Clean up before combining pipeline and
reservoir code
This commit is contained in:
@@ -1,31 +1,5 @@
|
||||
import numpy as np
|
||||
|
||||
def Volume_trend(influx, outflux, timestep=1, V_0=0):
|
||||
'''
|
||||
Returns the trend and the volume and the final volume, defined
|
||||
by influx and outflux patterns. The optional parameter timestep
|
||||
defines the time increment over which the fluxes are changing.
|
||||
'''
|
||||
net_flux = influx-outflux
|
||||
delta_V = net_flux*timestep
|
||||
V_trend = V_0+np.cumsum(delta_V)
|
||||
V_end = V_trend[-1]
|
||||
return V_end, V_trend
|
||||
|
||||
def Height_trend(V_trend, area=1, h_crit_low=-np.inf, h_crit_high=np.inf):
|
||||
'''
|
||||
Returns the trend and the height and the final height, defined
|
||||
by influx and outflux patterns as well as the crosssection area.
|
||||
The optional parameters h_crit_low/high indicate limits that the height
|
||||
should never exceed. If this occures, TRUE is returned in the corresponding
|
||||
h_crit_flag.
|
||||
'''
|
||||
h_trend = V_trend/area
|
||||
h_crit_flag_low = np.any(h_trend <= h_crit_low)
|
||||
h_crit_flag_high = np.any(h_trend >= h_crit_high)
|
||||
h_end = h_trend[-1]
|
||||
return h_trend, h_end, h_crit_flag_low, h_crit_flag_high
|
||||
|
||||
def get_h_halfstep(initial_height, influx, outflux, timestep, area):
|
||||
h0 = initial_height
|
||||
Q_in = influx
|
||||
@@ -45,7 +19,6 @@ def FODE_function(x, h, alpha, p, rho=1000., g=9.81):
|
||||
|
||||
def e_RK_4(yn, h, dt, Q0, Q1, A0, A1, p0, p1):
|
||||
alpha = (A1/A0-1)
|
||||
|
||||
h_hs = get_h_halfstep(h, Q0, Q1, dt, A0)
|
||||
p_hs = get_p_halfstep(p0, p1)
|
||||
Y1 = yn
|
||||
@@ -55,13 +28,3 @@ def e_RK_4(yn, h, dt, Q0, Q1, A0, A1, p0, p1):
|
||||
ynp1 = yn + dt/6*(FODE_function(Y1, h, alpha, p)+2*FODE_function(Y2, h_hs, alpha, p_hs)+ \
|
||||
2*FODE_function(Y3, h_hs, alpha, p_hs)+ FODE_function(Y4, h, alpha, p))
|
||||
|
||||
|
||||
|
||||
|
||||
## testing
|
||||
# if __name__ == "__main__":
|
||||
# influx = np.full([1, 100], 6)
|
||||
# outflux = np.full_like(influx, 4)
|
||||
# V_end, V_trend = Volume_trend(influx, outflux, timestep=0.5, V_0 = 100)
|
||||
# print(V_end)
|
||||
# print(V_trend)
|
||||
|
||||
@@ -1,67 +0,0 @@
|
||||
import numpy as np
|
||||
|
||||
def Volume_trend(influx, outflux, timestep=1, V_0=0):
|
||||
'''
|
||||
Returns the trend and the volume and the final volume, defined
|
||||
by influx and outflux patterns. The optional parameter timestep
|
||||
defines the time increment over which the fluxes are changing.
|
||||
'''
|
||||
net_flux = influx-outflux
|
||||
delta_V = net_flux*timestep
|
||||
V_trend = V_0+np.cumsum(delta_V)
|
||||
V_end = V_trend[-1]
|
||||
return V_end, V_trend
|
||||
|
||||
def Height_trend(V_trend, area=1, h_crit_low=-np.inf, h_crit_high=np.inf):
|
||||
'''
|
||||
Returns the trend and the height and the final height, defined
|
||||
by influx and outflux patterns as well as the crosssection area.
|
||||
The optional parameters h_crit_low/high indicate limits that the height
|
||||
should never exceed. If this occures, TRUE is returned in the corresponding
|
||||
h_crit_flag.
