§1. Course logistics: presentations and scheduling [00:09]
We will videotape the presentations. Even if you can't make it to class, you're expected to watch the presentations — all of them. A number of students from past years have watched them all because they liked it. There's usually some flexibility in the scheduling.
Last year we were actually able to get this room from 9:00 to 11:00, and on some days we did six presentations. The advantage of videotaping a live class is that some students can take this class even though they have conflicts with other classes. More and more faculty at this Institute don't just do Monday-Wednesday-Friday or Tuesday-Thursday at a particular hour. They might be lecturing at 11 o'clock and then throw in a 9 a.m. listed class, as if that doesn't screw up the schedule for another class.
So there are some students with conflicts who can't come on Wednesdays but would like to take the course. That's fine — you can watch the Wednesday lecture. These will go up on YouTube within a couple of class days once we get started. If we finish all the live lectures by early March, you've got plenty of time before grades go in in May. Some students take all three modules online, which is fine with me. You don't have to come to a live lecture. I think of it as sort of synchronous-asynchronous learning.
For the presentations: whatever modules you take, we'd like you to do a one-page summary of what you think you got out of the module. Bullet points of the key points. If you had to give a two-minute elevator pitch — you happen to get on the elevator with the CEO of a big company — you've got to be able to tell everything. Give us a short summary.
You're going to have ten minutes for the presentation itself, plus about another ten minutes for questions and answers from the other students. If you go over ten minutes, I will be timing it, and I will stand up. After eleven minutes I will slowly start walking toward you. By twelve minutes, you don't want to know me. You should not have more than ten PowerPoint slides — you can't do more than one PowerPoint slide a minute. That's a good rule of thumb and I've got a lot of empirical data to support it. Think about how you're going to organize the presentation. It's harder to make a short presentation than a long one. Some famous person once said, when asked to give a talk: if you're going to give me two hours, I can do it right now; but if you want 120 seconds, I'll need time to prepare.
Any topic you like. Some students will come in and present a thesis proposal, I don't care. Talk about something you did this summer. It's supposed to be informative to the other students. Any questions?
§2. Technology Review and the question: why no 3D printing of metals? [06:19]
Today I want to talk about a question that I was asked by Technology Review about two weeks ago. Technology Review is the magazine MIT puts out. James Rhyne Killian — former president of MIT, chairman of the corporation, 23 honorary doctorates, though he never got a real doctorate, and the first science advisor to the president — Killian started at MIT in 1925 as the editor of Technology Review. It's read by something like 60,000 people around the world, a lot of them congressional staffers and people in the know. If I want to look at what's new and upcoming in technology, I go to Technology Review, because it has the MIT imprimatur of being a little more in depth than the Wall Street Journal.
They came to me and said, we'd like to interview you on why there is no 3D printing of metals. That was the question.
This connects to the broader framing for the course. We're talking about the selection of structural materials. Structural materials are used in very large volumes, whereas functional materials — a catalyst, say — get used in micrograms. You hear a lot about nanotechnology; most of nanotechnology is going to be in functional materials: electronic, optical, chemical, magnetic, computer storage. The mechanical properties are what structural materials are mostly about, and that's what we're interested in.
§3. The origins of 3D printing: stereolithography and the Cima-Sachs printhead [09:11]
Does everybody know what 3D printing is? The original was called stereolithography. It was invented in the mid-1980s by a guy at the University of Texas. He had a bath of liquid photochromic polymer resin and two laser beams. Where the two lasers came together and intersected, the greater intensity of the UV would cause the liquid polymer to polymerize and solidify. He could build up a three-dimensional object by scanning. He was elected to the National Academy of Engineering for coming up with this idea.
About four or five years later, two guys here at MIT coined the term 3D printing — Professor Ely Sachs in mechanical engineering and Michael Cima, a materials engineer. They had initial funding from the Leaders for Manufacturing program. 2D dot-matrix printers were already available. They had the clever idea: instead of using ink, let's make our own ink that's a ceramic slurry, and 3D-print a layer of ceramic at a time. [Tom shows a complex 3D-printed ceramic shape from a web image.]
