mirror of
				https://github.com/KevinMidboe/linguist.git
				synced 2025-10-29 17:50:22 +00:00 
			
		
		
		
	
		
			
				
	
	
		
			325 lines
		
	
	
		
			9.9 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
			
		
		
	
	
			325 lines
		
	
	
		
			9.9 KiB
		
	
	
	
		
			Python
		
	
	
	
	
	
| #!/usr/bin/python
 | |
| # -*- coding: utf-8 -*-
 | |
| 
 | |
| # Copyright © 2013 Martin Ueding <dev@martin-ueding.de>
 | |
| 
 | |
| import argparse
 | |
| import matplotlib.pyplot as pl
 | |
| import numpy as np
 | |
| import scipy.optimize as op
 | |
| from prettytable import PrettyTable
 | |
| 
 | |
| __docformat__ = "restructuredtext en"
 | |
| 
 | |
| # Sensitivität der Thermosäule
 | |
| S = 30e-6
 | |
| 
 | |
| def phif(U):
 | |
|     return U / S
 | |
| 
 | |
| def main():
 | |
|     options = _parse_args()
 | |
| 
 | |
|     V = 1000
 | |
| 
 | |
|     data = np.genfromtxt("a-leer.csv", delimiter="\t")
 | |
|     t = data[:,0]
 | |
|     U = data[:,1] / V / 1000
 | |
|     U_err = 0.7e-3 / V
 | |
| 
 | |
|     offset = np.mean(U[-3:])
 | |
| 
 | |
|     x = np.linspace(min(t), max(t))
 | |
|     y = np.ones(x.size) * offset
 | |
|     pl.plot(x, y * 10**6, label="Offset")
 | |
| 
 | |
|     print "Offset: {:.3g} V".format(offset)
 | |
| 
 | |
|     pl.errorbar(t, U * 10**6, yerr=U_err * 10**6, linestyle="none", marker="+",
 | |
|                 label="Messdaten")
 | |
|     pl.grid(True)
 | |
|     pl.legend(loc="best")
 | |
|     pl.title(u"Bestimmung des Offsets")
 | |
|     pl.xlabel(ur"Zeit $t / \mathrm{s}$")
 | |
|     pl.ylabel(ur"Thermospannung $U / \mathrm{\mu V}$")
 | |
|     pl.savefig("Plot_a-leer.pdf")
 | |
|     pl.clf()
 | |
| 
 | |
|     V = 100
 | |
| 
 | |
|     data = np.genfromtxt("a-Lampe.csv", delimiter="\t")
 | |
|     t = data[:,0]
 | |
|     U = data[:,1] / V / 1000 - offset
 | |
|     U_err = 0.7e-3 / V
 | |
| 
 | |
|     x = np.linspace(min(t), max(t))
 | |
|     y = np.ones(x.size) * max(U) * 0.9
 | |
|     pl.plot(x, y * 10**6, label=ur"$90\%$")
 | |
| 
 | |
|     pl.errorbar(t, U * 10**6, yerr=U_err * 10**6, linestyle="none", marker="+",
 | |
|                 label="Messdaten")
 | |
|     pl.grid(True)
 | |
|     pl.legend(loc="best")
 | |
|     pl.title(u"Bestimmung der Ansprechzeit")
 | |
|     pl.xlabel(ur"Zeit $t / \mathrm{s}$")
 | |
|     pl.ylabel(ur"Thermospannung $U / \mathrm{\mu V}$")
 | |
|     pl.savefig("Plot_a-Lampe.pdf")
 | |
|     pl.clf()
 | |
| 
 | |
|     # Lesliewürfel
 | |
|     print """
 | |
| Lesliewürfel
 | |
| ============
 | |
| """
 | |
| 
 | |
|     glanz = np.genfromtxt("b-glanz.csv", delimiter="\t")
 | |
|     matt = np.genfromtxt("b-matt.csv", delimiter="\t")
 | |
|     schwarz = np.genfromtxt("b-schwarz.csv", delimiter="\t")
 | |
|     weiss = np.genfromtxt("b-weiss.csv", delimiter="\t")
 | |
| 
 | |
|     T0 = 19.0 + 273.15
 | |
|     T0_err = 1.0
 | |
| 
 | |
|     glanz[:,0] += 273.15
 | |
|     matt[:,0] += 273.15
 | |
|     schwarz[:,0] += 273.15
 | |
|     weiss[:,0] += 273.15
 | |
| 
 | |
|     glanz[:,1] /= 1000 * V
 | |
|     matt[:,1] /= 1000 * V
 | |
|     schwarz[:,1] /= 1000 * V
 | |
|     weiss[:,1] /= 1000 * V
 | |
| 
 | |
|     glanz[:,1] -= offset