|
||||
'''
|
||||
h_trend = V_trend/area
|
||||
h_crit_flag_low = np.any(h_trend <= h_crit_low)
|
||||
h_crit_flag_high = np.any(h_trend >= h_crit_high)
|
||||
h_end = h_trend[-1]
|
||||
return h_trend, h_end, h_crit_flag_low, h_crit_flag_high
|
||||
|
||||
def get_h_halfstep(initial_height, influx, outflux, timestep, area):
|
||||
h0 = initial_height
|
||||
Q_in = influx
|
||||
Q_out = outflux
|
||||
dt = timestep
|
||||
A = area
|
||||
|
||||
h_halfstep = h0+1/A*(Q_in-Q_out)*dt/2
|
||||
|
||||
def get_p_halfstep(p0, p1):
|
||||
p_halfstep = (p0+p1)/2
|
||||
|
||||
def FODE_function(x, h, alpha, p, rho=1000., g=9.81):
|
||||
f = x*abs(x)/h*alpha+g-p/(rho*h)
|
||||
return f
|
||||
|
||||
|
||||
def e_RK_4(yn, h, dt, Q0, Q1, A0, A1, p0, p1):
|
||||
alpha = (A1/A0-1)
|
||||
|
||||
h_hs = get_h_halfstep(h, Q0, Q1, dt, A0)
|
||||
p_hs = get_p_halfstep(p0, p1)
|
||||
Y1 = yn
|
||||
Y2 = yn + dt/2*FODE_function(Y1, h, alpha, p0)
|
||||
Y3 = yn + dt/2*FODE_function(Y2, h_hs, alpha, p_hs)
|
||||
Y4 = yn + dt*FODE_function(Y3, h_hs, alpha, p_hs)
|
||||
ynp1 = yn + dt/6*(FODE_function(Y1, h, alpha, p)+2*FODE_function(Y2, h_hs, alpha, p_hs)+ \
|
||||
2*FODE_function(Y3, h_hs, alpha, p_hs)+ FODE_function(Y4, h, alpha, p))
|
||||
|
||||
|
||||
|
||||
|
||||
## testing
|
||||
# if __name__ == "__main__":
|
||||
# influx = np.full([1, 100], 6)
|
||||
# outflux = np.full_like(influx, 4)
|
||||
# V_end, V_trend = Volume_trend(influx, outflux, timestep=0.5, V_0 = 100)
|
||||
# print(V_end)
|
||||
# print(V_trend)
|
||||
@@ -1,90 +0,0 @@
|
||||
from Ausgleichsbecken import FODE_function, get_h_halfstep, get_p_halfstep
|
||||
from pressure_conversion import pressure_conversion
|
||||
class Ausgleichsbecken_class:
|
||||
# units
|
||||
area_unit = r'$\mathrm{m}^2$'
|
||||
area_outflux_unit = r'$\mathrm{m}^2$'
|
||||
level_unit = 'm'
|
||||
volume_unit = r'$\mathrm{m}^3$'
|
||||
flux_unit = r'$\mathrm{m}^3/\mathrm{s}$'
|
||||
time_unit = 's'
|
||||
pressure_unit = 'Pa'
|
||||
|
||||
# init
|
||||
def __init__(self,area,outflux_area,level_min,level_max,timestep = 1):
|
||||
self.area = area # base area of the rectangular structure
|
||||
self.area_outflux = outflux_area # area of the outlet towards the pipeline
|
||||
self.level_min = level_min # lowest allowed water level
|
||||
self.level_max = level_max # highest allowed water level
|
||||
self.timestep = timestep # timestep of the simulation
|
||||
|
||||
def update_volume(self):
|
||||
self.volume = self.level*self.area
|
||||
# setter
|
||||
|
||||
def set_initial_level(self,initial_level):
|
||||
self.level = initial_level
|
||||
self.update_volume()
|
||||
|
||||
def set_influx(self,influx):
|
||||
self.influx = influx
|
||||
|
||||
def set_outflux(self,outflux):
|
||||
self.outflux = outflux
|
||||
|
||||
# getter
|
||||
def get_area(self):
|
||||
print('The base area of the cuboid reservoir is', self.area, self.area_unit)
|
||||
|
||||
def get_outflux_area(self):
|
||||
print('The outflux area from the cuboid reservoir to the pipeline is', \
|
||||
self.area_outflux, self.area_outflux_unit)
|
||||
|
||||
def get_level(self):
|
||||
print('The current level in the reservoir is', self.level , self.level_unit)
|
||||
|
||||
def get_crit_levels(self):
|
||||
print('The critical water levels in the reservoir are: \n',\
|
||||
' Minimum:', self.level_min , self.level_unit , '\n',\
|
||||
' Maximum:', self.level_max , self.level_unit )
|
||||
|
||||
def get_volume(self):
|
||||
print('The current water volume in the reservoir is', self.volume, self.volume_unit)
|
||||
|
||||
def get_timestep(self):
|
||||
print('The timestep for the simulation is' , self.timestep, self.time_unit)
|
||||
|
||||
def get_influx(self):
|
||||
print('The current influx is', self.influx, self.flux_unit)
|
||||
|
||||
def get_outflux(self):
|
||||
print('The current outflux is', self.outflux, self.flux_unit)
|
||||
|
||||
# methods
|
||||
|
||||
|
||||
def update_level(self,timestep):
|
||||
# dont update volume here, because update_level gets called to calculate h_halfstep
|
||||
net_flux = self.influx-self.outflux
|
||||
delta_V = net_flux*timestep
|
||||
new_level = (self.volume+delta_V)/self.area
|
||||
return new_level
|
||||
|
||||
|
||||
def e_RK_4(self):
