Das Functional Mockup Interface
zum Austausch Dynamischer Modelle
Martin Otter (DLR-RM)
Torsten Blochwitz (ITI)
Hilding Elmqvist (Dassault Systèmes – Dynasim)
Andreas Junghanns (QTronic)
Jakob Mauss (QTronic)
Hans Olsson (Dassault Systèmes – Dynasim)
ASIM Workshop, 4. - 5. März 2010
Contents
1. Functional Mockup Interface (FMI) – Overview
2. FMI - Motivation
3. FMI for Model Exchange – Overview
4. Model Distribution
5. Model Description Schema
6. Model C-Interface
7. Tool Support for FMI
8. Comparison with SIMULINK S-Function Interface
9. Outlook
10. Acknowledgements
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 2
1. Functional Mock-up Interface (FMI) – Overview
The FMI development is part of the ITEA2 MODELISAR project.
FMI development initiated, organized and headed by Daimler AG
Improved Software/Model/Hardware-in-the-Loop Simulation,
of physical models and of AUTOSAR controller models
from different vendors for automotive applications with
different levels of detail.
Open Standard
14 automotive use cases for evaluation in MODELISAR
etc.
Engine
with ECU
Gearbox
with ECU
Thermal
systems
Automated
Chassis components,
cargo door roadway, ECU (e.g. ESP)
functional mockup interface for model exchange and tool coupling
courtesy Daimler
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 3
Task is complex since the different parts are complex by themselves:
Model Exchange (ODE/DAE components without integrators)
Co-Simulation (ODE/DAE components with integrators)
Co-Simulation with PDE solver (MpCCI)
AUTOSAR (discrete components with complex communication)
Simulation Backplane
In Jan. 2010, the first version of "FMI for Model Exchange" was released.
It was mainly developed by Dassault Systèmes (Dynasim), DLR, ITI, QTronic.
www.functional-mockup-interface.org
specification
xml schema files
C header files
software development kit (QTronic)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 4
MODELISAR
ITEA2 project
3 years (2008 – 2011)
29 project partners
Coordinators
Dassault
Systèmes
Daimler AG
Budget / Funding
30M€ / 10M€
Funded by
Germany (BMBF)
France (DGCIS)
Sweden (VINNOVA)
Belgium
Austria
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 5
2. FMI - Motivation
supplier1
Problems / Needs
supplier2
supplier3
supplier4
supplier5
Component development by supplier
Integration by OEM
OEM
Many different simulation tools
?
Solution
supplier1
supplier2 supplier3
supplier4
supplier5
tool 4
tool 5
Reuse of supplier models by OEM:
DLL (model import) and/or
tool 1
tool 2
tool 3
Tool coupling (co-simulation)
Added Value
OEM
FMI
Protection of model IP of supplier
supplier1
!
OEM
Early validation of design
Increased process
efficiency and quality
slide from Nick Suyam, Daimler (adapted)
supplier2
supplier3
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 6
Example Scenario: Power Lift Gate
models solved with
SIMPACK integrators
(= central integration)
supplier1
supplier2
supplier3
AMESim
(Modelica)
Dymola
(Modelica)
SimulationX
(Modelica)
model export
FMI model exchange (DLL)
model import
OEM, dept. 1
Simpack (multi-body)
OEM, dept. 2
TargetLink (software)
model export
FMI tool coupling
FMI model exchange (C)
model import
co-simulation
Silver (system model) + TestWeaver (testing)
slide from Nick Suyam, Daimler (adapted)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 7
3. FMI for Model Exchange - Overview
Import and export of input/output blocks (FMU – Functional Mock-up
Unit)
described by
differential-, algebraic-, discrete equations,
with time-, state, and step-events
An FMU can be large (e.g. 100000 variables)
An FMU can be used in an embedded system (small overhead)
FMUs can be connected together
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 8
Signals of an FMU:
For example: 10 input/output signals (u/y) for connection and
100000 internal variables (v) for plotting
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 9
Mathematical description of an FMU:
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 10
4. Model Distribution
A model is distributed as one zip-file with extension ".fmu". Content:
XML model description file
All model information that is not needed during integration of model,
e.g., signal names and attributes. Advantage:
No overhead for model execution.
Tools can read this information (= complicated data structure)
with their prefered language (C++, C#, Java, ...)
Model equations defined by a small set of C-functions. In zip-file:
C source code and/or
Binary code (DLL) for one or more platforms (Windows, Linux, ...)
