Leverage all sources available: Although one of the previously mentioned sources may be the primary one from which a modeler will work, all available photographs, drawings, converted geometry, etc., should be used together to spot-check and verify the information given. Sometimes schematics are mislabeled, mistakes are made while measuring, or geometry from other CAD packages does not convert properly. The only way to catch some of these errors is to compare them against another source.

Get information while it is available: With the many data points involved in building complex geometry, it is not uncommon to find during geometry development that not all required information was obtained during the data measurement/collection phase. Important components and/or dimensions can be overlooked, and it may be inconvenient or impossible (e.g., with a combat vehicle) to recollect, remeasure, or reconvert what is missing. Thus, a modeler should be as thorough as possible when obtaining data. Even if it is not clear whether a piece of geometry or measurement will be required, it can always be discarded later if not needed. Also, in the spirit of the carpenter's maxim, it is a good idea to "measure twice and cut once" and double-check measured or converted geometry before it is placed in a model.

Get total lengths and total views: Modelers sometimes take relative measurements of objects across a face without measuring the entire length/width of the face. Unfortunately, at the end, the measurements do not always add up. It is much easier to "back out" missed or inaccurate measurements given total lengths/widths. Likewise, when photographing portions of a bigger object (e.g., a radiator on a truck), it is a good idea to also capture several "bird's eye" views of encompassing objects (e.g., the engine compartment or the entire truck) to help establish overall reference points.

Record measurements as clearly and consistently as possible: It is interesting how a "scribble" that is perfectly understandable to the measurement-taker who is still in front of the object can become indecipherable when it is later viewed back in the office (when the object is no longer accessible). Furthermore, despite the best laid plans, projects and personnel can change in midstream, and the person(s) taking measurements may wind up having little or no connection to the person(s) actually interpreting those measurements and building the model. Therefore, all drawings and notations should be sufficiently clear and consistent so that someone unfamiliar with the object could understand and work with the recorded measurements. A few recommendations are given as follows:

Use a top-down approach: It is a good idea to design the structure using a top-down approach, beginning with the largest, most encompassing, or most functionally significant parts/systems and working down from there. Once again, the model's mission is all-important here. If an armored tank is being modeled for a ballistic analysis, the model should probably be structured so that all the pieces connected to the turret are grouped together and, therefore, can move together when the turret is rotated.

Take advantage of established organizational conventions: It is wise to follow any traditional or widely used conventions that might be available in, say, an owner's or operator's manual. If a mechanic or user would normally expect a particular component to be part of a suspension system, then it is wise for the modeler to structure his model accordingly unless there is a good reason to do otherwise.

Use good naming practices: Closely associated with the idea of using good organizational conventions to structure geometry is the idea of using good naming conventions to name geometry. Although naming may appear to be a trivial matter, the fact is that it is not always easy to establish titles and schemes that are intuitive, robust, and useful in helping the end user know where he is in a potentially complex model.

There are few limitations to how files and objects can be named in BRL-CAD (other than that each file/object name must be unique). However, new modelers soon find that random or haphazard naming schemes lead to inefficiency and frustration. Thus, the following general recommendations are provided to help modelers efficiently organize and track what they build.

Include the right amount of detail: The structure should only be as deep as needed for the application. Obviously, every part, no matter how complex, could in theory be reduced down to the atomic or even subatomic level, but how cost efficient and useful would this be? More is not necessarily better. The modeler should use common sense and consult with the end user(s) when deciding how far to break down components and systems. Insufficient detail can diminish the model's usefulness and reduce user confidence, and yet too much detail can unnecessarily drain time and resources, slow down processing time of application codes, and frustrate users who have to wade through many parts that they do not need to get to what they do need.

mv  oldname newname
	      

mvall oldname newname
	      

Use location- and function-based groupings: Components should be grouped based on simple, logical categories such as location and/or functionality. For example, the structure of the simple radio that was built in Lesson 16 of Volume II of the BRL-CAD Tutorial Series (Butler et al., 2001) could be set up in several ways. Figure 4 shows a structure based on location, and Figure 5 shows a structure based on functionality.

