idk
103
ME_2046/HW3/simplifying.py
Normal file
@ -0,0 +1,103 @@
|
|||||||
|
import sympy as sm
|
||||||
|
import numpy as np
|
||||||
|
sm.init_printing()
|
||||||
|
|
||||||
|
s = sm.symbols('s')
|
||||||
|
z = sm.symbols('z')
|
||||||
|
T = sm.symbols('T')
|
||||||
|
|
||||||
|
#########################################################
|
||||||
|
print('PROBLEM 2:')
|
||||||
|
print('Part a:')
|
||||||
|
|
||||||
|
ZOH = (1-sm.exp(-s*T))/(s*T)
|
||||||
|
|
||||||
|
G_1_s = 1/s*ZOH
|
||||||
|
|
||||||
|
bilinear_s = 2/T *(z-1)/(z+1)
|
||||||
|
|
||||||
|
G_1_k = G_1_s.subs({s:bilinear_s}).simplify()
|
||||||
|
|
||||||
|
print("G_1_k = ")
|
||||||
|
sm.pprint(G_1_k)
|
||||||
|
|
||||||
|
print('Part b:')
|
||||||
|
|
||||||
|
theta_s_r_s = ZOH*1/s**2
|
||||||
|
theta_k_r_k = theta_s_r_s.subs({s:bilinear_s}).simplify()
|
||||||
|
print("theta_k_r_k = ")
|
||||||
|
sm.pprint(theta_k_r_k)
|
||||||
|
|
||||||
|
|
||||||
|
#########################################################
|
||||||
|
print('PROBLEM 3:')
|
||||||
|
A = sm.Matrix([[0, 1, 0, 0],
|
||||||
|
[0, 0, 1, 0],
|
||||||
|
[0, 0, 0, 1],
|
||||||
|
[1, 0, 0, 0]])
|
||||||
|
B = sm.Matrix([0, 0, 0, 1])
|
||||||
|
C = sm.Matrix([1, 0, 0, 0]).transpose()
|
||||||
|
D = sm.Matrix([1])
|
||||||
|
|
||||||
|
print('System Matricies A, B, C, D')
|
||||||
|
sm.pprint(A)
|
||||||
|
sm.pprint(B)
|
||||||
|
sm.pprint(C)
|
||||||
|
sm.pprint(D)
|
||||||
|
|
||||||
|
#Recursive Solution
|
||||||
|
k = sm.symbols('k', integer = True, real = True, positive = True)
|
||||||
|
|
||||||
|
def y(k,u):
|
||||||
|
term_1 = C * A**k * x_0
|
||||||
|
|
||||||
|
term_2 = sm.Matrix([0,0,0,0])
|
||||||
|
for j in range(np.size(u)):
|
||||||
|
term_2 = term_2 + A**(k-j-1) * B * u[j]
|
||||||
|
term_2 = C*term_2
|
||||||
|
|
||||||
|
term_3 = D*u[-1]
|
||||||
|
|
||||||
|
return term_1+term_2+term_3
|
||||||
|
|
||||||
|
print('Part c:')
|
||||||
|
x_0 = sm.Matrix([2, 1, 3, 0])
|
||||||
|
u = sm.Matrix([0])
|
||||||
|
|
||||||
|
output = y(k, u)
|
||||||
|
output = output[0].expand().simplify()
|
||||||
|
print('y(k) = ')
|
||||||
|
sm.pprint(output)
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
print('Part d:')
|
||||||
|
x_0 = sm.Matrix([0, 0, 0, 0])
|
||||||
|
u = sm.Matrix([2, 1, 3, 0])
|
||||||
|
output = y(k, u)
|
||||||
|
output = output[0].expand().simplify()
|
||||||
|
print('y(k) = ')
|
||||||
|
sm.pprint(output)
|
||||||
|
|
||||||
|
print('Part e:')
|
||||||
|
print('These are the same exact response between parts C and D. This makes sense, becasue we defined our states as just being delays in a chain. The result is that the input at timestep k trickles down through each state in k+1, k+2, and k+3. This means that our states save our input in a way, s.t. loading this initial state mathematically produces an identical result as loading those inputs in one time step at a time. \n \n This is reflected by the two algebraeic expressions being the same.')
