GSoC 2020 @SymPy
This repository showcases my proposal, and the work done (final report) during Google Summer of Code 2020 with the SymPy project. SymPy is a computer algebra system written in pure python.
It all started here:
The SymPy community is very welcoming and supportive. I sent an email to the SymPy mailing list around Feb 3rd, 2020, asking the community about the scope of adding a new Control Systems engineering package. Initially, their opinions were against this, and students were advised to focus on improving existing SymPy packages, instead of adding new packages (core or not).
I continued to show my interest in this project and guess what?? Here is the response I got finally (Thanks, Jason):
Project: Control Theory - Implement a control systems package
Student: Naman Gera (namannimmo10)
Official Mentors:
- Nikhil Maan (@Sc0rpi0n101)
- Jason K. Moore (@moorepants)
- Ishan Joshi (@ishanaj)
Since this was a big project and I had to develop a package from scratch, I got a lot of help from other members of the SymPy development team and even Control theory experts. I would also really like to thank S.Y. Lee, Oscar Benjamin, Eric Wieser, Ilhan Polat, and Richard Murray for constantly reviewing my PRs and discussing the API designs with me.
Work done:
My main focus was to implement in SymPy, a basic Control Systems functionality from scratch. This can be used by Control engineers or professors/students to solve various control theory related problems. Being a part of SymPy, this is purely symbolic in nature. The advantage of this package over other packages/libraries (which are great, btw!) like harold and python-control is that the solutions obtained from it are highly accurate and do not rely on numerical methods to approximate the solutions. The solutions obtained are in a compact form that can be used for further analysis. Documentation is available at Control API and Control Intro.
PULL REQUESTS -
Major additions:
- (Merged) sympy/sympy#19390 - Adds
TransferFunction,Series,Parallel, andFeedbackclasses for control package. - (Merged) sympy/sympy#19896 - Adds other useful methods in
TransferFunctionclass. - (Merged) sympy/sympy#19761 - Adds
TransferFunctionMatrixclass in physics.control.
Miscellaneous PRs opened in the summer:
- (Merged) sympy/sympy-bot#87 - Adds physics.control as valid submodule.
- (Merged) sympy/sympy#20008 - Modify docs of
is_stablemethod inTransferFunctionclass.
Future Work:
- Add
StateSpaceclass for creating State space models - Further improve the documentation of control package. Add a couple more examples to show us the users how to use the implemented functionality as they tend to use any software more if the docs are top-notch.
- Graphical analyses:
root_locus,pole_zero,bode, andnyquistplots
Examples:
This section consists of the main features that were added under sympy.physics.control.
A transfer function is used for representing linear, time-invariant (LTI) systems that can be strictly described by a ratio of polynomials. Here's how we can construct a transfer function:
>>> from sympy.abc import s, a
>>> from sympy.physics.control import TransferFunction
>>> numerator = 2*s + a
>>> denominator = s**2 + s + 1
>>> G = TransferFunction(numerator, denominator, s) # third arg is a complex variable of the Laplace transform
>>> G
TransferFunction(a + 2*s, s**2 + s + 1, s)
>>> G.num
a + 2*s
>>> G.den
s**2 + s + 1
>>> G.var
sNow using pretty-printing (which is pretty dope ;) to make it actually look like a ratio of polynomials:
>>> from sympy import pprint
>>> pprint(G)
a + 2⋅s
──────────
2
s + s + 1
>>> from sympy import init_printing
>>> init_printing(use_unicode=True)
>>> G2 = TransferFunction(s**4 - 2*s**3 + 5*s + 4, s + 4, s)
>>> -G2 # negate a transfer function
4 3
- s + 2⋅s - 5⋅s - 4
─────────────────────
s + 4
>>> G2**3 # take the integer power
3
⎛ 4 3 ⎞
⎝s - 2⋅s + 5⋅s + 4⎠
──────────────────────
3
(s + 4)
>>> _.expand() # now expand the num and den after taking the power
12 11 10 9 8 7 6 5 4 3 2
s - 6⋅s + 12⋅s + 7⋅s - 48⋅s + 12⋅s + 123⋅s - 30⋅s - 192⋅s + 29⋅s + 300⋅s + 240⋅s + 64
───────────────────────────────────────────────────────────────────────────────────────────────────
3 2
s + 12⋅s + 48⋅s + 64You can use a Float or an Integer (or other constants) as num and den:
>>> G3 = TransferFunction(1/2, 3, a)
>>> G3.num
0.500000000000000
>>> G3.den
3
>>> G3.var
aOther features used in computations:
>>> TransferFunction(s + 2, s**2 - 9, s).dc_gain() # compute the gain of the response as the freq approaches 0.
