Mirror of forum.nim-lang.org

5005 :: Simulation: FMI standard (FMU files)

• mantielero (orginal) [2019-07-11T22:19:53+02:00] view original

  I think Nim could shine for simulations.

  People use Simulink or OpenModelica with Modelica language. There are many softwares that can export those physical models as FMU files which follow the FMI standard.

  FMU are .zip files which contain libraries (.dll or .so) and some .xml description files. As far as I know, neither Python nor Julia can export those models easily to FMU files. Julia has the problem that the runtime is pretty big, so even when done manually, it requires more than 100Mb files.

  I am trying to do this in Nim. To achive it I would like to use: fmusdk <https://github.com/qtronic/fmusdk>. As an example, I would like to port the following example: bouncingBall. But I am not sure about what should I do.

  Looking at bouncingBall.c, it follows the following structure::

  #include "fmuTemplate.h"
  /*
   ...
  define a few C functions which represent the model
  ...
  */
  #include "fmuTemplate.c"
  Run

  Which one of the following approaches would be better?

  1. Implementing the model in Nim, export it into C moking above structure. Is this feasible?
  2. Creating bindings for fmuTemplate.h and calling it from a bouncingBall.nim? This is what I would do, but I don't what to do with that #include "fmuTemplate.c" at the end of the file. I don't feel confortable with C language.

  Regards

---

• juancarlospaco (orginal) [2019-07-12T01:08:23+02:00] view original

  I have no idea, but very interesting project, keep it up! :)

---

• rayman22201 (orginal) [2019-07-12T02:02:05+02:00] view original

  It seems like you could just use nimterop or c2nim to create nim bindings for the .h and .c file.

  https://github.com/nimterop/nimterop

  Also, ping @shashlick, the author of that library.

---

• mantielero (orginal) [2019-10-24T22:54:09+02:00] view original

  My first little success doing this. It is something probably very easy for you guys, but I have struggle A LOT (no pro-dev speaking).

  First I create bindings to fmuTemplate.h. I did so with a wrapper similar to the following:

  import os
  import nimterop/cimport
  
  const
    base = currentSourcePath.parentDir()
  
  static:
    cDebug()
  cDefine("DISABLE_PREFIX")
  
  cIncludeDir(base/"src/shared/include")
  cIncludeDir(base/"src/shared")
  cIncludeDir(base/"src")
  cImport(base/"src/fmuTemplate.h", recurse=true)
  Run

  I did:

  $ nim c wrapper.nim > fmutemplate.nim

  I had to manually modify fmutemplate.nim as suggested here.

  Then I ported a model. As an example I ported the inc model. I convert the file inc.c into:

  # nim c --nimcache:.cache --app:lib -o:inc.so inc.nim
  import fmuTemplate
  import strformat
  
  # Porting inc.c (a particular model)
  const
     MODEL_IDENTIFIER="inc"
     MODEL_GUID="{8c4e810f-3df3-4a00-8276-176fa3c9f008}"
     
     #define model size
     NUMBER_OF_REALS= 0
     NUMBER_OF_INTEGERS=1
     NUMBER_OF_BOOLEANS=0
     NUMBER_OF_STRINGS=0
     NUMBER_OF_STATES=0
     NUMBER_OF_EVENT_INDICATORS=0
  
  const
     counter = 0
  
  proc setStartValues(comp:ptr ModelInstance) {.exportc: "$1".} =
      comp.i[counter] = 1
  
  proc calculateValues(comp:ptr ModelInstance) {.exportc: "$1".}=
      if comp.state == modelInitializationMode:
          # set first time event
          comp.eventInfo.nextEventTimeDefined = fmi2True
          comp.eventInfo.nextEventTime        = 1 + comp.time
  
  proc eventUpdate(comp:ptr ModelInstance, eventInfo:ptr fmi2EventInfo, timeEvent:cint, isNewEventIteration:cint) {.exportc: "$1".} =
      if timeEvent != 0:
          comp.i[counter] += 1;
          if comp.i[counter] == 13:
              eventInfo.terminateSimulation  = fmi2True
              eventInfo.nextEventTimeDefined = fmi2False
          else:
              eventInfo.nextEventTimeDefined = fmi2True
              eventInfo.nextEventTime        = 1 + comp.time
  
  {.passC: "-Ifmi/shared/include -w -fmax-errors=5".}
  
  {.passC: "-DMODEL_IDENTIFIER=\\\"" & MODEL_IDENTIFIER & "\\\"".}
  {.passC: "-DMODEL_GUID=\\\"" & MODEL_GUID & "\\\"".}
  {.passC: "-DNUMBER_OF_REALS=" & $NUMBER_OF_REALS .}
  {.passC: "-DNUMBER_OF_INTEGERS=" & $NUMBER_OF_INTEGERS .}
  {.passC: "-DNUMBER_OF_BOOLEANS=" & $NUMBER_OF_BOOLEANS .}
  {.passC: "-DNUMBER_OF_STRINGS=" & $NUMBER_OF_STRINGS .}
  {.passC: "-DNUMBER_OF_STATES=" & $NUMBER_OF_STATES .}
  {.passC: "-DNUMBER_OF_EVENT_INDICATORS=" & $NUMBER_OF_EVENT_INDICATORS .}
  {.compile: "./src/fmuTemplate.c".}
  Run

  I manually modified added to fmuTemplate.c the line:

  include "fmuTemplate.h"
  Run

  I compiled the model into a library as follows:

  nim c --nimcache:.cache --app:lib -o:inc.so inc.nim

  This produced the file inc.so.

  Later, I used directly the fmusdk to create all the model's folder structure with all the files by doing:

  build_fmu me inc

  This creates a Model Exchange model for inc.

  I replaced the file inc.so created with fmusdk with the one created using inc.nim. The file is stored in fmu/binaries/linux64/inc.so.

  The whole folder structure is stored as:

  cd fmu
  zip -r ../inc.fmu ./

  The file inc.fmu is the model that can be integrated in OpenModelica, Dymola, Simulink, ...

  I have only tested it with FMUComplianceChecker and it looks good.

  As a proof of concept, I am happy.

  The way forward:

  • [ ] To create the folder structure in nim.
  • [ ] To create the modelDescription.xml file in nim.
  • [ ] To define the model in Modelica-like language using macros in order to build the models. Something like what is being done with Modia in Julia.

  I don't think I will be capable of doing the last bullet, but I will try.

---