|
||||||
|
|
||||||
|
#########################################################
|
||||||
|
print('PROBLEM 4:')
|
||||||
|
print('Part a:')
|
||||||
|
|
||||||
|
"""
|
||||||
|
x(k+2) - x(k+1) + 0.25 x(k) = u(k+2)
|
||||||
|
|
||||||
|
Applying Z transform:
|
||||||
|
z^2 X - z X + 0.25 X = z^2 U
|
||||||
|
X (z^2 - z + 0.25) = z^2 U
|
||||||
|
X/U = z^2 / (z^2 - z + 0.25)
|
||||||
|
"""
|
||||||
|
# Use SymPy to do partial frac
|
||||||
|
z = sm.symbols('z')
|
||||||
|
|
||||||
|
X_U = z**2/(z**2 - z + 0.25)
|
||||||
|
sm.pprint(X_U.apart())
|
||||||
|
|
||||||
|
|
||||||
1
NUCE_2113/lab5/.~lock.labviz.ods#
Normal file
@ -0,0 +1 @@
|
|||||||
|
,danesabo,danesabo-laptop,24.02.2025 15:04,file:///home/danesabo/snap/libreoffice/336/.config/libreoffice/4;
|
||||||
1051
NUCE_2113/lab5/2025-02-18LAB5-Cd.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-Cd.png
Normal file
|
After Width: | Height: | Size: 29 KiB |
1051
NUCE_2113/lab5/2025-02-18LAB5-Co.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-Co.png
Normal file
|
After Width: | Height: | Size: 33 KiB |
1051
NUCE_2113/lab5/2025-02-18LAB5-CoCs-CALIBRATION.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-CoCs-CALIBRATION.png
Normal file
|
After Width: | Height: | Size: 29 KiB |
1051
NUCE_2113/lab5/2025-02-18LAB5-Cs.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-Cs.png
Normal file
|
After Width: | Height: | Size: 25 KiB |
1051
NUCE_2113/lab5/2025-02-18LAB5-Na.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-Na.png
Normal file
|
After Width: | Height: | Size: 30 KiB |
1051
NUCE_2113/lab5/2025-02-18LAB5-UNKNOWN.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-UNKNOWN.png
Normal file
|
After Width: | Height: | Size: 27 KiB |
1052
NUCE_2113/lab5/2025-02-18LAB5-UNKNOWN_EX5.3.Spe
Normal file
BIN
NUCE_2113/lab5/2025-02-18LAB5-UNKNOWN_EX5.3.png
Normal file
|
After Width: | Height: | Size: 30 KiB |
BIN
NUCE_2113/lab5/combined_spectrum.png
Normal file
|
After Width: | Height: | Size: 70 KiB |
BIN
NUCE_2113/lab5/labviz.ods
Normal file
BIN
NUCE_2113/lab5/spectrum.png
Normal file
|
After Width: | Height: | Size: 29 KiB |
101
NUCE_2113/lab5/viz.py
Normal file
@ -0,0 +1,101 @@
|
|||||||
|
import glob
|
||||||
|
import os
|
||||||
|
import numpy as np
|
||||||
|
import matplotlib.pyplot as plt
|
||||||
|
|
||||||
|
def load_spectrum_data(filename):
|
||||||
|
"""
|
||||||
|
Load spectrum data from a text file in the provided format.
|
||||||
|
|
||||||
|
This function:
|
||||||
|
- Reads the file line by line.
|
||||||
|
- Finds the "$DATA:" section.
|
||||||
|
- Skips the first line after "$DATA:" (assumed to be the channel range header).
|
||||||
|
- Collects all subsequent lines (until the next section indicated by a line starting with '$').
|
||||||
|
- Parses these lines into a NumPy array.
|
||||||
|
|
||||||
|
Parameters:
|
||||||
|
filename (str): Path to the spectrum file.
