View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0001779 | OpenFOAM | Bug | public | 2015-07-09 13:03 | 2015-10-12 21:37 |
| Reporter | jkau | Assigned To | henry | ||
| Priority | normal | Severity | minor | Reproducibility | always |
| Status | closed | Resolution | fixed | ||
| Platform | GNU/Linux | OS | Ubuntu | OS Version | 15.04 |
| Summary | 0001779: Stability of sixDoFRigidBodyMotion | ||||
| Description | This is somehow related to report 0001763. I think in "sixDoFRigidBodyMotionSolver::solve()" and "sixDoFRigidBodyDisplacementPointPatchVectorField::updateCoeffs()" resp. updatePosition should be called after update acceleration, otherwise one gets hardly convergence within one timestep. With this modification, one can also increase the acceleration relaxation factor to higher values to obtain faster convergence without stability issues. With the original formulation, stability even decreases with increasing outer iterations! The attached example is from: http://www.cfd-online.com/Forums/openfoam-solving/153783-interdymfoam-floating-object-snappyhexmesh-not-solving.html It not a perfect test case, but should be good enough to demonstrate the general problem. Just run for one time step with high number of outer iterations. | ||||
| Steps To Reproduce | Download attached file float3.tgz extract with 'tar xvzf float.tgz' run Allrun run the case with interDyMFoam for one time step with something like 20 outer iterations plot the Linear or Angular velocity vs. iterations, e.g. in gnuplot: p "<grep Linear log.interDyMFoam | awk '{print FNR,$5}' | tr -d '()'" w l Modify sixDoFRigidBodyDisplacementPointPatchVectorField.C as desribed in Description. Run again. See (different) convergence of linear and angular velocity. | ||||
| Tags | No tags attached. | ||||
|
|
|
|
|
The ordering of updatePosition and update acceleration relates to the specification of the method to conserve both linear and angular momentum using a leap-frog approach. However, the method is explicit and should be called once per time-step. It is not clear how the method should be adapted to conserve both linear and angular momentum when applied iteratively with relaxation and also to maximize convergence rate. It would be useful if someone would analyse the method in detail and propose an alternative formulation which has the desired behavior. |
|
|
Ok, thanks for the feedback. I hope, I will find some time to go into details and can propose an alternative formulation. |
|
|
I have tested your proposal on some simple cases with analytical solutions and the results are TERRIBLE! While the simplectic nature of the algorithm is slightly compromised by iterating within a time-step the error is small and does not accumulate. However, changing the order of update of the velocity, acceleration and position completely destroys the algorithm resulting in serious phase and amplitude errors in all variables. In order construct an more stable algorithm consistent with iteration over the force and position updates the current simplectic approach would need to be completely replaced. If you have time to work on this please re-open this report with the details when you have something working. |
|
|
commit 57c32b05036b9157ce6de2cd8fd1f1b5e8523368 Author: Henry Weller <http://cfd.direct> Date: Mon Oct 12 21:27:42 2015 +0100 sixDoFRigidBodyMotion: Time integration now switches between symplectic and Crank-Nicolson For explicit motion (and the first iteration of iterative motion correction) the 2nd-order symplectic motion integrator is used. For iterative correction a form of lagged Crank-Nicolson is used in which the current time-step values correspond to the current iteration. This converges to a 2nd-order implicit solution. |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2015-07-09 13:03 | jkau | New Issue | |
| 2015-07-09 13:03 | jkau | File Added: float3.tgz | |
| 2015-07-09 13:42 | henry | Note Added: 0005056 | |
| 2015-07-09 14:22 | jkau | Note Added: 0005058 | |
| 2015-10-09 17:50 | henry | Note Added: 0005387 | |
| 2015-10-09 17:50 | henry | Status | new => closed |
| 2015-10-09 17:51 | henry | Assigned To | => henry |
| 2015-10-09 17:51 | henry | Resolution | open => fixed |
| 2015-10-12 21:37 | henry | Note Added: 0005389 |
