The OpenSees Community

Please find them below:

****************************************************
M A T E R I A L S
****************************************************

# --------------------------------------------------------------------------------------------------------------------------------------------------
# U N I A X I A L : S T E E L 0 2
# --------------------------------------------------------------------------------------------------------------------------------------------------
# uniaxialMaterial Steel02 $matTag $Fy $E $b $R0 $cR1 $cR2 [$a1 $a2 $a3 $a4 $sigInit]
# --------------------------------------------------------------------------------------------------------------------------------------------------
uniaxialMaterial Steel02 800 465000 2E+08 0.0005 18.5 0.925 0.15;
uniaxialMaterial MinMax 100 800 -min -0.035 -max 0.1 ;

uniaxialMaterial Steel02 801 465000 2E+08 0.0005 18.5 0.925 0.15;
uniaxialMaterial MinMax 101 801 -min -0.1 -max 0.1 ;

# --------------------------------------------------------------------------------------------------------------------------------------------------
# n D M A T E R I A L : P L A N E U S E R S T R E S S M A T
# --------------------------------------------------------------------------------------------------------------------------------------------------
# nDMaterial PlaneStressUserMaterial $matTag 40 7 $fc $ft $fcu $epsc0 $epscu $epstu $stc
# --------------------------------------------------------------------------------------------------------------------------------------------------
nDMaterial PlaneStressUserMaterial 2000 40 7 31250 1960 -6275 -0.0024 -0.006 0.001 0.08;
nDMaterial PlaneStressUserMaterial 2001 40 7 40370 1960 -8074 -0.0085 -0.044 0.001 0.08;

# --------------------------------------------------------------------------------------------------------------------------------------------------
# n D M A T E R I A L : P L A T E F R O M P L A N E S T R E S S
# --------------------------------------------------------------------------------------------------------------------------------------------------
# nDMaterial PlateFromPlaneStress $newmatTag $matTag $OutofPlaneModulus
# --------------------------------------------------------------------------------------------------------------------------------------------------
nDMaterial PlateFromPlaneStress 2500 2000 1.0584E+07;
nDMaterial PlateFromPlaneStress 2501 2001 1.0584E+07;

# --------------------------------------------------------------------------------------------------------------------------------------------------
# n D M A T E R I A L : P L A T E R E B A R
# --------------------------------------------------------------------------------------------------------------------------------------------------
# nDMaterial PlateRebar $newmatTag $matTag $sita
# --------------------------------------------------------------------------------------------------------------------------------------------------
nDMaterial PlateRebar 3000 100 90;
nDMaterial PlateRebar 3001 101 0;


****************************************************
S E C T I O N
****************************************************
section LayeredShell 2 8 2500 0.025 3001 0.000335 3000 0.000443 2500 0.0991 2500 0.0991 3000 0.000443 3001 0.000335 2500 0.025;

Okay, I seemed to have found the source of my own problem. I'll write it down here just for the records.

In my model I had some SHELLMITC4 elements and they inherited LayeredShell sections. Apparently, Mumps doesn't like it. Soon as I switched to ElasticMembranePlateSection, things went back to normal.

Ömer.

I am encountering the the exact same problem. I have two different models. One of them runs fine on all threads but the other one can only run on one core?

Does anybody have an idea what might be causing such error?

Ömer.

That is exactly what I did, many thanks Frank for the clarification!

fmk wrote:
> the numbering of the dof does not effect the answer (at least if numerical
> roundoff is not an issue) .. the error you see from the force beam column
> is coming from the formulation, it is force based and not displacement
> based .. as a consequence at each trail set of displacements the element
> does some iteartion to find equi which a displacement based ele does not
> have to do .. the iterations require that the section flexibilty be
> inverted .. if it cannot invert the flexibility matrix you get the error
> you see .. to avoid the error make sure that the matrix can be inverted
> (think about adjusting material properties, i.e. if elastic perfectly
> plastic give the material a small bit of hardening) .. the theory can be
> found in the references:
> http://opensees.berkeley.edu/wiki/index ... mn_Element

You cannot capture such phenomenon using beam-column elements. You can read the works of D. E. Lehman, L. N. Lowes and J. Pugh on modelling these shear wall elements.

The only option that I am aware that you can consider is to use shell elements. There is this multi-layered shell element (MITC4) developed by Xinzheng Lu. However, I tried using it in the past and it has major problems.

