beahviour of NCNV command in PyMAPDL

[vymlatilp]

Hello,
I would like to calculate nonlinear analysis using PyMAPDL. If the analysis does not converge, it should behave depending on KSTOP parameter of NCNV command as follows:

0	—	Do not terminate the analysis if the solution fails to converge.
1	—	Terminate the analysis and the program execution if the solution fails to converge (default).
2	—	Terminate the analysis, but not the program execution, if the solution fails to converge.

If I use KSTOP=1, and solution does not converge, I get message "Run continued at user request.", see output and next substep proceeds (without bisection!). So the behavior is like KSTOP=0.

     HT FLOW CONVERGENCE VALUE=  0.4393E-05  CRITERION=  0.4778E-04 <<< CONVERGED
     TEMP CONVERGENCE VALUE   =   2.585      CRITERION=  0.9824    
    EQUIL ITER  49 COMPLETED.  NEW TRIANG MATRIX.  MAX DOF INC= -0.3814    
     HT FLOW CONVERGENCE VALUE=  0.4373E-05  CRITERION=  0.4778E-04 <<< CONVERGED
     TEMP CONVERGENCE VALUE   =   2.589      CRITERION=  0.9824    
    EQUIL ITER  50 COMPLETED.  NEW TRIANG MATRIX.  MAX DOF INC=  0.3817 

 *** WARNING ***                         CP =   14536.250   TIME= 16:05:28
 Solution not converged at time 831600 (load step 21 substep 3).         
  Run continued at user request.                                         
 *** LOAD STEP    21   SUBSTEP     3  COMPLETED.    CUM ITER =    583
 *** TIME =   831600.         TIME INC =   3600.00    
 *** AUTO STEP TIME:  NEXT TIME INC =  3600.0      UNCHANGED

     HT FLOW CONVERGENCE VALUE=  0.1954E-01  CRITERION=  0.1088E-03
     TEMP CONVERGENCE VALUE   =   3.661      CRITERION=  0.9824    
    EQUIL ITER   1 COMPLETED.  NEW TRIANG MATRIX.  MAX DOF INC= -0.3898    
     HT FLOW CONVERGENCE VALUE=  0.5629E-05  CRITERION=  0.4776E-04 <<< CONVERGED
     TEMP CONVERGENCE VALUE   =   2.648      CRITERION=  0.9824    
    EQUIL ITER   2 COMPLETED.  NEW TRIANG MATRIX.  MAX DOF INC=  0.3953    
     HT FLOW CONVERGENCE VALUE=  0.4530E-05  CRITERION=  0.4775E-04 <<< CONVERGED

Can it be, that mapdl.ncnv(1) command issued before mapdl.solve() doesn't influence the program behavior upon non-convergence?
According my experience from MAPDL, the NCNV,2 skips all commands until next finish is found. We use it very often to postprocess simulations (f.g. simulations to determine load carrying capacity), where the non-convergence is expected. Can this work in pymapdl?

In my opinion the options KSTOP 0 or 1 would be sufficient and then the python code can decide how to proceed based on mapdl.solution.converged.

[mikerife]

Hi @vymlatilp Which MAPDL version are you using? Can you show/send the solution commands used? I'd like to test this out with the specific solution setup if possible.
Mike

[vymlatilp]

Hi @mikerife,

regarding versions:

Python 3.9.9 (tags/v3.9.9:ccb0e6a, Nov 15 2021, 18:08:50) [MSC v.1929 64 bit (AMD64)]
Type 'copyright', 'credits' or 'license' for more information
IPython 8.4.0 -- An enhanced Interactive Python. Type '?' for help.
Product: Ansys Mechanical Enterprise
MAPDL Version: 21.2
ansys.mapdl Version: 0.62.1

I cannot share project from my previous post. I created simple example which can be used to testing, see code below. This example end with non-convergence.

from ansys.mapdl.core import launch_mapdl
import os

run_location = os.getcwd()


mapdl = launch_mapdl(run_location=run_location)
print(mapdl)
mapdl.finish()
mapdl.clear()
prep = """
/prep7
ET,1,SOLID185 
MPTEMP,,,,,,,,  
MPTEMP,1,0  
MPDATA,EX,1,,200000 
MPDATA,PRXY,1,,0.3  
TB,BISO,1,1,2,  
TBTEMP,0
TBDATA,,300,0,,,,   
BLC4,0,0,100,1000,100
LSEL,S,LOC,Y,0
LSEL,A,LOC,Y,1000
LESIZE,ALL, , ,2, , , , ,1
LSEL,S,LOC,Y,1,999
LESIZE,ALL, , ,20, , , , ,1
VSEL,ALL
VSWEEP,ALL
NSEL,S,LOC,Y,0
D,ALL,UY
NSEL,S,LOC,Y,0
NSEL,R,LOC,X,0
D,ALL,UX
NSEL,S,LOC,Y,0
NSEL,R,LOC,Z,0
D,ALL,UZ
ASEL,S,LOC,Y,1000
SFA,ALL,1,PRES,-500
"""

output = mapdl.run_multiline(prep)
print(output)

mapdl.slashsolu()
mapdl.allsel(labt="all")
mapdl.antype(antype="static")
mapdl.nlgeom(key="off")
mapdl.time(time="1")
mapdl.ncnv(kstop="2")
mapdl.neqit(neqit="20")
mapdl.nsubst(nsbstp='20', nsbmx='20', nsbmn='20')
mapdl.outres(item="ALL",freq="1")
mapdl.solve()

mapdl.post1()
mapdl.set(lstep="last")
result = mapdl.result
result.plot_nodal_displacement(6,"Y")
result.plot_principal_nodal_stress(5,"SEQV")
mapdl.exit()

I hope it will help you to check behaviour of ncvn command.

Petr

[mikerife]

Hi @vymlatilp Petr - could you please share only the solution commands from your project. This could be a logic issue i.e. traversing through functions A -> B -> C lead to the behavior but A -> H -> C does not. So I would like to test the specific solution settings you had just in case. As well as a general case.
Mike

[mikerife]

Hi @vymlatilp
I moved the "allsel" to the prep comment block then ran the solve and finish commands non-interactively like so:

(Sorry, this GitHub text entry gets the better of me all the time...not sure why I can't figure out how to indent regular text)

The non_interactive writes a input file of the commands that follow and issues a /input to MAPDL to read the file. This way if/when the solution does error out there is a finish command waiting for it (the ncnv command). This should then result in a run-time error, but the MAPDL instance still be running and can now process more commands:

Mike

[mikerife]

Hi @germa89 It does look to be related to /NERR. /NERR,,,-1 is issued according to the apdl log (thanks for that!). But that's a bit confusing as the MAPDL docs say that field is unused. But if I change it to /NERR,,,0 then the NCNV,1 behavior is as Petr expects. I'll if I can track down what that field does...unless you happen to know?
Mike

[vymlatilp]

Hi @germa89 , @mikerife ,
thanks for the hints. The /NERR,,,0 helps to stop executions of solve command. However, due to noncovergence is the process abborted. If I use try block, and ncvn,2 I still can continue with postprocessing commands after solve command. With ncnv,1 the is mapdl process terminated, but that is in line with ncnv commad. So my final code is (without preprocessing block):

mapdl.slashsolu()
mapdl.allsel(labt="all")
mapdl.antype(antype="static")
mapdl.nlgeom(key="off")
mapdl.time(time="1")
mapdl.ncnv(kstop="2")
mapdl.neqit(neqit="20")
mapdl.nsubst(nsbstp='2', nsbmx='20', nsbmn='2')
mapdl.outres(item="ALL",freq="1")

try:
    with mapdl.non_interactive:
        mapdl.run("/NERR,,,0")
        mapdl.run("solve")
        mapdl.run("finish")
except Exception as e:
    print(e)


mapdl.post1()
mapdl.set(lstep="last")
result = mapdl.result
result.plot_nodal_displacement(2,"Y")
result.plot_principal_nodal_stress(2,"SEQV")
mapdl.exit()

One additional note: not all errors are reason to terminate. Some errors are be solved by bisection, at least the exception message is not telling the true reason why the execution was stopped. The first error message was taken. Maybe it is not necessary to raise exeption at all, because in nonlinear analysis is the noncovergency (and related errors) expected.

Best regards,
Petr

[germa89]

Hi all

I'm very happy we could track this down to the NERR command. Thank you @mikerife for this work. Very insightful!

@vymlatilp I understand that raising an exception is... annoying.

PyMAPDL raises an exception every time that it finds an error (*** ERROR ***). This might be "annoying", and I understand that. But I think it is not a bad approach regarding keeping the user aware. We should probably expand the error classes, to be more expecific about the type of errors, and not just throwing a general exception.

[germa89]

Pinging @akaszynski for further opinion about this.

I'm tempting to keep this as it is because of:

https://github.com/pyansys/pymapdl/blob/06f90911a08431957436a408fe7e3e7638469ef5/src/ansys/mapdl/core/mapdl_grpc.py#L528-L533

[vymlatilp]

Hi @germa89, @mikerife,

the proposed solution woks fine and try-exept block can be used to handle different situations, at least in our project is fully sufficient.
Thanks for your hints!
Petr

[germa89]

For convenience, I'm quoting here the answer given by @mikerife

Hi @germa89 It does look to be related to /NERR. /NERR,,,-1 is issued according to the apdl log (thanks for that!). But that's a bit confusing as the MAPDL docs say that field is unused. But if I change it to /NERR,,,0 then the NCNV,1 behavior is as Petr expects. I'll if I can track down what that field does...unless you happen to know?

Link: #1230 (reply in thread)