Matlab interface Draft

[Pimss]

Just finished developing a draft version of automatic matlab interface. The interface is generic and can be used by pathsim

...
mw = MatlabWrapper()
mw.import_func('sq.m')
mcallab = mw.get_mat_callable()
Matfunction = Function(mcallab)
...

This is rather not what I had first in mind but it therefore can be used in Wrapper and Function class.
Two class are used MatlabWrapper to provide a callable with added features. And MatlabParser to parse .m files. It leverages the matlab engine class allowing calling matlab function from python

It parses .m files as a way to know input and output sizes and add path to matlab. I have limited the use to only float and 1D inputs and outputs (vectors will most likely raise an error).

This version uses an old version of matlab and I wish to use a more recent version

Here is a sample of code and associated outputs

sq.m file:

function [y] = sq(x)
y=x*x;
end

pathsim_matlab_test.py file:

import numpy as np
import matplotlib.pyplot as plt
from matlab_wrapper import MatlabWrapper

from pathsim import Simulation, Connection
from pathsim.blocks import Source, Scope, Function
from pathsim.solvers import RKCK54


def f_time(t):
    return t

time=Source(f_time)
sco = Scope(labels=['t', "t**2 matlab"])

mw = MatlabWrapper()
mw.import_func('sq.m')
mcallab = mw.get_mat_callable()
Matfunction = Function(mcallab)

blocks = [time,Matfunction, sco]

connections = [
    Connection(time,Matfunction,sco[0]),
    Connection(Matfunction,sco[1])]

#simulation initialization
Sim = Simulation(blocks, connections, Solver=RKCK54)


# Run Example ===========================================================================

if __name__ == "__main__":

    #run the simulation for some time
    Sim.run(5)
    sco.plot(lw=2)
    plt.show()

[Pimss]

Integration in pathsim would be tedious as is requires a matlab install and an env with matlab engine. Not sure how to proceed...
unittest are running on my side but I have pretty low confidence in the possibility that those tests will be successful on GitHub CI

[Pimss]

Performances are seriously impacted. There is a way to do async call, haven't look much into it yet

[milanofthe]

Thats cool! Are you familiar with FMI/FMU? I was planning on adding FMI support to PathSim via rumoca, haven't found the time yet. This would enable model exchange with other simulation tools and also simulink.

[milanofthe]

Thats how I would like to add cross compatibility to PathSim.

[Pimss]

I haven't developed any workflow using FMI/FMU. But some friends of mine have. I might take a look soon as a way to use simulink / Dymola / OpenModelica models within Python. I don't have the experience on that subject that I have with Matlab but I know I need to find some time to dig into that.

[Pimss]

After taking some time to look around for fmu/fmi tools, fmipy looks like a good solution.

• First step would be to parse all inputs and outputs.
• Create a WrapperFmui (not inherited from Wrapper class ?) with attribute input_names and outputs_names and match them with parsed data (hopefully all parameters have default values ? Not sure)
• Get pathsim inputs and pass it to fmu (given input_names)
• Simulate fmu step by step by passing pathsim time step to WrapperFmui (ensure no backward time is sent to fmu/fmi ?)
• Get fmu outputs and match it to pathsim (given output_names)
• ...
• Profit ?
  I might look that up. Not sure when

[Pimss]

That would allow any user to select specific inputs and outputs to be used in pathsim. Fmu files can have large numbers of input parameters and output results

[milanofthe]

That would be for co-simulation, similar to the the Wrapper block. I think we should go with rumoca for the FMI integration. They have already proven to work with collimator (which is similar to PathSim but no longer supported).

Id like to integrate FMI more holistically to enable export of blocks, subsystems and whole simulations as FMUs. And also importing FMUs with two modes, co-simulation (thats what you are refering to, where the FMUs maintain their own internal solvers and synchronize with PathSim at predefined time intervals, similar to the Wrapper) and continuous simulation where the internal ODE / DAE is exposed to the PathSim solvers.