Algebraic Loops over FMUs failing

[AnHeuermann]

Description

I have a non-linear algebraic loop with three variables and equations

  a   + b   + c    = 0;
  2*a - 3*b + 2*c  = 9;
  a*a + b*b + c*c  = 5;

with Solution a≈0.635425, b≈-1.8, c≈1.16458 or a≈1.16458, b≈-1.8, c≈0.635425.
When splitting the loop into two FMUs OMSimulator detects the loop, but can't solve it.

Steps to reproduce the behavior

I used this mos script to test:

loadModel(Modelica);

loadString("model loopOverFMUsA
  Real a, b, c;
  Modelica.Blocks.Interfaces.RealOutput a_out;
  Modelica.Blocks.Interfaces.RealOutput b_out;
  Modelica.Blocks.Interfaces.RealInput c_in;
equation
  c = c_in;
  // Eq. 1 and 2 of loop
  a + b + c=0;
  2*a - 3*b + 2*c=9;
  a_out = a;
  b_out = b;
end loopOverFMUsA;
"); getErrorString();

loadString("model loopOverFMUsB
  Real a, b, c;
  Modelica.Blocks.Interfaces.RealInput a_in;
  Modelica.Blocks.Interfaces.RealInput b_in;
  Modelica.Blocks.Interfaces.RealOutput c_out;
equation
  a = a_in;
  b = b_in;
  // Equation 3 of loop
  a*a + b*b + c*c=5;
  c_out = c;
end loopOverFMUsB;
"); getErrorString();

buildModelFMU(loopOverFMUsA, fmuType="me"); getErrorString();
buildModelFMU(loopOverFMUsB, fmuType="me"); getErrorString();

writeFile("loopsOverFMUs.lua","
oms_setTempDirectory(\"./temp/\")
oms_newModel(\"loopOverFMUs\")
oms_addSystem(\"loopOverFMUs.root\", oms_system_sc)

-- instantiate FMUs
oms_addSubModel(\"loopOverFMUs.root.A\", \"loopOverFMUsA.fmu\")
oms_addSubModel(\"loopOverFMUs.root.B\", \"loopOverFMUsB.fmu\")

-- connections: A -> B
oms_addConnection(\"loopOverFMUs.root.A.a_out\", \"loopOverFMUs.root.B.a_in\")
oms_addConnection(\"loopOverFMUs.root.A.b_out\", \"loopOverFMUs.root.B.b_in\")
oms_addConnection(\"loopOverFMUs.root.A.c_in\", \"loopOverFMUs.root.B.c_out\")

-- simulation settings
oms_setResultFile(\"loopOverFMUs\", \"loopOverFMUs.csv\")
oms_setStartTime(\"loopOverFMUs\", 0)
oms_setStopTime(\"loopOverFMUs\", 0.1)

oms_instantiate(\"loopOverFMUs\")
--oms_setReal(\"loopOverFMUs.root.system1.x_start\", 2.5)

oms_initialize(\"loopOverFMUs\")
oms_simulate(\"loopOverFMUs\")
oms_terminate(\"loopOverFMUs\")
oms_delete(\"loopOverFMUs\")
"); getErrorString();

system("OMSimulator loopsOverFMUs.lua"); getErrorString();

Output:

$ omc loopsOverFMUs.mos
 omc loopsOverFMUs.mos
true
true
""
true
""
"/mnt/d/workspace/EMISIYS/Demonstratores/myTest/loopOverFMUsA.fmu"
""
"/mnt/d/workspace/EMISIYS/Demonstratores/myTest/loopOverFMUsB.fmu"
""
true
""
info:    Set temp directory to    "/mnt/d/workspace/EMISIYS/Demonstratores/myTest"
info:    Set working directory to "/mnt/d/workspace/EMISIYS/Demonstratores/myTest"
info:    Set temp directory to    "/mnt/d/workspace/EMISIYS/Demonstratores/myTest/temp"
info:    New model "loopOverFMUs" with corresponding temp directory "/mnt/d/workspace/EMISIYS/Demonstratores/myTest/temp/loopOverFMUs-qwaomow6"
error:   [fmiLogger] module FMI2XML: XML element 'ScalarVariable': could not parse value for enumeration attribute 'causality'='none'
error:   [fmiLogger] module FMI2XML: XML element 'ScalarVariable': could not parse value for enumeration attribute 'causality'='none'
info:    model doesn't contain any continuous state
info:    Result file: loopOverFMUs.csv (bufferSize=1)
warning: Alg. loop (size 3)
error:   [solveAlgLoop] max. number of iterations (10) exceeded at time = 0.100000
Segmentation fault
139
""

In addition to the failing algebraic loop OMSimulator fails to unload all data and has a segmentation fault. Output from gdb:

