Residual Monitoring for unsteady problem
 

Hello all. While modelling Pressure pulses travelling down a straigt pipe, I encountered inacurate propagation velocities and found that the best way to eliminate these was to use the Explicit Unsteady time formulation. However when doing this The residuas do not tend to converge but remain relatively constant. Could anyone shed some light on what the residuals are showing and how should I monitor solution accuracy. Thanks in advance for your help

Re: Residual Monitoring for unsteady problem
 

Hi,

do you calculate only with turbulence model, or multiphase model or something else? If it's mixture multiphase model, the couple solver is not available. Do I understand properly?

Emma

Re: Residual Monitoring for unsteady problem
 

I am not sure how accurate this actually is, but the simplest expination I have heard for what the residuals are showing is the the difference between either side of the navier stokes equation. so your residuals are are x-mom, y-mom, pressure...etc, one for all the PDE's that you are showing. If the residual is, say, 3*10^-3 then the the difference between the values on either side of that equation is 3*10-3. Convergence is a tricky one to judge. When you say they do not converge, do you mean they drop to a low value, say 1*10^-5 and stick there, because that may be an acceptiable solution to your problem. Weather or not a solution is converged depends very much on what problem you are trying to solve.

Re: Residual Monitoring for unsteady problem
 

Emma and Andrew, Firstly, thank you very much for you replies. The residuals do not drop by much, if I normalise them they only drop by around 1 order of magnitude and then stay relatively constant. The solution that i get gives fairly accurate wavve propagation speeds but I would usually make sure that the residuals drop to ensure problem convergence. Also the problem is modelled is single phase with the basic K-e turbulence model. Any sugestions? Thanks again. James

Re: Residual Monitoring for unsteady problem
 

Residual computes the differences in the variables of interest between each iterations (n+1 and n). e.g. for the continuity, residual may be (rms of) SUM(rho(x,n+1)-rho(x,n))/(number of nodes). Depending on the integration scheme(explicit, implicit, 1st order, 2nd order) the residual may drop (if the CFL number is good for the integration) or may oscillate. But the rule of the thumb is that the residual be small (like 1e-5, 1e-6) between these iterations, guaranteeing that the solution is converging to a certain quantity that the CFD discretized equation is trying to solve for. Also, the pressure wave propagation needs to use the turbulence model?? it can be easily modeled as inviscid since the pressure pulse is very small and won't create any flow (if the pulse duration is taken very small). I would guess the wave equation would do the job (without getting into the CFD models...)