So those are your two original examples: stereolithography, which was plastics, and then Cima and Sachs, who coined the term 3D printing — now a generic term — using a $300 2D printhead and an old table that would traverse up and around to make parts. Some of the first parts they made twenty-five years ago were ceramic molds for medical devices like artificial hips. Why would that be an early adoption for this technology?
Student: Customizable to the individual.
Customizable to the individual, that's one. Cost is the other — these are very high-value-added products. They're selling artificial hips for three, five, ten thousand dollars depending on complexity. So you can afford to do something that's very slow. They couldn't make metal parts back then because metal melts at very high temperatures, but it's easy to put down a little ceramic slurry. The big challenge was the CAD system — taking the design from the computer and turning it into a program that runs the machine to make the ceramic molds that you cast metals in. That was the early adoption.
There's a related story from the same era. They were trying to make little enzyme strips for testing blood sugar — those plastic strips diabetics use, where you prick your finger, put a drop of blood on the strip, and read it on a little meter. The enzyme on that strip costs ten thousand dollars a gallon. They were having terrible quality-control problems getting the enzyme onto the strips. I said, why don't you get a 2D printhead and just print the enzyme onto these strips, however much you want? Well, it turns out one of the second big applications for 3D printing wasn't enzyme strips — that company refused to listen, though I'll bet you now they're using the technology — it was making pharmaceutical pills. Most of the filling in a pill is calcium carbonate, limestone, something not toxic — sometimes magnesium carbonate. That's the carrier for the real drug.
One of the problems in popping out pills is making sure the powders are mixed uniformly. They're not always homogeneous when you start making little pills at large volumes. So they were 2D-printing the drug onto the pills to get a uniform dose.
§4. Early attempts at metal 3D printing [15:18]
Some of the early attempts were at metals. In fact I had a research contract — I didn't need the money, so I gave it to someone over in mechanical engineering who wanted to do 3D printing of printed circuit boards.
[Tom holds up a printed circuit board.] Here's a printed circuit board. This is a thick one — about 21 layers, some metal on some plastic, a bunch of layers through the thickness. This was cut from a panel about 24 inches by 36 inches. This is basically what Cisco puts their little router chips on. These things are worth a fortune. They wanted to be able to customize one-of-a-kind printed circuit boards. I was also working with Motorola — they had a little operation down south, and another here on Route 128 — and they were trying to do the same thing. It never worked. People would still love to do this.
So that's why now, 25 years later, they're asking me the question: why don't they do this? Well, I had a project back in the mid-1990s to try to make some parts by 3D printing of metals. [Tom passes samples around.] This is nickel-aluminum-vanadium superalloy, electron-beam 3D-printed. We laid down strips. This other one is aluminum-bronze. The Navy was paying for this — they'd like to make propellers and things. We came up with a way to do it, but the equipment would have cost on the order of ten million dollars to make big parts. That's what we were focusing on.
So now, 25 years later, people are still not making a lot of 3D-printed metal parts. Why? You can ask, as the Japanese do, the five whys. Has anyone heard of the five whys before? What are they?
I think I talked about this last time. If you want to get to the root cause of a problem, you keep asking why. A few years ago in Philadelphia, an Amtrak train jumped the rails coming into 30th Street Station. Why did the train jump the rails? The first answer: he was doing 50 miles an hour in a 30-mile-an-hour zone. Around the curve, you go too fast, you jump the rails. Why was he going 50? He was on drugs. You can do the same thing with the MBTA cars around here. There's nothing magic about five whys — it could be four, it could be six. If you keep asking why, like a three-year-old asks their mother, and you keep answering, eventually you get to the root cause.
