the hydrocarbonstate is stored in the reservoir state and
used instead of the primalvariable_
The initial hydrocarbonstate is computed using
initHydroCarbonState(...);
the results look okay, while the running for flow_solvent needs further
investigation even the results with flow_solvent actually look okay.
With two different version of
computePropertiesForWellConnectionPressures, flow_solvent produces the
same results. This is something needs further investigation.
The current implementation requires a copy of
computeWellConnectionPressure in StandardWells and StandardWellsSolvent.
That means probably we need to introduce the asImpl() for the Wells.
- the computation of well potentials in the model class calculates the
well potentials using computeWellFlux()
- in this way the well potential calculations also handle well where
some perforations are closed by the simulator due to cross-flow.
- the well potentials pr perforation and phase is stored in the well
state.
Computation of properties used in computeConnectionPressureDelta
is factored out to computePropertiesForWellConnectionPressures
The motivation is to be able to use a modified version of
computePropertiesForWellConnectionPressures in the solvent model
The following changes are done in order to remove the duplicated code in
assemble().
- extractWellPerfProperties takes SolutionState as input (only used in
the solvent model)
- the computation of effective parameters is moved to computeAccum()
With these changes the solvent model can use assemble() from the base
model.
Density and viscosity are given as input instead of calculated inside
computeMassFlux. This allow for modifying the properties prior to
calling computeMassFlux which avoids code duplication in the solvent
implementation.
The convergenceReduction() method no longer takes a number-of-wells argument,
instead it infers the number of well fluxes to check from the size of the
well_flux_eq member of the residual.
After this, a custom getConvergence() method is no longer required for the
multi-segment well model.
This means that details such as calling assemble(), solveJacobianSystem(),
updateState() etc. are now left to the model class. This will make it easier
to create new model classes with different behaviour (such as sequential models).
When running parallel one cannot use Eigen::Array::mean() for this
as the it is just a local part of the complete array. With this commit
we correctly compute the number of global cells in the grid and use this
together with a parallel reduction to compute a global mean value.
