[kveki]

Hello!

I apologize for bad English, I use google translator.

I ask for help in one question. I will demonstrate it in the model problem of the two-dimensional gas flow problem. I use my little cfd-program. After several iterations, the velocity field has the form in Fig. 1. The matrices for velocities are shown in Fig. 2. It is seen that for cell 8, the central coefficient is several orders of magnitude lower than for other cells. To calculate the system, I use the simple Gauss method of subsequent exclusion of variables (Fig. 3). Obviously, for cell 8, when divided by the elements of the main diagonal, extremely high speeds are obtained. I use the SIMPLE pressure correction method, it corrects the situation. However, at the next iterations, the effect occurs again in other cells and the problem falls apart.
I described a model problem, but the same effects are also manifested in other problems that I have considered. Low central coefficients are obtained for cells when there are oppositely directed velocities at the boundaries, as for cell 8 in Fig. 1. When solving these problems on Fluent, I do not get such effects.

My question is how to overcome this problem.

[duri]

Try initial value u=1 and v=1. If this is your complete domain two inlet and two outlet you should get converged in 1 iteration otherwise something wrong with your boundary condition.

[kveki]

Originally Posted by duri

Try initial value u=1 and v=1. If this is your complete domain two inlet and two outlet you should get converged in 1 iteration otherwise something wrong with your boundary condition.

You are absolutely right. If you set the initial conditions u=1, v=1 m / s, then everything converges in one iteration. It also works well if we take the initial field equally directed with boundary velocities. However, if we take the initial conditions, when the direction of the initial velocity field does not coincide with the direction of the boundary conditions (for example, u=-1, v=-1 m/s or u=1, v=0 m/s, etc.), the case I described occurs. The field tries to line up in the right direction, but in the process of solving, pictures appear, as in the picture, when the speeds in the cells are directed in the opposite direction. Then very high speeds occur during iteration in these cells and the task falls apart. If you solve similar tasks in fluent, for example, nothing like this happens.

[duri]

The non convergence with bad initial guess is common in any solver. The robustness of the solver changes with the numerical schemes. To benchmark against fluent you need to benchmark against the solver settings and the boundary condition implementation against fluent. Usually having lower time step, relaxation, etc. gives better stability. Try to mimic your solver in fluent with right multistage coef., multigrid, gradient estimation, limiter, flux scheme, ...