Layered Manufacturing For Scientific Visualization

From the moment we’re born, we learn to use and depend on our sense of touch to gain knowledge and understanding of the world around us. Today, it can also be exploited to understand scientific data. Rapid prototyping, or layered manufacturing (LM), is a mainstream technology in engineering product development, enabling engineers to create prototype parts from new designs before their organizations invest in final production. The Center for Visualization Prototypes (CVP) at Oregon State University (originally at the San Diego Supercomputer Center) uses LM technology to create 3D visualization hard copies, collaborating with geologists, astronomers, oceanographers, cartographers, biomedical researchers, chemists, and other scientists to enhance data understanding. During its 10 years of operation, it has fabricated more than 1,000 physical models, helping them visualize their data.

Figure

Large-scale manufacturing is typically characterized by subtractive processes, starting with, say, a block of material, then removing pieces of it with a milling machine until only the desired part remains. LM, on the other hand, is characterized by additive processes. Starting with nothing, it adds material layer-by-layer or drop-by-drop until the 3D part is complete. Additive manufacturing for scientific visualization provides three significant advantages over its subtractive manufacturing counterparts:

• Handles extremely complex parts. A subtractive manufacturing element (such as a milling machine bit) snakes its way through complex geometry where it is likely to gouge part surfaces along the way;
• Runs to completion. Part manufacturing can run to completion without intervention, with no need to flip or otherwise reorient parts during the process; and
• Makes impossible parts. Topologically impossible 3D parts can be fabricated. It is even possible to produce already-assembled interlocking parts (such as chains and ball-in-socket joints).

The prototype manufacturing market includes many LM technologies and vendors [4]. CVP uses a Z Corporation Z406 machine (www.zcorporation.com) that selectively applies a bonding agent to a thin layer of powder and sometimes also applies color to the model’s outer surface. Color is especially important in visualization applications; parts not only convey shape but can also map scalar data values to the surface.

Back to Top

Model Input File

The de facto geometric information transfer standard for all LM machines is the stereolithography, or STL, file format, often called a “bucket of triangles.” It has been useful for users who want to invest in software tool development but who also want to be reassured that their tools will work on many different devices and be usable well into the future.

An STL file describes 3D solids by listing the triangles bounding their exteriors (see the table for an example STL file). While it is easy to create and write a file using STL, the file lacks much of the robustness that geometric modeling practitioners have come to expect from digital geometric representation. For example, there is no guarantee that its set of triangles form a legal solid, continuously bound on all sides, that there are no T-intersections, that the triangle vertices are all oriented consistently, or that the surface normal vectors are all outward-facing. Even many STL files originating from robust solid modeling computer-aided design systems are not always correctly formatted or represent some stable, manufacturable object. STL files generated from scientific research software are even less likely to be correct because scientists make mistakes or submit incomplete files. Thus, before fabrication technology could be used productively, better ways had to be found to deal with the inherent limitations of the STL file format. CVP researchers have developed a program called Graphics Library STL to preview and repair STL files (see the article by Sara McMains in this section), in the following ways:

• Import ASCII and binary STL files;
• Import color STL files using a CVP-developed ASCII color STL extension;
• Interactively transform files provided by scientists in 3D while inspecting the part;
• Display the part as points, lines, reduced lines, shaded surfaces, or shrunk triangles;
• Scale and reorient parts;
• Detect and fix model cracks and reversed triangles;
• Display parts using ChromaDepth 3D glasses and imaging; and
• Export parts in the STL, color STL, or binary color polygon formats used by the color Z Corporation machine.

The most important aspect of the effort to automatically preview and repair STL files was turning the bucket of triangles into a robust winged-edge adjacency data structure, allowing CVP researchers to check triangle connectivity. The program does this by accumulating unique vertices in a balanced tree upon reading the STL file. Once it determines the common vertices, it locates common triangle edges, automatically determining the triangle adjacency information for itself.

In scientific data, STL flaws take one of two forms: surface normals pointing in the wrong direction and missing triangles forming cracks in the surface of the part to be manufactured. The normals are fixed in two stages. The first divides the triangles into a binary space partition (BSP) structure, a way of dividing a 3D scene into manageable pieces; the second processes each triangle by drawing an infinite ray from its center to determine which other triangles it intersects. The program intersects the ray with each of the BSP boxes. If the ray hits a box, then the program adds every triangle in the box to a list of triangles to be checked for collisions. In order to ensure the program doesn’t check the same triangle twice, a flag is automatically set in each triangle after it’s been tested. If the ray hits an odd number of faces behind the triangle and an even number of faces in front of it, then the triangle was facing outward, which is correct, and is what the STL format requires. If not, the normal was facing the wrong direction, and the program flips it into position.

To find cracks, the program looks for edges with only one attached triangle. The program adds these edges to a list, then looks for two edges in the list with a common point. Finding them, it completes the third edge between them to create a new crack-patching triangle. If the newly created edge does not complete another triangle, it is added to the bounds-a-single-triangle list, which is processed until empty. These methods alleviate almost all of the problems that typically arrive in STL files scientists generate from their research software. Automatically correcting errors means the CVP can accept almost any surface data a research project might want to send.

