
                             SLFFEA 1.1 Non-Linear Brick README

                    Copyright (C) 1999 San Le

                    email:     slffea@juno.com
                    homepage:  http://www.geocities.com/Athens/2099/slffea.html


In the slffea-1.0/data/br/ directory, there is the file brdynam.dat.  This is where
you enter in values for dt, cc, and iteration_max in that order.  I have a few
examples prepared for the data files "ball", "rubber", "torsion" and "tensile" in
this same directory.  The purpose of these variables is explained below.

As with most non-linear FEA software, running the non-linear brick requires
a little art in the selection of parameters which will keep the program stable
as it iterates toward the solution.

The method of iteration I have chosen is called Updated Lagrange where the mesh
is updated between time steps along with the internal stress which is updated
using the Jaumann Stress Rate.  Once the internal stress times the B matrix added
together for each element achieves equilibrium with the external forces, the
calculation is finished.  This is expressed as:

          P_global = P_global + ( [B transpose] [stress] for each element)

          if P_global = force, stop iterations

P_global is never exactly equal to the force so you will have to decide on the
tolerance acceptable for your needs. 

The numerical integration method I have chosen is dynamic relaxation.  This
involves formulating the problem as a pseudo dynamic 2nd order ordinary differential
equation with damping, i.e.,

         [m] * [d^2x/dt^2] + [c] * [dx/dt] + [x] = [Force]

which becomes(after decoupling):

         d^2y/dt^2 + cc * dy/dt + w^2 * y = force

where "cc" is the damping and "w" is the angular frequency.  There are as many
frequencies and modes as there are degrees of freedom in a mesh.  We are concerned
only with the largest and smallest.  The rule for stability is that:

                              dt*(w max) < 2.0

and for best performance:

                              cc = 2*(w min)

which makes the system critically damped and produces the fastest convergence.  This makes
up the three quantities you have to input; time step, dt; damping, cc; and the number of
iterations.  Normally, you would also have to choose a pseudo mass.  I currently have
the program use the diagonal of the global stiffness times 100.

When I run the code, I usually have to experiment with all three quantities since
calculating (w max) and (w min) requires a bit of computation in itself.  It is 
possible using techniques such as the power method and inverse iteration to get (w max)
and (w min), and I may add this feature someday.  For now, you'll just have to try
different values for dt, cc, and iteration_max. 

I recommend doing only a few iterations at first, like 3 or 4 to see if your time
step is good.  If it becomes unstable-and this will usually happen within the first
few steps-you can slightly decrease the time step.  For the value of cc, any mesh
having more than 10 elements usually has a very low (w min) so you may just want to pick
something small for cc.  Underdamping is almost always better than overdamping.  It
should also be noted that depending on the problem, especially how large a load you
want to use, the above criteria may not be sufficient to ensure stability.  Again, the
best thing to do is experiment with all the input parameters.