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
numpy.ndarray: Array of count values.
|
||||||
|
"""
|
||||||
|
with open(filename, 'r') as f:
|
||||||
|
lines = f.readlines()
|
||||||
|
|
||||||
|
data_lines = []
|
||||||
|
in_data_section = False
|
||||||
|
skip_first_data_line = True # flag to skip the channel-range header
|
||||||
|
|
||||||
|
for line in lines:
|
||||||
|
# Look for the beginning of the data section
|
||||||
|
if line.strip().startswith("$DATA:"):
|
||||||
|
in_data_section = True
|
||||||
|
continue
|
||||||
|
if in_data_section:
|
||||||
|
# If we hit a new section header, exit the data section
|
||||||
|
if line.strip().startswith("$"):
|
||||||
|
break
|
||||||
|
# Skip the first line after "$DATA:" if it contains exactly two numbers (channel range header)
|
||||||
|
if skip_first_data_line:
|
||||||
|
parts = line.strip().split()
|
||||||
|
if len(parts) == 2:
|
||||||
|
skip_first_data_line = False
|
||||||
|
continue
|
||||||
|
skip_first_data_line = False # even if not two numbers, do not skip subsequent lines
|
||||||
|
# Append non-empty lines
|
||||||
|
if line.strip():
|
||||||
|
data_lines.append(line.strip())
|
||||||
|
|
||||||
|
# Combine the collected lines into one string and parse numbers
|
||||||
|
data_str = " ".join(data_lines)
|
||||||
|
# Convert the string of numbers into a NumPy array
|
||||||
|
data = np.fromstring(data_str, sep=' ')
|
||||||
|
return data
|
||||||
|
|
||||||
|
# Get a list of all .Spe files in the current directory
|
||||||
|
spe_files = glob.glob("*.Spe")
|
||||||
|
|
||||||
|
# To store data for the combined plot
|
||||||
|
combined_data = []
|
||||||
|
|
||||||
|
for file in spe_files:
|
||||||
|
# Load the spectrum data
|
||||||
|
spectrum_data = load_spectrum_data(file)
|
||||||
|
# Create an array for channel numbers (one per data point)
|
||||||
|
channels = np.arange(len(spectrum_data))
|
||||||
|
|
||||||
|
# Store the data for the combined plot (use file name without extension for legend)
|
||||||
|
base_name = os.path.splitext(os.path.basename(file))[0]
|
||||||
|
combined_data.append((base_name, channels, spectrum_data))
|
||||||
|
|
||||||
|
# Plot individual spectrum
|
||||||
|
plt.figure(figsize=(10, 6))
|
||||||
|
plt.step(channels, spectrum_data, where='mid', color='blue')
|
||||||
|
plt.xlabel('Channel')
|
||||||
|
plt.ylabel('Counts')
|
||||||
|
plt.title(f"Spectrum: {base_name}")
|
||||||
|
plt.grid(True)
|
||||||
|
plt.tight_layout()
|
||||||
|
|
||||||
|
# Save individual figure as PNG
|
||||||
|
output_filename = base_name + '.png'
|
||||||
|
plt.savefig(output_filename)
|
||||||
|
plt.close() # Close the figure to free memory
|
||||||
|
|
||||||
|
# Now create a combined plot with all spectra
|
||||||
|
plt.figure(figsize=(12, 8))
|
||||||
|
for name, channels, spectrum_data in combined_data:
|
||||||
|
# Do not specify a color so that the default cycle gives different colors for each curve
|
||||||
|
plt.step(channels, spectrum_data, where='mid', label=name)
|
||||||
|
|
||||||
|
plt.xlabel('Channel')
|
||||||
|
plt.ylabel('Counts')
|
||||||
|
plt.title("Combined Spectrum Data")
|
||||||
|
plt.grid(True)
|
||||||
|
plt.legend()
|
||||||
|
plt.tight_layout()
|
||||||
|
|
||||||
|
# Save and display the combined plot
|
||||||
|
plt.savefig('combined_spectrum.png')
|
||||||
|
plt.show()
|
||||||