-2/9
>>> TransferFunction(a, s, s).dc_gain()
∞⋅sign(a)
>>> G4 = TransferFunction((1 - s)**2, (s + 1)**2, s)
>>> G4.is_stable() # checks for the asymptotic stability
True
>>> G4.poles()
[-1, -1]
>>> G4.zeros()
[1, 1]
>>> G4.is_proper
True
>>> G4.is_biproper
True
>>> G4.is_strictly_proper
False
>>> from sympy import symbols
>>> c, p, d0, d1, d2 = symbols('c, p, d0:3')
>>> num, den = c*p, d2*p**3 + d1*p**2 - d0
>>> G5 = TransferFunction(num, den, p)
>>> G5.subs({c: 2, d0: 3, d1: 2, d2: 5})
2⋅p
───────────────
3 2
5⋅p + 2⋅p - 3
>>> G5.subs({c: 2, d0: 3, d1: 2, d2: 5}).evalf()
2.0⋅p
─────────────────────
3 2
5.0⋅p + 2.0⋅p - 3.0
>>> G5.xreplace({c: 3.0})
3.0⋅p
───────────────────
2 3
-d₀ + d₁⋅p + d₂⋅p
>>> from sympy import factor
>>> factor(TransferFunction(s - 1, s**2 - 2*s + 1, s))
s - 1
────────
2
(s - 1)
>>> TransferFunction((p + 3)*(p - 1), (p - 1)*(p + 5), p).simplify()
p + 3
─────
p + 5Now, let's do some SISO transfer function algebra:
>>> init_printing(use_unicode=False)
>>> G6 = TransferFunction(s + 1, s**2 + s + 1, s)
>>> G6
s + 1
----------
2
s + s + 1
>>> G7 = TransferFunction(s - p, s + 3, s)
>>> G7
-p + s
------
s + 3
>>> G8 = TransferFunction(4*s**2 + 2*s - 4, s - 1, s)
>>> G8
2
4*s + 2*s - 4
--------------
s - 1
>>> G9 = TransferFunction(a - s, s**2 + 4, s)
>>> G9
a - s
------
2
s + 4
>>> G6 + G7 # just adding two TFs will leave them unevaluated
s + 1 -p + s
---------- + ------
2 s + 3
s + s + 1
>>> # `.doit()` will evaluate the result above
>>> # `.rewrite(TransferFunction)` does the same.
>>> _.doit()
⎛ 2 ⎞
(-p + s)⋅⎝s + s + 1⎠ + (s + 1)⋅(s + 3)
───────────────────────────────────────
⎛ 2 ⎞
(s + 3)⋅⎝s + s + 1⎠
>>> G8 - G9
2
4⋅s + 2⋅s - 4 -a + s
────────────── + ──────
s - 1 2
s + 4
>>> G6 * G9
⎛ s + 1 ⎞ ⎛a - s ⎞
⎜──────────⎟⋅⎜──────⎟
⎜ 2 ⎟ ⎜ 2 ⎟
⎝s + s + 1⎠ ⎝s + 4⎠
>>> G8 * G9 + G6 - G7 # do all at the same time
⎛ 2 ⎞
⎜4⋅s + 2⋅s - 4⎟ ⎛a - s ⎞ s + 1 p - s
⎜──────────────⎟⋅⎜──────⎟ + ────────── + ─────
⎝ s - 1 ⎠ ⎜ 2 ⎟ 2 s + 3
⎝s + 4⎠ s + s + 1
>>> (G8 * G9 + G6).rewrite(TransferFunction)
⎛ 2 ⎞ ⎛ 2 ⎞ ⎛ 2 ⎞
(a - s)⋅⎝s + s + 1⎠⋅⎝4⋅s + 2⋅s - 4⎠ + (s - 1)⋅(s + 1)⋅⎝s + 4⎠
────────────────────────────────────────────────────────────────
⎛ 2 ⎞ ⎛ 2 ⎞
(s - 1)⋅⎝s + 4⎠⋅⎝s + s + 1⎠A class for representing negative feedback interconnection between input/output systems was also added - Feedback.
