There was a time when Michael Bailey, an expert in computer graphics development, toed the party line: he told people that a two-dimensional image of an object on a computer screen could convey as much detail and enlightenment as the real thing. But over the past few years he’s become something of a heretic. Now I have to tell those people that I was wrong, he says. It’s embarrassing, but graphics are just not as good as having an object to touch and hold.
Bailey has rediscovered a need that we all, as babies, instinctively recognized. When babies see something new, they immediately want to touch it, hold it, turn it over in their hands, explore its crannies with their tiny fingers. It’s the way they acquire a solid working knowledge of the physical world. Now a growing number of scientists are discovering, along with Bailey, that a picture—even the most sophisticated 3-D computer simulation—just won’t do for their purposes, either. No matter how many hours of processing you pour into an image, it will never be more than an image. It will never gain the solidity of a real object.
Of course, not all objects of scientific scrutiny are suitable for tactile discovery: an individual molecule is too small to be clutched and eyeballed; tectonic plates are too large. But objects large and small can be represented with physical models—and these are Bailey’s new passion. Walk into his office at the San Diego Supercomputer Center and you can’t help noticing them littering his desk and the tops of his file cabinets. Most have the color, heft, and polished softness of carved wood. His collection includes a human heart, fist-size molecules, a three-dimensional representation of a mathematical function of complex numbers, and a strangely flattened jack-o’-lantern.
They look like solid wood, but all were crafted from paper and glue. The sculptor was a copy machine–size device called the Helisys 1015 Laminated Object Manufacturing Rapid Prototyping System—or lom. The lom is the star of the center’s Tele-Manufacturing Facility, which Bailey runs. (The lom, a computer in Bailey’s office, and a link to the Internet constitute the entire Tele-Manufacturing Facility.)
The Tele-Manufacturing Facility represents the latest chapter in the decade-long history of the attempt to turn computer graphics into solid models—a process known as rapid prototyping. There are a number of rapid prototyping systems, and they variously build their models out of paper, resin, metal, or plastic. Engineers, rather than scientists, have used them to make quick three-dimensional models for objects as varied as diesel engine housings, gears, missiles, toy cars, and golf clubs. The engineers can examine the models for design flaws or use them to generate molds for mass production.
But Bailey argues with missionary zeal that rapid prototyping has the power to transform science. We’ve found that even before you pick up the model and touch it, you’ve already gathered more information than if you just looked at it on a flat computer screen, Bailey says. We’re trying to figure out why, but my hunch is that it has something to do with the way your eye can vary its focus on a solid model.
Of course, models have a long, august history in science. In seventeenth-century Europe, beautifully detailed wax models of human anatomy offered medical students a far better understanding of how the body fits together than could the best book illustrations. And in our own time, ball-and-stick models of molecules have helped bring about some important discoveries in chemistry and biology—such as that of the double-helix structure of dna. But whereas these models had to be built by hand, machines can now be programmed to make exquisitely detailed scientific models in a matter of days or hours.
Mitra Fattahipour, a geologist at San Diego State University, is among those already convinced. Fattahipour, working with Bailey, San Diego State geologist Eric Frost, and computer visualization specialist Andres Polit, used elevation data gathered by the U.S. Geological Survey to build a scale model of Death Valley. The model is only seven inches square, but it’s nothing like the plastic relief maps that hang on grade-school walls. It’s loaded with detail, showing 11,000-foot mountains towering over deep gorges as much as 282 feet below sea level (Fattahipour exaggerated the vertical relief 14 times to enhance the differences). Death Valley was created by a detachment fault that is slowly pulling apart the underlying rock; in the model the fault is clearly visible along the eastern edge of the mountain range, intersecting both the mountains and the valley floor.
Although we have the data on computers—and that is helpful—you get a whole different sense of things when you have a physical model you can touch and turn around in your hands, Fattahipour says. It lends itself to ways of experimentation that we might not have thought of before. For example, in Death Valley you can easily see the alluvial fan sedimentation that is deposited into the valley from the mountains. So one thing we thought to do is drop sand and clay onto the model, and videotape how the sediment falls. Not only is that great for teaching, but three-dimensionally, you get another sense of how things move.
For John Johnson, a structural biologist at the Scripps Research Institute, a lom-created model of a protein turned out to be a revelation. Johnson studies the black beetle virus, a simple pathogen that infects insects, so that he can better understand more complex viruses that plague humans. Viruses have the same problems to solve whether they are hiv or insect viruses, Johnson explains, and this one is very easy to work with because it has only two genes.
After 15 years of careful work, Johnson thought that he and his colleagues completely understood the black beetle virus. They knew, for example, that 180 identical copies of one protein linked together to form the virus’s geodesic dome–shaped outer capsule, they knew how those links were forged, and they could program a computer to show it. At one level we knew everything, Johnson says. We could generate tables of distances and all the nitty-gritty that describes the way these proteins interact with one another, and we could look at it on a graphics system. In fact, we were satisfied with our results—up to the time we got the solid model.