§5. The five whys: why metals, and why not plastics or ceramics? [20:17]
So let's do the five whys on metals. Why metals? Why are people interested in metals at all? Why don't you just go with ceramics and glasses? Why not plastics?
Plastics just can't take the heat. What's the highest-temperature plastic? Silicone rubber, those little kitchen tools that pay a premium for being in silicone rubber — those will go to 500 degrees Fahrenheit. Try building a heat engine at 500 degrees Fahrenheit maximum. You'd like something that goes to higher temperatures. Well, ceramics can take higher temperatures than even metals. Why don't we use ceramics?
Back in the mid-1980s, people in the ceramics community were telling the world: why are you using metals? Ceramics aren't subject to corrosion, they can go to even higher temperatures, we want to build jet engines out of ceramics. I spent 1984-85, my first sabbatical year, in Tokyo, Japan. They had something they called ceramics fever, and a ceramics high-tech showcase in Shinjuku. The Japanese were just enamored with fine ceramics. We're going to take over the world. Metals were consistently called a thing of the past. I'll show you some other examples from that era and afterwards where people said, we don't need metals, ceramics are our world. We're going to talk about why you don't use ceramics for critical structural applications.
It turns out metals have very good fracture toughness, and it's not just strength that matters. They used to build railroads out of cast iron — stronger than fine, they used steel too. They built railroads out of a lot of brittle materials. People learned to do tensile tests in the 1880s to measure the strength. But ceramics can be extremely brittle. You know how to cut glass: you take a carbide wheel or a diamond scribe, and you just scratch it. It's so brittle that a scratch a thousandth of an inch deep — 25 microns — and a small whack right there, you don't have to hit it hard, will break it right where you scratched. Watch a glassblower cut glass off the blowpipe — they take a metal file, score the hot glass, tap it on the table, and it breaks clean.
When I came back from Japan, I talked about how ceramics weren't very good, because ceramicists hadn't learned about the property fracture toughness. Fracture toughness was first discovered around 1925 by a guy named Griffith in England. Griffith was studying the fracture of glass, and he came up with the fundamental equation of fracture mechanics: the stress must be greater than something proportional to one over the square root of the crack length — σ√πa. We didn't really pay attention to fracture mechanics until World War II. Does anyone know what happened in World War II that started us on fracture mechanics? Actually the formal work was in the 1950s, but the impetus was earlier. The Navy had a problem —
Student: The Liberty ships?
Yes, the Liberty ships. We built about 5,000 of them to carry troops and supplies out of North America to where we were fighting the war. I have a 1946 report with a picture of a whole ship that broke in two. The Navy had an inquiry and found that fracture toughness was just as important as strength. That's one of the things you're going to learn when we talk about structural materials. Strength is the force of fracture in terms of tensile stress; toughness is the energy of fracture. Both are important to a structural material. Strength is the strength of the material; toughness is the ability to resist fracture. The guy who was head of materials for aircraft engines at General Electric once said: physicists think structure controls the properties of materials, but metallurgists know that defects — the little flaws Griffith was studying — control fracture. Those defects caused the fracture of the steel Liberty ships the Navy built in World War II. They didn't all fail that way, but out of five thousand ships, something like forty had major cracks, and several had total fracture.
So fracture toughness and strength are both important. We don't use plastics for all applications because they don't have high-temperature capability. We don't use ceramics in many cases because they're brittle. Metals have ten times the strength of plastics at high temperatures. Well, why not metals for 3D printing? Why can't we do it? They have certain properties that make 3D printing difficult at best. It's been 25 years since Ely Sachs and Mike Cima first did 3D printing, and thirty years since the first stereolithography.
