Stephen
M. Hollister
New
Wave Systems, Inc.
A surface is developable
if it is a combination of flat, cylindrical, and conical sections. A developable surface is one where you can
lay a straight edge anywhere on the surface and find a direction where it
completely touches the surface at all parts of the straight edge. This straight section of the surface is
called a ruling line. Note that
for flat, cylindrical, and conical sections, this test would always be
true. A mathematical description of
this condition is when the Gaussian curvature of the surface is zero at
all points. This is just a fancy way of
saying that for all points on the surface there is always one direction that is
straight (has a curvature of zero).
The purpose of creating
developable surfaces is to allow you to unwrap or flatten out the surface so
that you can cut it out of flat material, like plywood, aluminum, steel, and
even cloth. If you cut the pattern out
using CNC (Computer Numerical Control), you can save a lot of time
building the 3D structure. In reality,
however, people have found that the surface to be constructed out of flat
material does not have to be perfectly developable to be buildable. For many materials, it is relatively easy to
introduce stretch, twist or compound curvature in the 2D pattern as you wrap it
into its 3D shape. This allowance for
twist opens up a huge range of additional 3D shapes that can be built from 2D
patterns. The problem is that there is
no exact answer to how much twist you can have before the stretching or
twisting can’t be done or before the material fails. As soon as you introduce stretching or twisting, there is no one
solution to the process of unwrapping and flattening out the 3D surface into
its 2D pattern shape. As you can
imagine, the 3D shape that is created from the 2D pattern depends on how
you stretch and twist the pattern. We
have found, however, that if the 3D model shape is nearly developable,
then you should have no problems with the unwrapped 2D pattern. Unfortunately, it may take some
trial-and-error practice to determine what nearly means.
As the 3D shape becomes
more twisted or contains more compound curvature, there is no magic way for a
program to tell you when the full 3D surface can’t be built from one 2D
pattern. The amount of twist that is
allowed (i.e. to be buildable) depends on the type of material, its size, its
thickness, the machine you use to introduce the added curvature, and the skill
of the machine operator. Remember that
a 2D pattern can be stretched or twisted into an infinite number of 3D
shapes. You have to develop your own
criteria for determining how much twist you can introduce. One test is to plot out small 2D patterns
and try constructing the 3D shape from cardboard or some material that
simulates the properties of material that you use for construction. If there is any doubt, then you should go
back to the 3D model and change its shape to be more developable. If you have a perfectly developable surface,
then you should have no problems with the 2D patterns. The problems arise when twist is introduced
into the shape. Another trick you could
use for twisted shapes is to cut out the 2D patterns with a little bit of extra
material along two of the edges. Then
fit the material and scribe and cut the two edges for a perfect fit. This works, but this refit process is really
what you’re trying to eliminate.
When a 3D surface has so
much twist or compound curvature that it cannot be built out of one piece of
flat material by simple stretching or twisting techniques, then the surface is
said to be expandable. Rather
than plate development, this process is called plate expansion. There is no well-defined point at which the
process changes from plate development to plate expansion. Generally speaking, the term expansion is
used when there is a lot of stretching or twisting distortion required to get
the 2D pattern to fit the 3D shape, or if you have to subdivide the full 3D
surface into smaller pieces before unwrapping them into 2D patterns. Keep in mind that even a sphere (which has a
lot of double curvature) can be built out of simple 2D pieces without a lot of
distortion if the 2D pieces are small enough.
Pilot3D uses the same
technique to unwrap or flatten out the 3D model shape into the 2D pattern for
both developable (or nearly so) and expandable (doubly curved) surfaces. If the surface is perfectly developable, the
program will give you the exact, no twist 2D pattern. Your job is to create the desired surface using the twist or
curvature information provided by the program.
We suggest that you calibrate this twist or curvature information in you
own mind by testing the program out on an object that you have already built.
If the object that you
are designing needs to be developable, or nearly so, then we recommend that you
use the ruling line techniques discussed below. These ruling lines are automatically calculated and drawn between
two curve entities and show you the amount of twist by using different colors
for the ruling lines. The twist is
defined by the angle that the material must be rotated along the entire ruling
line. By moving the edit points on the
two boundary curves, you can dynamically shape the surface and view the change
in the ruling line twist colors. Once
the desired shape and level of twist has been achieved, the program can fit the
ruling lines with a NURB surface. The
ruling lines become the columns of the NURB surface. If you have some understanding of the proper orientation of
ruling lines for your surface, you can create the surface first and orient the
surface columns as ruling lines yourself.
After you have had some experience with developable surfaces and ruling
lines, you might find this direct approach faster.