Figure 4.1. Location-based structure of the radio in Volume II.

Figure 4.2. Function-based structure of the radio in Volume II.

The structuring phase, of course, gets trickier and more subjective as the model gets more complex. Regardless of whether the structure is based on location, function, or something else, it is not always clear which parts belong to which structures. In fact, some parts are clearly designed to interface between parts or systems, and so the modeler must choose where he should place them in the tree structure. A consistent treatment of these parts within the model is an important part of the user's ability to understand and use the model.

It is also important to remember that the tree structure in MGED is independent of the geometry created. The structure is simply a tool to help the user organize and work with the database. Accordingly, the tree structure can be manipulated to suit whatever needs the user(s) may have. Consider the example of a model of a room containing a table and a cup on top of the table. If one wanted to relocate the table (along with the cup) next to a wall, one could create a temporary combination containing the table and the cup. This combination could then be used to move the two objects together to their new location. After the objects are in position, the temporary combination could be "pushed" (see discussion of the push command in Section 5) and then deleted using the kill command (see Appendix A of Volume II [Butler et al., 2001]).

Build the main structure first: As mentioned previously, it is a good idea to start building with the "main" object of a model. This could be the largest piece, the piece most central to the rest of the model, or a piece whose location represents a prominent corner or point. Much like on the assembly line of an automobile manufacturer, building the main frame first provides an overall model coordinate system for the rest of the smaller, secondary parts to reference. Also, for projects in which multiple modelers work on separate pieces simultaneously, starting with the main structure allows other pieces to be built in place or put in position immediately upon completion. This practice is more efficient in that it eliminates having extra parts floating around waiting to be positioned, and it provides a better picture of model completion throughout the project.

Know the four modeling levels and their differences: As shown in Table 4 (and discussed in Volume II [Butler et al., 2001]), all models built in BRL-CAD are built within the confines of its four modeling levels: (1) the primitive level, (2) the combination level, (3) the region level, and (4) the assembly level. Knowing the characteristics of these modeling levels is one of the first keys to developing effective geometry.

Know the major primitives and their constraints: Another key to good model building is to understand the required inputs, editing options, geometric characteristics, and relative advantages/disadvantages of the package's basic "building blocks"--the primitive shapes. Although the package currently has more than 20 primary primitives (as well as another dozen developmental/special-use primitives), only a few of these primitives are used on a regular basis (see Table 5). That is not to say, of course, that less common shapes could not be used, but with some experience, users can begin to "see geometry" in a relatively small set of primitives and understand the data needed to produce accurate models.

Table 5.2. The major BRL-CAD primitives and parameters.

Primitive Shape/BRL-CAD Abbreviation	Input Parameters/Definitions[a]
Arbitrary convex polyhedron, 8 pts (arb8)	8 vertices, 6 faces, and 12 edges Each face must be a plane. Illegal variations include the following:
Arbitrary convex polyhedron, 6 pts (arb6)	6 vertices, 5 faces, and 9 edges. Stored as an arb8, where point 8 is coincident with point 5 and point 7 is coincident with point 6
Sphere (sph)	Vertex, V. Radii, A, B, C. Stored as an ellipsoid (ell). Vectors A, B, and C are mutually perpendicular.
Right circular cylinder (rcc)	Vertex, V. Radii, A, B, C, D. Height vector, H (base-to-top distance). Stored as a truncated general cone (tgc). Vectors A and B have equal lengths, C and D have equal lengths, and all vectors are perpendicular to H.
Torus (tor)	Vertex, V (center of hole). Normal direction for the plane of the ring. Radius 1 (radius from V to center of tube). Radius 2 (radius of tube).
Pipe (pipe)	Outer diameter (OD). Inner diameter (ID). Bend radius (equivalent to an r1 value of a torus). Each point contains X, Y, Z coordinates, OD, ID, and bend radius data. Is effectively a subregion combination of cylinders and bounded tori whose path is defined by a series of coordinates.
[a] To maximize database efficiency, some shape types are stored as other types (e.g., all arbs are stored as arb8's), but this behavior is invisible to the user.