 | |
|     matt[:,1] -= offset
 | |
|     schwarz[:,1] -= offset
 | |
|     weiss[:,1] -= offset
 | |
| 
 | |
|     glanz_phi = phif(glanz[:,1])
 | |
|     matt_phi = phif(matt[:,1])
 | |
|     schwarz_phi = phif(schwarz[:,1])
 | |
|     weiss_phi = phif(weiss[:,1])
 | |
| 
 | |
|     T_err = 0.3
 | |
| 
 | |
|     sigma = 5.670373e-8
 | |
| 
 | |
|     def boltzmann(T, epsilon, offset):
 | |
|         return epsilon * sigma * T**4 + offset
 | |
| 
 | |
|     glanz_popt, glanz_pconv = op.curve_fit(boltzmann, glanz[:,0], glanz_phi)
 | |
|     matt_popt, matt_pconv = op.curve_fit(boltzmann, matt[:,0], matt_phi)
 | |
|     schwarz_popt, schwarz_pconv = op.curve_fit(boltzmann, schwarz[:,0], schwarz_phi)
 | |
|     weiss_popt, weiss_pconv = op.curve_fit(boltzmann, weiss[:,0], weiss_phi)
 | |
| 
 | |
|     glanz_x = np.linspace(min(glanz[:,0]), max(glanz[:,0]))
 | |
|     glanz_y = boltzmann(glanz_x, *glanz_popt)
 | |
|     pl.plot(glanz_x, glanz_y, label="Fit glanz", color="gold")
 | |
| 
 | |
|     matt_x = np.linspace(min(matt[:,0]), max(matt[:,0]))
 | |
|     matt_y = boltzmann(matt_x, *matt_popt)
 | |
|     pl.plot(matt_x, matt_y, label="Fit matt", color="yellow")
 | |
| 
 | |
|     schwarz_x = np.linspace(min(schwarz[:,0]), max(schwarz[:,0]))
 | |
|     schwarz_y = boltzmann(schwarz_x, *schwarz_popt)
 | |
|     pl.plot(schwarz_x, schwarz_y, label="Fit schwarz", color="black")
 | |
| 
 | |
|     weiss_x = np.linspace(min(weiss[:,0]), max(weiss[:,0]))
 | |
|     weiss_y = boltzmann(weiss_x, *weiss_popt)
 | |
|     pl.plot(weiss_x, weiss_y, label="Fit weiss", color="gray")
 | |
| 
 | |
|     print "glanz ε = {:.3g} ± {:.3g}".format(glanz_popt[0], np.sqrt(glanz_pconv.diagonal()[0]))
 | |
|     print "glanz offset = {:.3g} ± {:.3g}".format(glanz_popt[1], np.sqrt(glanz_pconv.diagonal()[1]))
 | |
|     print "matt ε = {:.3g} ± {:.3g}".format(matt_popt[0], np.sqrt(matt_pconv.diagonal()[0]))
 | |
|     print "matt offset = {:.3g} ± {:.3g}".format(matt_popt[1], np.sqrt(matt_pconv.diagonal()[1]))
 | |
|     print "schwarz ε = {:.3g} ± {:.3g}".format(schwarz_popt[0], np.sqrt(schwarz_pconv.diagonal()[0]))
 | |
|     print "schwarz offset = {:.3g} ± {:.3g}".format(schwarz_popt[1], np.sqrt(schwarz_pconv.diagonal()[1]))
 | |
|     print "weiss ε = {:.3g} ± {:.3g}".format(weiss_popt[0], np.sqrt(weiss_pconv.diagonal()[0]))
 | |
|     print "weiss offset = {:.3g} ± {:.3g}".format(weiss_popt[1], np.sqrt(weiss_pconv.diagonal()[1]))
 | |
| 
 | |
|     pl.errorbar(glanz[:,0], glanz_phi, xerr=T_err, yerr=U_err/S,
 | |
|                 label="glanz", color="gold", linestyle="none")
 | |
|     pl.errorbar(matt[:,0], matt_phi, xerr=T_err, yerr=U_err/S,
 | |
|                 label="matt", color="yellow", linestyle="none")
 | |
|     pl.errorbar(schwarz[:,0], schwarz_phi, xerr=T_err, yerr=U_err/S,
 | |
|                 label="schwarz", color="black", linestyle="none")
 | |
|     pl.errorbar(weiss[:,0], weiss_phi, xerr=T_err, yerr=U_err/S,
 | |
|                 label="weiss", color="gray", linestyle="none")
 | |
| 
 | |
|     header = ["T / K", "Phi/F in W/m^2", "Fehler T", "Fehler Phi/F"]
 | |
| 
 | |
|     print """
 | |
