How to Reduce Computation Time for Earthquake Simulation?

[Zouatine]

Dear STKO Team,

I hope you are doing well.

I wanted to kindly ask for your proposals to solve my issue. I have a quite complicated model, and for a time history analysis in a single direction, it takes 4 hours because the earthquake duration is quite long. (Note: I cannot shorten it or take only the portion where the earthquake is strong.) I am running this analysis, which takes 4 hours, on a good computer with 8 processors. Despite this, it still takes 4 hours, which is not practical for me because I am planning to run around 1000 analyses. However, in the model results, I am only interested in the diaphragm node displacement and acceleration on each story.

Could you please provide me with a solution to my problem?

Thank you.

[kesavapraba]

Hi
I am wondering why the time region with near-zero values in the attached figure are included in your analysis. These regions, even though seemingly insignificant, might still contain valuable information, as you may have some reaons for that.


I would suggest the following to be tried:
Unpartitioned Model: Since your model isn't very large and you're already using parallel processing with 8 processors, you could run each of the 1000 analyses on separate processors using an unpartitioned model.
Trade-off: While each individual analysis might take slightly longer than the 4 hours in your previous case, you'll be able to complete 8 analyses simultaneously with your current setup. Ideally, with 1000 processors, you could finish all analyses in roughly the same time as a single analysis previously.

You can use amazon cloud services (Dr. Scott's page on it given below) or designsafe for availing such a large number of processors (video link below).

Dr. Scott's page:
https://portwooddigital.com/2023/04/16/ ... the-cloud/

How to use STKO on designsafe:
https://www.youtube.com/watch?v=_8vQ1FH ... 4&index=37

This approach would significantly reduce the total analysis time compared to using a partitioned model with the full 200-second record.

[Zouatine]

Your response is much appreciated ,

I believe the most reliable solution for me is to use Design_Safe to run this analysis. However, when I tried to do so, I encountered an error, as shown in the attached image. It seems to be blocked.

Problem_Desing_Safe.png (127.65 KiB) Viewed 5377 times

I have a question regarding the model: I do not want to create the model again as demonstrated in the webinar. I simply want to upload the model and run it. Is this possible? Additionally, I am curious about the payment method because I did not enter the customer ID for STKO anywhere.

Regarding computation time, I looked at my generated file from a 4-hour analysis. The file size is around 6 GB, which is impractical if I want to run 1,000 calculations. I believe the large file size results from recording the displacement and acceleration of every node. For this analysis, I only need the displacement and acceleration of the diaphragm node. Would requesting only this data reduce the analysis time and size of the computed file?

Picture_File_Size.PNG (69.66 KiB) Viewed 5377 times

Besides the above problem, what do you think about reducing the mesh size to affect computation time?

In addition to questions about computation time, I wanted to ask about the output of the displacement. As far as I know, the multisupport excitation records the absolute responses. To get the relative displacement between the top floor and the base, I need to subtract them. I performed this operation and obtained the following plot, which I cannot understand from a physical point of view.

Absolute_Displacements.PNG (87.99 KiB) Viewed 5377 times

Relative_Displacement.PNG (52.83 KiB) Viewed 5377 times

Thank you , and i am waiting for your response,

[kesavapraba]

Hi,

I have a question regarding the model: I do not want to create the model again as demonstrated in the webinar. I simply want to upload the model and run it. Is this possible? Additionally, I am curious about the payment method because I did not enter the customer ID for STKO anywhere

You will have to contact Design-safe help center for STKO access and payment related queries. Please remember design-safe also has other computational frameworks, including OpenSees (sequential and parallel versions). You can generate Tcl files from your local system and upload and run on design-safe using OpenSeesMP (without STKO). You can explore this aspect with design-safe community.

Besides the above problem, what do you think about reducing the mesh size to affect computation time?

It will not be an issue if you make use of parallel versions of OpenSees. Only couple of hours will be more, which is Ok. Because, you will be running the model for as many number of earthquake records as the number of cores available for you on the cloud. Also, you can do mesh convergence study to decide on optimum mesh size.

In addition to questions about computation time, I wanted to ask about the output of the displacement. As far as I know, the multisupport excitation records the absolute responses. To get the relative displacement between the top floor and the base, I need to subtract them. I performed this operation and obtained the following plot, which I cannot understand from a physical point of view.

Relative displacement is one order less than the absolute displacement. The trend in relative displacement indicates that a residual dislacement (inelastic) kicks in as the relative displacement stays above the zero line. This happens at the very early stage of the analysis, and this does not seem to be Ok as the strong motion was not started yet in that stage. I would expect the model is still in elastic regime. Please check again the model input, including the earthquake ensemble.

Please also check how you input type of your ground motion record, as the documentation below says there should displacement time history as input or combination other two time histories (accn+displacement, accn+vel, etc)

[Zouatine]

Hello,

Thank you, Sir, for your kind reply.

I have checked the model again to see if I made any mistakes in the modeling. From the perspective of the OpenSees command, it seems fine to me. However, I am first applying the gravity analysis and then the earthquake motion. For this reason, I have some initial displacement at the start. I am not sure if this displacement is due to some nonlinearity (my explanation).

Furthermore, I went ahead with the model and included some springs at the base to represent the soil. The model runs for the linear stage but not for the nonlinear case (e.g., increasing GM intensity to 4g ). I am getting the following error:

Error opensees.PNG (199.82 KiB) Viewed 5325 times

Note that , the model was running just fine when i used a fixed base with intensity 4g of ground motion.

i did check the dispalcement at the top floor and the base of the building and at the one fixed ( added recently to model the soil) and i am having the following out puts ( i am not sure why there is this peak at the end of the running case ????)

Absolute_Displacements.PNG (87.99 KiB) Viewed 5325 times

I also did calculate the relative displacement between the top node and the base node ( and this peak is also there )

relative_displacemnt.PNG (26.94 KiB) Viewed 5325 times

I will highly appreciate if you can take a look at the model , i attached the model along with this post.


Thank you and waiting for your kind reply,

[STKO Team]

I highly suggest you to review your model.
There are a lot of inconsistency that will make the analysis fail.
You have areas with fine mesh and areas with very coarse meshes, meshes are incompatible, GM are applied only at some vertices and not at the base, etc..