I wonder if there's some experience out there, how much performance can be gained from using the Simulink Coder to generate models over running the model in (rapid) accelerator mode?
For instance, how does it compare to the quite dramatic improvements from not-accelerated to accelerated models?
I know this is kind of vague and probably depends on the specifics of a model, but I'd be happy with examplary numbers.
I don't know if you will really find speed improvement anecdotes that helpful, as your results are going to be highly dependent on your own implementation.
That said, the following reference on How Acceleration Modes Work might be useful to you.
As described in this link...
The Accelerator mode generates and links code into a
C-MEX S-function. Simulink uses this acceleration target code to
perform the simulation...
...the model methods are separate from the Simulink
software and are part of the Acceleration target code. A C-MEX
S-function API communicates with the Simulink software, and a MEX API
communicates with MATLAB. The target code executes in the same process
as MATLAB and Simulink.
The link goes on to describe Rapid Acceleration, as such:
The Rapid Accelerator mode creates a Rapid Accelerator standalone
executable from your model. This executable includes the solver and
model methods, but it resides outside of MATLAB and Simulink. It uses
External mode (see Host/Target Communication) to communicate with
Simulink.
(Note that the actual link has some nice diagrams that help to illustrate this architecture).
I think that there are a few important things to takeaway from these descriptions. First, in Accelerator Mode, code is only generated for the model methods; however, in Rapid Accelerator Mode, code is also generated for the solver. In either case there will be some overhead due to communication between the generated code and Simulink (though in general that overhead would be greater in non-rapid accelerator mode).
Now, whether you are going to see significant differences between these acceleration modes and full-on code generation will most likely depend on how you would be using the generated code. You had stated that your application is intended to essentially analyze/visualize the outputs of simulink models. Do you intend for you application, itself, to be a Simulink model? If so, I imagine your workflow would be that you generate code for the "plug-in" model and then incorporate this into your "parent" model as an s-function. Based on the descriptions given above, my gut feeling is that you would not see much of a performance improvement compared to accelerator mode.
However, if your application is to be implemented outside of Simulink, then code generation would definitely be the way to go. In this case, you've completely eliminated the overhead of communicating with Simulink. In this workflow, you could, for example, generate a stand-alone executable from the "plug-in" model, execute that model to get the outputs, and then simply load those outputs for analysis and visualization into your application. Your application could be written in any language you want. In fact, you could even still write your application in MATLAB -- the key point, though, being that your application runs once the stand-alone executable (generated from the "plug-in" model) has finished executing, rather than interfacing with the plug-in model while it is executing.
Related
we want to publish an Open-Source for integrating Reinforcement Learning to Smartgrid optimization.
We use OpenModelica as GUI, PyFMI for the import to Python and Gym.
Nearly everything is running, but a possibility to connect or disconnect additional loads during the simulation is missing. Everything we can do for now is a variation of the parameters of existing loads, which gives some flexibility, but way less then the possibility to switch loads on and off.
Using the implemented switches in OpenModelica is not really an option. They just place a resistor at this spot, giving it either a very low or very high resistance. First, its not really decoupled, and second, high resistances make the ODE-system stiff, which makes it really hard (and costly) to solve it. In tests our LSODA solver (in stiff cases basically a BDF) ran often in numerical errors, regardless of how the jacobian was calculated (analytically by directional derivatives or with finite differences).
Has anyone an idea how we can implement a real "switching effect"?
Best regards,
Henrik
Ideal connection and disconnection of components during simulation
requires structure variability, which is not fully supported
by Modelica (yet). See also this answer https://stackoverflow.com/a/30487641/8725275
One solution for this problem is to translate all possible
model structures in advance and switch the simulation model if certain conditions are met. As there is some overhead involved, this approach only makes sense, when the model is not switched very often.
There is a python framework, which was built to support this process: DySMo. The tool was written by Alexandra Mehlhase, who made a lot of interesting publications regarding structure variability, e.g. An example of beneficial use of
variable-structure modeling to enhance an existing rocket model.
The paper Simulating a Variable-structure Model of an Electric Vehicle for Battery Life Estimation Using Modelica/Dymola and Python of Moritz Stueber is also worth a look. It contains a nice introduction about variable structure systems and available solutions.
I was wondering if it is possible to use simlink's PID architecture in matlab to actually control hardware - rather than modelling it.
I have been playing around a little with the PID tuning in matlab - and have worked out the correct gains for my system (or at least good enough).
Now I want to implement the actual control loop in matlab using the real hardware. In this case I am tying to stabilise a laser using a measurement from a wavemeter - although from the point of view of matlab essentially I have two functions - one which returns the current wavelength, and another which alters the wavelength with an input from 0-100.
How do I get the fancy PID objects to work in a loop where for example I input data each loop, and get the required output to stabilise things given my gains - or should I just hardcode the equations in?
I have tried looking through some of the simlink examples - but there doesn't seem to me much actual implementation - mostly just modelling.
Thanks in advance for any advice.
MATLAB (without Simulink, or with Simulink in purely a simulation mode) is not a real-time environment. Trying to use it to control hardware in real-time won't work unless you have very slow sample rates.
If you do have slow sample rates, then you'd connect to your hardware to MATLAB using device drivers such as those in the Data Acquisition Toolbox
You haven't found any info on real-time implementation in the Simulink documentation because from Simulink the mechanism for real-time implementation is via Simulink Coder. You would need to use it in conjunction with a real-time environment such as Simulink Desktop Real-Time (if your sample rates are relatively slow), Simulink Real-Time, or one of many other 3rd party real-time form factors.
