The new visualization algorithm is based on
the reverse process of auto meshing. There are three steps in this algorithm:
first, solve the above equations with the known and unknown reversed,
then search among the big elements, and finally, compute the value with
proper interpolation functions.
In a single big element, the current problem is that
is known and it is necessary to find .
In other words, the known and unknown are reversed in the above equations
compared with meshing.
It is very difficult to reverse these equations analytically,
partially because the shape function
N is different from one element type to another.
The easiest way is to use the Newton-Raphson method to solve these equations
numerically.
It can produce sufficiently precise values of , and it does not care
about what is the shape function.
The initial guess for is , because and
. The iteration is terminated when
and . This convergence
condition ensures that the error is less than one unit in real regular space.
In solving the equations, the Newton-Raphson method definitely
converges, because the inverse of the Jacobian matrix exists, and
its coefficients are the first derivatives of the shape function.
In FEM analysis,
this inverse matrix is also used in projection back and forth between those two
coordinate systems.
The rules in FEM guarantee
existence of the inverse matrix [1] [2] [8].
Also, the Jacobian matrix can be analytically derived and evaluated,
because N is a set of simple polynomials.
The Newton-Raphson method converges quadratically, and the initial
solution is very close to the solution of the equations.
In numerical experiments, it only takes three or
four iterations to find the solutions.
By now, a single big-element has successfully been projected to
regular space.
For the whole structure, the only way is to search among big-elements,
because there is no requirement for
how a structure is built up of big-elements.
It will need a few more iterations, but usually after one iteration, it can be
known if a point is inside a big-element. A structure mostly consists of
dozens of big-elements. That seems to be
a drawback of this visualization algorithm, but on the other hand
the scale of the problem is reduced greatly.
The search is now performed among dozens of items
rather than among thousands.
The advantage is obvious.
The next step is to transform the solution to real regular
space. It is a matter of simply scaling and transforming.
By the way, this method can also determine whether a given point is
inside a big-element or not,
because the ranges of and are already known.
So it can also judge if a point is inside the volume.
Using the above method, the lower left corner point coordinates can be found
in regular space for a given
point. Then, the rest of the surrounding points can be
very easily found by increasing one unit in every direction.
The pseudo code of this process is as follows:
Using the Newton-Raphson method to solve the equations;
Exit, if the point is inside the current big element;
Continue, if outside
End of loop
If the point is inside the volume then
Transform the solution into regular space;
Return
Else
Return not found;
End of if
Loop for all big elements
The value at point can be computed
by interpolation with the consideration
of what kind of element is used in the analysis phase, after
its surrounding points are found. This step seems very trivial,
but it is very important to reduce errors.
It is incorrect to use linear interpolation to compute this value
in all cases, because
the shape function is not always a linear function.
The shape function used in FEM analysis must be used as the
interpolation function here,
because the distribution pattern was determined
when the element type was chosen in the analysis phase.
Any other interpolation function can not be used. Otherwise, the distribution
graphs will be very odd. At least there will be abnormal jumps
(not continuous in their first derivatives) at boundaries of cells.
That will give a false feedback to designers,
because in a good design stress changes reasonably smoothly
and stress jumps signify a serious design defect.
In this visualization process, there is no need to preprocess
the whole data set,
because projecting a point into regular space is very fast.
Whenever a point needs to be projected into regular space ,
just call this function. This way, not much memory space is needed. Then
a visualization method, such as ray tracing, can be used
to render the whole volume.