Source code for tespy.tools.fluid_properties.helpers

# -*- coding: utf-8

"""Module for fluid property helper functions.


This file is part of project TESPy (github.com/oemof/tespy). It's copyrighted
by the contributors recorded in the version control history of the file,
available from its original location
tespy/tools/fluid_properties/helpers.py

SPDX-License-Identifier: MIT
"""

import CoolProp.CoolProp as CP
import numpy as np

from tespy.tools.global_vars import ERR
from tespy.tools.helpers import central_difference
from tespy.tools.helpers import newton_with_kwargs
from tespy.tools.logger import logger


def _is_larger_than_precision(value):
    return value > ERR


def _check_mixing_rule(mixing_rule, mixing_functions, propertyfunction):
    if mixing_rule not in mixing_functions:
        msg = (
            f"The mixing rule '{mixing_rule}' is not available for "
            f"the fluid property functions for {propertyfunction}. Available "
            f"rules are '" + "', '".join(mixing_functions.keys()) + "'."
        )
        logger.exception(msg)
        raise KeyError(msg)


[docs] def get_number_of_fluids(fluid_data): return sum([1 for f in fluid_data.values() if _is_larger_than_precision(f["mass_fraction"])])
[docs] def get_pure_fluid(fluid_data): for f in fluid_data.values(): if _is_larger_than_precision(f["mass_fraction"]): return f
[docs] def single_fluid(fluid_data): r"""Return the name of the pure fluid in a fluid vector.""" if get_number_of_fluids(fluid_data) > 1: return None else: for fluid, data in fluid_data.items(): if _is_larger_than_precision(data["mass_fraction"]): return fluid
[docs] def get_molar_fractions(fluid_data): molarflow = { key: value["mass_fraction"] / value["wrapper"]._molar_mass for key, value in fluid_data.items() } molarflow_sum = sum(molarflow.values()) return {key: value / molarflow_sum for key, value in molarflow.items()}
[docs] def inverse_temperature_mixture(p=None, target_value=None, fluid_data=None, T0=None, f=None): # calculate the fluid properties for fluid mixtures valmin, valmax = get_mixture_temperature_range(fluid_data) if T0 is None or T0 == 0 or np.isnan(T0): T0 = (valmin + valmax) / 2.0 valmax *= 2 function_kwargs = { "p": p, "fluid_data": fluid_data, "T": T0, "function": f, "parameter": "T" , "delta": 0.01 } return newton_with_kwargs( central_difference, target_value, val0=T0, valmin=valmin, valmax=valmax, **function_kwargs )
[docs] def get_mixture_temperature_range(fluid_data): valmin = max( [v["wrapper"]._T_min for v in fluid_data.values() if _is_larger_than_precision(v["mass_fraction"])] ) + 0.1 valmax = min( [v["wrapper"]._T_max for v in fluid_data.values() if _is_larger_than_precision(v["mass_fraction"])] ) - 0.1 return valmin, valmax
[docs] def calc_molar_mass_mixture(fluid_data, molar_fractions): return sum([x * fluid_data[fluid]["wrapper"]._molar_mass for fluid, x in molar_fractions.items()])
[docs] def fluid_structure(fluid): r""" Return the checmical formula of fluid. Parameters ---------- fluid : str Name of the fluid. Returns ------- parts : dict Dictionary of the chemical base elements as keys and the number of atoms in a molecule as values. Example ------- Get the chemical formula of methane. >>> from tespy.tools.fluid_properties.helpers import fluid_structure >>> elements = fluid_structure('methane') >>> elements['C'], elements['H'] (1, 4) """ parts = {} for element in CP.get_fluid_param_string( fluid, 'formula').split('}'): if element != '': el = element.split('_{') parts[el[0]] = int(el[1]) return parts