Back to Top

Color in Part Visualization

Color is a mainstay in visualization data display. Although some LM devices, including the Z406, are capable of making parts in color, the STL file format does not allow for color specification. Getting around this limitation, Z406 software expects color parts to be specified in a polygon binary file, a format different from STL. So, to make it possible to specify color from existing STL-based applications, CVP added its own extension to STL. Thus, a line of the form

color r g b

can be included anywhere in an STL file as many times as needed. The r, g, and b values are in the range (0.,1.). The effect is the coloring of all triangles after the line, until the next color command is encountered in the STL file. The result is that scientists can preview color parts, then export them in the format required by the LM machine. This is useful in scientific visualization because scientists can display additional scalar information on the model surface. Examples include the objects in Figure 1: a Mars globe with surface coloring based on research by Mike Malin of Malin Space Science Systems; a molecule showing electric charge based on research by Lynn Ten Eyck of the University of California, San Diego; a terrain visualization of the state of Oregon with superimposed satellite image based on my own research; and 3D mathematical surfaces showing various types of curvature, also based on my research.

Back to Top

Three Model Categories

CVP researchers find that physical visualization models fall into three categories—viewing, touching, and collision behavior—depending on how they might be used.

Viewing. Portability is a useful attribute of any 3D physical model, which can be taken anywhere and shown to anyone. It is relatively straightforward to generate and animate 3D graphics on a display monitor and somewhat straightforward to apply stereographics to them, but doing so is useful only when the user is sitting at the monitor. A physical model can function as a portable stereographics display. It can also be put on display for scientists, students, and the casually curious to enjoy and to learn from. Along these lines, Figure 2a shows a visualization I created of a series of Southern California fires in October 2003 based on NASA satellite data, and 2b shows a map of the U.S. based on my research that is now part of the U.S. Library of Congress map collection.

Touch. Many 3D shapes are so complex they defy complete understanding even by trained scientists using 3D stereographics display systems. Physical models allow users to touch, hold, rub, poke, pinch, rotate, and zoom the shape in order to understand them. Figure 2c shows part of an anthrax molecule produced for a drug-search project by Graham Richards and Meir Glick at Oxford University. The nooks and crevices in the molecule’s structure, indicating what other structures might be able to bind with it, are much more understandable when running one’s fingers over them.

Collision behavior. CVP researchers have found physical models especially useful for understanding collision behavior between two objects; molecular interaction is an example. While good collision-detection algorithms are available, many nonconvex points of contact make it difficult to model full collision behavior accurately. Moreover, even when the points are detected, haptic systems find it difficult to completely replicate the forces produced by multiple points of contact. Physical models finesse the issue. Figures 2d and 2e show parts of the Black Beetle virus being fit together [3, 6]. These particular models helped provide the insight needed by biomolecular researcher Jack Johnson at the Scripps Research Institute to understand the function of the docking and structural relationships of the virus components.

Back to Top

From Volume Data to Physical Models

Isosurfaces are a standard way to extract a geometric surface from nongeometric volume data. However, to be manufacturable, the surface must encompass a solid region bounded on all sides. For example, a user can produce an isovolume by specifying scalar values—S_min and S_max.¹ Turning them into a legal solid involves a two-step process: Compute each one’s corresponding isosurface and, at the boundaries of the volume, cap the gap between the isosurfaces.

The current graphics and visualization literature is generally concerned with polygon decimation, seeking to eliminate detail that is perceptually unnecessary to achieve better graphics performance. Polygon decimation works well for computer graphics where displays exploit smooth shading and dynamics tricks to get away with less detail. But physical solids provide no such trickery. Large polygons that look smooth on a graphics display create fabricated surfaces that look coarse and chunky. Fortunately, when fabricating isovolumes, display speed is not relevant. Whereas interactive graphics encourage the trading of display quality and accuracy for speed, fabrication encourages maximum quality display, regardless of polygon count. CVP researchers use the term “triangle incremation” to describe the addition of such scene detail through adaptive subdivision [2]. Figures 3a, 3b, and 3c show three models resulting from volume data [5].

---

The nooks and crevices in the molecule’s structure, indicating what other structures might be able to bind with it, are much more understandable when running one’s fingers over them.

---

Back to Top

Slope Enhancement

In dealing with the everyday physical world, we gather considerable information about an object’s shape by noting how it interacts with light. In the case of Lambertian shading, or why surfaces facing a light source are brighter than other surfaces, the more the surface’s perpendicular normal vector points toward the source, the brighter the surface will be. Figure 3d is a model of the Grand Canyon. The intensity of the light on the canyon walls is obviously different from the intensity of the light on the nearly horizontal surface above the walls. But, for visualization, the difference between light intensities is inadequate for conveying the difference in slope.

CVP researchers exaggerate the intensity difference to enhance scientists’ understanding of the model. In Figure 3e, the intensity has been raised to a power, resulting in a faster intensity drop-off with slope. The steep walls of the canyon are thus more apparent. This exaggerated intensity also enhances subtler features (such as river tributaries and mesas). Figure 3f shows the super-shaded terrain using satellite image data.

CVP researchers have also used LM to create physical models to aid education at all levels. One notable experiment in 1998 involved using them in Anna Wilder’s ninth grade Earth science class at La Costa Canyon High School in Carlsbad, CA. A model of U.S. topography showed her students how the shapes of mountains are a clue to their age. They compared the Rockies and the Appalachians in the continental U.S. in terms of both appearance and feel. Which do you think are older—the Rockies or the Appalachians? (See footnote ² for the answer.)

Back to Top

Conclusion

We’d like to stretch our ability to see and touch the physical 3D world to include exploring and understanding abstract scientific data as well. 3D visualization hard copies serve as research aids and as teaching tools. Because CVP researchers and their scientist collaborators have found that producing physical models from scientific data is far more challenging than they would like, they’ve had to develop a number of tools and tricks to convert general 3D data into scientifically valid physical models. They have been pleasantly surprised how well this new way of representing and visualizing scientific data has been received among scientists and engineers, as well as how useful it’s been producing scientific insight.

Back to Top

Back to Top

Back to Top

Back to Top

Back to Top

Back to top