Free and open-source control system modeling software are first step to design and a good method to learn foundations in software design. Such software may not have complete reliability, and usability compared to commercial software (with AMC), but these are outcome of first year course in computer engineering.

COMSOL is excellent in electromagnetic FEA, Octave in musical mathematics, LabView in interactive software utilization, SIMULINK in block design approach, Python in general purpose terse dynamic machine readable programming, Wolfram Mathematica in mathematically intensive higher order discrete differential equations, and OpenModelica for modeling, simulating, optimizing and analyzing complex dynamic systems.

It makes sense that OpenModelica (with all advanced features) have not gained popularity despite that it is open-source.

As Kumar suggests, the students will learn different things from different platforms. They are not really competitors, but rather complement each other. Identifying the best one, is a lost cause.

Obviously, you forgot Haskell on your list.  Why?

SciLab should also be mentioned. While Octave strive to be matlab compatible, thus copying the flaws along with any advantages, SciLab is a more modern design.

Modelica is an equation-based modelling language, and will thus support a correct mathematical understanding of most physical and mechanical systems. It is not the language to implement control algorithms, but excellent to simulate the system being controlled. Compared to the imperative languages, the models will be closer to mathematical textbook descriptions. Note that OpenModelica is not the most mature implementation of Modelica. Dymola (commercial) seems to be a better choice if you can afford it.

Haskell var ikke populært i mine dager på skolen. Jeg gjorde ikke glemme det; MATLAB har overvunnet begrensninger og er ganske populært i disse dager. Begrensning kan være at det er mer plasskrevende, mens du kjører nøyaktig samme ting som Dymola. Usability, interaktivitet, og pålitelighet er fortsatt primære bekymring som deles av flere kontrollingeniører.

I suggest 3 questions. How elegant is the language? ("Elegant" can be a multi-modal construct). What purpose do you require (e.g. real-time)? What extras do you gain beyond just the language? I suppose you could be crass and ask about commercial interests too.

On elegance alone and links to C (still vitally important though not mentioned by anyone, and able to represent the mathematics faithfully) I am enjoying new languages like Julia and Rust. Having discussed Haskell though never used it, my concern is that you learn a lot about programming but maybe get distracted from the control, and I am not certain about the control purposes it satisfies (others may comment with more expertise).

For particular purpose, Labview (block-diagram "programming") and Modelica have strong claims. I have been impressed by Labview-based robotics competitions.

For extras though, MATLAB/SIMULINK is peerless, though SCILAB is priced right! What the language lacks in elegance it gains in ease of learning.

I would go first for the easy of learning a language and implementing the solutions. I would also go for the integration of the platform/language you would use. I am claiming this two points because I would preserve the students from distracting activities, which are (i) learning a rather difficult and complex language together with the user interface of the compiler and of project management, (ii) having to work with different tools to accomplish the final goal (e.g., having to rely on a distinct graph tool to get the diagrams of simulated variables produced by a simulation tool which lacks any graphical features).

For these reasons I would neglect all the real programming languages suggested above and go for an easier software development tool which integrates all the features necessary to carry out the control simulation activities. This is the case of Matlab and possibly Simulink. They are NOT perfect, they are rather simplified tools whose numerical algorithms are decorous and rather good for didactic purposes but not for real research activities where serious integration routines (developed either in Fortran or in C++) can solve quite challenging problems (e.g., stiff ODE system) with a more robust approach and in a fraction of CPU time taken by Matlab. However, these last two point are minor for didactic purposes.

MATLAB and Simulink are good platforms to develop control system models and quite easy to learn. There are several inbuilt functions in MATLAB that comes handly while studying control engineering.

You make a very good point, Davide. If the students know or need no other programming language or environment, then Matlab is the easier option, minimising the extraneous cognitive load.

Once you start on comlex, composite systems, Matlab is not an easy option, and much of what you learn in Matlab is hardly transferrable to other environments. Students studying control theory, are likely to need to program embedded systems, interface with embedded systems, and interface with models from other tools. For one, this is hard to do in Matlab. Secondly, it requires an understanding of software engineering which require study of more general-purpose environments. Hence, in the context of a degree programme, Matlab may no longer have the advantages it has in an isolated module.

Clearly, Matlab remains the easier choice for the teacher, who can copy ready-made examples and exercises which have been tried and tested already. Moving to a less wide-spread platform imposes a start-up cost.

However, if you are concerned that Matlab is not fast enough or accurate enough, I very much doubt that you have found the best way to do it in Matlab.

Wow, thanks to everyone for their thoughtful comments! I only just now saw these comments. 

I’m also wondering if Labview can be a good platform to teach control engineering cours.