I've read that I can generate code from Simulink models/block diagrams. I am curious whether Simulink always converts a model to (c/c++/java) code prior to running a simulation in the Simulink software, and then execute that code? I mean, whenever I'm running a simulation is Matlab converting the block diagram to (c/c++/java) code and running that code behind the scene. In this case, simulation in Simulink directly depends on running some code; this information is important to me in some way.
Generating and running code for a complete model seems plausible, as we can write s-functions using C/Matlab code and use them as custom blocks. So simulating a model involves running code in some degree. Again, since we can write Matlab code as well, simulation may involve interpreting Matlab code in some environment. It makes me curious whether these information are available - how tightly running a simulation in Simulink depends on executing native code in user's machine.
I did some search before posting and found this SO question: How does simulation engine work? Discussion in this question does not answer my question directly.
The answer depens on which mode you chose.
In the normal mode Simulink will run the model as it is primarily using the MATLAB execution engine. No code is generated. Native implemented parts (e.g. S-Functions) are used as individual binaries called by the MATLAB interpreter.
In the accelerator mode Simulink generates model code. This means your full model (except parts where code generation is impossible) is generated and compiled into one binary.
In the rapid accelerator mode not only your model but also the solver is generated and compiled into one binary, now running in a separate process.
For more details refer to the official documentation
I have a large simulink model with many source and sink blocks, many with only elementary arithmetic operations in between. I have been asked to document the equations behind the model. I am currently doing this manually and I am finding it rather slow and there is a relatively high chance of errors in the process.
Is there any way for Simulink to generate the equations (in MATLAB syntax for example) automatically?
There is no utility in MATLAB/Simulink that can do exactly what you are looking for (and I personally don't know of any third-party tools that can do this, either).
However, I think that your best bet might be to make use of Simulink Coder. This will allow you to convert your Simulink model to C code. From that code, you may be able to extract the equivalent equations more easily than you can by analyzing the Simulink model by hand.
The catch, though, is that Simulink Coder is an add-on package to base Simulink, so you may or may not have this tool available to you.
As far as I know, stateflow and simulink are often used at the same time, and are both environments developed by MathWorks, who make Matlab. May I know what's the difference between them?
Simulink is largely a controls oriented solution. It graphically depicts math like products, sums, integrals, etc. However, it's conditional logic facility is lacking. Any kind of if construct quickly becomes terse and unmanageable in my opinion. I've seen many models, and there is a clear line here where Simulink should end and Stateflow should start.
Stateflow deals extremely well with logic and, of course, state machines. Now with the addition of the Simulink Function blocks within Stateflow, we have a powerful combination to allow the state machine in Stateflow direct the rest of the program.
As far as functionality goes, they are both functionally complete, meaning anything you can code in C, you can code in Simulink or Stateflow. However, I would not recommend coding a PID loop in Stateflow, but it's possible. You could also easily create a state machine in Simulink, but I'd advise against it.
As far as code generation; in the beginning of the meld, the Stateflow and Simulink had separate code generators that were sewed together with more Simulink generated C code at code generation time. Then came CGIR (Code Generation Intermediary Representation), which unifies the code generator between Stateflow and Simulink. It came around 2007, and has continued to deliver substantial increases in performance. Generated code has increased in on target performance to a point where companies can use the code in their embedded systems and actually get a performance benefit rather than take a small hit. Also, the time to generate the code has also decreased substantially. CGIR is a replacement for the Target Language Compiler, however, never fear, TLC API will still be available, perhaps forever for those who have developed massive libraries of proprietary code generation libraries.
Hope this helps, let me know if something sounds fishy or if I need to clarify.
Quoting a Stateflow Webinar:
Simulink is used to respond to continuous changes in dynamic changes.
Stateflow is used to respond to instantaneous changes in dynamic changes.
Real-world systems have to respond to both continuous and instantaneous changes.
Use both Simulink and Stateflow so that you can use the right tool for the right job. Examples: Suspensions dynamics of a car are modelled with Simulink whereas, the gear transmission is modelled with Stateflow.
Learn more about Stateflow in general at:
http://www.mathworks.com/products/stateflow/examples.html
Stateflow has been updated for making it very easy to create state machines and flow charts in R2012b.
The major updates include a new graphical editor, state transition tables, MATLAB as the action language and an integrated debugger. Find short videos for these features and how they can be used at:
http://www.mathworks.com/products/stateflow/whatsnew.html
best,
Siddharth
I'm also currently involved with both simulink and stateflow. Till now I did everything in Simulink but once you have to implement logic( if, case) your model becomes visually difficult to be analysed after coding. But I think theoretically you can do everything also just in Simulink (correct me if I'm wrong).
The answer of macduff explains very good the differences.
Regards,
GR
I have worked on both on simulink and state flow environment both are Matlab tool. When u want to design a reactive system or event driven system at that time we should prefer stateflow instead of simulink.
because we can control an event in stateflow in better way compare to simulink .though debugging is easy in simulink but readability and code generation are easy in stateflow.
Complex state machine: Stateflow
Complex mathematical process: Simulink blocks
Others: both of Stateflow and Simulink are fine
Another major benefit of Stateflow is the integration and usage of external C code. This can be done by selecting the Action Language C within the Chart settings.
As commonly known the user can include external Code within the "Simulation Target" options. While a normal Simulink model need small workarounds to access the external C code (ceval(), Simulink.Parameter for globals etc.), Stateflow can directly access functions, defines, etc.
Problems like array of buses/structs can also be avoided by coding these arrays in external code.
For a lot of use cases I love the flexibility to directly interact with C code within Stateflow.