Use the best command to build primitives: In addition to understanding the package's basic building blocks and modeling levels, it is important to understand the behavior and advantages/disadvantages of its basic building "tools" (see Table 6). Using the right building command at the right time can maximize modeling efficiency by, in some cases, taking advantage of data from previously built geometry and saving measurement and/or input time.

Build objects in the most convenient location: Although coordinate systems vary according to the type of situation (e.g., converted geometry or group modeling, where a particular orientation has been established), BRL-CAD models are generally centered at the origin (x y z = 0 0 0), where the +X axis is front, the +Y axis is left, and the +Z axis is up.

For objects that are symmetrical in nature, this practice can take advantage of BRL-CAD's mirroring operations and can provide simpler reference numbers for objects that are more complex in composition and/or orientation. In some cases, however, the modeler will find it makes more sense to build objects in place in the model. These include cases in which previously created objects offer convenient reference numbers for the object's location/orientation and cases in which tangencies and other necessary calculations would be more difficult to derive with the object at the origin.

Note that there are traditional coordinate system conventions that some organizations use for their target descriptions (Ellis, 1992; Robertson et al., 1996; Winner et al., 2002). For a turreted vehicle, the origin is traditionally located at the intersection of the axis of the turret rotation and the ground surface. The +X axis points to the front of the vehicle, the +Y axis points toward the vehicle's left, and the +Z points up (see Figure 6). For a nonturreted vehicle, the axes are the same, but there is no axis of rotation to provide a definitive reference point. So, the origin is located at the intersection of the ground surface and a convenient point along the left-right, mid-plane of the vehicle (see Figure 7). For fixed-wing and rotary-wing aircraft, the axes are the same, but the origin is located on the front nose of the airframe (see Figures 8 and 9).

Figure 5.1. Coordinate axes of a turreted ground vehicle.

Figure 5.2. Coordinate axes of a nonturreted ground vehicle.

Figure 5.3. Coordinate axes of a fixed-wing aircraft.

Figure 5.4. Coordinate axes of a rotary-wing aircraft.

Build multiple occurrences of objects in the most advantageous manner: Sometimes a modeler will have to make several occurrences of an object. For example, imagine modeling a box of new, identical pencils. Wouldn't it be convenient to take advantage of the similarities involved? There are two basic techniques for constructing such collections. The first involves actually replicating geometry; the second involves referencing shared geometry.

Regardless of the technique used, the modeler typically starts by creating a prototype of the object. In the first technique (illustrated in Figure 10), the modeler creates complete copies of the object to be replicated. Each copy is then positioned within the model. In the second technique (illustrated in Figure 11), a "reference" combination that contains only the prototype is created. This combination is then positioned within the model.

As shown in Table 7, there are trade-offs to be considered when using each of these approaches. Construction effort is one of them. If the prototype consists of many objects or layers of structure, replication could be a tedious task. In the box of pencils, for example, all of the structure of the pencil would have to be duplicated, including the wood, eraser, barrel, and lead. On the other hand, if the referencing approach is used, then a relatively minor amount of work is needed to create the multiple occurrences.

Figure 5.5. Building multiple occurrences through replication.

Figure 5.6. Building multiple occurrences through referencing.

Also, if the modeler wants to make a change to all of the objects (e.g., sharpening the point of the pencil), then the referencing approach has definite advantages. The prototype object is edited to incorporate the change, and all occurrences automatically reflect that change. However, if only one object is to be modified, then a copy of the prototype must be made, and the reference for that item must now refer to the copy. Not surprisingly, when this type of operation is to be performed often, the replication approach has definite advantages over the referencing approach.

Referencing also has the advantage that it can reduce the amount of disk space needed to store multiple copies of complex objects. The extra space needed to store each new occurrence on disk consists of the transformation matrix and the name of the object and reference combination. This can be significantly smaller than the replication of all the geometry that makes up the prototype.

It should be noted, however, that because some analysis codes require a unique identifier for each object in the database, some agencies require that all occurrences be replicated to the primitive level without matrices.

There are several other tools that can make the duplication process easier--namely, the Build Pattern tool and the keep and dbconcat commands. The Build Pattern tool, which is discussed in Appendix E, can help the modeler automatically generate multiple copies of geometry in rectangular, spherical, or cylindrical patterns. The keep command can be used to save portions of geometry, and the dbconcat command can be used to concatenate (add) them to other geometries or reinsert them into the existing database as copies.

Use the push command to eliminate matrices from replicated geometry: When the replication technique has been used to copy a particular piece of geometry, the push command is frequently used to walk the geometry tree from a specified top to the primitive level and collect the matrix transformations (i.e., any translations, rotations, or scales applied to the new assembly using matrix edits). The push command applies the matrix transformation to the parameters of the primitives, eliminating the need for storing the matrices. One disadvantage of this operation is that any local coordinate system used in constructing objects is lost.

Use the best method for exporting and importing pieces of a database: Sometimes a modeler will want to save a portion of a model to be added to another database, to be reinserted into the original database as a copy, to be saved for future use, or to be edited as a new database (e.g., using a crew member or engine from one database in a different database). There are two commonly used methods to export and import geometry in BRL-CAD: (1) using the keep and dbconcat commands from the command line, or (2) using the export and import commands from the GUI.

For the first method, the keep command exports data either creating a new database file or appending objects to an existing database. The form of the command is as follows:

The dbconcat command adds the contents of an existing database file to the database currently open. The user may import the database as is or choose to rename each element of the geometry by specifying a prefix. The user may alternatively use the -s or -p option to add a computer-generated suffix (-s) or prefix (-p). The form of the command is as follows:

As mentioned previously, every BRL-CAD object must have a unique name; however, when combining geometry from more than one database, there may be duplicate names (especially if a modeler uses standard naming conventions in all of his models). If there are name collisions, the package will automatically add computer-generated prefixes to the duplicate items in the concatenated geometry. The default prefix names are of the form A_, B_, C_, etc. Note that these prefixes will not be added to the member names in existing combinations in the database. This allows the user to edit or remove this geometry independently of existing data, preventing unintentional overwriting of the existing database items.

Another way to move data to and from separate databases is by using the export and import commands in MGED's GUI. Located under the File menu, these commands allow the user to choose either ASCII or binary objects. They perform the same functions as their command-line counterparts. (When exporting, if no objects are selected, the default objects will be any that are currently displayed in the graphics window.)

It is good modeling practice to check for duplicate names before inserting new geometry into your database. To check for duplicates, use the dup command from the command line. This command compares external database file object names with current database file object names and reports duplicate names. The form of the dup command is as follows:

Note that there is currently no GUI equivalent to the dup command.

Keep bounding primitives as small and compact as possible: Although it is possible to use large primitives to achieve intersected or subtracted shapes in BRL-CAD (e.g., using a large sphere to create the relatively flat curve of a radar dish), using bounding or subtraction primitives that extend significantly beyond the outer boundaries of the positive volume of the region is generally not recommended because it slows down raytracing applications and can make wireframe geometry more difficult to view, especially in a complex database.

