5. Precautions for use#

Here is a set of indispensable tips for the proper use of recalibration.

  • Experimental curves are defined as tables with two columns: one for the abscissa and one for the ordinates.

  • The experimental curves must be functions: an x-axis must correspond only to one ordinate. If an experimental curve includes cycles (for example stress as a function of charge-discharge deformation), it is then necessary to split this parameterized curve into two curves, expressing on the one hand the abscissa, on the other hand the ordinates of the cyclic curve as a function of the parameter (for example deformation as a function of time and stress as a function of time).

_images/Shape8.gif

  • We have to recalculate \(N\) curves calculated on \(N\) experimental curves.

  • The first calculated curve will be reset to the first experimental curve, the second calculated curve will be adjusted to the second experimental curve, and so on in the order specified for the operands RESU_EXP and RESU_CALC.

  • The calculated quantities entered under the RESU_CALC operand must come from POST_RELEVE_T (except for dynamics where you can have tables from CREA_TABLE and MAC_MODES)

  • The reset parameters must be declared in bulk at the beginning of the slave command file. For example:

DEBUT ()

DSDE__ = 200.

YOUN__ = 8.E4

SIGY__ = 10.

...

  • The initial values of the reset parameters are those entered for operand LIST_PARA and not those present in the user’s slave file.

  • At each readjustment iteration, the calculations defined in the slave file must converge. In the context of recalibrating non-linear calculations, it is therefore strongly recommended to use automatic time step cutting.

In the context of recalibrating non-linear calculations with automatic time step cutting, it is**indispensable* to define an archiving list under operand LIST_ARCH.

Recalibration is a powerful way to obtain parameter values from tests. However, it is not a miraculous step: the**experiment* curves must contain enough information to identify the parameters. For example, it is impossible to identify elastoplastic parameters with a test remaining in the elastic domain. Experimental tests must therefore excite the parameters to be identified.

In the same logic, it is desirable for the**experiment* curves to contain sufficient numbers of points to properly describe the action of the parameters to be identified.

  • Finally, in the case of using several experimental curves, the fact that they have the same number of points balances the information they provide.