|
||||
# Update to deal with non constant pipeline pressure!
|
||||
yn = self.outflux/self.area_outflux
|
||||
h = self.level
|
||||
dt = self.timestep
|
||||
p,_ = pressure_conversion(self.initial_pressure,self.pressure_unit,'Pa')
|
||||
p_hs,_ = pressure_conversion(self.initial_pressure,self.pressure_unit,'Pa')
|
||||
alpha = (self.area_outflux/self.area-1)
|
||||
h_hs = self.update_level(dt/2)
|
||||
Y1 = yn
|
||||
Y2 = yn + dt/2*FODE_function(Y1, h, alpha, self.initial_pressure)
|
||||
Y3 = yn + dt/2*FODE_function(Y2, h_hs, alpha, p_hs)
|
||||
Y4 = yn + dt*FODE_function(Y3, h_hs, alpha, p_hs)
|
||||
ynp1 = yn + dt/6*(FODE_function(Y1, h, alpha, p)+2*FODE_function(Y2, h_hs, alpha, p_hs)+ \
|
||||
2*FODE_function(Y3, h_hs, alpha, p_hs)+ FODE_function(Y4, h, alpha, p))
|
||||
|
||||
self.outflux = ynp1*self.area_outflux
|
||||
@@ -1,145 +0,0 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 4,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import numpy as np\n",
|
||||
"from Ausgleichsbecken_class_file import Ausgleichsbecken_class\n",
|
||||
"import matplotlib.pyplot as plt\n",
|
||||
"from pressure_conversion import pressure_conversion"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 5,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# define constants\n",
|
||||
"initial_level = 5. # m\n",
|
||||
"initial_influx = 1. # m³/s\n",
|
||||
"initial_outflux = 0. # m³/s\n",
|
||||
"initial_pipeline_pressure = 1.\n",
|
||||
"initial_pressure_unit = 'bar'\n",
|
||||
"conversion_pressure_unit = 'mWS'\n",
|
||||
"\n",
|
||||
"area_base = 1. # m²\n",
|
||||
"area_outflux = 0.5 # m²\n",
|
||||
"critical_level_low = 0. # m\n",
|
||||
"critical_level_high = 10. # m\n",
|
||||
"simulation_timestep = 0.001 # s\n",
|
||||
"\n",
|
||||
"# for while loop\n",
|
||||
"total_min_level = 0.01 # m\n",
|
||||
"total_max_time = 150 # s"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 6,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"%matplotlib qt\n",
|
||||
"\n",
|
||||
"V = Ausgleichsbecken_class(area_base, area_outflux, critical_level_low, critical_level_high,simulation_timestep)\n",
|
||||
"V.set_initial_level(initial_level) \n",
|
||||
"V.set_influx(initial_influx)\n",
|
||||
"V.set_outflux(initial_outflux)\n",
|
||||
"\n",
|
||||
"V.initial_pressure, V.pressure_unit = pressure_conversion(initial_pipeline_pressure,input_unit = initial_pressure_unit, target_unit = conversion_pressure_unit)\n",
|
||||
"\n",
|
||||
"time_vec = np.arange(0,total_max_time,simulation_timestep)\n",
|
||||
"outflux_vec = np.empty_like(time_vec)\n",
|
||||
"outflux_vec[0] = initial_outflux\n",
|
||||
"level_vec = np.empty_like(time_vec)\n",
|
||||
"level_vec[0] = initial_level\n",
|
||||
" \n",
|
||||
"i_max = -1\n",
|
||||
"\n",
|
||||
"for i in range(np.size(time_vec)-1):\n",
|
||||
" V.e_RK_4()\n",
|
||||
" V.level = V.update_level(V.timestep)\n",
|
||||
" V.update_volume()\n",
|
||||
" outflux_vec[i+1] = V.outflux\n",
|
||||
" level_vec[i+1] = V.level\n",
|
||||
" if V.level < total_min_level:\n",
|
||||
" i_max = i\n",
|
||||
" break\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"fig1, (ax1, ax2, ax3) = plt.subplots(3, 1)\n",
|
||||
"fig1.set_figheight(10)\n",
|
||||
"fig1.suptitle('Ausgleichsbecken')\n",
|
||||
"\n",
|
||||
"ax1.plot(time_vec[:i_max],level_vec[:i_max], label='Water level')\n",
|
||||
"ax1.set_ylabel(r'$h$ ['+V.level_unit+']')\n",
|
||||
"ax1.set_xlabel(r'$t$ ['+V.time_unit+']')\n",
|
||||
"ax1.legend()\n",
|
||||
"\n",
|
||||
"ax2.plot(time_vec[:i_max],outflux_vec[:i_max], label='Outflux')\n",
|
||||
"ax2.set_ylabel(r'$Q_{out}$ ['+V.flux_unit+']')\n",
|
||||
"ax2.set_xlabel(r'$t$ ['+V.time_unit+']')\n",
|
||||
"ax2.legend()\n",
|
||||
"\n",
|
||||
"# plt.subplots_adjust(left=0.2, bottom=0.2)\n",
|
||||
"ax3.set_axis_off()\n",
|
||||
"cell_text = np.array([[initial_level, V.level_unit], \\\n",
|
||||
" [initial_influx, V.flux_unit], \\\n",
|
||||
" [initial_outflux, V.flux_unit], \\\n",
|
||||