Cheers.
Ömer

Dear forum,

I would like ask a quick question about a particular error that the force-based element spits. The error message is "ForceBeamColumn3d::update() -- could not invert flexibility\n". You might be familiar with this too. This means that it cannot obtain the element stiffness matrix. I am not entirely sure what is the reason for this but I am usually running into this problem with elements that have a large sectional dimension proportions (h/b). And apparently it is quite sensitive to how the fiber section is created. In some cases I could avoid this problem by increasing the number of fibers along the short side axis of the section but not always. The strange thing is that I run into this problem even right before the structure is in free vibration after a strong shaking (meaning that the error does not arise during the actual loading)... Does the numbering + the solver scheme and how the ax=b is handled affect this? Does anyone have any consistent solution to this problem?

Many thanks.
Ömer

Thank you very much for the prompt reply! I had the same doubt for a long while untill I ran the same analysis using only the core-wall configuration where the same elements still exist. And it runs just fine.

I am wondering if it's the diaphragm causing problems together with the gravity beams across plan (which are modelled as beamWithHinges elements as I mentioned). Because some of them connect to another beam instead of a column joint. I put up the following visuals for clarity, please check them out.

Plan view: https://drive.google.com/open?id=1SdLXT ... Scw22cy0jO
3D view of the first story (OS model): https://drive.google.com/open?id=1K_YK- ... a6B-70KnRJ

Hello all,

I wish to consult with you expert Opensees users about a particular problem I have. I am first putting up a general description of my model and then I will explain the type of problem I am encountering. Finally, I will share some observations that might trigger a light-bulb moment for the reader. Before I start however, I'd like to point out that the model does run both gravity and eigenvalue analysis. These results seem to agree with basic hand calculations. The challenge is to run a successful dynamic analysis.

Plan view: https://drive.google.com/open?id=1SdLXT ... Scw22cy0jO
3D view of the first story (OS model): https://drive.google.com/open?id=1K_YK- ... a6B-70KnRJ

Note: I am willing to share the full model with those who think that they can help.

As the title suggests, I have a 3D model of an RC tall building in Opensees. To summarize the critical properties of my model:

***********************************************************************************************************************************************************
MATERIALS
***********************************************************************************************************************************************************
Steel: Steel02, nothing crazy. Concrete: Concrete01
<E.g., uniaxialMaterial Steel02 100 462000 2E+08 0.01 18.5 0.925 0.15;>


***********************************************************************************************************************************************************
ELEMENTS
***********************************************************************************************************************************************************
Beams/Columns:
------------------
'beamWithHinges' element with standard modified Gauss-Radau plastic hinge integration method. E.g.,
<< Nothing special is going on here. I have the elastic properties of my frame section assigned at the interior. At the hinge IPs, 'Aggregator' sections with linear torsional and shear force-deformation relationships are assigned>>

Shear Walls:
------------------
1) 'forceBeamColumn' elements with five integration points. The elements inherit 'Aggregator' sections similar to the beam/column members. Linear uniaxial materials are used to mimic tors. and shear F-D relationships.
2) Here comes the tricky part. Each planar wall segment is modeled by a 'forceBeamColumn' plus twoNodeLink elements stretching across the two ends of the cross section (like a fishbone) to be able to simulate the actual geometry of the shear walls. This way I connect my beams by respecting the acutal geometry.
3) The twoNode link elements are assigned with linear elastic force-deformation relationships that have large stiffness values (3e+8 in all DOFs. Units: kN-m, so it is a large stiffness value in all DOFs)
<E.g., element twoNodeLink 675 439 402 -mat 2 2 2 2 2 2 -dir 1 2 3 4 5 6 -orient 0 0 1 -mass 0; >

Ficticious Columns:
------------------------
These are simple twoNodeLink elements that level the entire building height with the coordinates (plan coord, i.e., x,y) of the centre of mass at each storey. These are essentially truss-like elements that are quite stiff in only the axial DOF. Seismic mass are assigned (two translational and one rotational around the vertical axis) at the top node of these elements. The reason I am using twoNodeLink element instead of a truss element is that, as far as I am aware, they don't participate in formation of the damping matrix so they won't cause any spurious damping forces.
<E.g., element twoNodeLink 2222 1323 1349 -mat 2 3 3 3 3 3 -dir 1 2 3 4 5 6 -orient 1 0 0 -mass 0;>
<where 2: stiff elastic uniaxial mat. & 3: flexible elastic uniaxial mat.>

