No Correspondence between Equivalent Load Steps

[fran.cesc]

Hi all,

I would like to thank anybody in advance for any help you could provide me.

I have got a panel with an aerofoil shape to which I applied a pretension as "initial force". I defined 2 load sets: both load sets are characterised by a ratio 1by2 between the force in x direction and the force in y direction (defined with respect to the element coordinate system). In particular:

a) 5x10

b) 10x20


I asked the solver to solve the equations in 10 load increments, 10 steps. Now, I thought that both the last step of the 5x10 case and the 5th step of the 10x20 case would have given me the same results: which is the result for the case in which a 5x10 force is applied to my model. So I picked a node and - with my surprise - I found out that the same node has got different displacements for the last step of the 5x10 case and for the 5th step of the 10x20 case.

I thought that this was due to the different amplitude of the load step between the 2 cases, so I tried to launch the case 10x20 with 20 increments so that, also in this case, each load step is equal to 0.5kN for the force in x direction and 1kN for the force in y direction, but again I did not find any correspondence.

I thought that you could study the load increments to see the behaviour of the model when a certain percentage of the total load is applied, but apparently there is something which I don't get. Was I wrong to assume that I could see the behaviour of the model when loaded at 50% of the total force simply by studying the right increment?

I played a bit with the software: I tried to use different convergence criteria, to tick the double precision option, to increase the number of load steps (30000), to change the BC (both constraints and load type), the material (I also used simply isotropic steel), the mesh refinement, the iteration procedure (Full Newton Raphson, Initial Stiffness, Modified Newton Raphson, Arc Length Method), I tried different meshing technique, but nothing. Only when I un-ticked the geometry non-linearity option, the two results become coincident, but this result is totally different to that which I get with the geometry non linearity option activated. For some reason, the intermediate results seem to be influenced by the final 100% load. I run different tests with simple geometries (beams, rectangular panel to which I applied the same forces as on my model) and here the correspondence between the results of equivalent load steps is perfect.


Have you got any idea about why the results of the same model when subject to the same load should depend on if this load is the 100% load rather than the 50% load? That is why the results at 100% load of the 5x10 case are not the same as those of the 50% load of 10x20 case?

PS: I am using a software which is called Midas NFX. I am attaching a picture of the model.

Thanks very much

 

[ChadV]

Are you using the same number of load steps for each case? That is, do you have 10 load steps to go to 5x10 in case (a) and 10 load steps to go to 10x20 in case (b)? If so, then model (a) will take 10 steps to get to 5x10, and model (b) will take 5 steps to get to 5x10. If the results differ by much, then the time step may be too coarse. I would expect the model (a) result at 5x10 to be more accurate. What happens when you increase the number of load steps? How much different are the results?

Have you tried allowing model (b) to take twice as many steps as model (a)? Then when they both get to 5x10 they should have the same results (unless there is some other issue in play).

 

[IDS]

Have you tried allowing model (b) to take twice as many steps as model (a)? Then when they both get to 5x10 they should have the same results (unless there is some other issue in play).

Unless I misread it, that's exactly what he said he did in the second stage check.

If the same load increments give different results there must be some other difference, other than that one analysis has more steps defined. The computer doesn't take any account of the future steps until it gets to them.

Was the preload the same in both cases?

Doug Jenkins
Interactive Design Services

http://newtonexcelbach.wordpress.com/

 

[ChadV]

Whoops, sorry IDS / fran - I completely missed the paragraph about changing the resolution to be the same load increment. Teach me to post on Eng-Tips when my mind is elsewhere!

 

[fran.cesc]

Thanks IDS, thanks ChadV

Unless I misread it, that's exactly what he said he did in the second stage check

Yes you are right. I tried both to set load step amplitudes equivalent between the two cases and a very very high number of load steps for both cases (30000 load steps).

Was the preload the same in both cases?

The only load which I am considering is that of the cases a and b. No loads or pre loads other than these.

there must be some other difference, other than that one analysis has more steps defined

This is what I thought: I thought that I have accidentally set differently some parameters between the 2 analysis. So I checked lots of times the two cases. Moreover I launched copied/pasted analyses where I simply un-checked the geometry non linearity option and I obtained the same results, so I do not really think that I made a mistake in defining the analysis case.


I applied exactly the same BC on a rectangular plate and it works fine. Is it possible that this non-equivalence is due to the particular curvature of my shell? If not, what could cause such a problem?

Thanks again