" [round(V.initial_pressure,2), V.pressure_unit], \\\n",
|
||||
" [simulation_timestep, V.time_unit], \\\n",
|
||||
" [area_base, V.area_unit], \\\n",
|
||||
" [area_outflux, V.area_unit]])\n",
|
||||
"\n",
|
||||
"row_labels =['initial_level', \\\n",
|
||||
" 'initial_influx', \\\n",
|
||||
" 'initial_outflux', \\\n",
|
||||
" 'initial_pipeline_pressure', \\\n",
|
||||
" 'simulation_timestep', \\\n",
|
||||
" 'area_base', \\\n",
|
||||
" 'area_outflux']\n",
|
||||
"\n",
|
||||
"plt.table(cellText=cell_text, \\\n",
|
||||
" cellLoc='center', \\\n",
|
||||
" colWidths=[0.3,0.1,0.3], \\\n",
|
||||
" rowLabels=row_labels, \\\n",
|
||||
" loc = 1, \\\n",
|
||||
" rowLoc='left', \\\n",
|
||||
" fontsize = 15.)\n",
|
||||
"\n",
|
||||
"fig1.tight_layout() "
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3.8.13 ('Georg_DT_Slot3')",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.8.13"
|
||||
},
|
||||
"orig_nbformat": 4,
|
||||
"vscode": {
|
||||
"interpreter": {
|
||||
"hash": "84fb123bdc47ab647d3782661abcbe80fbb79236dd2f8adf4cef30e8755eb2cd"
|
||||
}
|
||||
}
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
}
|
||||
File diff suppressed because one or more lines are too long
@@ -1,77 +0,0 @@
|
||||
# convert to Pa
|
||||
def bar_to_pa(p):
|
||||
return p*1e5
|
||||
|
||||
def mWS_to_pa(p):
|
||||
return p*9.80665*1e3
|
||||
|
||||
def torr_to_pa(p):
|
||||
return p*133.322
|
||||
|
||||
def atm_to_pa(p):
|
||||
return p*101.325*1e3
|
||||
|
||||
def psi_to_pa(p):
|
||||
return p*6894.8
|
||||
|
||||
# convert from Pa
|
||||
def pa_to_bar(p):
|
||||
return p*1e-5
|
||||
|
||||
def pa_to_mWS(p):
|
||||
return p*1/(9.80665*1e3)
|
||||
|
||||
def pa_to_torr(p):
|
||||
return p/133.322
|
||||
|
||||
def pa_to_atm(p):
|
||||
return p*1/(101.325*1e3)
|
||||
|
||||
# converstion function
|
||||
|
||||
def pa_to_psi(p):
|
||||
return p/6894.8
|
||||
|
||||
def pressure_conversion(pressure, input_unit = 'bar', target_unit = 'Pa'):
|
||||
p = pressure
|
||||
if input_unit.lower() == 'bar':
|
||||
p_pa = bar_to_pa(p)
|
||||
elif input_unit.lower() == 'mws':
|
||||
p_pa = mWS_to_pa(p)
|
||||
elif input_unit.lower() == 'torr':
|
||||
p_pa = torr_to_pa(p)
|
||||
elif input_unit.lower() == 'atm':
|
||||
p_pa = atm_to_pa(p)
|
||||
elif input_unit.lower() == 'psi':
|
||||
p_pa = psi_to_pa(p)
|
||||
elif input_unit.lower() == 'pa':
|
||||
p_pa = p
|
||||
else:
|
||||
raise Exception('Given input unit not recognised. \n Known units are: Pa, bar, mWs, Torr, atm, psi')
|
||||
|
||||
if target_unit.lower() == 'bar':
|
||||
return pa_to_bar(p_pa), target_unit
|
||||
elif target_unit.lower() == 'mws':
|
||||
return pa_to_mWS(p_pa), target_unit
|
||||
elif target_unit.lower() == 'torr':
|
||||
return pa_to_torr(p_pa), target_unit
|
||||
elif target_unit.lower() == 'atm':
|
||||
return pa_to_atm(p_pa), target_unit
|
||||
elif target_unit.lower() =='psi':
|
||||
return pa_to_psi(p_pa), target_unit
|
||||
elif target_unit.lower() == 'pa':
|
||||
return p_pa, target_unit
|
||||
else:
|
||||
raise Exception('Given target unit not recognised. \n Known units are: Pa, bar, mWs, Torr, atm, psi')
|
||||
|
||||
# testing_pressure_conversion
|
||||
if __name__ == '__main__':
|
||||
p = 1
|
||||
|
||||
unit_dict = ['Pa','Bar','Torr','Atm','MWS','psi']
|
||||
|
||||
for input_unit in unit_dict:
|
||||
for target_unit in unit_dict:
|
||||
converted_p = pressure_conversion(p,input_unit,target_unit)
|
||||
print(input_unit,target_unit)
|
||||
print(converted_p)
|
||||
@@ -1,12 +0,0 @@
|
||||
import numpy as np
|
||||
from math import pi
|
||||
|
||||
|
||||
def Hagen_Poiseuille(P_above,P_below,dx,constants=[1,1]):
|
||||
dP = P_above-P_below
|
||||
r = constants[0]
|
||||
vis = constants[1]
|
||||
Q = (pi*r**4)/(8*vis)*dP/dx
|
||||
return Q
|
||||
|
||||
|
||||
@@ -1,47 +0,0 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 1,
|
||||
"metadata": {},
|
||||
"outputs": [
|
||||
{
|
||||
"name": "stdout",
|
||||
"output_type": "stream",
|
||||
"text": [
|
||||