§6. Mechanical properties that defeat metal 3D printing: thermal expansion, volume change, residual stress [28:33]
One reason is mechanical. Certain properties of metals — thermal expansion and volume change on solidification — make 3D printing hard. If you want to know more about these properties, go to a book by Mike Ashby, now in its fourth edition. Ashby was a professor at Harvard, British by background, went back to Cambridge, retired from Cambridge — brilliant materials scientist and mechanical engineer in terms of design. He wrote Materials Selection in Mechanical Design. If I had to pick a single book to be the textbook for this course, it would be this. We'll be going through it. He came up with what are now called Ashby plots. The first edition cost $300, and it had a nice pamphlet with no copyright on it — he wanted you to be able to copy these plots and pass them out to students. Ashby likes to plot things on log-log scale. Professor Sadoway has said: one of his teachers at University of Toronto told him, if you plot something on a log-log scale and you don't get a straight line, you don't have enough data. But not everything is a straight line.
Here's a plot of thermal expansion versus thermal conductivity. These plots typically go over about five orders of magnitude. Thermal conductivity in watts per meter-kelvin, from one-hundredth to a thousand — all materials fall into this simple five-orders-of-magnitude plot. Linear expansion coefficient in microstrains per degree kelvin on the other axis. Engineering ceramics have low coefficients of thermal expansion — that's good. Metals are higher than ceramics. Elastomers and rubbers have the largest coefficients of thermal expansion, but they have low thermal conductivity. Metals have one of the highest combinations of thermal conductivity and thermal expansion, and that's one of the things that kills you when you're trying to melt little pieces of metal on top of other pieces. So we can do ceramics, we can do polymers at room temperature — polymers have lousy thermal conductivity over here. But because of these particular properties of metals, 3D printing them is very difficult.
The other problem is solidification. You're going to melt the metal and it has to solidify. For aluminum, that's a six percent volume change on solidification; for iron, three and a half percent. You probably never had the experience of looking up volume change on solidification — you won't find a table; nobody knows how to find it in the literature directly. You look up the density of the liquid, you look up the density of the solid — the inverse of each is the volume — and you calculate it from those. You won't find the volume change tabulated, but you can get it. That's a pretty big change, and it's going to lead to residual stresses in the final part. If you try to make a big part — three-quarters of an inch by three-quarters of an inch — the odds are it may crack from internal residual stresses once you get much bigger than about a centimeter, because that big volume change leads to residual stress, and residual stress leads to cracking even in the material itself.
§7. Surface chemistry and surface tension: the Langmuir and the Bond number [33:29]
So we've got a residual stress problem. Those are the mechanical problems. Then there are melting-process problems. The first one is that some of these metals have very tenacious oxides. Aluminum forms an oxide. You can do the whole thing in a vacuum system, but if you take the welding module — does anyone know how long it takes for the oxide to form on a piece of aluminum? What I usually do is take a piece of chalk, fracture it right before your eyes — I form two new surfaces. How long does it take for a monolayer of gas atoms to contaminate that new surface? Out of sight: about 10 nanoseconds, 10 to the minus 8 seconds. The term is called the Langmuir, after Irving Langmuir, an engineer at General Electric around the turn of the 20th century. He studied surfaces, trying to understand why tungsten didn't work better for light bulbs because of evaporation. Langmuir won the Nobel Prize for studying evaporation and surface contamination.
If you go through the kinetic theory of gases at 1 atmosphere pressure, the Langmuir is about 10 to the minus 8 atmosphere-seconds. That's how long it takes for gas molecules at one atmosphere to form a monolayer on the surface of a metal. It doesn't take very long. If you're going to do 3D printing, you can reduce the pressure to a high-vacuum atmosphere, but then you have to do all your 3D printing in one second to avoid oxide formation. So you're not going to get rid of the oxides. That doesn't mean it's hopeless — people in nanotechnology have learned that you can use oxide-dispersion strengthening. Some of these oxides are very stable; aluminum oxide is extremely stable, so is magnesium oxide.
