set of custom progressive taps from Widell Industries. Progressive taps cut threads in three steps, increasing the thread depth as they go. As a result, the cutting force required is generally much lower.
Even using progressive taps, I was shocked at how difficult tapping the second part was. It was incredibly slow going, and produced a lot of heat. I used cutting fluid liberally, and 45 minutes later was done.
I should note here that titanium is a hard metal regardless of how it’s processed. Moreover, alpha case is preventable; in this case, it’s simply the result of the stress relief process being done in a furnace that contains some trace oxygen. Annealed titanium 6/4 has a typical Vickers hardness of about 349, but when a part has been stress relieved in an oxygenated environment, that number might jump to more than 412. By comparison, 4130 steel and 6061-T6 aluminum (both of which are used extensively in the bicycle industry) have Vickers hardnesses of around 207 and 107, respectively. In future prototypes, I would probably specify that the stress relief should happen in a full vacuum; that would at least make the tapping a bit easier.
Regardless: Finally, the part was ready to assemble:
After a total of eight build iterations, I could finally have the part tested — and learn whether my underlying design worked.
To help understand if my design would handle real world performance requirements, I worked with EFBE Prüftechnik, a German bicycle & component testing facility. EFBE tested the part to ISO 4210–9:2014, 4.5. That test entails:
- Clamping the seatmast topper onto a 34.8mm pillar angled at 73°, and fitting a dummy saddle rail into the saddle clamp.
- Applying 100,000 cycles of a test force of 1230 N, at a distance of 70mm to the center of the rail clamp, with the saddle rail tilted down/backwards by 10°.
- Applying a vertical static load of 2050 N to the center of the saddle rail clamp.
Marcus Schröder, managing director of EFBE, put my part through the dynamic test first. It passed.
Before he went through the static load test, Marcus asked whether I wanted to make sure I got an intact part back — or if I would rather find the failure mode in the static test. In the latter scenario, he would apply the maximum force his rig could handle and see if he could get the part to break — allowing me to redesign it accordingly. Wanting to know as much about my design as possible, I chose the latter option.
Marcus’s test fixture was capable of applying 3750 Newtons to the part. My part withstood the whole thing.
Marcus used penetrating dye to confirm that the part didn’t have any micro fractures, and it came back negative. The part had met and exceeded the requirements for parts like it.
It’s worth noting that this test is simply that: a test. It’s meant to simulate real world conditions and guarantee that the part meets generally accepted standards. But it simulates those conditions only generally; manufacturers of these kinds of parts will often have their own in-house spec that to tune the characteristics they optimize for. But in general, a designer needs to choose a test, and then optimize his design such that the part fails just beyond the test’s requirements. If I trust the ISO spec implicitly then it stands to reason that I should remove more material from the part; after all, it passed the test with a wide margin.
Regardless, my part could be further optimized. What I’ve done to date was prove a basic concept: That metal powder bed fusion can be used to make thin walled bicycle parts. Which, to be honest, isn’t a particularly surprising result; after all, metal powder bed fusion has been used by others to create all kinds of bike parts, including both road and mountain frames.
The question is: Can I make it commercially viable?
With the current design and an order quantity of 10 pieces, the as-printed parts cost about $500 to make. Meanwhile, the most expensive commercially available seatmast topper I’m aware of (made by No22) costs $300, and the fanciest seatpost I’ve ever seen (made by AX Lightness) was under $600.
Now, there are a number of interesting things to note here. First, I’m able to buy in fairly low quantities. It’s not unreasonable for me to purchase parts in batches of 10, which is about as low as any non-stock commercial product in the world— and much lower than most products that involve forging, casting or CNC machining. If I can sell my part at a high end price point, then it wouldn’t take much cost reduction before I’ve got a reasonable margin — even with a strikingly low order volume. And there are a number of ways that I can reduce cost on this part:
- Even keeping the part’s design the same, I can reduce the cost by 25–40% by doubling the layer thickness. This will result in a rougher surface finish, but it’s possible that the difference will be acceptable.
- A significant amount of time and effort can be saved by redesigning the underlying model so that the inner diameters need very little — or even zero — post processing. It’s unclear exactly how much work this will take, but it could reduce the price significantly.
- Moreover, the entire part can be redesigned in order to reduce both the end mass *and* the amount of support structures necessary. Both of these have a big effect on price, though it will be time consuming to find an optimal design.
All of this assumes that I stick with a laser powered process. Electron beam melting (which I’ve been experimenting with) might reduce cost further, though the design constraints there are quite different.
It’s also possible that while this part isn’t the absolute best application for additive manufacturing, there’s another place on high end bicycles where additive works better. This is key: The cost of this part is in the right order of magnitude as my goal. Any number of small changes — whether to the machine’s build parameters, the design, or the underlying cost of the technology — could easily put the numbers in my favor.
As I hope you would expect by now, a few things have jumped out to me in this past few months of work — some relating to longer term questions I have about the industry.
As a designer, having more — and earlier — access to support structure generation software is incredibly helpful. Today, countless design decisions are made on little more than a hunch; it’s not until much later that the ramifications of those decisions become evident. This leads to an inefficient design-for-manufacturing process, where it’s difficult for product designers and manufacturing engineers to communicate all of the nuance needed. Until professional grade design software (and here I’m looking at Autodesk, Dassault, Siemens, and PTC) allows these implications to be understood early on, this will be a big problem. In other words, I should be able to set build orientation and design support structures directly in my CAD modeling environment.