"11\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"source": [
|
||||
"import numpy as np\n",
|
||||
"import plotly\n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"interpreter": {
|
||||
"hash": "84fb123bdc47ab647d3782661abcbe80fbb79236dd2f8adf4cef30e8755eb2cd"
|
||||
},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3.8.13 ('Georg_DT_Slot3')",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.8.13"
|
||||
},
|
||||
"orig_nbformat": 4
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
}
|
||||
@@ -1,9 +0,0 @@
|
||||
class Person:
|
||||
def __init__(self, name, age):
|
||||
self.name = name
|
||||
self.age = age
|
||||
|
||||
p1 = Person("John", 36)
|
||||
|
||||
print(p1.name)
|
||||
print(p1.age)
|
||||
@@ -1,89 +0,0 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 8,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"%matplotlib qt\n",
|
||||
"import numpy as np\n",
|
||||
"import matplotlib.pyplot as plt\n",
|
||||
"\n",
|
||||
"n = 10000\n",
|
||||
"\n",
|
||||
"t = np.linspace(0,200*np.pi,n)\n",
|
||||
"omega = 2\n",
|
||||
"f_t = np.sin(omega*t)*np.cos(50/n*t)\n",
|
||||
"\n",
|
||||
"dt_max = 100\n",
|
||||
"x_s = np.full([dt_max,],np.NaN)\n",
|
||||
"y_s = np.full([dt_max,],np.NaN)\n",
|
||||
"\n",
|
||||
"# fig_ref = plt.figure()\n",
|
||||
"# ax_ref = fig_ref.add_subplot(111)\n",
|
||||
"# line_obj_ref = ax_ref.plot(t,f_t, marker='.')\n",
|
||||
"# plt.show(block=False)\n",
|
||||
"# plt.pause(3)\n",
|
||||
"# plt.close(fig_ref)\n",
|
||||
"\n",
|
||||
"fig = plt.figure()\n",
|
||||
"ax1 = fig.add_subplot(211)\n",
|
||||
"ax2 = fig.add_subplot(212)\n",
|
||||
"ax1.set_xlim([0,t[100+50]])\n",
|
||||
"ax1.set_ylim([-1.05,1.05])\n",
|
||||
"ax2.set_xlim([t[0],t[100+50]])\n",
|
||||
"ax2.set_ylim([-1.05,1.05])\n",
|
||||
"line_obj1, = ax1.plot(0,0, marker='.')\n",
|
||||
"line_obj2, = ax2.plot(0,0, marker='.')\n",
|
||||
"plt.show(block=False)\n",
|
||||
"plt.pause(0.01)\n",
|
||||
"\n",
|
||||
"for i in range(n):\n",
|
||||
" if i <= dt_max:\n",
|
||||
" x_s[:i] = t[:i]\n",
|
||||
" y_s[:i] = f_t[:i]\n",
|
||||
" else:\n",
|
||||
" x_s = t[i-dt_max:i]\n",
|
||||
" y_s = f_t[i-dt_max:i]\n",
|
||||
" ax1.set_xlim([t[i-dt_max],t[i]+t[50]])\n",
|
||||
" ax2.set_xlim([t[0],t[i]+(t[i]-t[0])/2])\n",
|
||||
" \n",
|
||||
" line_obj1.set_xdata(x_s)\n",
|
||||
" line_obj1.set_ydata(y_s)\n",
|
||||
" line_obj2.set_xdata(t[:i])\n",
|
||||
" line_obj2.set_ydata(f_t[:i])\n",
|
||||
" ax1.set_title(str(i))\n",
|
||||
" fig.canvas.draw()\n",
|
||||
" plt.pause(0.001)\n",
|
||||
"\n",
|
||||
" \n"
|
||||
]
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"interpreter": {
|
||||
"hash": "84fb123bdc47ab647d3782661abcbe80fbb79236dd2f8adf4cef30e8755eb2cd"
|
||||
},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3.8.13 ('Georg_DT_Slot3')",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.8.13"
|
||||
},
|
||||
"orig_nbformat": 4
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
}
|
||||
@@ -1,138 +0,0 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import numpy as np\n",
|
||||
"import pandas as pd\n",
|
||||
"import plotly.express as px\n",
|
||||
"from plotly.subplots import make_subplots\n",
|
||||
"import plotly.graph_objects as go\n",
|
||||
"from flow_patterns import return_flux_profiles,make_flux_df\n",
|
||||
"from volume_change import V_h_test_2,h_V_test_2"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# #constant flows\n",
|
||||
"# #number of steps\n",
|
||||
"# n = 100\n",
|
||||
"# #input identifiers\n",
|
||||
"# i_i_1 = 0\n",
|
||||
"# #output identifiers\n",
|
||||
"# o_i_1 = 0\n",
|
||||
"# # influx and outflux offset\n",
|
||||
"# i_o = 10\n",
|
||||
"# o_o = 10\n",
|
||||
"# #outflux delay\n",
|
||||
"# o_d = 10\n",
|
||||
"\n",
|
||||
"# influx_profile,outflux_profile = return_flux_profiles(n,i_i_1,o_i_1,i_o,o_o,o_d)\n",
|
||||
"# flux_df = make_flux_df(influx_profile,outflux_profile)\n",
|
||||
"\n",
|
||||
"# fig = make_subplots(2,1)\n",
|
||||
"\n",
|
||||
