RadFrac

Emulates Aspen's RadFrac column. It gathers data from feed stream, and combine additional unit-specific input to make predictions on duties and bottom & distillate streams.

Distillation(
    no_stages: Union[int, List[int]] None
    pressure: Union[float, List[float]] = None
    dist_rate: Union[float, List[float]] = None
    reflux_ratio: Union[float, List[float]] = None
    bott_rate: Union[float, List[float]] = None
    reflux_rate: Union[float, List[float]] = None
    boilup_rate: Union[float, List[float]] = None
    boilup_ratio: Union[float, List[float]] = None
    dist_to_feed_ratio: Union[float, List[float]] = None
    bott_to_feed_ratio: Union[float, List[float]] = None
    condensor_duty: Union[float, List[float]] = None
    reboiler_duty: Union[float, List[float]] = None
    free_water_reflux_ratio: Union[float, List[float]] = None
    feed_stage: Union[int, List[int]] = None
    calc_type: Literal["equil", "rate-based"] = None
    condensor_type: Literal["total", "partial-v", "partial-vl", "none"] = None
    reboiler_type: Literal["kettle", "thermosiphon", "none"] = None
    valid_phases: Literal["vl", "vll", "vl-fwc", "vl-fwns", "vl-dwc", "vl-dwas"] = None
    convergence: Literal["std", "petro/wb", "snil", "azeo", "cryo", "custom"] = None
    feed_stage_convention: Literal["above", "on", "vapor", "liquid"] = None
    rate_units: str = "kmol/hr"
    duty_units: str = "kJ/hr"
    pres_units: str = "atm"
    property_method: Literal["nrtl"] = "nrtl"
    verbose: bool = False
    check_data: bool = True
    object_id: str = None
)(feed_stream: MaterialStream, **kwargs)

Data Checks & Manipulations

• Will raise an error if no_stages and pressure aren't specified.
• If feed_stage isn't specified, will default to round(no_stages/2). Also, if feed_stage > no_stages + 1, it wll trigger an error.
• User can input the units of the input data via rate_units, duty_units and pres_units. The corresponding quantities will be converted to default units.
• 2 out of operating configurations (dist_rate, reflux_rate, etc.) must be specified.
• Input must obey type hints (enforced by Pydantic).

Support for Batch Predictions

Distillation currently has limtied support for broadcasting when

• The incoming feed stream has 1 row of data while current unit data has multiple.
• The incoming feed stream has multiple rows of data while the current unit data has 1.

The best way to to take advantage of batch predictions is to prepare a dataframe of input parameteras and input individual columns.

Useful Attributes

• object_id: unique id of object. Can be set during initialization, and will be checked against a global registry to prevent duplicates.
• before: list of feed streams.
• after: list of output streams.
• data: a pandas df containing all data used for ML.
• condensor_duty: predicted given input.
• reboiler_duty: predicted given input.
• status: numpy array specifying which row of ML df should be trusted. Similar to how Aspen indicates Errors.