[docs] def darcy_friction_factor(re, ks, d): r""" Calculate the Darcy friction factor. Parameters ---------- re : float Reynolds number re / 1. ks : float Pipe roughness ks / m. d : float Pipe diameter/characteristic lenght d / m. Returns ------- darcy_friction_factor : float Darcy friction factor :math:`\lambda` / 1 Note ---- **Laminar flow** (:math:`re \leq 2320`) .. math:: \lambda = \frac{64}{re} **turbulent flow** (:math:`re > 2320`) *hydraulically smooth:* :math:`\frac{re \cdot k_{s}}{d} < 65` .. math:: \lambda = \begin{cases} 0.03164 \cdot re^{-0.25} & re \leq 10^4\\ \left(1.8 \cdot \log \left(re\right) -1.5 \right)^{-2} & 10^4 < re < 10^6\\ solve \left(0 = 2 \cdot \log\left(re \cdot \sqrt{\lambda} \right) -0.8 - \frac{1}{\sqrt{\lambda}}\right) & re \geq 10^6\\ \end{cases} *transition zone and hydraulically rough:* .. math:: \lambda = solve \left( 0 = 2 \cdot \log \left( \frac{2.51}{re \cdot \sqrt{\lambda}} + \frac{k_{s}}{d \cdot 3.71} \right) - \frac{1}{\sqrt{\lambda}} \right) Reference: :cite:`Nirschl2018`. Example ------- Calculate the Darcy friction factor at different hydraulic states. >>> from tespy.tools.fluid_properties.helpers import darcy_friction_factor >>> ks = 5e-5 >>> d = 0.05 >>> re_laminar = 2000 >>> re_turb_smooth = 5000 >>> re_turb_trans = 70000 >>> re_high = 1000000 >>> d_high = 0.8 >>> re_very_high = 6000000 >>> d_very_high = 1 >>> ks_low = 1e-5 >>> ks_rough = 1e-3 >>> darcy_friction_factor(re_laminar, ks, d) 0.032 >>> round(darcy_friction_factor(re_turb_smooth, ks, d), 3) 0.038 >>> round(darcy_friction_factor(re_turb_trans, ks, d), 3) 0.023 >>> round(darcy_friction_factor(re_turb_trans, ks_rough, d), 3) 0.049 >>> round(darcy_friction_factor(re_high, ks, d_high), 3) 0.012 >>> round(darcy_friction_factor(re_very_high, ks_low, d_very_high), 3) 0.009 """ if re <= 2320: return 64 / re else: if re * ks / d < 65: if re <= 1e4: return blasius(re) elif re < 1e6: return hanakov(re) else: l0 = 0.02 function_kwargs = { "function": prandtl_karman, "parameter": "darcy_friction_factor", "reynolds": re } return newton_with_kwargs( prandtl_karman_derivative, 0, val0=l0, valmin=0.00001, valmax=0.2, **function_kwargs ) else: l0 = 0.002 function_kwargs = { "function": colebrook, "parameter": "darcy_friction_factor", "reynolds": re, "ks": ks, "diameter": d, "delta": 0.001 } return newton_with_kwargs( central_difference, 0, val0=l0, valmin=0.0001, valmax=0.2, **function_kwargs )
[docs] def blasius(re): """ Calculate friction coefficient according to Blasius. Parameters ---------- re : float Reynolds number. Returns ------- darcy_friction_factor : float Darcy friction factor. """ return 0.3164 * re ** (-0.25)
[docs] def hanakov(re): """ Calculate friction coefficient according to Hanakov. Parameters ---------- re : float Reynolds number. Returns ------- darcy_friction_factor : float Darcy friction factor. """ return (1.8 * np.log10(re) - 1.5) ** (-2)
[docs] def prandtl_karman(reynolds, darcy_friction_factor, **kwargs): """ Calculate friction coefficient according to Prandtl and v. Kármán. Applied in smooth conditions. Parameters ---------- re : float Reynolds number. darcy_friction_factor : float Darcy friction factor. Returns ------- darcy_friction_factor : float Darcy friction factor. """ return ( 2 * np.log10(reynolds * darcy_friction_factor ** 0.5) - 0.8 - 1 / darcy_friction_factor ** 0.5 )
[docs] def prandtl_karman_derivative(reynolds, darcy_friction_factor, **kwargs): """Calculate derivative for Prandtl and v. Kármán equation.""" return ( 1 / (darcy_friction_factor * np.log(10)) + 0.5 * darcy_friction_factor ** (-1.5) )
[docs] def colebrook(reynolds, ks, diameter, darcy_friction_factor, **kwargs): """ Calculate friction coefficient accroding to Colebrook-White equation. Applied in transition zone and rough conditions. Parameters ---------- re : float Reynolds number. ks : float Equivalent sand roughness. d : float Pipe's diameter. darcy_friction_factor : float Darcy friction factor. Returns ------- darcy_friction_factor : float Darcy friction factor. """ return ( 2 * np.log10( 2.51 / (reynolds * darcy_friction_factor ** 0.5) + ks / (3.71 * diameter) ) + 1 / darcy_friction_factor ** 0.5 )