| Tabellen für den Lesliewürfel-Plot
 | |
| ----------------------------------
 | |
| """
 | |
| 
 | |
|     print "Glanz"
 | |
|     glanz_table = PrettyTable(header)
 | |
|     for row in zip(glanz[:,0], glanz_phi, np.ones(glanz[:,0].size)*T_err, np.ones(glanz_phi.size)*U_err/S):
 | |
|         glanz_table.add_row(row)
 | |
|     print glanz_table
 | |
|     print
 | |
| 
 | |
|     print "Matt"
 | |
|     matt_table = PrettyTable(header)
 | |
|     for row in zip(matt[:,0], matt_phi, np.ones(matt[:,0].size)*T_err, np.ones(matt_phi.size)*U_err/S):
 | |
|         matt_table.add_row(row)
 | |
|     print matt_table
 | |
|     print
 | |
| 
 | |
|     print "Schwarz"
 | |
|     schwarz_table = PrettyTable(header)
 | |
|     for row in zip(schwarz[:,0], schwarz_phi, np.ones(schwarz[:,0].size)*T_err, np.ones(schwarz_phi.size)*U_err/S):
 | |
|         schwarz_table.add_row(row)
 | |
|     print schwarz_table
 | |
|     print
 | |
| 
 | |
|     print "Weiß"
 | |
|     weiss_table = PrettyTable(header)
 | |
|     for row in zip(weiss[:,0], weiss_phi, np.ones(weiss[:,0].size)*T_err, np.ones(weiss_phi.size)*U_err/S):
 | |
|         weiss_table.add_row(row)
 | |
|     print weiss_table
 | |
|     print
 | |
| 
 | |
|     epsilon = 0.1
 | |
| 
 | |
|     x = np.linspace(min([min(x) for x in [glanz[:,0], matt[:,0], schwarz[:,0],
 | |
|                                           weiss[:,0]]]),
 | |
|                     max([max(x) for x in [glanz[:,0], matt[:,0], schwarz[:,0],
 | |
|                                           weiss[:,0]]]),
 | |
|                     100)
 | |
|     offset = - epsilon * sigma * T0**4
 | |
|     print "ideal offset = {:.3g}".format(offset)
 | |
|     y = boltzmann(x, epsilon, offset)
 | |
|     pl.plot(x, y, label=ur"$\epsilon = 0.1$")
 | |
| 
 | |
| 
 | |
|     pl.grid(True)
 | |
|     pl.title(u"Lesliewürfel")
 | |
|     pl.xlabel(ur"Temperatur $T / \mathrm{K}$")
 | |
|     pl.ylabel(ur"Strahlungsfluss $\frac{\Phi}{F} / \mathrm{\frac{W}{m^2}}$")
 | |
|     pl.legend(loc="best", prop={"size": 12})
 | |
|     pl.savefig("Plot_b.pdf")
 | |
|     pl.clf()
 | |
| 
 | |
|     # Aufgabe c
 | |
|     print """
 | |
| Aufgabe c
 | |
| =========
 | |
|     """
 | |
| 
 | |
|     data = np.genfromtxt("c-erste.csv", delimiter="\t")
 | |
|     d = data[:,0] / 100
 | |
|     U = data[:,1] / V
 | |
|     phi = phif(U)
 | |
| 
 | |
|     def c(x, a, b):
 | |
|         return a*x + b
 | |
| 
 | |
| 
 | |
|     dx = d**(-2)
 | |
|     dy = phi
 | |
| 
 | |
|     dx_err = np.abs(-2 * d**(-3)) * 0.001
 | |
|     dy_err = 0.001 / S
 | |
| 
 | |
|     popt, pconv = op.curve_fit(c, dx, dy)
 | |
|     x = np.linspace(min(dx), max(dx))
 | |
|     y = c(x, *popt)
 | |
|     pl.plot(x, y, label="Fit")
 | |
| 
 | |
|     print "Fitparameter"
 | |
|     print "a", popt[0], "±", np.sqrt(pconv.diagonal()[0])
 | |
|     print "b", popt[1], "±", np.sqrt(pconv.diagonal()[1])
 | |
| 
 | |
|     pl.errorbar(dx, dy, xerr=dx_err, yerr=dy_err, linestyle="none",
 | |
|                 marker="+", label="Messdaten")
 | |
|     pl.grid(True)
 | |
|     pl.title(u"Halogenlampe bei verschiedenen Abständen")
 | |
|     pl.xlabel(ur"Abstand $d^{-2} / \mathrm{m^{-2}}$")