Here's how solve a basic block diagram problem (having a negative feedback) with that class:
>>> from sympy.physics.control import Feedback
>>> G = plant = TransferFunction(2*s**2 + 5*s + 1, s**2 + 2*s + 3, s)
>>> C = controller = TransferFunction(5*(s + 2), s + 10, s)
>>> velocity = Feedback(plant, controller)
>>> velocity
⎛ 2 ⎞
⎜2⋅s + 5⋅s + 1⎟
⎜──────────────⎟
⎜ 2 ⎟
⎝ s + 2⋅s + 3 ⎠
───────────────────────────────
⎛ 2 ⎞
1 ⎜2⋅s + 5⋅s + 1⎟ ⎛5⋅s + 10⎞
─ + ⎜──────────────⎟⋅⎜────────⎟
1 ⎜ 2 ⎟ ⎝ s + 10 ⎠
⎝ s + 2⋅s + 3 ⎠
>>> velocity.doit() # this gives the resultant closed-loop transfer function
⎛ 2 ⎞ ⎛ 2 ⎞
(s + 10)⋅⎝s + 2⋅s + 3⎠⋅⎝2⋅s + 5⋅s + 1⎠
──────────────────────────────────────────────────────────────────────
⎛ ⎛ 2 ⎞ ⎛ 2 ⎞⎞ ⎛ 2 ⎞
⎝(s + 10)⋅⎝s + 2⋅s + 3⎠ + (5⋅s + 10)⋅⎝2⋅s + 5⋅s + 1⎠⎠⋅⎝s + 2⋅s + 3⎠
>>> velocity.doit().simplify() # we can further cancel poles and zeros
⎛ 2 ⎞
(s + 10)⋅⎝2⋅s + 5⋅s + 1⎠
────────────────────────────────────────────────────
⎛ 2 ⎞ ⎛ 2 ⎞
5⋅(s + 2)⋅⎝2⋅s + 5⋅s + 1⎠ + (s + 10)⋅⎝s + 2⋅s + 3⎠My GSoC Experience and Learnings:
I've been writing weekly blog-posts documenting my journey during GSoC. The blog is hosted at namannimmo.me/emerald. This experience was so amazing and complete and I learned lots of different things; I can't write them all in here.
But, here are a few:
- Having to adjust to my team members' different timezones was definitely a challenge. I also loved the challenge of working with a remote team and leaving my comfort zone of talking in person to work on something
- I also learned written communication. Writing accurately is difficult, so I had to learn how to convey my thoughts well enough, because we were working remotely, right?
- I had not implemented any package from scratch before, so this was a challenge in itself. But the mentors were helpful enough and discussed everything with me - they were readily available for discussions
- If could repeat this again, I would do so. This community is so welcoming, even if you come up with a naïve approach of solving a particular problem, they will help you to improve the code even further because we had to adhere to the SymPy standards
- Since a lot of people get confused in the beginning, I've written a detailed answer on Quora for students preparing for GSoC. Here's my answer to "How can I apply and prepare for Google Summer of Code?" - hope this helps!
Edit: I'm mentoring for SymPy in GSoC 2021.
:wq for now.
Naman Gera (namannimmo)
Email | GSoC blog | StackOverflow | Quora