CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CFX (http://www.cfd-online.com/Forums/cfx/)
-   -   Time step for combustion simulation (http://www.cfd-online.com/Forums/cfx/70221-time-step-combustion-simulation.html)

Time step for combustion simulation

how can i calculate or estimate the time step for a combustion simulation?
im using flamelet, k-e RNG, methane-air,(its a tubular combustion chamber)

and what numeric scheme its more appropiate for this? high resolution? agressor?

best regards

 ghorrocks November 17, 2009 21:05

The CFX documentation has the best overall guide, have a read.

In general aim to get 3-5 iterations per timestep. Maybe a few more for multiphase stuff, up to 10 iterations. This is easily put in as adaptive timestepping so the simulation can find its own timestep size.

Combustion is particularly sensitive to this sort of thing due to the very rapid timescale of the chemical reactions. This means you may have to use much smaller timesteps around the combustion time (if transient combustion like an IC engine) or small timesteps when you set up a continuous combustion case.

 joey2007 November 18, 2009 14:43

Thats true for EDM, FRC, BVM and ECFM. For pre non-premixed flamelet the mixing is dominating not the reaction. So it may be possible to work with larger time steps. Adaptive time stepping may solution in any way.

Ghorrocks and Joey2007 thanks very much for your answars.
im reading the documentation of ansys for time-step, and i think the better choice is to use physical timestep, and that its defined like L/2U , but i dont have te value of U?, how can a set an apropiate physical timestep?, i should make a simulation with auto timescale and gete from there an average of U?.

(my simulation its a steady state combustion), the formulation of the time-scale uses in CFX its fully implicit, that's reffered to crank-nicolson method?

thanks very much:D

 ghorrocks November 24, 2009 07:07

CFX does not use Crank-Nicholson, just first and second order Eularian schemes. For steady state runs it uses a first order Eularian scheme but with a number of the transient terms removed so that is why it is psuedo-time (as it is not a true time advancement scheme).

For a steady state run start with any time step size which will converge and once it has converged for a while ramp the timestep up to accelerate convergence. I recommend using physical timesteps as a first preference, and local timescale factor only if the physical timesteps are not converging. But either way the final convergence should be obtained with physical timesteps.

thanks veryvery much for the hepl ,im understanding know. but my doubt its
: how can know the values for the physical timesteps? or the local timescale factor?, how can i choose this times???

for this simulation i have:

pressure inlet for gas: 14.31bar
pressure inlet for air(primary,secundary,dilution,refrigeration,swirl ing): 6.40bar

outlet condition: im working on it

thanks very much

 ghorrocks November 24, 2009 17:44

This is discussed in the documentation under obtaining convergence. Also the following may help http://www.cfd-online.com/Wiki/Ansys...gence_criteria

thanks very much =)

best regards

do you know if, flamelet use a log normal distribution for the scalar dissipation rates Xst?, or flamelet just, use the beta pdf for z, zvariance and Xst?

best regards

 zhiyuam December 2, 2009 22:01

Lognormal for dissipation rate, and beta pdf (get from farver average Z and its variance) to calculate the reactive quantity. This has been proved in someones paper but I do not remember which one.