Adaptive Time Stepping: extrange behavior!

Freeman

I'm running 2D simulations of a flow over an automobile. I set up my problem as follows:

-Vel inlet from 10 to 50m/s (independent simulations with each velocity)

-I start with k-e standard, 1st order schemes in momentum, k and epsilon. Then I swich to k-e Realizable with 1st order scheme.

-Then, I run an unsteady Realizable with 2nd order scheme (because if I continue running steady simulations I don't get any convergence: ALL parameters remain fluctuating periodically: residuals, Cd, Cl...). I activate NITA and PISO options to make iterations run faster. Finally, I activate Adaptive Time Stepping (ATS) and I start with a time step of 0.001.

While the simulation runs, Fluent reduces the time step by a factor of 2, and it doesn't seem to change. If I start the simulation again with a time step of 1e-20... Fluent still reduces this step even more (and so on)!

Where is my mistake? Has anyone successed using ATS? Which was your initial time step and how did you calculate it (Formula?)?

Many thanks!