Program received signal SIGSEGV, Segmentation fault.
__GI___libc_free (mem=0x7ff8000000000000) at malloc.c:3103
3103    malloc.c: No such file or directory.
(gdb) bt
#0  __GI___libc_free (mem=0x7ff8000000000000) at malloc.c:3103
#1  0x00007ffff3a3d185 in deInitializeDataStruc (data=0x9f9620) at simulation/solver/model_help.c:1166
#2  0x00007ffff3a325eb in fmi2FreeInstance () from /mnt/d/workspace/EMISIYS/Demonstratores/myTest/temp/loopOverFMUs-906c32y3/temp/0002_B/binaries/linux64/loopOverFMUsB.so
#3  0x00000000005e7e55 in fmi2_capi_free_instance (fmu=0x9f9250) at /mnt/d/ubuntu/OpenModelica/OMSimulator/3rdParty/FMIL/src/CAPI/src/FMI2/fmi2_capi.c:402
#4  0x00000000005779fd in oms::ComponentFMUME::terminate (this=0x9f2650) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/ComponentFMUME.cpp:525
#5  0x000000000051d3ca in oms::SystemSC::terminate (this=0x9e7230) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/SystemSC.cpp:334
#6  0x000000000059c08e in oms::Model::terminate (this=0x9e6ff0) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/Model.cpp:599
#7  0x00000000004bfc8f in oms_terminate (cref_=<optimized out>) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/OMSimulator.cpp:1123
#8  0x00000000004a0167 in OMSimulatorLua_oms_terminate (L=0x9dc688) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/../OMSimulatorLua/OMSimulatorLua.c:936
#9  0x0000000000630bf6 in luaD_precall (L=0x9dc688, func=0x9dbdb0, nresults=0) at ldo.c:434
#10 0x000000000063c4cd in luaV_execute (L=0x9dc688) at lvm.c:1134
#11 0x0000000000630e54 in luaD_call (L=0x9dc688, func=<optimized out>, nResults=<optimized out>) at ldo.c:499
#12 luaD_callnoyield (L=0x9dc688, func=<optimized out>, nResults=<optimized out>) at ldo.c:509
#13 0x000000000063008e in luaD_rawrunprotected (L=0x9dc688, f=0x62e5b0 <f_call>, ud=0x7fffffffd910) at ldo.c:142
#14 0x000000000063142b in luaD_pcall (L=0x9dc688, func=0x0, u=0x9f0bf0, old_top=16, ef=140737281170361) at ldo.c:729
#15 0x000000000062e520 in lua_pcallk (L=0x9dc688, nargs=<optimized out>, nresults=0, errfunc=<optimized out>, ctx=<optimized out>, k=0x9f8e00) at lapi.c:968
#16 0x00000000004bb0d5 in oms_RunFile (filename_=<optimized out>) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/OMSimulator.cpp:1101
#17 0x0000000000595578 in oms::Flags::SetCommandLineOption (cmd=...) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/Flags.cpp:165
#18 0x00000000004a5b4c in oms_setCommandLineOption (cmd=0x9d8be0 "loopsOverFMUs.lua") at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulatorLib/OMSimulator.cpp:373
#19 0x000000000049caee in main (argc=<optimized out>, argv=0x7fffffffdd98) at /mnt/d/ubuntu/OpenModelica/OMSimulator/src/OMSimulator/main.cpp:45

EDIT: This segmentation fault is caused by the FMU exporting tool OpenModelica.

Expected behavior

Version and OS

• Version: OMSimulator v2.1.0-dev-152-g63ee20c-linux
• OS: Ubuntu 18.04 on WSL2 on Windows 10

[AnHeuermann]

I tested again with a newer version of OpenModelica and OMSimulator (was using an old version for the first output).

The problem is the iteration process for the non-linear loop between FMUs A and B.

The generated code for the FMUs is looking good:
FMU A:

c := c_in;
b := 3/5*c + 0.4*c -1.8 -c; [ = -1.8]
a := -b -c;
b_out := b;
a_out := a;

FMU B:

a := a_in;
b := b_in;
c := abs(sqrt(5-a*a + b*b));
c_out := c;

The newest problem:
The iteration process is going into the wrong direction. It tries with some values for a and b with 5-(a*a+b*b)<=0 what will result in a assert.

warning: Alg. loop (size 3)
warning: B (logStatusWarning): The following assertion has been violated during initialization at time 0.000000
assert            | debug   | Model error: Argument of sqrt(5.0 - (a_in ^ 2.0 + b_in ^ 2.0)) was -1.48 should be >= 0
getBestJumpBuffer got mmc_jumper=(nil), globalJumpBuffer=(nil)
Aborted
134

TODOs:

• Add (basic) handling of asserts inside algebraic loops.
• Report which variables are inside the loop, so users can choose appropriate start values.
• Add user flag to choose max. number of iterations. (Adding user flag to set maximum number of loop iterations #811)
• Add better non-linear solvers for loops between components. (Better non-linear solver for algebraic loops between components. #812)

[sjoelund]

It shouldn't matter that it's going in the wrong direction. Fix the generated FMU; we should never call abort() inside of an FMU :)

[AnHeuermann]

Good point. But I'm not sure what the best approach would be for that. I've created a ticket for this to continue the discussion there: https://trac.openmodelica.org/OpenModelica/ticket/6125#ticket

[AnHeuermann]

Related PR: OpenModelica/OpenModelica#6757

[lochel]

@AnHeuermann What is the status?

[AnHeuermann]

I should re-run my example. But it is not really a good one.
We still have asserts and are not treating the returned error codes in a good way. Ideally it should be possible for OMSimulator to solve this loop, even though the initial iteration will trigger the asserts because 5-a*a + b*b<0.