Learning Forum

[ncalzolano]

I'm currently creating an assignment to simulate laminar flow (Re = 500) in a straight, axisymmetric pipe. The purpose of the assignment is to show the benefit of using inflation layers to obtain near-wall accuracy (in lieu of a uniform mesh) in terms of computational time to obtain the solution. The assignment has the students run cases with both inflated and uniform mesh. In both cases, the continuity residual fails to converge. I've sifted through other posts regarding continuity residual convergence (of which there are quite a few) and haven't come across anything useful (this problem is so simple, most of what I've read isn't even applicable).

The pipe being simulated has a radius of 0.001m and a length of 0.1m. It is represented as 2D axisymmetric by a rectangular domain. Both mesh configurations are all quad with zero average skewness and an average orthogonal quality of one. The first cell height next to the wall was estimated through the y+ method using y+=1 for the cell next to the wall. The mesh configurations are shown below.

[YasserSelima]

The questions are:nAre you getting the expected velocity profile or not?nDoes the solution remain the same after running more iterations? Did you try monitors?nResiduals in order of 10^-3 are not bad if the solution doesn't change.nOne question to you, why is y+ important if you have laminar flow?.

[Karthik R]

Hi,nAlso, try coupled pseudo-transient with hybrid initialization.nIs the pipe long enough to get a fully-developed flow? nKarthikn

[ncalzolano]

Hello, thank you for your responses!nVelocity profile is as expected. I monitored convergence of area-weighted average velocity over several cross sections of the domain and they appear to converge. More iterations do not seem to lessen the residual. It simply hangs around one value.nUsing coupled pseudo-transient with hybrid initialization allowed convergence to 10^-3 for continuity (which is what I had set as my criterion), but the same thing happens with the residual trends just after 10^-3. I'm still a bit confused why residuals would behave like this for such a simple problem.nn

[Rob]

Residuals are a function of the error in the solution, and scale at about iteration 5 (it's in the Solver theory). We use those along with monitors and fluxes to check convergence. If the initial conditions are very similar to the converged solution residuals don't always behave as in the text book. nGiven the monitors and post processing look right I'd be fairly confident that the results are correct. n

[YasserSelima]

nIf this is the case, I would trust the solution given that I have done sensitivity study.n