If the surface has too
much twist or compound curvature (expandable), or if you want to skip the
ruling line approach, you can create and edit the shape of the surface
directly. To check for the amount of
twist or compound curvature in the surface, you need to display its Gaussian
curvature (Surf-Kpat commands). Since
the Gaussian curvature calculation is difficult and slow, the program gives you
control over the location and color density of the curvature display. This is good for curvature feedback while
you are performing detailed shape editing of the surface. This program encodes the Gaussian curvature
of zero (perfectly developable) as a dark blue. As the surface contains more compound curvature, the colors
change to light blue, to green, and then to yellow and red, for the highest
double curvature. This makes the
colored Gaussian curvature display of the surface look like a finite element
analysis stress map. Just like the
ruling line technique, we recommend that you test this process with an object
that you have already constructed to develop a feel for the colors displayed
for the Gaussian curvature. You can
also use the Surf-Develop Patch command to force part of a surface to be
perfectly developable.
Once you have a surface (however it was created), Pilot3D will apply a finite element type of calculation to flatten out the plate and determine all of the internal strains created during the process. (Note that the flattened 2D plate shape can also be marked with cross-section trace lines.) For perfectly developable surfaces, the 2D layout will be exact and the internal strains will be zero because there is no stretching or twisting. For plates with twist or compound curvature, the program calculates and sums these strains to give you an idea of how much stretching or twisting is required. Since there is no unique solution to this compound curvature problem, the program requires the user to enter the amount that the perimeter of the 3D surface edges stretches (or shrinks) when it is flattened out. This is enough information for the program to uniquely solve the problem. But how much edge stretching does occur? You have to study your construction process. If the edges of the 2D pattern are constrained while the shape is forced into the material, then you would want to keep the perimeter length of the 2D pattern equal to the perimeter length of the 3D model shape and specify no stretching or shrinking of the edges. Some sources, however, say that the best expansion is one where the total of all strains is minimized. To do this in the program, you would have to repeat the calculation with several +/- edge stretch percentages to see which one minimized the sum of all internal strains.
Pilot3D lets you to tell
the program to draw ruling lines between any two curves. As you shape and fair the two curves, the
program dynamically recalculates and draws the ruling lines with colors that
indicate the amount of twist in the ruling lines. When you achieve the desired shape and level of twist, the
program allows you to fit a NURB surface through all of the ruling lines. Then you can unwrap or flatten out the
surface into its 2D pattern.
For this ruling line
calculation, the goal is to calculate the best ruling lines that can be nicely
fit by a NURB surface. Nice ruling
lines are ones that are fairly evenly spaced over the lengths of the curves and
have no gaps or missing ruling lines.
This is not as easy as it might seem, since the ruling line with the
smallest twist angle may not be the best ruling line. In addition, there may be areas where there is a good amount of
twist in the surface and you do not want to search for the ruling line with the
smallest twist.
First, you need to know
what an exact ruling line is. It is a
straight line crossing the surface from edge to edge that has zero twist. Imagine laying a sheet of plywood between
two curves in space. If you construct a
line on the plywood connecting the places on the two curves where the plywood touches
the curves, then that line would be an exact ruling line. Next, imagine gluing two pencils, one at
each end of the ruling line, normal or perpendicular to the plywood. Then, if two people (at each end of the
ruling line) twist the plywood in opposite directions (introducing twist in the
plywood), the two perpendicular pencils will rotate in opposite
directions. Looking down the ruling
line, the twist angle is the angle between the two pencils. When there is no twist in the plate, this
angle is zero. As you twist the plate
in opposite directions, the twist angle increases. Pilot3D lets you set a “Desired Twist Angle” value that you want
to achieve. You have to imagine
yourself putting that amount of twist angle in the material you are using all
along the plate.
The problem is that you
cannot have the program search anywhere for ruling lines where the twist angle
is very near zero. Ruling line twist
angles are very sensitive to the shapes of the two curves. If you always look for the lowest twist
angle, the final set of ruling lines between the two curves will be uneven,
they may cross each other, or they might be missing altogether (no solution),
even though the twist might be well within building tolerances. As you fair or smooth the two boundary
curves, the ruling lines might change shape dramatically with a very small
change in curve shape. In addition, for
flat portions of surfaces, any line crossing the surface will have a twist
angle of zero. In those areas, the
program must try to spread the ruling lines nicely over the surface.
Pilot3D tries to avoid
these problems by starting with nice ruling lines and then changing their
positions as little as possible to minimize the twist angle. The nice starting
ruling lines are found by dividing each curve arc-length-wise into the “Number
of Ruling Lines” input value (the default value is 50 ruling lines), and then
connecting each associated point on each curve with an initial guess ruling
line.
