Hi Vesna,
We said analytical time step is ok as long as it is less than 1/10-1/20 of natural period.
Most of the examples in OpenSee example manual is making DtAnalysis equate to ground motion time step. And myself used the time step of ground motion as the analytical time step also. It works pretty good.
However, when I reduced the analytical time step, result deviated. I didn't expect this except a quicker convergence. Do you have any idea on it?
Thank you.
Yuhao
Time steps and convergence
Moderators: silvia, selimgunay, Moderators
Time steps and convergence
Yuhao Feng
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Vesna,
I have significant deviations. I reduced the time step because model cant converge at a specific time during dynamic analysis.
I also checked the norm unbalance force and incremental displacement at the point which fail to converge(i am using EnergyIncr convergence test). Interestingly, residual force is large and the incremental displacement is very small. ie. deltaX=0.0003, deltaR=1.26e+008.
It doesn't make sense having a large residual force. I think there must be an error in the numerical process. Also, the same error may lead to significant deviation on responses with different time steps.
Do you have any idea about this?
Thank you.
Yuhao
[quote="vesna"]Yuhao,
As you reduce the time step of analysis compared to ground motion time step the responses will not be exactly the same but they should be almost the same. Is this what you are getting, or you have significant deviations?[/quote]
I have significant deviations. I reduced the time step because model cant converge at a specific time during dynamic analysis.
I also checked the norm unbalance force and incremental displacement at the point which fail to converge(i am using EnergyIncr convergence test). Interestingly, residual force is large and the incremental displacement is very small. ie. deltaX=0.0003, deltaR=1.26e+008.
It doesn't make sense having a large residual force. I think there must be an error in the numerical process. Also, the same error may lead to significant deviation on responses with different time steps.
Do you have any idea about this?
Thank you.
Yuhao
[quote="vesna"]Yuhao,
As you reduce the time step of analysis compared to ground motion time step the responses will not be exactly the same but they should be almost the same. Is this what you are getting, or you have significant deviations?[/quote]
Yuhao Feng
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Vesna,
Yes, my tolerance is 10^-8.
Yuhao
[quote="vesna"]Yuhao,
What is the tolerance for the test? For EnergyIncr convergence test (residual force)*(incr disp)<tol. This would mean that your tolerance is on the order of 10^4. Please check on this one.[/quote]
Yes, my tolerance is 10^-8.
Yuhao
[quote="vesna"]Yuhao,
What is the tolerance for the test? For EnergyIncr convergence test (residual force)*(incr disp)<tol. This would mean that your tolerance is on the order of 10^4. Please check on this one.[/quote]
Yuhao Feng
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Frank,
Yes. I am using force based Beam with hinge element. But acturally, I am reducing the time step and the result varies. Do you have any suggestion on selecting the time step?
Yuhao
[quote="fmk"]a problem with increasing time step is that the changes in the response get larger at each step (dU) .. certain elements (force Beam Column) and the solution algorithm (if discontinuities exist) can have problems depending on the state.[/quote]
Yes. I am using force based Beam with hinge element. But acturally, I am reducing the time step and the result varies. Do you have any suggestion on selecting the time step?
Yuhao
[quote="fmk"]a problem with increasing time step is that the changes in the response get larger at each step (dU) .. certain elements (force Beam Column) and the solution algorithm (if discontinuities exist) can have problems depending on the state.[/quote]
Yuhao Feng
Research Assistant & graduate student
CCEE Dept. North Carolina State University
Research Assistant & graduate student
CCEE Dept. North Carolina State University
in examples we typically use the time step from the record because for any points in between results in linear interpolation between 2 enclosing data points and the dT is usually small enough given the periods .. sometimes the dT has to be made smaller so as to capture certain periods in the structure.
unless you are using Penalty constraints and have mp constraints in the model you might want to see why the unbalance is so large .. if the model is small enough a 4 flag at the end will print the values from the X and B in he system .. and a print of nodes will show the id or each dof (the mapping into these vectors)
unless you are using Penalty constraints and have mp constraints in the model you might want to see why the unbalance is so large .. if the model is small enough a 4 flag at the end will print the values from the X and B in he system .. and a print of nodes will show the id or each dof (the mapping into these vectors)