"# fig.add_trace(go.Scatter(x=flux_df['time'],y=flux_df['influx']),row=1,col=1)\n",
|
||||
"# fig.add_trace(go.Scatter(x=flux_df['time'],y=flux_df['outflux']),row=2,col=1)\n",
|
||||
"# fig.show()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# #linear increasing flows\n",
|
||||
"# #number of steps\n",
|
||||
"# n = 100\n",
|
||||
"# #input identifiers\n",
|
||||
"# i_i_2 = 'lin_0010'\n",
|
||||
"# #output identifiers\n",
|
||||
"# o_i_2 = 'lin_0010'\n",
|
||||
"# # influx and outflux offset\n",
|
||||
"# i_o = 10\n",
|
||||
"# o_o = 10\n",
|
||||
"# #outflux delay\n",
|
||||
"# o_d = 10\n",
|
||||
"\n",
|
||||
"# influx_profile,outflux_profile = return_flux_profiles(n,i_i_2,o_i_2,i_o,o_o,o_d)\n",
|
||||
"# flux_df = make_flux_df(influx_profile,outflux_profile)\n",
|
||||
"\n",
|
||||
"# fig = make_subplots(2,1)\n",
|
||||
"\n",
|
||||
"# fig.add_trace(go.Scatter(x=flux_df['time'],y=flux_df['influx']),row=1,col=1)\n",
|
||||
"# fig.add_trace(go.Scatter(x=flux_df['time'],y=flux_df['outflux']),row=2,col=1)\n",
|
||||
"# fig.show()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# #sawtooth flows\n",
|
||||
"# #number of steps\n",
|
||||
"# n = 100\n",
|
||||
"# #input identifiers\n",
|
||||
"# i_i_3 = 'st_0010_0010'\n",
|
||||
"# #output identifiers\n",
|
||||
"# o_i_3 = 'st_0010_0010'\n",
|
||||
"# # influx and outflux offset\n",
|
||||
"# i_o = 10\n",
|
||||
"# o_o = 10\n",
|
||||
"# #outflux delay\n",
|
||||
"# o_d = 10\n",
|
||||
"\n",
|
||||
"# influx_profile,outflux_profile = return_flux_profiles(n,i_i_3,o_i_3,i_o,o_o,o_d)\n",
|
||||
"# flux_df = make_flux_df(influx_profile,outflux_profile)\n",
|
||||
"\n",
|
||||
"# fig = make_subplots(2,1)\n",
|
||||
"\n",
|
||||
"# fig.add_trace(go.Scatter(x=flux_df['time'],y=flux_df['influx']),row=1,col=1)\n",
|
||||
"# fig.add_trace(go.Scatter(x=flux_df['time'],y=flux_df['outflux']),row=2,col=1)\n",
|
||||
"# fig.show()"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": []
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"interpreter": {
|
||||
"hash": "84fb123bdc47ab647d3782661abcbe80fbb79236dd2f8adf4cef30e8755eb2cd"
|
||||
},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3.8.13 ('Georg_DT_Slot3')",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.8.13"
|
||||
},
|
||||
"orig_nbformat": 4
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
}
|
||||
@@ -1,67 +0,0 @@
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
import plotly.express as px
|
||||
from plotly.subplots import make_subplots
|
||||
import plotly.graph_objects as go
|
||||
|
||||
|
||||
def return_flux_profiles(number_of_steps = 1,influx_identifier = 0, outflux_identifier = 0,influx_offset=0,outflux_offset=0, outflux_delay = 0):
|
||||
''' Identifier patterns:
|
||||
0 ... constant
|
||||
'lin_SSSS' ... linear increase with slope int(SSSS)
|
||||
'st_SSSS_PPPP' ... sawtooth pattern with slope int(SSSS) and period int(PPPP) steps
|
||||
'''
|
||||
|
||||
# case identifiers for if statment
|
||||
i = influx_identifier
|
||||
o = outflux_identifier
|
||||
|
||||
n = number_of_steps
|
||||
#starting value for the influx and outflux
|
||||
i_o = influx_offset
|
||||
o_o = outflux_offset
|
||||
# number of steps, the outflux is held at 0 at the beginning
|
||||
o_d = outflux_delay
|
||||
|
||||
|
||||
# get base profile for the influx (offset will get applied later)
|
||||
if i == 0:
|
||||
influx_profile = np.zeros(n)
|
||||
elif 'lin' in influx_identifier:
|
||||
k = int(influx_identifier[-4:])
|
||||
influx_profile = np.linspace(0,k*(n-1),n)
|
||||
elif 'st' in influx_identifier:
|
||||
k = int(influx_identifier[3:7])
|
||||
p = int(influx_identifier[-4:])
|
||||
influx_profile = np.tile(np.linspace(0,k*(p-1),p),int(np.ceil(n/p)))
|
||||
|
||||
|
||||
# apply influx offset
|
||||
influx_profile = influx_offset + influx_profile
|
||||
|
||||
if o == 0:
|
||||
outflux_profile = np.zeros(n)
|
||||
elif 'lin' in outflux_identifier:
|
||||
k = int(outflux_identifier[-4:])
|
||||
outflux_profile = np.linspace(0,k*(n-1),n)
|
||||
