Add combine() and split() functions for transfer matrices
#1073
Conversation
There was a problem hiding this comment.
Thanks for the contribution! This will make a nice addition.
Lots of little suggestions herein, mainly to keep things consistent with other code in the package.
One bigger picture question is on the function names: since these only work for transfer functions, I wonder if the function names shouldn't be combine_tf and split_tf (using 'tf' rather than transfer_function since the factor function is tf).
Also, I think there needs to be some checking for size consistency. For example, consider the following code:
import control as ct
sys1 = ct.tf(ct.rss(2, 1, 2))
sys2 = ct.tf(ct.rss(2, 2, 1))
sys = ct.combine([[sys1, sys2]])
The first system has 1 output and the second system has two outputs, so I think this should generate an error. Instead, I get a 3 input (correct) by 1 output (incorrect) system. If instead I do
sys = ct.combine([[sys2, sys1]])
I get an error message:
587 for col in row:
588 for j_in in range(col.ninputs):
--> 589 num_row.append(col.num[j_out][j_in])
590 den_row.append(col.den[j_out][j_in])
591 num.append(num_row)
IndexError: list index out of range
Shouldn't both forms generate error messages (perhaps saying that there are an inconsistent number of outputs in row 1)?
| try: | ||
| for row in tf_array: | ||
| for tfn in row: | ||
| dt_list.append(getattr(tfn, "dt", None)) |
There was a problem hiding this comment.
dt will always exist for an InputOutputSystem, so this can just be tfn.dt.
There was a problem hiding this comment.
The combine_tf() function allows NumPy arrays to be input as well, so you can do something like
control.combine_tf([
[np.eye(2), np.zeros((2, 1))],
[np.zeros((1, 2)), control.TransferFunction([1], [1, 0])],
])So in that case, the NumPy arrays do not have a dt attribute. The dt will be taken from the TransferFunction object in this case.
I've found this useful when creating weighting functions for optimal control problems.
|
Thanks for the feedback. Adding I've also fixed the dimension checking and added a couple unit tests for that case I missed. |
RE #325
I've run into a few times where it would be convenient to construct MIMO
TransferMatrixobjects from SISO ones, or the reverse. In particular, this came up as I've been working on a D-K iteration package,dkpy. The specific implementation here is from utilities.py indkpy.I saw there was a related issue (#325 ) and people seemed open to it. Specifically, I've added
control.combine()that takes anArrayLikeofTransferFunctionobjects, scalars, and NumPy arrays, and creates a MIMO transfer matrix, andcontrol.split()that takes a MIMOTransferFunctionand returns a NumPy array with SISOTransferFunctionobjectss.I'm new to the
python-controlcodebase, so I'm not 100% sure if I've placed things in the correct files, but I've added unit tests and would appreciate some feedback.