What if the calculation fails?
Contents
1. Card BPG 5 What if the calculation fails 1 2 What if the calculation fails The first things to do in case of a failure of the calculation are the following Browse the log file listing and the documentation user manual Check that the version of the code is the expected one and that it is effectively accessible on the machine Simplify the case remove as many user defined functions as possible to better identify the source of the failure for example the simulation should be checked with constant physical properties before applying variable values If the user is not perfectly familiar with the computer the case should be transferred to a machine that the user is more familiar with so as to avoid any cause associated with the operating system or with the installation of the code If the simulation fails after the log file of the kernel of the code has been produced increase the level of verbosity IWARNI to try and better identify the source of the problem If the problem persists one may try and obtain a first converged result that could potentially be improved later with the following parameters and choices One should check that the initial state is reasonable it is not essential to prescribe an initial solution close to the solution that is searched for however an unphysical initial estimate of the solution may lead to a failure of the computation for example initial velocity oriented in the opposit2. e direction initial turbulent variables incoherent between each other too large initial density gradients Some difficulties can be associated with the choice of the initial state they are normally characterized by a transient that can be evacuated through the outlet of the domain by setting a time step value corresponding to a large Courant number during the first few time steps for example the standard value of the time step may be multiplied by a factor of 10 If one suspects a difficulty associated with the mesh quality one may use the gradient computation based on an extended stencil restricted to risky regions IMRGRA 3 For tetrahedral grids the gradient computation based on a fully extended stencil IMRGRA 2 is the most reliable approach If one observes oscillations on the advected variables it is advised to modify the convective scheme as indicated in Card_BPG_4 in the worst most unstable case a first order upwind scheme should be used for all the advected variables BLENCV 0 0 If one observes that the pressure iterative solver takes a long time to converge or that overshoots of the velocity occur locally where the quality of the mesh may be questionable it is advised to under relax the pressure increments RELAXV IPR 0 7 If one observes that the convergence is difficult on a low quality mesh the stability of a RANS computation with a turbulent viscosity model k epsilon et k omega may be improved by switching off t
3. he flux reconstruction for the variables for the turbulence k epsilon omega IRCFLU IVAR 0 If spurious velocities appear close to a non vertical wall for a computation with gravity and variable density it is advisable to use the extrapolation of the pressure gradient at the wall EXTRAG IPR 1 0 If spurious velocities appear in a region where the pressure gradient is discontinuous because of the presence of a stratification or of a head loss zone it is advisable to use the option IPHYDR and to take care that the parameter ICALHY is properly set when there is a head loss on cells adjacent to the outlet of the domain When this parameter is activated it is useless to extrapolate the pressure gradient at the wall with EXTRAG IPR If spurious oscillations appear at the outlet it is advisable to use a Dirichlet condition for pressure to impose a flat profile as long as this hypothesis is physically sound Card BPG 5 What if the calculation fails 2 2 The previous analysis may lead to the conclusion that it is necessary to modify the mesh An example of a safe mode set of parameters is provided hereafter with this choice the priority is given to robustness and low CPU cost over accuracy Turbulence model k epsilon with linear production Convective scheme SOLU for the velocity UPWIND for the turbulence and the scalars No flux reconstruction for the turbulence IRCFLU 0 for k and epsilon Threshold for the co
4. nvergence of the iterative solvers 10 Pressure increments under relaxation RELAXV IPR 0 7
Download Pdf Manuals
Related Search