Learning Forum

[npayne1]

Greetings,

I have been developing a multi-species model where CO2 is injected through a small circular inlet (orifice) into an air-dominated box domain. The inlet is centered on one surface of the box while the outlet takes up the entire opposing surface. The simulation is transient with velocity inlet, outflow outlet, and walls as the boundary conditions. Over time the CO2 sinks in the domain resulting in a near-floor flow with higher concentration of CO2. Near the outlet and near the floor, the continuity error is significantly larger than in the rest of the domain. This high error means results in an overall higher continuity residual on the order of 1e-1 or 1e-2. 

I know that a simulation is usually considered converged when all residuals are below 1e-5 but have gotten advice from an Ansys representative suggesting that convergence can be ensured in other ways for exceptional cases. Here, strangely enough, the individual residuals for CO2 and air are low even though the overall continuity residual is high. Furthermore, this error is localized to the region described near the outlet. This error does not seem to significantly decline with increasing mesh refinement. 

Could you provide advice on whether or not this behavior is abnormal or typical for multi-species flows and make suggestions for how residuals might be reduced if this behavior is abnormal. So far, I have ensured a Courant number of 1, parametrically varied pressure and species relaxation factors, and attempted mesh refinement without it much impacting the errors or improving the flow field for the better.

 

[Rob]

Have a look at monitors and flux balance. You may also want to plot contours of "stuff" every some iterations to see how the flow is behaving. I can't tell you it's converged, I can only suggest things to check: you need to judge if the solution is good enough for what you need. 

As an aside, it's 3 orders of magnitude for most fields, 5 for species & 6 for energy: it's covered somewhere in the Intro training and likely in at least one course on Learning.