Termination of Analysis

[taneryilmaz]

Hi to all,

I'm dealing with a 3D bridge model and my problem is having the nonlinear time history analyses terminated before reaching their finishing times.
I'm using the following elements:
Columns are modeled with displacement-based fiber elements.
Pile groups are modeled with linear elastic beam-column elements attached to the columns at the top with rigid links.
Soil springs are modeled with zero-length elements being assigned with pySimple1 materials in two orthogonal directions.
Abutment-backfill interaction is modeled with zero-length elements with EPP material.
At the in-span hinge, there are zero length elements with EPPgap and EPP materials in long. direction.

I'm using a loop in which I use different solution algorithms one by one similar to the given time-history examples (at the last try, I use a large number of iterations for Newton-Raphson and force to continue analysis even if convergence is not met)


- With the complete model (all foundation elements are included), I have tried the "BandGeneral", "UmfPack" and "SuperLu" solvers for the dynamic analysis and all of them yielded the termination of analysis.
When I use "UmfPack", the error message says: "UMD2FA: matrix is singular. Only 20790 pivots found. WARNING UmfpackGenLinSolver::solve,void.- 4 returned in factorization UMD2FA"
When I use "BandGeneral", the error message says: "LAPACK routine returned 19"
When I use "SuperLU", the error message says: "Error 829 returned in factorization dgstrf<>"

- Having such a problem may be due to the foundation elements (such as soil springs), I removed all of the foundation elements (having fixity at the bottom of columns), then my analysis was completely finished (I used "UmfPack" solver)

- I have modeled the piles (and soil springs also) at each bent/abutment one by one having fixity for all other bent/abutment locations. Then I got my analysis finished when I only model abutment piles, and the piles at one of the bentsone of the bents. But again for all other cases (modeling only the piles at other bent locations) ended up with failure of analysis.

I could not see any consistency why I get my analyses unfinished. I checked my boundary conditions many times. I don't see any problem in the values I assigned for the p-y spring elements.

I need guidance to overcome this problem.

Thank you in advance.

Taner Yilmaz.

[fmk]

if all the solvers are failing it means the system of equations is becoming singular, i.e. rigid body modes have developed. if the problem is with the springs for the piles and you want the analysis to work, look at modifying the materials to have some nominal positive stiffness at all times, i.e. you may put it in parallel with some elastic material with a small enough E to overcome the problem. It should overcome the numerical problem of the singularity, however it is only a numerical trick and whether it is an appropriate modeling change is up to you.

[taneryilmaz]

Thank you so much for the reply Frank,

I will apply the numerical trick as a last method. I am now checking y50 and pult numbers again for my p-y springs again to be completely sure on them. I'd like to note that once I used a single set of y50 and pult values for all of my spring elements at a bent, the analysis finished smoothly.
Taner.