Imagine that a user wants half of a sphere for the target geometry (see Figure 12). In some cases, the user might want to use a large primitive that already exists in the database because it is in the proper location/orientation or because it requires no edits. The user should recognize, however, that whenever this object is rendered, any rays that pass through the large bounding primitive will have to do the extra calculation to determine whether or not the ray is in the positive volume for that region (see Figure 13). Therefore, whenever possible, the use of smaller, more compact bounding primitives is recommended (see Figure 14).

Figure 5.7. Target geometry.

Figure 5.8. Example of an Overly Large Bounding Primitive.

Figure 5.9. Example of a compact bounding primitive.

Consider the possibility of articulations, animations, and presentations: Sometimes models need to be able to simulate movement in parts and personnel or to show unique views for presentation purposes. Unfortunately, the modeler (or even the user) cannot always predict all the possible uses at the outset of a project. Therefore, it is wise, especially in organizations that use many different types of model applications, to try to design and build models with the thought that they may need to be articulated, animated, or presented in different configurations at some point.

For articulation and animation, this generally means that objects that normally move together (e.g., components on a helicopter rotor, tank turret, etc.) should be grouped together in assembly combinations (as shown in Figure 15).

Figure 5.10. Example of grouping objects for articulation.

In the example shown in Figure 15, we would want to create a turret_asy assembly with turret_armor, main_gun, and commanders_hatch in it.

Also, as discussed in Lesson 16 of Volume II (Butler et al., 2001), specialty models or assemblies can be made to simulate changes in model configuration (e.g., personnel hatches opened/closed, crew compartments occupied/unoccupied, fuel tanks full/half full/empty, etc.) or to show views not normally seen (e.g., transparent skin or cross-sectional cutouts to show internal components, similarly colored components to show subsystem categorization, etc.). Specialty models usually involve copying the original model or assembly, altering the copy to achieve the special effect, and then substituting in the copy as needed.

Understand and use Boolean operations properly: Because Boolean operations play such a vital role in building geometry, it is important that the modeler possesses a good understanding of them. As shown in Table 4, a combination is the BRL-CAD database record that stores Boolean operations. It can take one of three forms:

Figure 5.11. Sample Boolean operations.

Combinations can be created with a variety of commands, depending on the user's requirements. These commands include the following:

In addition, there are several general recommended practices when dealing with Boolean operations. They are as follows:

Figure 5.12. Properly (top) and improperly (bottom) ordered regions.

Follow or develop standardized conventions for colorizing objects: When displaying a complex model, it is sometimes difficult for the user to visually differentiate one system, subsystem, or component from another. Also, it is not always clear as to which components belong to which systems/subsystems. Therefore, if possible, it is good practice to follow a standardized RGB (red-green-blue) color scheme for commonly modeled/analyzed systems (e.g., engine, suspension, communications, etc.).

Table 8 shows some RGB colors traditionally used in MGED (out of a possible 17 million color combinations between black [0 0 0] and white [255 255 255]) (Applin et al., 1988). Table 9 shows some commonly used system-color assignments for various ground and air target descriptions (as drawn in a graphics display window with a black background) (Robertson et al., 1996; Winner et al., 2002).

Take advantage of advanced/automation modeling tools: BRL-CAD offers many tools that can help users perform advanced functions or automate complex or tedious aspects of the geometry development process. Examples of some these tools, which are discussed in Appendices A-F, include the pipe primitive (which can automate the building of wiring or hydraulic lines), the projection shader (which can paste words or images onto geometry instead of having to build them), the extruded bitmap (which can turn two-dimensional objects [e.g., a building floor plan] into 3-D geometry [e.g., walls]), the .mgedrc file (which can create customized shortcuts for many MGED operations), the Build Pattern tool (which can automatically replicate objects in a specified pattern), and the build_region command (which can automatically build regions by grouping together similarly named objects).

Evaluate early and often: As mentioned previously, evaluations should be performed both on individual objects as they are built and on sections of the model as they are assembled. In general, problems identified early are more easily isolated and fixed than if "buried" in a host of other problems. Early and continuous evaluation also reduces the amount of evaluation that needs to be done at the end. Final evaluation at the end helps ensure that the individual pieces are all working together in the model.