The designs that are loaded onto a metal 3D printer are often very different from the finished part, whether due to the addition of stock material to be removed subsequently (like my female threads), or the support structures necessary to hold the part on the build platform. But today’s modeling software generally only shows one state for each 3D model; those intermediate, near net shapes aren’t linked to the end design. This makes the design for manufacturing process even more disjointed and awkward, as it means that I’m never working on the same design documents as my manufacturing partner is. This communication structure is bound (cf. Conway’s Law) to be replicated in the end product, and the result is bad. To fix it, we need a new class of software that blends CAD (computer aided design) and CAM (computer aided manufacturing), allowing designers to understand a part’s production cycle with perfect clarity.
True design optimization
In the work I’ve described so far, I’ve relied exclusively on conventional volumetric and NURBS modeling techniques. But a new wave of design tools is out there — topology optimization and lattice generation, for instance — and they promise to create designs directly from functional requirements. Presumably these techniques would be just as applicable to my part as they are to the applications they’re already used on (mainly aerospace, biomedical, and other high tech applications). I’ve begun to explore this space, but the fact of the matter is that I’m not aware of a single consumer product today that was designed with these tools. My hope is that they both remove weight and make the part more visually appealing, but it could take a lot of work and some expensive (and experimental) software to find whether I’m correct.
Blatant editorializing: Gongkai for industrial additive manufacturing
Today, the viability of additive processes is totally opaque, producing a chilling effect on the creativity of both designers, service bureaus, and machine manufacturers. It’s my strong belief that that will only change by better understanding the efficiencies (and inefficiencies) of the additive manufacturing toolchain, and through my own work I hope to do just that.
In a series (one, two) of recent blog posts, Bunnie Huang describes the way that Chinese electronics manufacturers have been able to drastically decrease the cost of consumer electronics. To me, they provide an example for how additive manufacturing could advance much more quickly:
My most striking impression was that Chinese entrepreneurs had relatively unfettered access to cutting-edge technology, enabling start-ups to innovate while bootstrapping. Meanwhile, Western entrepreneurs often find themselves trapped in a spiderweb of IP frameworks, spending more money on lawyers than on tooling. Further investigation taught me that the Chinese have a parallel system of traditions and ethics around sharing IP, which lead me to coin the term “gongkai”… this copying isn’t a one-way flow of value, as it would be in the case of copied movies or music. Rather, these documents are the knowledge base needed to build a phone using the copyright owner’s chips, and as such, this sharing of documents helps to promote the sales of their chips. There is ultimately, if you will, a quid-pro-quo between the copyright holders and the copiers.
It would be an understatement to say that industrial additive manufacturing hasn’t adopted gongkai. Today, trade secrets & patents are the name of the game; the access I’ve been permitted (by Layerwise, DRT Medical-Morris, and countless other friends across the industry) is rarely afforded to others. It’s my feeling that this is bad, both for the technology as a whole and for the long term interest of individual players within it.
For instance: Anyone with experience could, given a part geometry and its build orientation, surmise more or less what the support structures will look like. If you have a physical part in hand, it’s even easier to reverse engineer; a part’s layer boundaries reveal its build orientation, and even with careful cleanup it’s generally possible to tell which surfaces have had support structures removed from them. In short, manufacturing forensics is, with enough experience and care, fairly reliable. And yet orientation and support structure setups are almost always treated as closely guarded secrets.
With so much uncertainty in industrial additive manufacturing today, firms are caught in something of a prisoner’s dilemma; their protective IP strategies prevent the industry from moving forward in the way that everyone ultimately wants it to.
All around the world, intelligent, hardworking people trying to solve the most basic problems in additive manufacturing. Everyone in the industry knows what these are, and all of the major players are fighting tooth and nail to solve them first. And though it seems a long way off, I think most of them genuinely believe that they’ll see fully automated orientation & support structure generation within the next decade or two.
But today, the process of printing a part is decidedly hands-on; expertise is critical. “The people who are good at this stuff are good at it because they’ve been doing it for eight years,” one industry veteran told me recently. For sure, this industry is full of valuable intellectual property. But in most cases, *craftsmanship* is central to most firms’ bottom lines — and it is protected at all expense.
I believe that industrial additive manufacturing needs far, far more knowledge sharing. We need an environment closer to the one that Bunnie describes:
We need “a ‘network’ view of IP and ownership: the far-sight necessary to create good ideas and innovations is attained by standing on the shoulders of others, and as such there [should be] a network of people who trade these ideas as favors among each other.”
In the coming moths, I’m looking forward to working on just that.
This post is just one in a series, and I remain convinced of what I’m working towards. With any luck — and more open collaboration with intelligent, committed people in the industry — I’ll have more to report soon :)
First, thanks to Martijn Vanloffelt and Tom De Bruyne, of Layerwise, for both their hard work and their willingness to help me understand how they work.
Second, thanks to Marcus Schröder, of EFBE, for both his hard work and the enthusiasm he had for testing my part — and discussing at length the testing & engineering cycle he sees in the industry today.