For each ruling line of
the initial set of nice ruling lines, the following is done:
1. Fix the curve 1 end
of the ruling line.
2. Vary the curve 2 end
of the ruling line back and forth from its initial position by the incremental
search step size (curve length divided by the number of Search Steps on Curves)
to find the nearest curve 2 end of the ruling line that meets the
Desired Twist Angle input value. If,
after searching the Maximum Search steps either side of the initial guess
location, the Desired Twist Angle condition is not found, the program uses the
one ruling line with the smallest twist angle.
If this twist angle is still greater than the Maximum Twist Angle input
value, then the program will not display the ruling line.
This algorithm will
always try to find the nicest spread of ruling lines between the curves, since
the ultimate goal is to fit a NURB surface to the ruling lines and to generate
the 2D pattern. Since the calculation
and display of the ruling lines and twist angles are automatically done as you
move either curve, you get immediate feedback on the developability of the
surface. The calculated ruling lines
behave very well for either large or small changes to the model, and for large
or small ruling line twist angles. This
makes it much easier to create and develop surfaces that have a good amount of
twist.
Once you have the curve
and ruling line shapes you desire, Pilot3D allows you to fit the ruling lines
with one NURB surface that can be unwrapped or flattened out into a 2D pattern.
Note 1: When you cut all of the developed plates and
support frames exactly from the computer-generated patterns, you have to be
extremely careful about the set-up of the frames and the wrapping of the plates
onto the frames. One little error or
mis-position of a frame can cause a progressive error to creep in. These little errors grow until a plate can
be off by more than an inch at the ends of the structure. If the plate has already been cut, then many
people are not going to be happy. If
you cut out frames and plates exactly, then you must be very careful and
accurate during setup. The old fix-up
techniques of cut-and-fit do not work anymore.
Note 2: Make sure the plate to be developed is
buildable, but there
are no neat mathematical formulas that the program can use to tell you if you
are not close enough. If the Gaussian
curvature display shows that the surface is almost exactly developable, then
there is no problem, but as you add more twist to the surface or ruling lines,
you may wish to add additional scrap material to two of the sides of the plate
so that you can scribe and fit the plate exactly. Another check is shown in the Layout Numbers view when the
PlaneCut trace lines are turned on. The
program calculates and compares the 2D developed PlaneCut trace girths with the
full-size, actual 3D girths along the plate.
The two numbers should be the same if the plate is developable. However, if there is any doubt about
developability, cut and fit.
Note 3: Make sure you use “enough” supports to
construct the object, even if you have to use temporary supports. If you do not use enough supports, the plate
will have a hard time assuming the shape it had in the computer.
Note 4: Make sure that you have the supports spaced
very accurately and that the supports are aligned perfectly. If you cut out both the supports and the
developed plates by NC control, then the construction process must be done very
carefully. This is not the time to use
the old cut and fit tricks.
Note
5: The goal of all of this work
is to accurately build objects that form the foundation for cutting out the
rest of the parts in the object. The
more accurate the initial fit of supports and plates, the more accurate and
quickly the rest of the object will go together. In addition, you will gain experience in building accurately and
be able to pre-cut more pieces ahead of time with less recut and fit
problems. This has a significant domino
effect throughout the rest of the construction process.
Developability Check
As an aid to determine
the developability
of the plate, the developed YZ trace lines include a calculation at the end
which compares the developed (2D) girth length of the YZ PlaneCut on the
developed plate with the (3D) curved girth length of the YZ PlaneCut along the
plate. Since the frame trace line is
supposed to be wrapped onto the curved 3D shape of the cut-out YZ PlaneCut,
then their two lengths should be the same.
What the program calculates are both the 2D developed YZ PlaneCut girth
length and the actual 3D shape of the cut on the surface. These two numbers should be equal if the
plate is perfectly developable. If it
isn’t, you can evaluate the difference in girth length to estimate whether you
can stretch or twist the plate into shape or to determine about how much scrap
material to leave along one edge. Note
that this is just one estimate of
how close a plate is to being perfectly developable. It still can be difficult to judge whether you can cut the plate
out of 2D material and twist it into shape.
We recommend that you take some examples that you have already built
(both developable and non-developable) and put their shapes into the program
and evaluate what the program thinks about their developability, both in terms
of the Gaussian curvature and in terms of these 2D and 3D girth
differences. Another common technique
is to construct a model out of cardboard (plot scale versions of the frames and
plates) and see if it can be built.
Obviously, this is tedious, but not as bad as making a mistake on the
full-size structure and wasting a lot of time and perhaps a big sheet of
aluminum or steel.