Mea column improvements #1077
Mea column improvements #1077
Conversation
andrewlee94
added
bug
idaes/core/base/property_base.py
Outdated
| ): | ||
| if iscale.get_scaling_factor(v) is None: # don't replace if set | ||
| name = v.getname().split("[")[0] | ||
| index = v.index() | ||
| sf = self.config.parameters.get_default_scaling(name, index) | ||
| if sf is not None: | ||
| iscale.set_scaling_factor(v, sf) | ||
|
|
||
| for con in self.component_data_objects(Constraint, descend_into=False): | ||
| # if con.name == 'stripper_section.stripper.liquid_phase.properties[0.0,0.8].appr_to_true_species[Liq,CO2]': |
There was a problem hiding this comment.
This comment needs to be removed.
idaes/core/base/property_base.py
Outdated
| index = con.index() | ||
| sf = self.config.parameters.get_default_scaling(name, index) | ||
| if sf is not None: | ||
| iscale.constraint_scaling_transform(con, sf, overwrite=False) |
There was a problem hiding this comment.
This is a fairly significant change to what this method does - we probably need to discuss the implications of this.
There was a problem hiding this comment.
Thinking a bit more (and repeating a comment from Slack for posterity) - I think this is the wrong way to go.
IPOPT should be considered a special case, not the general case, in which case we need to set the scaling suffix for the constraints and then let the user decide what to do with them. IPOPT users can use the constraint_scaling_transformation function if they wish, but the generic workflow would use the Pyomo scaling transformation (which relies on having the suffixes).
There was a problem hiding this comment.
Removing the default status of IPOPT and constraint_scaling_transform in the long run is fine, but it's also a complete revolution in how scaling is treated in IDAES models, which make heavy use of constraint_scaling_transform. If this change is reverted, I'll just as likely use constraint_scaling_transform somewhere else to do the desired scaling: probably in electrolyte_states.py.
There was a problem hiding this comment.
It is fine to still use constraint_scaling_transform, but it should not be done in the base classes (i.e. anything in idaes.core.base). Ideally, we would only do this in the model level calculate_scaling_factors method.
The point being this is not something we want baked in deep as it will be harder to change later - ideally we want to limit the use of this to the calculate_scaling_factors methods so that when we make the transition, we only need to change those (or if we are lucky, just switch to new methods and delete those as they are no longer needed).
| @@ -321,11 +321,11 @@ def build(self): | |||
| flow_basis = self.liquid_phase[t_init].get_material_flow_basis() | |||
| if flow_basis == MaterialFlowBasis.molar: | |||
| fb = "flow_mole" | |||
| elif flow_basis == MaterialFlowBasis.molar: | |||
| fb = "flow_mass" | |||
| # elif flow_basis == MaterialFlowBasis.mass: | |||
|
This model will eventually (soon) be accompanied by a surrogate model. This PR might or might not be merged in the meantime. |
Summary/Motivation:
Miscellaneous changes/fixes in order to improve scaling and convergence of column models.
Legal Acknowledgement
By contributing to this software project, I agree to the following terms and conditions for my contribution: