Same script, different results

[edumalta]

Good evening fellow colleagues,

I am using a python script to simulate a buckling problem using ABAQUS static/Riks. The problem is that I am getting different results for every time i run the SAME script. Sometimes I get the result I expect and sometimes I get rubbish. I have already looked into the Riks / Arc-Length formulation and I can`t find a random factor that could be causing this variability. I had the same difficulty using ANSYS APDL and I thought it was a bug, then I turned to ABAQUS. Do you have a clue on what is going on?

Thank you

 

[mechfeeney]

I'd check for mesh convergence and make sure that the number of increments, initial, minimum, and maximum arc length increment values are appropriate. Each iteration of the buckling calculation steps the displacement by the arc length increment value. The arc length increment value may be selected "randomly" for each independent iteration, let alone, an independent simulation all together. My feeling is that this is where the "rubbish" is coming from. If I remember correctly, you need to drop the minimum arc length increment value down from it's default, and increase the number of allowable increments. Also tighten up your mesh just in case.

Let me know if this helped. Good Luck.

 

[edumalta]

Firstly, thank you mechfeeney. I made a few tests with different values of increment values. My results are still different from each other, but more close in terms of precision. I also managed to avoid the "rubbish" with that. You said that the increment values are selected "randomly", but in its essence, the Riks method doesn't have a random factor. I don't know how it is implemented in ABAQUS or ANSYS, but I guess the problem is there. Can you appoint me a reference where I can garantee that this random factor exists?

 

[mechfeeney]

Try one more thing before we go on any further. In your RIKS set up change the NLGEOM option from off to on (if it isn't already). You may have really large displacements. Let's see what this does first...then continue the conversation.

 

[edumalta]

Yes, it was ON, all the time.

 

[mechfeeney]

In the Riks setting under the other tab there is an option for 'Extrapolation of previous state at start of each increment.' Perhaps the default 'Linear' option isn't high enough order. Is it possible that you are getting extrapolation errors...yielding the inconsistencies?

 

[mechfeeney]

Here's a line taken directly out of the user guide from Abaqus under the Riks section. This is referring to editing the Riks setup under the 'Basic' tab (where NLgeom is located)

Choose a Type option:

Choose Automatic to allow Abaqus/Standard to choose the size of the arc length increments based on computational efficiency.

Choose Fixed to specify direct user control of the incrementation. Abaqus/Standard uses an arc length increment that you specify as the constant increment size throughout the step. This method is not recommended for a Riks analysis since it prevents Abaqus/Standard from reducing the arc length when a severe nonlinearity is encountered.

Abaqus will "choose" the increment value...I bet there is some for of randomness in this process.

 

[mechfeeney]

Try selecting Fixed and see if you still get inconsistent solutions.

 

[edumalta]

I tried "Parabolic", but didn't change much. I'm calling the day. Tomorrow I continue. I appreciate all the help you gave me till now.
Good evening to you