Discussion Forum

I am new to comsol (v3.5a) and am trying to teach myself. I would like to (eventually) model vortex shedding of a tapered pole, working through some intermediate steps. I have looked at the cylinder flow and obstacle in fluid models as a starting point. I have tried extending the cylinder flow model into 3-D and using just a cylinder in the obstacle model, but I have not been able to get a transient solution. I am also unable to get any solution for a RE >~500. Also, I would like to introduce motion into the cylinder flow model to simulate vibration of the pole. Any suggestions?

thanks

Hi Brian,

Can you be more specific on what didn't work?

3D cylinder in the flow should not be a problem. Movement of the cylinder is tricky, but using ALE eventually you can do that too. Try and when you get stuck post here, and we can check what possibly went wrong.

Danial

I've attached a few models.

For the 3d cylinder (based on obstacle in fluid model), it solves as a stationary model, but if I try transient I get 2 different errors. I tried adding flc1hs(t[1/s]-0.1,0.1) to the velocity boundary condition. With it, I get the error

------------------------------------------------
Time 0.009000000000000001:
Nonlinear solver interrupted since convergence rate above limit.
Last time step is not converged.
------------------------------------------------

Without it I get
------------------------------------------------
Failed to find consistent initial values.
Segregated_group_X1

Sqrt_of_negative
Function: sqrt
Failed_to_evaluate_expression
Expression: nojac((1/max(((2*rho_ns)*sqrt(emetric_ale((u2-xt),(v2-yt),(w2-zt)))),((eta_ns*48)/(h_ale^2)))))
Failed_to_evaluate_variable
Variable: taum_ns, Defined_as: nojac((1/max(((2*rho_ns)*sqrt(emetric_ale((u2-xt),(v2-yt),(w2-zt)))),((eta_ns*48)/(h_ale^2)))))
Failed_to_evaluate_expression
Expression: nojac((min((gijgij_ns^-0.25),((U_ale_ns*rho_ns)/(eta_ns*48)))*(0.5*min((sqrt(((((res_v2_ns*taum_ns)^2)+((((res_p2_ns*tauc_ns)/(nojac((rhoint_g1_ns/vol_g1_ns))*nojac(sqrt((u2int_g1_ns/vol_g1_ns))))))^2+((taum_ns*res_u2_ns))^2))+((res_w2_ns*taum_ns)^2)))*ck_ns),U_ale_ns))))
Failed_to_evaluate_symbolic_derivative_variable
Variable: shock_capt_ns@VDN${test@15}, Defined_as: ((rho_ns*(-nojac((min((gijgij_ns^-0.25),((U_ale_ns*rho_ns)/(eta_ns*48)))*(0.5*min((sqrt(((((res_v2_ns*taum_ns)^2)+((((res_p2_ns*tauc_ns)/(nojac((rhoint_g1_ns/vol_g1_ns))*nojac(sqrt((u2int_g1_ns/vol_g1_ns))))))^2+((taum_ns*res_u2_ns))^2))+((res_w2_ns*taum_ns)^2)))*ck_ns),U_ale_ns))))))*((g31_ns*u2z)+((g21_ns*u2y)+(g11_ns*u2x))))
Failed_to_evaluate_expression
Expression: d((((-2*eta_ns*u2x+p2)*test(u2x)-eta_ns*(u2y+v2x)*test(u2y)-eta_ns*(u2z+w2x)*test(u2z)-rho_ns*(u2*u2x+v2*u2y+w2*u2z)*test(u2)+gls_ns+shock_capt_ns)-(0))*(dvol_ale),{test@15})
---------------------------------------------------------

For the 3d cylinder 2 (based on cylinder flow model), it also solves as a stationary model, but not transient

With the flc1hs function, I get
------------------------------------
Error:
Time 2.103739315846686e-008:
Nonlinear solver did not converge.
Last time step is not converged.
-------------------------------------------

Without it, I get
----------------------------------------
Error:
Failed to find consistent initial values.
Last time step is not converged.
-------------------------------------------

I'm guessing I need to use the flc1hs function (or a variant) and figure out how to make it converge.

thanks

As far as motion goes, can I do a small cyclic displacement in 2d to simulate vibration?

thanks

As far as motion goes, can I do a small cyclic displacement in 2d to simulate vibration?

thanks

For 3D cylinder cases, I would suggest you to first learn what you're doing. There is another post here where I mentioned "obstacle in the flow" model and its definition. Then, comes the solver issues (I suggest remove segregated solver), and finally the fact that FSI in 3D is tricky and slow, extremely slow.

If you just need to move a cylinder up and down, you do not need a full FSI implementation. You can just use the Moving Mesh and apply a velocity to the moving boundaries.

Hello,

I am using COMSOL v4.0a to solve the F=free vibration of elastically mounted cylinder in Re=100 flow.

I am thankful to COMSOL v4.0a user friendly interface, I have set up the problem and able to simulate the vibration of cylinder. I am using mesh motion coupled by displacement to an ODE representing spring mounted cylinder.

Results look good except for the magnitude of the cylinder displacement. Literature suggests the maximum displacement at Re=100 for a mass ratio =10 is around 0.6*dia of cylinder. But in my simulation I am seeing displacements of ~1.5*dia of cylinder. I am thinking I should tune some parameters so that the solution represents actual physics.

I have tried both solver, gen alpha and BDF and various time step and accurately level which is not helping much. I am not sure how consistent/inconsistent stabilization affects mesh motion. Can someone let me know of how to tune the mesh motion and fluid flow to represent physics accurately.

I have tried both solver, gen alpha and BDF and various time step and accurately level which is not helping much. I am not sure how consistent/inconsistent stabilization affects mesh motion. Can someone let me know of how to tune the mesh motion and fluid flow to represent physics accurately.

It looks like a simple physics but actually even changing tolerance/solver parameters/Artificial diffusion (eventually I convinced support that there was something wrong about it in v3.5a) can change your result. I had played with oscillating cylinders long time ago just to get used to comsol. But without an actual model to play with I can not say much.

Dear daniyal, (salam agha daniyal)
I put my question regarding FSI convergence problem on COMSOL forum a few days ago. However I haven't received any answer for it yet. I was wondering if you could give me some advice on how I can sort out the convergence issue. My question is:
There is an example of Fluid-Structure interaction problem given in Comsol model library which is an oscillating beam. I tried to give some change to that model. I separated the beam from the cylinder to investigate the detached splitter plate, removed the Gaussian pulse function and the point load, and applied a fixed displacement constraint to the plate's edge opposite the cylinder. The solution diverged. even when I assumed that the plate is rigid; the solution did diverge . If I run this simulation for an splitter plate attached to the cylinder without point load and Gaussian pulse function, the solution is convergent. Can anyone suggest why my simulation is not convergent in the other two cases?
Many thanks,
Bryan
(mamnoon misham rahnamayim koni)