# buoyancy augmented k-eps model

 Register Blogs Members List Search Today's Posts Mark Forums Read

 June 15, 2009, 12:31 #3 Senior Member   Aram Amouzandeh Join Date: Mar 2009 Location: Vienna, Vienna, Austria Posts: 186 Rep Power: 9 Hi!! I was playing around with the discretisation/linearisation of the source terms and did some changes. The implementation showed below allows the simulation to run longer but crashes at time 6s (instead of 0.2s) due to bounding k (no bounding of epsilon). The so far obtained flow field looks plausible. volScalarField P = (3.0/2.0)*rhoRef_*gMAG_*alphah_*mut_*diffGrad.component (1)/sqr(rho_); No inclusion of k for calculating P; this is done in the source terms, so the source term for the dissipation equation is discretised in the following way: + fvm::SuSp(C1_*(1-C4_)*P/sqr(k_), epsilon_) Befor discretising the source for k, I linearised it: S_k = P/k = 2*P/k - (P/k^2)*k and hence the source for the k equation is discretised as follows: + 2*P/k_ - fvm::Sp(P/sqr(k_), k_) I would greatly appreciate your comments as I m not sure if this makes sens; at least it allowed my simulation not to blow up imediatly ... Thx in advance!! Aram

 June 25, 2009, 10:55 #4 Senior Member   Aram Amouzandeh Join Date: Mar 2009 Location: Vienna, Vienna, Austria Posts: 186 Rep Power: 9 Dear all!! I played around with different ways of formulating and implementing the source terms for my buoyancy modified k-eps model and was able to set up a non-crashing simulation. In the new implementation I first substituted mu_t in the source term P with its definition for the k-eps model. Then I set v'v' = k (suggested in [1]). So I get: P = (3/2)*g*(rhoRef/rho)*(Cmu/sigma_h)*(k/epsilon)*(u'v'*drho/dx + k*drho/dy + v'w'*drho/dz) The implemetation looks as follows: volSymmTensorField RS = R(); volScalarField diffGrad = RS.component(1)*gradRho.component(0) + k_*gradRho.component(1) + RS.component(4)*gradRho.component(2); volScalarField P = (3.0/2.0)*gMAG_*(rhoRef_/rho_)*alphah_*Cmu_*diffGrad; I did not include the quotiont "k/epsilon", so it cancels out for the additional source term in the dissipation eqn.: S_eps = C1*(1-C4)*(epsilon_/k_)*(k/epsilon)*P = C1*(1-C4)*P I tried out two ways + fvm::SuSp(C1*(1-C4)*P/epsilon_, epsilon_) and + C1*(1-C4)*P the latter showed better behaviour. The additional source term for the k eqn. is added like: + fvm::SuSp(P/epsilon_, k_) The simulation with the coarsest grid is running but I encounter some serious problems though: _ a strong sensitivity with respect to the domain size. Simulations crash when domain size gets too small or too big. _ sensitivity with respect to the initial value of epsilon (varied it by one to two orders of magnitude) where only 2.9e-5 (for my case) reuslts in a stable simulation. _ when I refine the grid (from a grid spacing of 0.025 m to 0.01 m) the simulation crashes after 3s (my main problem now!). I greatly appreciate any comment on my problem!! Thx in advance!! Aram [1] Van Maele and Merci: "Application of two buoyancy modified k-eps turbulence models to different types of buoyant plumes"