Reducing Initial Time Step Benefits for FE Analysis

[pukb]

What would be the practical application of reducing the initial time step in a FE analysis.
By Initial, I mean - the commercial software always calculates stable timestep based on geometry and material properties. There is always a provision given to use a lesser time step value at start of the run.

Where would this be useful.

 

[AlephZero]

You don't give enough context (type of problem being solved, name of FE program, etc) to give a good answer, but in general, any "simple" estimate of the behavior of a sufficiently nonlinear model will be wrong, so the user needs the option to override it.

Also, for a problem in the time domain, the required step size might be determined by the loading, not the structural response. For example consider a mass on a spring with a vibration frequency of 1 Hz, but an applied load with a frequency spectrum that extends to 1000 Hz. The automatic time step estimate for transient dynamics is likely to be of the order of 1 second, not 0.001 seconds.

 

[pukb]

The problem solved is EXPLICIT in PAMCRASH solver.
Since the problem is explicit, we are not looking at frequency response.

There is a provision to give a lower timestep (lower to the stable one calculated by solver) throughout the run - it is understood
also, as i mentioned earlier htere is a provision to give a lower timestep for only the beginning of the run - i don't understand why this would be required - cannot think of a practical scenario

 

[AlephZero]

i don't understand why this would be required - cannot think of a practical scenario

Suppose the response at the end of the first time step involves some nonlinearities in the model (contact, plasticity, etc). The program may not be able to estimate a sensible size for the first step based on the (linear) initial conditions.

If the first step is inaccurate because it is too big (even if it is mathematically "stable"), the errors may propagate through the rest of the solution.

An extreme example might be something like a string fixed at the ends, with zero elastic stiffness and no initial tension, subject to a transverse force (which will deform the string, create some tension in it, and therefore give the string some stiffness). A stability analysis might decide that the first time step could be infinitely large, since the initial stiffness is 0.

 

[pukb]

Hi AlephZero
Can you please give a more practical example for this.

 

[AlephZero]

Make you own model in PAMCRASH. Model an elastic band (with no initial tension) pinned at each end, with beam elements. Initial displacements and velocities all zero. Apply a constant transverse force at the mid point, big enough to stretch the band sideways to an angle of about 45 degrees. I predict the initial time step will be much too big for an accurate dynamic solution. If you get a solution at all (i.e. you get at least one complete cycle of the oscillation before anything blows up) reduce the initial step size by a factor of 10 or 100 and compare the answers.

Note, I don't use PAMCRASH, so it might be smarter than I think. But their sales team have tried (and failed) to sell it to me a few times...