The other problem is surface tension. Metals have about ten times the surface energy of other materials. The surface energy of a metal is typically about one joule per square meter. It doesn't really matter whether it's a low-temperature metal like a tin-lead solder you'd use on a printed circuit board, or high-temperature steel. The steel might be half a joule per square meter, the solder might be 0.7, but they're within a small range. There's a dimensionless number — I always thought it was the Bond number, but apparently in Europe it's called the Eötvös number — Eötvös first, Wilfred Noble Bond a little later. It's the ratio of gravity forces to surface tension forces of the liquid. For a typical metal, gravity force equals surface tension force at about an eighth of an inch — three millimeters. [Tom shows a sketch of solder going through a via in a printed circuit board.] If you put more and more solder on a wire going through a via — say the wire is a millimeter — with a little bit of solder, the surface tension defies gravity. The Bond number goes as length squared. At a millimeter, surface tension forces are ten times the gravity forces. As it gets larger, gravity starts to dominate, and the two forces balance at about three millimeters. L is length, sigma is surface tension.
So you now know why you can't 3D-print a circuit board. You try to put a drop of metal on the board, and it balls up from surface tension. That drop is much smaller than an eighth of an inch — much less. Surface tension dominates over gravity, and unless you shoot it at a cold substrate to make it go splat, you won't get a nice thin layer the way you want for a printed circuit board. Motorola tried it. Intel tried it. You name the companies, everybody tried it 25 years ago, and everyone gave up because the surface tension of metal is so different from these other materials.
§8. Heat transfer: Fourier and Prandtl numbers, and why you need lasers or electron beams [40:16]
Besides the Bond number, there's the Fourier number. Joseph Fourier was a scientist in the 1800s studying heat transfer. The Fourier number is the thermal diffusivity over length squared — how fast heat can diffuse into a material. There's also the Prandtl number, the ratio of viscous diffusion rate to thermal diffusion rate. The metal is going to flow and solidify. K is thermal conductivity, K over rho-C-p is thermal diffusivity.
If you compare materials on heat transfer, metals are in a class by themselves. They conduct heat really well compared to most other materials. If you go back to Ashby, you can see the ratio: metals are very high in thermal conductivity and very low in viscosity. Ceramics we can't even melt. So thermal conductivity is also a problem.
Fourier's first law says heat flux equals minus the thermal conductivity times temperature gradient: Q = -K dT/dx. You're splatting molten metal against a metal substrate. The thermal conductivity of metals is the highest of any class of materials except diamond. If your substrate is cold, dT/dx is huge — you're sucking heat out quickly. If K is large and dT/dx is large, you'd better have a very intense heat source — Q must be large. That means your only choice is lasers or electron beams to do 3D printing of metals.
I didn't expect you to know all this beforehand, Sara, but consider the numbers. If you look at heat intensity on a surface, Q has units of watts per square centimeter. A little plumber's torch, an air-fuel-gas flame, is around 100 watts per square centimeter — that would burn your finger if you put it in the flame. The sun on a sunny day — we haven't had one of those recently — is about 0.3 watts per square centimeter, two and a half orders of magnitude smaller than a plumber's torch. An arc weld is around ten thousand watts per square centimeter. Lasers and electron beams are up around a million watts per square centimeter. So we're talking two or three million times more intensity than the sun on a sunny day. You can get more than that on a sunny day if you concentrate — kids burn ants on the sidewalk with a magnifying glass. I lived in Pahrump, Nevada once, and it got up to 130, 140 degrees. People say you can crack an egg on the sidewalk, but I wouldn't want to cook my breakfast there.
§9. The Navy particle-beam-weapons-to-3D-printing pivot [45:31]
So heat transfer is another problem in all of this. Why don't we do 3D printing of metals? There are some applications I can think of. One is printed circuit boards — that's defeated by surface tension. But for bigger parts, there's what I was trying to do for the Navy.
In 1992, peace broke out with the former Soviet Union. The US Navy had been spending money — about a quarter of a billion dollars — on particle beam weapons. Physicists building multi-megavolt electron beams. The idea was to shoot down incoming missiles fired at an aircraft carrier. Take this multi-megavolt electron beam, fire it at a missile, knock it out of the sky 30 miles away. They built them, and they could fire about 30 miles — they actually got a straight beam for about 30 inches. Then peace broke out, so they didn't really need them.
