I have a model of a fuel tank which undergoes large scale plastic deformation during a vehicle crash. I have set up the internal volume of the fuel tank with a solid mesh through which I am performing an ALE calculation to show the fuel/tank interaction. As the tank wall undergoes large scale plastic deformation, the solids collapse creating instabilities, this causes the model to fail early in the analysis. Any ideas how to overcome this?

Just a guess at the obvious (which may be wrong!):

Cut your time step drastically and rerun.

Thanks Jim,

I agree, but the model is already close to its limit. I already refined the mesh in the critical areas and hence reduced the time step. Reducing it any further will send the cpu time through the roof. Please let me know if you can think of any other suggestions.

Regards

Christian

The ALE code I worked with at Los Alamos (in the 80's ...) had a limiter on the time step that prevented each node from being moved more than the local cell dimension. Usually one set a parameter that was a ratio of the allowed movement to the cell dimension (0.3 as an example). Trouble was, during the iterative process, the provisional movement might momentarily become too large anyway and the calculation crashed.

Having the time step cut by refining the mesh wouldn't (I think) change the ratio. If your code allows you to do so (or even has such a ratio), that's where I'd try to impose the cut.

We're now thoroughly into the area beyond my knowledge! Good luck!

If the topology of the problem is naturally highly skewed or there are "physical" collapses involved, it will not make a difference how small of a time-step you take. You will eventually end up with the same problem!

There are two solutions that I know of: Apply remeshing or use meshless methods.

Remesh: Every few timesteps you take the solution and project it onto a new set of grid points which are basically uniform and "well behaved/distributed"

Meshless: This is more complicated and time-consuming in your case, but the idea behind meshless solutions (such as in finite element meshless methods) is that there is no connectivity between a node and its neighbors. This eliminates most of the problems that you are experiencing now. Of course, there are other issues involved which are beyond the scope of this note.

Adrin Gharakhani

Thanks once again Jim,

I can limit the time step, but as its an explicit code I have to pay a mass penalty. The problem is that as the solids collapse distance between the faces reduce and no mater how much I refine it I will end up in the same boat eventually.

Thanks for your assistance and best regards

Christian.

Thanks Adrin,

I can't remesh using this package. I am reading up on SPH methods just now but its going to take a while to get up to scratch. I was hoping that someone might have a method which could help me move on a bit with the ALE simulation while I get on top of the SPH option.

Many thanks

Christian

Be careful with SPH. There are lots of issues you need to understand well. The issue of boundary condition has been one of the more difficult altough I have seen some good progress in recent years. If you have the source code for the ALE program you may want to try modifying it using meshless finite element methods (or equivalently the moving-least-squares approach)

I have been doing similar fuel tank crash simulation involving fuel sloshing inside the tank at the same time a couple years ago. If you have local high distorsions of the tank skin, the problem obviously become very complex for the ALE mesh. Here are two things that can help: -Make sure your initial fluid mesh is as coarse and as regular as possible. By doing so, the collapse of elements will be avoided or at least appear later -Use what we call 'ALE rigid links'. With our code, RADIOSS-CFD, one can use those links to localy rigidify the mesh. Therefore, the mesh distortion can be spread further in the model and it becomes smoother.

Regards,

Dimitri Nicolopoulos