elif 'st' in outflux_identifier:
|
||||
k = int(outflux_identifier[3:7])
|
||||
p = int(outflux_identifier[-4:])
|
||||
outflux_profile = np.tile(np.linspace(0,k*(p-1),p),int(np.ceil(n/p)))
|
||||
|
||||
#apply outflux offset and delay (delay means, that the first o_d steps, the outflux will be 0)
|
||||
outflux_profile = np.concatenate((np.zeros(o_d),outflux_profile[:-o_d]+o_o))
|
||||
|
||||
return influx_profile,outflux_profile
|
||||
|
||||
def make_flux_df(influx_profile,outflux_profile, time = 0):
|
||||
if time == 0:
|
||||
time = np.arange(0,len(influx_profile))
|
||||
flux_df = pd.DataFrame(np.transpose([time, influx_profile, outflux_profile]), \
|
||||
columns=['time', 'influx', 'outflux'])
|
||||
return flux_df
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
influx_profile,outflux_profile = return_flux_profiles(100,influx_identifier='st_0010_0010',influx_offset=10)
|
||||
print(influx_profile)
|
||||
@@ -1,92 +0,0 @@
|
||||
{
|
||||
"cells": [
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 12,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"import numpy as np \n",
|
||||
"from pressure_propagation import pressure_update"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": 13,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": [
|
||||
"# Ausbreitungsgeschwindigkeit\n",
|
||||
"u = 1 # m/s\n",
|
||||
"# Rohrlänge\n",
|
||||
"l = 100 # m\n",
|
||||
"# maximal simulierte Zeitspanne\n",
|
||||
"t_max = 60 # s\n",
|
||||
"\n",
|
||||
"# Zeitschritt\n",
|
||||
"delta_t = 0.1 # s\n",
|
||||
"# Diskretisierungslänge = Ausbreitungsgeschwindigkeit*Zeitschritt\n",
|
||||
"delta_x = u*delta_t\n",
|
||||
"\n",
|
||||
"# Anzahl der örtlichen Diskretisierungsintervalle\n",
|
||||
"n_x = int(np.floor(l/delta_x))\n",
|
||||
"# Anzahl der zeitlichen Diskretisierungsintervalle\n",
|
||||
"n_t = int(np.floor(t_max/delta_t))\n",
|
||||
"\n",
|
||||
"\n",
|
||||
"#initiale Druckverteilung (excl hydrostatischer Drucks)\n",
|
||||
"p_0 = np.ones([n_x,1])\n",
|
||||
"# np.array das den Verlauf der Druckverteilungen speichert\n",
|
||||
"pressure_profiles = np.tile(p_0,[1,n_t])\n",
|
||||
"\n",
|
||||
"pressure_profiles[-1,0] = 2 # for testing\n",
|
||||
"# loop\n",
|
||||
"for i in range(1,n_t): # start at 1 because i reference i-1 in the loop over the control-volumina\n",
|
||||
" #get boundary pressure from outflux-change and hydrostatic pressure in the pool\n",
|
||||
" pressure_profiles[-1,i] = 2 # for testing\n",
|
||||
" \n",
|
||||
" for j in range(n_x-2,1,-1): # leave out the first and last control-volume because their pressure\n",
|
||||
" # is set by the boundary conditions\n",
|
||||
" p = pressure_profiles[j,i-1]\n",
|
||||
" p1 = pressure_profiles[j+1,i-1]\n",
|
||||
" p2 = pressure_profiles[j-1,i-1]\n",
|
||||
" pressure_profiles[j,i] = pressure_update(p,p1,p2)\n",
|
||||
" \n",
|
||||
"\n",
|
||||
"\n"
|
||||
]
|
||||
},
|
||||
{
|
||||
"cell_type": "code",
|
||||
"execution_count": null,
|
||||
"metadata": {},
|
||||
"outputs": [],
|
||||
"source": []
|
||||
}
|
||||
],
|
||||
"metadata": {
|
||||
"interpreter": {
|
||||
"hash": "84fb123bdc47ab647d3782661abcbe80fbb79236dd2f8adf4cef30e8755eb2cd"
|
||||
},
|
||||
"kernelspec": {
|
||||
"display_name": "Python 3.8.13 ('Georg_DT_Slot3')",
|
||||
"language": "python",
|
||||
"name": "python3"
|
||||
},
|
||||
"language_info": {
|
||||
"codemirror_mode": {
|
||||
"name": "ipython",
|
||||
"version": 3
|
||||
},
|
||||
"file_extension": ".py",
|
||||
"mimetype": "text/x-python",
|
||||
"name": "python",
|
||||
"nbconvert_exporter": "python",
|
||||
"pygments_lexer": "ipython3",
|
||||
"version": "3.8.13"
|
||||
},
|
||||
"orig_nbformat": 4
|
||||
},
|
||||
"nbformat": 4,
|
||||
"nbformat_minor": 2
|
||||
}
|
||||
@@ -1,2 +0,0 @@
|
||||
def pressure_update(p,p1=-1,p2=-1):
|
||||
return 1/4*(2*p+p1+p2)
|
||||
@@ -1 +0,0 @@
|
||||
import plotly
|
||||
File diff suppressed because it is too large
Load Diff
@@ -1,140 +0,0 @@
|
||||
|
||||
# Testvolume
|
||||