They called me up and said, we're trying to find an application, our funding is gone — could we use these things to weld a submarine? I looked at it. To weld a submarine you need about 10, 20, 30 kilowatts of power. These things had a megawatt. I said, well, maybe you could melt submarines and weld them with these. Then I got to thinking. When you build something like an aircraft carrier, it costs about fifteen billion dollars — about five billion for the ship, ten billion for the nuclear weapons and aircraft. But that five billion for the ship is still a fair amount of money, and you spend a couple hundred million buying spares: great big valves, great big castings, things with a one- or two-year lead time. If a major seawater valve cracked, or a tail shaft, or a propeller, you can't go to Home Depot. If you went to order one from a forge shop, they'd quote you about a year and a half. Even using the Defense Production Act, which allows the Defense Department to come in and take over a factory, you still couldn't get one in less than three or four months. And you can't have a capital ship out of service for three or four months because it has no propeller.
So my idea was, maybe we could use these multi-megavolt electron beams to overcome some of the problems in 3D printing and build up big parts.
§10. Why electron beams might have worked: 500 pounds per hour, and the Grüneisen stress relief [48:42]
One of the problems in 3D printing of metals — and it's not that we haven't done it some way — I mentioned this to Dr. Boehm last week. [Tom holds up a sample.] This is a laser-printed 3D metal part. It's got some holes in it, kind of a rough surface from the melting and surface tension, and on the backside it's nice flat where it was on the build plate. He said this cost $500, and I figured it must have taken two or three hours to make. You've got to lay out very thin layers of metal powder and then melt them so they stay thin and don't ball up on you. Your melt pool has to be very small. If you try to go to higher power, the Langmuir kicks in and you start evaporating, you start forming welding fume in the atmosphere. So you have a thermal-conductivity limitation on your productivity. How many grams is that, $500 to make it? It's thousands of dollars a pound. You're not going to do end-product manufacturing of a lot of parts at that price. Cost is still a problem.
Pratt Whitney developed the 25-kilowatt laser back in 1975, sort of a research-type thing. They said, why don't you make a turbine disc? They gave them a million dollars or something. It took them a month to make a six-inch turbine disc. They had a manual setup and it went around and around. But because the laser has such a high heat intensity, if you try to put in more heat you just evaporate more off the surface. So you can't get more heat in, and with the very thin layer they were laying down, it took a month to make a five-inch-diameter turbine disc. They did do 3D printing of metals back in 1975 — just unaffordable financially.
So I decided we could use multi-megavolt electron beams. Because electron beams, unlike lasers — which are photons depositing heat at the surface — million-volt electrons actually deposit their heat about a half a millimeter in. So I could melt much faster. Rather than doing 10 microns at a time on each pass, I could melt something much thicker. We did the calculations: I could lay down 500 pounds an hour of melt. So instead of buying a couple hundred million dollars of spares, you could have just a computer disk of the CAD file for the tail shaft, and when someone needed it, you build up a new one at 500 pounds an hour — in a week's time you'd have a cast part.
Except for stresses and cracking. But we had a way to take care of that. We were going to use the laser or electron beam to send a shockwave through it — the Grüneisen effect. If you hit it with enough power, about 10 to the 8th watts per square centimeter, you can send a 400-ksi shockwave through it that eliminates the residual stresses, like a sharp stress relief. We proved you could do that — we showed you can get rid of the residual stresses. It works. Just no one wanted to give me ten billion dollars to build the thing — you'd have to dig a pit about four stories in the ground to shield yourself from the x-rays. So we could do it.
I'll tell you a little more tomorrow. A year ago I was visiting a place using a laser up in Boulder, Massachusetts, where they were making aerospace parts. We can make 3D networks.