Evaluate at low resolution before high resolution: Especially in large and complex models, running rtcheck at high resolution can be computation- and time-intensive. Therefore, it is recommended that the modeler initially set a lower number of rays to be fired, fix any significant overlaps, and then increase the number of rays and/or zoom in on particular areas as fewer overlaps are found.

Use multiple views: Because rtcheck finds overlaps by firing individual rays at geometry, it can miss overlaps that occur between rays (e.g., when viewing a face edge-on or with low obliquity). Therefore, to ensure the highest evaluation accuracy possible, it is a good idea to use several of BRL-CAD's standard views (e.g., top; az 35, el 25; etc.) as well as at least one arbitrary or randomly selected view (e.g., az 72, el 23).

Set the proper eye point: It is also possible for a user to miss detecting overlaps when the eye point is set in the middle or front of an object (e.g., when Z clipping is turned on). This does not mean that rtcheck is not catching them; it just means that they are not being displayed. So, before an evaluation is run, it is recommended that Z clipping be turned off and the eye point be sufficiently offset from the geometry so that the rays intersect the entire breadth of geometry or portion of geometry the modeler wants to evaluate and display.

Chunk big problems into smaller problems: Experienced modelers in BRL-CAD know that "killing" overlaps is simply a part of the modeling process. However, dealing with large amounts of overlaps can be overwhelming, especially to a new modeler or a modeler who has carefully built each piece and expects rtcheck to find few, if any, problems. Fortunately, in many cases, what appears to be extensive overlapping might just be one section of geometry (e.g., a wall of buttons and switches) that is slightly out of position, and a simple translation or rotation can simultaneously fix many problems. In other cases, overlaps are the result of simple miscalculations (e.g., a 2-in vs. a 1.5-in radius) that are not likely to be noticed until positioned with surrounding geometry. Whatever the case, the best approach to extensive overlaps is not to try to fix them all at once but to divide the problem into smaller problems, concentrate on individual pieces, and use the display to help identify and fix errors. For example, rather than starting with "all," start with, say, "engine," and then add "chassis." One can then continue this process and work up to evaluating the entire model.

Who will be using the model?

What are the user(s) going to do with the model and why?

What information can I give about the model that might save the user some time or frustration?

Tell what it's got and what it's not: It is useful to record not only what components have been included in the model but also what components/details have not been included (and why). This will remove doubt as to what the end user does and does not have.

Tell the purpose of the model: Although the primary end user may know exactly what the intended application of the model may be, the documenter should consider the possibility of other users becoming involved during a model's lifetime. Furthermore, these users may not know the model's original purpose and may try to use the model in a way in which it was not intended. The documentation should explain the choices the modeler made as well as identify, if applicable, ways in which the model has been designed for articulation and/or animation.

Tell when it was built: This information identifies the time period during which the model was built. This information can be especially important when modeling developmental items with continuously changing design or model specifications. If a model undergoes a radical redesign during any stage of the development process, the period of performance can help identify why the model does or does not reflect given redesigned components as well as identify when changes were implemented.

Tell from what sources it was built: It is important to note the type of data sources from which the model dimensions were collected (i.e., field measurements, special tools, mechanical drawings, or converted geometry). If a modeler physically measures an object, it is also important to note any specific manufacturing information (e.g., make and model, year of production, factory, etc.), any special designators or insignias, and any damaged or missing items.

Tell how it was built: This information records the significant techniques and configurations that were used to build the given geometry as well as any irregular or specialized constructions (e.g., for articulation, an intended ballistic impact scenario, etc.). Also included here are basic tree structures and any naming conventions used.

Tell the level of detail to which it was built: This information documents the specific tolerances and level of accuracy used to construct the model, as well as ways in which accuracy was checked (e.g., rtcheck and rtweight).