# Depth of the whole structure is constant and given by the variable d
|
||||
#
|
||||
#
|
||||
# { x_1*d*h for h <= h_1
|
||||
# V(h) = { x_1*d*(h-h_1)+(x_2-x_1)*d*(h-h_1)**2/(2*(h_2-h_1) + V(h_1)) for h_1 < h <= h_2
|
||||
# { x_2*d*(h-h_2)+(x_3-x_2)*d*(h-h_2)**2/(2*(h_3-h_2) + V(h_2)) for h_2 < h <= h_3
|
||||
# { x_3*d*(h-h_3) + V(h_3) for h_3 < h
|
||||
#
|
||||
#
|
||||
# { V/(x_1*d) for V <= V_1
|
||||
#h(V) = { (-b_2+sqrt(b_2**2-4*a_2*c_2)/(2*a_2)) for V_1 < V <= V_2
|
||||
# { (-b_3+sqrt(b_3**2-4*a_3*c_3)/(2*a_3)) for V_2 < V <= V_3
|
||||
# { (V-V_3)/(x_1*d) for V_3 < V
|
||||
#
|
||||
# with
|
||||
# a_2 = 0.5*((x_2-x_1)*d)/(h_2-h_1)
|
||||
# a_3 = 0.5*((x_3-x_2)*d)/(h_3-h_2)
|
||||
#
|
||||
# b_2 = x_1*d-((x_2-x_1)*d*h_1)/(h_2-h_1)
|
||||
# b_3 = x_2*d-((x_3-x_2)*d*h_2)/(h_3-h_2)
|
||||
#
|
||||
# c_2 = ((x_2-x_1)*d*h_1**2)/(h_2-h_1)-h_1*x_1*d-(V-V_1)
|
||||
# c_3 = ((x_3-x_2)*d*h_2**2)/(h_3-h_2)-h_2*x_2*d-(V-V_2)
|
||||
#
|
||||
#
|
||||
#
|
||||
#
|
||||
#
|
||||
# _____
|
||||
# | | |
|
||||
# | | |
|
||||
# | | | h_4 - h_3
|
||||
# | | _|_
|
||||
# __| _ _ |__ |
|
||||
# / x_3 \ |
|
||||
# / \ |
|
||||
# / \ |
|
||||
# / \ | h_3 - h_2
|
||||
# / \ |
|
||||
# / \ |
|
||||
# / \ |
|
||||
# / \ |
|
||||
# / \ _|_
|
||||
# <-----------------------------> |
|
||||
# \ x_2 / | h_2 - h_1
|
||||
# \ / |
|
||||
# \ _ _ _ _ _ _ _ _ _ _ _ / _|_
|
||||
# | x_1 | |
|
||||
# | | | h_1
|
||||
# | | |
|
||||
# |_____________________| _|_
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
def test_1_parameters():
|
||||
h_1 = 10
|
||||
h_2 = 5 + h_1
|
||||
h_3 = 5 + h_2
|
||||
|
||||
x_1 = 100
|
||||
x_2 = 101
|
||||
x_3 = 30
|
||||
|
||||
d = 5
|
||||
|
||||
vol_1 = x_1*d*h_1
|
||||
vol_2 = x_1*d*(h_2-h_1)+(x_2-x_1)*d*(h_2-h_1)**2/(2*(h_2-h_1)) + vol_1
|
||||
vol_3 = x_2*d*(h_3-h_2)+(x_3-x_2)*d*(h_3-h_2)**2/(2*(h_3-h_2)) + vol_2
|
||||
|
||||
a_2 = 0.5*((x_2-x_1)*d)/(h_2-h_1)
|
||||
a_3 = 0.5*((x_3-x_2)*d)/(h_3-h_2)
|
||||
|
||||
b_2 = x_1*d-((x_2-x_1)*d*h_1)/(h_2-h_1)
|
||||
b_3 = x_2*d-((x_3-x_2)*d*h_2)/(h_3-h_2)
|
||||
|
||||
c_2 = ((x_2-x_1)*d*h_1**2)/(2*(h_2-h_1))-h_1*x_1*d
|
||||
c_3 = ((x_3-x_2)*d*h_2**2)/(2*(h_3-h_2))-h_2*x_2*d
|
||||
|
||||
return h_1,h_2,h_3,x_1,x_2,x_3,d,vol_1,vol_2,vol_3,a_2,a_3,b_2,b_3,c_2,c_3
|
||||
|
||||
def V_h_test_1(h):
|
||||
h_1,h_2,h_3,x_1,x_2,x_3,d,vol_1,vol_2,vol_3,a_2,a_3,b_2,b_3,c_2,c_3 = test_1_parameters()
|
||||
if h <= h_1:
|
||||
V = x_1*d*h
|
||||
elif (h_1 < h) and (h <= h_2):
|
||||
V = x_1*d*(h-h_1)+(x_2-x_1)*d*(h-h_1)**2/(2*(h_2-h_1)) + vol_1
|
||||
elif (h_2 < h) and (h <= h_3):
|
||||
V = x_2*d*(h-h_2)+(x_3-x_2)*d*(h-h_2)**2/(2*(h_3-h_2)) + vol_2
|
||||
elif (h_3 < h):
|
||||
V = x_3*d*(h-h_3) + vol_3
|
||||
|
||||
return V
|
||||
|
||||
def h_V_test_1(V):
|
||||
h_1,h_2,h_3,x_1,x_2,x_3,d,vol_1,vol_2,vol_3,a_2,a_3,b_2,b_3,c_2,c_3 =test_1_parameters()
|
||||
if V <= vol_1:
|
||||
h = V/(x_1*d)
|
||||
elif (vol_1 < V) and (V <= vol_2):
|
||||
h = (-b_2+(b_2**2-4*a_2*(c_2-(V-vol_1)))**0.5)/(2*a_2)
|
||||
elif (vol_2 < V) and (V <= vol_3):
|
||||
h = (-b_3+(b_3**2-4*a_3*(c_3-(V-vol_2)))**0.5)/(2*a_3)
|
||||
elif (vol_3 < V):
|
||||
h = (V-vol_3)/(x_3*d)+h_3
|
||||
return h
|
||||
|
||||
|
||||
def test_2_parameters():
|
||||
x = 10.
|
||||
d = 10.
|
||||
return x,d
|
||||
|
||||
def V_h_test_2(h):
|
||||
x,d = test_2_parameters()
|
||||
return x*d*h
|
||||
|
||||
def h_V_test_2(V):
|
||||
x,d = test_2_parameters()
|
||||
return V/(x*d)
|
||||
|
||||
def show_parameters(test_version):
|
||||
h_1,h_2,h_3,x_1,x_2,x_3,d,vol_1,vol_2,vol_3,a_2,a_3,b_2,b_3,c_2,c_3 = test_1_parameters()
|
||||
x,d = test_2_parameters()
|
||||
|
||||
if test_version == 1:
|
||||
print('h_1: ', h_1)
|
||||
print('h_2: ', h_2)
|
||||
print('h_3: ', h_3)
|
||||
print('x_1: ', x_1)
|
||||
print('x_2: ', x_2)
|
||||
print('x_3: ', x_3)
|
||||
elif test_version == 2:
|
||||
print('x: ', x)
|
||||
print('d: ', d)
|
||||
|
||||
|
||||
|
||||
File diff suppressed because one or more lines are too long
Reference in New Issue
Block a user