Last year Johnson handed his data over to Bailey, who then built nine copies of the protein, each 20 million times actual size. When Johnson got his hands on the protein models, he assembled three into a clump, just as they assemble themselves in nature. Then, turning the proteins over in his hands, Johnson saw a hole where the three molecules met. In all his years of studying the proteins, this hole had eluded him; computer graphics couldn’t give him enough sense of depth to notice it. As soon as Johnson saw the hole in the model, he recognized that it was a perfect fit for a peglike protrusion on another surface of the three-protein chain.
There is an alignment, just like you would align things if you were fitting together pieces of wood, Johnson says, and we hadn’t realized it. We didn’t even know we had missed it. Now it is possible to design small molecules that can fit into that hole and block the assembly of the virus. As soon as I saw the model, I was running around like a crazy man, showing everyone how stupid we had been. I think that is going to be a fairly universal experience when it comes to looking at the assembly of subunits with these models.
In some ways Bailey’s lom method works like the standard technique for rapid prototyping: stereolithography. That process uses a liquid polymer that solidifies when hit by a laser to build objects from the bottom up, one horizontal cross section at a time. Bailey prefers his lom method to stereolithography, however, because unlike models made of plastic or resin, his paper-and-glue constructions can be sanded and painted, have nails pounded into them, or be threaded with screws (not to mention that they look, in Bailey’s words, warm and pretty).
LOM begins building an object by creating the bottommost layer. Paper fed off a roll is glued to a metal platform inside the lom, and a carbon-dioxide laser then traces an outline on the paper. Any paper lying outside the outline then gets crosshatched by the laser, and the metal platform drops down a fraction of an inch. Another section of paper is then rolled out on top of the first layer, and as they are heated to 400 degrees, the two layers fuse. The laser now cuts a new outline for the second layer.
The machine repeats this process hundreds or thousands of times, and when the last layer has been rolled and cut, Bailey can pry the model off the platform with a putty knife. The model is a solid block, but the crosshatched areas surrounding the object fall away easily. The smaller pieces wedged into cracks and crevices can be fished out with tools that look like dental picks. The process can take anywhere from a few hours, for simple models like the human heart in Bailey’s office, to more than a day for complicated molecules like Johnson’s black beetle virus protein.
Because the scorched edges of the paper form the exterior surfaces of the parts, the finished product has the appearance of aged wood, which can be accentuated by a coat of lacquer. The hue can vary across the surface of a model: relatively horizontal areas have fewer scorched edges, and so are lighter in color than the dark brown vertical areas, where there is a high concentration of edges. On geologic models, like the Death Valley one, the edges serve as ready-made contour lines.
With the help of Dru Clark, a graduate student at the University of California at San Diego, Bailey has created hundreds of models. Some are exotic: topographical maps of the seafloor and Venus, a swirling hurricane. Others are traditional manufacturing models, like gears and aircraft parts. Some are simply prototyping parlor tricks: a chain whose links never had to be forged together, and a set of three rings, each ring intersecting the other two but with none linked together.
The data for building these models generally come to Bailey’s office over the Internet. Before a model can be built, though, the data must be translated into a geometric representation of the object’s surface—a mesh of as many as 600,000 tiny triangles. The translation isn’t always smooth. The problem is that the computer graphics on the screen doesn’t care about solidity, Bailey says. You can do all kinds of things with graphics and the computer will happily draw it up because you are simply modeling appearance. But you can build only those physical objects that can exist in three dimensions. There are mathematical rules that must be followed. If a gap is hiding in the data, for example, it can become a crack in the model.
If there is a crack, the machine can get confused about what is inside and what is outside, says Bailey. For instance, if I drew a circle on a piece of paper and asked you what was inside, well, that’s easy. But what if I drew an arc? What’s inside? I don’t know. It could be the area on the bottom of the line, or it could be the area on top. If a file has a mistake like this, the machine will hazard a guess about what’s in and what’s out. Most of the time it does a good job, but sometimes it cross-hatches things you want to keep, or fails to crosshatch things you really want to fall away, Bailey says, so you can end up with little gouges, or a wart on the outside that you know is not supposed to be there.
Bailey has built a program that will let his computer automatically check data for these sorts of errors, with the ultimate goal of making the lom produce finished models without his supervision. Researchers will be able to go to the tmf Web site and simply put their files into the processing queue. They won’t hear from us until the part is done, Bailey says. For now, however, Bailey and Clark must still intervene before a new model is made, pulling the model out of the machine and cleaning up the nooks and crannies.
These subtleties of turning strings of ones and zeros into a piece of sculpture are pretty much lost on the researchers who send Bailey their data files, but he doesn’t mind their indifference. The scientists don’t want to know—they just want the model. They are only interested in the science, Bailey says. And that is the purpose of the project—scientists don’t have to be mathematicians or engineers.
