A drastic example for the danger of non-converged solutions is the following flow around a generic airplane geometry with high-lift devices (slats upstream and flaps downstream of the wing employed) deployed. The simulations for the JAXA geometry were carried out in the framework of the 3rd AIAA CFD High Lift prediction Workshop (https://hiliftpw.larc.nasa.gov/Workshop3/geometries.html). A first simulation with the SST model using default model and steady state settings resulted in a substantial under-prediction of the maximum lift coefficient () due to a massive separation on the main wing. This can be seen in Figure 12.159: Figure 96: Wall shear stress for high-lift airplane simulation for =18 . Left: Simulation in 'steady-state mode'. Right: Simulation in 'unsteady mode'. (left) showing the skin friction on the surface for the 'steady-state' set-up for an angle of attack of . At first, this was attributed to the SST model's separation prediction capabilities, which were apparently interpreted as being overly 'aggressive' for this case. For this reason, the coefficient in the SST model was increased to 1 to reduce separation sensitivity. However, it was found that convergence for the case with the default SST model setting was poor and the case was therefore re-run with unsteady solver settings. The difference due to this change in numerical settings was dramatic as shown in Figure Figure 96 (right) which shows the skin friction distribution for the unsteady run. The unsteady set-up allowed the simulation to 'escape' from the incorrect flow topology and move to a more realistic and much less separated condition.

The effect on the lift coefficient can be seen in Figure 12.160: Lift coefficient, CI, for high-lift aircraft simulation versus angle of attack (AOA). RANS-SST with default and 'steady state mode', SST with a1=1.0 and 'steady state mode' and URANS-SST with default SST and 'unsteady mode' set-up. which contains three curves. One for SST-default and 'steady state mode' one for SST with (instead of default ) and again 'steady state mode' and finally with SST default () and 'unsteady mode'. The change in the a1 coefficient results in less separation and therefore a better agreement with the exp. data. However, such a change in is not justified from a physics standpoint, when considering at 2D airfoil performance (see 4.2.2). Alternatively, the solution was also improved significantly by just switching from the steady-state set-up to the unsteady set-up. The lift for the URANS simulation is still somewhat lower than the experiment, but by a much lower factor compared to the steady run. There is also indication that including laminar-turbulent transition in the simulation will further increase the lift curve, so that the remaining difference can partly be attributed to this effect. Besides such details, the important message here is that the difference between the steady and unsteady solutions with identical turbulence model settings can be larger than the differences between strongly different turbulence model settings.

Figure 12.159: Figure 96: Wall shear stress for high-lift airplane simulation for =18 . Left: Simulation in 'steady-state mode'. Right: Simulation in 'unsteady mode'.

Figure 12.160: Lift coefficient, CI, for high-lift aircraft simulation versus angle of attack (AOA). RANS-SST with default and 'steady state mode', SST with a1=1.0 and 'steady state mode' and URANS-SST with default SST and 'unsteady mode' set-up.

It is important to stress that situations like the one shown above appear very seldomly, but they can happen in CFD. The exact cause for such scenarios is not easy to determine. However, there are some contributing factors: