Pile-soil Interaction Convergence

[[email protected]]

Dear STKO Team,

I have been trying to get a simple pushover analysis going in clay for a wind turbine monopile using the penalty constrained BeamContact3D element. The pile settlement seems to be alright, however when I apply lateral load the pile either blows up or it moves randomly (not according to the loading).

Can you give me some pointers about what causes the convergence issue and how it may be solved.

I am attaching my stko file.

Thank you in advance for your help! Regards,
Bence

[STKO Team]

The pile settlement seems to be alright, however when I apply lateral load the pile either blows up or it moves randomly (not according to the loading).

It's mainly due to the fact that in some of your analyses, you are using a Linear algorithm, while you have nonlinearities (contact).

Here is a working copy of your example.

Pushover1.zip

(269.6 KiB) Downloaded 232 times

[[email protected]]

Dear STKO Team,

Thank you kindly for the suggestions and the example file. I have a few questions and follow up issues.

1) I see you suggested me to use Krylov-Newton method, while on another occasion I have been suggested to always use Newton with line search for contact problems. Is there a way to predict which would be better algorithm or it's just a conclusion based on trial and error?

2) Can you explain why the increase of pile diameter is necessary in the BeamEndContact3dp element? I understand that due to the meshing of the circular diameter the nominal size might not be enough to create a contact on the side for the BeamContact3dp element, however I do not understand why this would be the case for the BeamEndContact3dp element where the end of the pile is flush with the soil.

3) Just to confirm, the use of plain constrain handles is okay for the analysis even though we have contact constrains because the contact is a zero length element/material which is connected using a homogeneous and identity constraint (in this case EDOF)?

4) If I try to scale the mesh (densify) or change it in any way my model will not run. It stops without warning after the initial stage analysis is turned off with the prompt: "press any key to continue". This has happened to me on several occasions before with other models as well. Sometimes redrawing the geometry from scratch resolved the issue but not in this case. I have attached a refined mesh model for reference.

5) Based on the model you attached I updated mine and it converged. The only thing I had to modify is to use standard brick element because when I remove the elements from the pile's volume the SSP element will retain considerable residual displacements however, stdBrick does not (just an FYI as it might be useful tip for other forum users).
However, from the results I see that the pile does not exhibit gapping even though 0 cohesion and 0 tension strength was specified for the contact material and the pile displaces sufficiently to exhibit gapping. (attached picture you can see the displacements in the loading direction. What may cause this error?


Thank you very much for your time in advance!

Bence

[marafini.f]

Dear bence,
I am sorry we haven't get back to you yet, we have been a bit busy this week and your questions require a bit of time to be answered properly.
Dr. Petracca will answer you as soon as possible.
Thank you for your participation in our forum.
Francesca

[[email protected]]

Dear STKO Team,

No worries. However, in the meanwhile I have another issue regarding openseesMP. I have converted my sequential model to work with openseesMP according to one of your early webinars. It works well for gravity analysis, however when I try to remove the solid elements (within the pile volume) the solver crashes with the attached attached error message. I assume the message passing between sub-domains is not set up for analyses such as element removal. I also can not control the partitioning in STKO (to put all to be removed elements and their neighbors in one partition to avoid this issue). How can this issue be bridged to enable parallel computing of my model?

I also append the element removal tcl code that you provided to me in a prior forum discussion in case it's relevant to you.

Thank you very much in advance!

Bence

Element removal code:

# Remove elements (and nodes) in $region_remove.
# However, some nodes in the elements in $region_remove
# are shared with the outer soil domain, and they should be kept.
# So we remove nodes in elements in $region_remove that are not in $region_keep

# a dictionary to keep track of nodes already removed
set removed_nodes [dict create]
set all_nodes [dict create]

foreach ele_tag $region_remove {
# get nodes from this element
set ele_nodes [eleNodes $ele_tag]
# remove element
remove element $ele_tag
# remove nodes
foreach node_tag $ele_nodes {
# add it to all_nodes, not necessary, just for check
dict set all_nodes $node_tag ""
# only if not already removed ...
if {![dict exists $removed_nodes $node_tag]} {
# only if not on the boundary (not in region_keep)
if {[lsearch -exact $region_keep $node_tag] < 0} {
# set as removed in the dictionary
dict set removed_nodes $node_tag ""
# remove any SP constraint on this node
for {set dof 1} {$dof <= 4} {incr dof} {
remove sp $node_tag $dof
}
# remove node from OpenSees
remove node $node_tag
}
}
}
}

# convert dictionaries to lists
set removed_nodes [dict keys $removed_nodes]
set all_nodes [dict keys $all_nodes]
set num_node_processed [llength $all_nodes]
set num_node_removed [llength $removed_nodes]
set num_ele_removed [llength $region_remove]
puts "Inner hole removal:"
puts " Removed elements: $num_ele_removed"
puts " Removed nodes: $num_node_removed of $num_node_processed"

# domain has changed
domainChange

[[email protected]]

Dear STKO team,

Update on the parallel opensees issue. I have confirmed that the removing of elements causes the issue for openseesMP. But sadly, I do not have the expertise to modify my element removal tcl code to ensure correct MPICH message passing of the removed element and node IDs across multiple processors. May I ask your help on how to modify it?

Moreover, I have tried to run the same model on openseesSP without removing the elements, however, I receive an MPI error when I am changing the model subset (before 2nd stage i.e. adding the pile). The mpco outputs are the same for all cores and it only contains one partition (subdomain) of the model after visualization in STKO post-processor. Is there a conflict between openseesSP and construction stage analysis type models or am I making some mistake here?
I am providing the opensees and stko prompts for the openseesSP run (without removing elements).

Thank you for your kind help in advance! Regards,
Bence

OpenSees Output:

MPI_Channel::recvID() - incorrect number of entries for ID received: 1 exptected: 4
ShadowSubdomain::revertToStart ERROR ERROR
MPI_Channel::recvID() - incorrect number of entries for ID received: 1 exptected: 4
ShadowSubdomain::revertToStart ERROR ERROR


STKO output when I try to load the model:

Command: OpenDatabase
Database "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-0.mpco" correctly open
Database "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-1.mpco" correctly open
Partition "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-1.mpco" has inconsistent number of Model Stages (it will be skipped).
Current = 3, next = 1.
Database "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-1.mpco" correctly closed
Database "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-2.mpco" correctly open
Partition "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-2.mpco" has inconsistent number of Model Stages (it will be skipped).
Current = 3, next = 1.
Database "D:/Simulation/OpenSees/STKO/Second Run/Platsic_Update_test1/Results.part-2.mpco" correctly closed
End Command: OpenDatabase

[STKO Team]

It works well for gravity analysis, however, when I try to remove the solid elements (within the pile volume) the solver crashes with the attached error message. I assume the message passing between sub-domains is not set up for analyses such as element removal. I also can not control the partitioning in STKO (to put all to be removed elements and their neighbors in one partition to avoid this issue). How can this issue be bridged to enable parallel computing of my model?

Yes, the TCL custom script that I gave you works for sequential analyses, and it should be changed for parallel analyses. However, the required changes are quite difficult to be explained here (it requires a lot of additional code to make the processes communicate information about removal).
But don't worry, we will implement a built-in tool in STKO version 2.0.3 (will be released soon before the next webinar, 20th of May). It will be the counter-part of the addModelSubset (a removeModelSubset) and it will automatically take care of the parallel case.

In the meantime, if you cannot wait until the 20th of May, you can "somehow" control the way STKO does the partitioning. For example, if you want to make sure the entire solid part that should be removed (call it SOLID_A), be on 1 process, you can add a dummy interaction (do not assign any condition to it) that connects 1 vertex of SOLID_A to the whole SOLID_A, using the interaction type "General Link".
It should be enough to ensure that the whole SOLID_A be on a single process. Then, when you do your partition, you can visualize the partition SOLID_A belongs to, and you can add an if-statement to make sure the removal happens only on that process.

[[email protected]]

Thanks a lot for the advice! I will try the trick you mentioned. I can't wait for the next webinar! I hope you guys can touch on lateral dynamic pile loading as well.

[Clarabella]

Great! We are looking forward to seeing you at the webinar!

[[email protected]]

Dear STKO Team,

I've been playing around with the "wall-supported excavation" case that was discussed during last week's seminar. When using non-linear soil models there is an issue at the pile toe during the reset of displacements (i.e. InitialStateAnalysis off command). The displacements directly under the pile toe persist after the OFF command is issued. I was thinking this may be an related to the ENDBeamContact element. I have tried playing with the element (e.g. parameters, and initial contact on/off) but without any success. Hence, I am trying to set up a staged addition of the pile and its contact elements in a way that the bottom few meters of the pile is only added after the 'Initial State Analysis off' command was issued, while the top portion is added before the removal of soil elements to restrict lateral displacements in the vertical soil wall during excavation.

However, when I do such staged addition of the pile my system becomes singular. I am writing to ask some tips on what may cause this issue. I have attached my model file for reference.

Thanks in advance!
Bence