***********************************************************************************************************************************************************
MASS ASSIGNMENTS - CONNECTIVITY - DIAPHRAGMS
***********************************************************************************************************************************************************
1) Nodes at each story are slaved to a master node, i.e., the top node of the fictitious column of the respective story. Constraints are created by the 'rigidDiaphragm' command.
2) The mass is lumped at each story to this slave node as two terms in translational DOFs and one in rotational.
<E.g. rigidDiaphragm 3 1 2 3 4 . . . 403 407;

***********************************************************************************************************************************************************
DAMPING
***********************************************************************************************************************************************************
Applying modal damping for all (15) modes calculated.
modalDamping 0.015;

***********************************************************************************************************************************************************
ANALYSIS SETUP
***********************************************************************************************************************************************************
The following are the start-up settings for dynamic analysis. Obviously I do switch between different algorithms and test schemes and play around with the number of iterations and the tolerance. It is also worth noting that I am using the analysis time step as the time step of the ground motion. Along the way, upon non-cnvergence, I do try to run using a smaller time step.

numberer Plain;
system UmfPack;
constraints Transformation;
test NormDispIncr 1e-3 100 2
algorithm KrylovNewton
integrator Newmark 0.5 0.25;
analysis Transient;

***********************************************************************************************************************************************************
O B S E R V A T I O N S
***********************************************************************************************************************************************************
1) The elements are loaded successfully in the end of gravity analysis.
2) Eigenvalue analysis runs fine and the plotted modes/mode shapes look reasonable.
3) There is a tedious problem caused by the use of rigid diaphragm together with RC fiber beams. Elements develop significant axial forces due to the constraints. This is caused by the shift of neutral axis in bending leading to excess axial strains at the section level, and consequently, at the element level. Since the element is not allowed to shorten/elongate, it causes axial forces to develop. This behaviour was resulting in significant (~0.5*fc*A) axial forces to develop in my coupling beams ultimately transfer very large moments/shears to the walls rendering the analyses very difficult to converge after a point. As a temporary solution, I reduced the area assignment of the beamWithHinges interior IPs to a thousandth of its original value. This doesn't lead to any significant change in behaviour other than allowing for the strains to develop at the element interior, getting rid of the excess axial forces.
4) Upon performing dynamic analysis on the model, I can barely make it get through the first 3 seconds at best. This point corresponds to an acceleration value of 0.01g. The structure does a full swing to one side, starts coming back and right before it reaches back to zero top displacement it hits a non-convergence point. I did try different ground motions, and I did scale them down quite a lot to see if it will be any good. But none of it worked.
5) Both pushover and dynamic analyses do run fine for the same model with only the core-walls+coupling beams modelled and the frames excluded. I am getting reasonable results in the end of these analyses. The core wall configuration is a pretty standard one. It is more or less symmetric in Y-Y plan dimension. The Left hand side mirror counterpart looks like the letter 'E'.

Any kind or form of comment or feedback is appreciated! I hope that I am committing a terribly obvious mistake that can be picked up from these lines. Otherwise I'd be happy to exchange e-mails with anyone willing to help. My e-mail address: omer.odabasi@iusspavia.it

Many thanks,
Ömer

I am aware that this topic was posted a couple times in the past but I could not find any clarification/solution to this issue.

I am trying to use CUSP solver after following the steps explained in OpenSees wiki page. However the only different action that I took was that I installed CUDA v9 instead of 5.5. This is simply because when I tried to install CUDA toolkit v5.5 it just says that my system is not compatible even though I have CUDA supported hardware (NVIDIA GTX 1060). I'm guessing it's because my OS is Windows 10 and CUDA v5.5 is meant for W8.1 as the latest supported Windows OS. I don't run into this problem with v9. In the end, I'm getting the following error: "Function load error: Version of .dll may be incorrect".

Long story short, my question is whether we absolutely need CUDA Toolkit v5.5 (which apparently only works for W10?) or is there a way to make it work using more recent CUDA versions?

Many thanks in advance!
Ömer

Dear David,

Could you get it to work by any chance? I am getting the same error?

Cheers.
Ömer

dgale wrote:
> Hello shashabshamlou,
> I haven't known how to install libreries for CuSP, could you send me all
> files you downloaded, please? david.gale@upm.es
> And if it is possible a little explanation about how to install it in this
> post in order to everyone knows
> Thanks in advance