 | |
|     pl.ylabel(ur"Strahlungsfluss $\frac{\Phi}{F} / \mathrm{\frac{W}{m^2}}$")
 | |
|     pl.legend(loc="best")
 | |
|     pl.savefig("Plot_c-erste.pdf")
 | |
|     pl.clf()
 | |
| 
 | |
|     print
 | |
|     print "Tabelle für Aufgabe c"
 | |
|     fields = ["d^-2 in m^-2", "Phi/F in W/m^2", "Fehler d^-2", "Fehler Phi/F"]
 | |
|     table = PrettyTable(fields)
 | |
|     table.align = "l"
 | |
|     for row in zip(dx, dy, dx_err, np.ones(dy.size)*dy_err):
 | |
|         table.add_row(row)
 | |
|     print table
 | |
|     print
 | |
| 
 | |
|     data = np.genfromtxt("c-zweite.csv", delimiter="\t")
 | |
|     U1 = data[:,0]
 | |
|     I1 = data[:,1]
 | |
|     U2 = data[:,2] / V
 | |
| 
 | |
|     U_err = 0.001
 | |
|     I_err = 0.01
 | |
| 
 | |
|     p = U1 * I1
 | |
|     R = U1 / I1
 | |
|     R_err = np.sqrt(
 | |
|         (1/I1 * U_err)**2
 | |
|         + (U1/I1**2 * I_err)**2
 | |
|     )
 | |
| 
 | |
|     phi = phif(U2)
 | |
|     phi_err = U_err / S
 | |
| 
 | |
|     alpha = 4.82e-3
 | |
|     beta = 6.76e-7
 | |
| 
 | |
|     R0 = 0.35
 | |
|     R0_err = 0.05
 | |
| 
 | |
|     T = (-alpha*R0 + np.sqrt(R0)*np.sqrt(4*beta*R + alpha**2*R0 - 4*beta*R0) +
 | |
|  2*beta*R0*T0)/(2*beta*R0)
 | |
| 
 | |
|     popt, pconv = op.curve_fit(boltzmann, T, phi, sigma=phi_err)
 | |
|     x = np.linspace(min(T), max(T))
 | |
|     y = boltzmann(x, *popt)
 | |
|     pl.plot(x, y, label="Fit")
 | |
| 
 | |
|     epsilon = popt[0]
 | |
|     epsilon_err = np.sqrt(pconv.diagonal()[0])
 | |
| 
 | |
|     print "ε = {:.3g} ± {:.3g}".format(epsilon, epsilon_err)
 | |
| 
 | |
|     f1 = (1/(np.sqrt(R0)*np.sqrt(4*beta*R + alpha**2*R0 - 4*beta*R0))) * R_err
 | |
|     f2 = T0_err
 | |
|     f3 = ((-alpha + ((alpha**2 - 4*beta)*np.sqrt(R0))/( 2*np.sqrt(4*beta*R + alpha**2*R0 - 4*beta*R0)) + np.sqrt( 4*beta*R + alpha**2*R0 - 4*beta*R0)/(2*np.sqrt(R0)) + 2*beta*T0)/( 2*beta*R0) - (-alpha*R0 + np.sqrt(R0)*np.sqrt(4*beta*R + alpha**2*R0 - 4*beta*R0) + 2*beta*R0*T0)/( 2*beta*R0**2)) * R0_err
 | |
| 
 | |
|     T_err = np.sqrt(f1**2 + f2**2 + f3**2)
 | |
| 
 | |
|     pl.errorbar(T, phi, xerr=T_err, yerr=phi_err, label="Messdaten",
 | |
|                 linestyle="none", marker="+")
 | |
|     pl.grid(True)
 | |
|     pl.legend(loc="best")
 | |
|     pl.title(u"Halogenlampe bei verschiedenen Leistungen")
 | |
|     pl.xlabel(u"Temperatur $T / \mathrm{K}$")
 | |
|     pl.ylabel(ur"Strahlungsfluss $\frac{\Phi}{F} / \mathrm{\frac{W}{m^2}}$")
 | |
|     pl.savefig("Plot_c-zweite.pdf")
 | |
|     pl.clf()
 | |
| 
 | |
| def _parse_args():
 | |
|     """
 | |
|     Parses the command line arguments.
 | |
| 
 | |
|     :return: Namespace with arguments.
 | |
|     :rtype: Namespace
 | |
|     """
 | |
|     parser = argparse.ArgumentParser(description="")
 | |
|     #parser.add_argument("args", metavar="N", type=str, nargs="*", help="Positional arguments.")
 | |
|     #parser.add_argument("", dest="", type="", default=, help=)
 | |
|     #parser.add_argument("--version", action="version", version="<the version>")
 | |
| 
 | |
|     return parser.parse_args()
 | |
| 
 | |
| 
 | |
| if __name__ == "__main__":
 | |
|     main()
 |