Automation of surrogate model, when it is out of range

[karimimp]

Assuming the connection between the CFD code and MD tool; The independent variables of the detailed model and their maximum and minimum values are passed to the data base (or known). The framework should identify the values of the independent variables for which the detailed model is run. In other words: The information provided by the macro-scale model:

• definitions of the dependent and independent variables,
• the minimum and maximum values of the independent variables in the detailed model. The framework should be able to:
• Identifying the number of points for probing the detailed model(s)
• Fit the detailed model with a surrogate model

[sigveka]

@henrus can we make this a "feature request"

@karimimp

You want to be able to do something like this:

• Define a surrogate function (same-old same-old)

f = CFunction(                                                      
    Ccode= '''Omitted, working example contains a C-function in this string''', 
    inputs={  # Global/physical bounds of the model                                        
        'input_1': { 'min': 0, 'max': 9e99, 'argPos': 0 },
        'input_2': { 'min': -9e99, 'max': 9e99, 'argPos': 1 },
        'input_3': { 'min': 1, 'max': 10, 'argPos': 2 },
    },
    outputs={              
        'output_1': { 'min': -9e99, 'max': 9e99, 'argPos': 0 },
    },
    parameters={        
        'param0': { 'min': -10.0, 'max': 10.0, 'argPos': 0 },
        'param1': { 'min': 0.0, 'max': 1.0, 'argPos': 1 },
    },
)

• Define a backward mapping model with a new initialisationStrategy

m = BackwardMappingModel(
    _id= 'md',
    surrogateFunction= f,
    exactTask= MolecularDynamics(),
    substituteModels= [ ],
    initialisationStrategy= Strategy.InitialRange(
    { # Local bounds
        'input_1': { 'min': 5, 'max': 9 },
        'input_2': { 'min': -9, 'max': 10 },
        'input_3': { 'min': 0, 'max': 0 },
    }
    ),
    ... ... ... ... etc.
)

The initialisation strategy should perform DoE/parameter fitting of the surrogate model only inside the range provided in initialRange.

Is this somewhat close to what you want?

[karimimp]

@sigveka
Thanks. This is, I believe, what we even have an example of it in the private repo under /applications/PUfoam/units_density_reaction_mixture . The exact task runs the detailed model for the MD code and the parameters should be re-fitted.