Intermediate results while STAT_NON_LINE is running

Caster

Bonjour

I tried to search the forums but didn't find definitive answer. Answer to my question is almost certainly "no", but I try anyway.

Is it possible to save intermediate results while STAT_NON_LINE is running? Like equilibrium path. Does the new MECA_NON_LINE change anything here? Especially with larger models where solving takes a long time and solver parameters are difficult to choose properly in the first try, mainly when using LONG_ARC. Many times it happens with LONG_ARC that it starts to follow a wrong path, turns backwards or gets stuck, but you cannot really deduct anything by just looking at the residuals while your solver is running. I know that the fields are calculated with CALC_CHAMP, but maybe there is an unorthodox way to save these intermediate results.

Merci d'avance !

AsterO'dactyle

1/ No. Except using OBSERVATION
2/ YEs, in 2027 or 2028. A major project is underway to overhaul the entire memory management system. Among the expected results will be the ability to output data (including fields during calculations) as well as more robust handling of large volumes of data.

In addition, a doctoral student is currently working on trajectory control for continuation methods (using machine learning methods);

Caster

Nice, looking forward to future advancements of Code_Aster 🙂 Having continuation methods implemented like LONG_ARC is one of the key reasons why I use Code_Aster, so seeing advancements in this field is great news.

jacob

For equilibrium path you can plot displacements of a point or other fields when running using SUIVI_DDL in STAT_NON_LINE.
SUIVI_DDL=(_F(EVAL_CHAM='MAX',

                                 EVAL_CMP='FORMULE',

                                 FORMULE=formula,

                                 GROUP_MA=('BOLT', ),

                                 NOM_CHAM='DEPL',

                                 NOM_CMP=('DX', 'DY'),

                                 TITRE=('BOLT', )),

                             _F(EVAL_CHAM='MAX',

                                 EVAL_CMP='FORMULE',

                                 FORMULE=formula,

                                 GROUP_MA=('NUT', ),

                                 NOM_CHAM='DEPL',

                                 NOM_CMP=('DX', 'DY'),

                                 TITRE=('NUT', )),    

                              _F(EVAL_CHAM='MAX',

                                 EVAL_CMP='VALE',

                                 GROUP_MA=('F_1', ),

                                 NOM_CHAM='DEPL',

                                 NOM_CMP=('DY', ),

                                 TITRE=('F_1_DY', ))))

For whole intermediate results you can print the result after a chosen solution step and then restart the analysis. This can be automated similarly as I have done in my topology optimization "software". However, not very practical for complicated studies.

Caster

jacob

Nice! I will try to have time to test this in practice later this year. I will get back to it to report how it went. Just seeing a displacement of a certain node during nonlinear run is very useful.

SantiagoMR

Hello Caster! I’m not sure if this fits your specific use case, but I’m sharing a small code snippet I use to monitor my simulations.

The key is running a loop inside STAT_NON_LINE / DYNA_NON_LINE by iterating over an array of target times and steps. Once the resolution of an interval is complete according to the requested ARCHIVAGE, I extract the displacements (or any required variable; for example, running a prior CALC_CHAMP if you need to extract SIEQ_NOEU).

Another key is "Pythonizing" the output of POST_RELEVE_T by converting it directly into a Pandas DataFrame using .EXTR_TABLE(), making it easy to export clean CSV files via df.to_csv(). In this implementation, the file gets updated/overwritten while maintaining previous time steps, though you can also parameterize the output path or filename if you prefer to save each step separately.

Hope this helps. Cheers!


# Save last result from previous analysis stage
RESNL_LAST = NONLIN  
RESNL = [None] * RESNL_STEPS

# General output directory 
OUTPUT_DIR = os.path.join(os.path.expanduser('~'), 'project_output', 'extract_depl')
os.makedirs(OUTPUT_DIR, exist_ok=True)

for j in range(RESNL_STEPS):
    
    # Define initial state and time increment steps
    if j == 0:
        initial_state = [_F(EVOL_NOLI=RESNL_LAST)]
        increment_step = [_F(
            LIST_INST=l_times_3,
            NUME_INST_FIN=NUMBER_OF_SAVES[j],
            INST_INIT=paso_anterior
        )]
    else:
        initial_state = [_F(EVOL_NOLI=RESNL[j-1])]
        increment_step = [_F(
            LIST_INST=l_times_3,
            NUME_INST_FIN=NUMBER_OF_SAVES[j],
            NUME_INST_INIT=NUMBER_OF_SAVES[j-1]
        )]

    # Non-linear dynamic analysis computation
    RESNL[j] = DYNA_NON_LINE(
        ARCHIVAGE=_F(LIST_INST=steps_3),
        CHAM_MATER=FIELDMAT,
        reuse=NONLIN,
        ETAT_INIT=initial_state,
        COMPORTEMENT=COMPORTAMIENTO,
        EXCIT=excit_nonlin2,
        INCREMENT=increment_step,
        AMOR_RAYL_RIGI='ELASTIQUE',
        SCHEMA_TEMPS=_F(
            SCHEMA='HHT',
            FORMULATION='DEPLACEMENT',
            MODI_EQUI='OUI',
            ALPHA=-0.1
        ),
        CONVERGENCE=_F(
            ITER_GLOB_MAXI=30,      # Increase iterations for generalized Newton solver
            RESI_GLOB_RELA=1.E-6    # Standard convergence threshold
        ),
        METHODE='NEWTON',
        MODELE=MECA_MOD,
        SOLVEUR=SOLV_STABLE,
        **contact_keywords
    )

    # Partial results extraction
    SCL_TABLE = POST_RELEVE_T(
        ACTION=(
            _F(
                INTITULE='N_EXTR_PLAC', 
                OPERATION='EXTRACTION', 
                TOUT_ORDRE='OUI', 
                RESULTAT=RESNL[j],
                NOM_CHAM='DEPL', 
                TOUT_CMP="OUI", 
                GROUP_NO='N_EXTR_PLAC'
            ),
            _F(
                INTITULE='N_EXTR_TOP', 
                OPERATION='EXTRACTION', 
                TOUT_ORDRE='OUI', 
                RESULTAT=RESNL[j],
                NOM_CHAM='DEPL', 
                TOUT_CMP="OUI", 
                GROUP_NO='N_EXTR_TOP'
            ),
        )
    )

    # Convert extracted table to DataFrame and export to CSV
    table_data = SCL_TABLE.EXTR_TABLE()
    df_result = pd.DataFrame(table_data.rows)

    output_file = os.path.join(OUTPUT_DIR, f'{test_id}-DISPL_LOAD.csv')
    df_result.to_csv(output_file, index=False)