Resources
Documentation (html files)
Model icon (bitmap file)
Maps and tables (read by model during initialization)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 11
Structure of an FMU zip-file
modelDescription.xml
model.png
documentation
// Description of model (required file)
// Optional image file of model icon
// Optional directory containing the model
// documentation
_main.html
// Entry point of the documentation
<other documentation files>
sources
// Optional directory containing all C-sources
// all needed C-sources and C-header files to compile and link the model
// with exception of: fmiModelTypes.h and fmiModelFunctions.h
binaries
// Optional directory containing the binaries
win32 // Optional binaries for 32-bit Windows
<modelIdentifier>.dll
// DLL of the model interface implementation
VisualStudio8
// Microsoft Visual Studio 8 (2005)
<modelIdentifier>.lib
// Binary libraries
gcc3.1
// Binaries for gcc 3.1.
win64
// Optional binaries for 64-bit Windows
...
linux32 // Optional binaries for 32-bit Linux
...
resources // Optional resources needed by the model
< data in model specific files which will be read during initialization >
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 12
control
simulator
GUI
reads
XML schema (.xsd)
defined by the
FMI specification
references
model.dll
run 1 or many
modelDescription.xml
.fmu
simulator
solver
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 13
5. Model Description Schema
All model information not needed for execution is stored in one xml-file
(modelVariables.xml in zip-file) defined by xml schema files.
Advantage:
Complex data structures give still simple interface, and tool can use its favorite
programming language for reading (e.g., C++, C#, Java).
Definition of display units
Definition of type defaults
Default stop time, tol. etc.
Tool specific data
Variable names and attributes
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 14
Model attributes. Most important
modelIndentifier is a C-name that is
used as prefix for the C-functions
(model interface)
guid is a globally unique identifier
("fingerprint" of all releveant information
in the xml file) that is also stored in the
C-functions to gurantee consisteny
Number of continuous states and
of event indicators; numbers are fixed
(meaning of states can change dynamically
during simulation)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 15
ModelVariables
data types
ordered set of scalar variables
(arrays, records, etc. must be
mapped to scalars when
generating code).
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 16
Attributes of ModelVariables
unique name
handle to identify
variable in C-functions
...
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 17
Data types allow to store all (relevant) Modelica attributes, including units.
Defaults from TypeDefinitions
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 18
Example
modelDescription.xml
<?xml version="1.0" encoding="UTF8"?>
<fmiModelDescription
fmiVersion="1.0"
modelName="Modelica.Mechanics.Rotational.Examples.Friction"
modelIdentifier="Modelica_Mechanics_Rotational_Examples_Friction"
guid="{8c4e810f-3df3-4a00-8276-176fa3c9f9e0}"
...
numberOfContinuousStates="6"
numberOfEventIndicators="34"/>
<UnitDefinitions>
<BaseUnit unit="rad">
<DisplayUnitDefinition displayUnit="deg" gain="57.2957795130823"/>
</BaseUnit>
</UnitDefinitions>
<TypeDefinitions>
<Type name="Modelica.SIunits.AngularVelocity">
<RealType quantity="AngularVelocity" unit="rad/s"/>
</Type>
</TypeDefinitions>
<ModelVariables>
<ScalarVariable
name="inertia1.J"
valueReference="16777217"
description="Moment of inertia"
variability="parameter">
<Real declaredType="Modelica.SIunits.Torque" start="1"/>
</ScalarVariable>
...
</ModelVariables>
</fmiModelDescription>
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 19
6. Model C-Interface
Two C-header files
Platform dependent definitions (basic types)
C-function interfaces
18 core functions
6 utility functions
no macros
C-function name: <ModelIdentifier>_<name>, z.B. "Drive_fmiSetTime"
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 20
Example:
// Set input arguments
fmiSetTime(m, time);
fmiSetReal(m, id_u1, u1, nu1);
fmiSetContinuousStates(m, x, nx);
// Get results
fmiGetContinuousStates(m, derx, nx);
fmiGetEventIndicators (m, z, nz);
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 21
Caching for efficient model evaluation
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 22
Clean Definition of State Events
A state event occurs when an event indicator z changes its domain from
z>0
to
z≤0
or
vice versa
Always well defined!
Usual definition z(ti)*z(ti-1) ≤ 0 is not always well-defined
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 23
7. Tool Support for FMI (planned for 2010 in MODELISAR)
7.1 FMU import and export
AMESim
(Modelica)
CATIA Systems
(CAD/Modelica)
Dymola 7.4
(Modelica)
already available
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 24
EXITE ACE
(co-simulation of
software/AUTOSAR)
SimulationX
(Modelica)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 25
7.2 FMU import
EXITE
(co-simulation)
Silver/TestWeaver
(FMU integration,
cosimulation, testing)
Simpack
(multi-body)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 26
8. Comparison with SIMULINK S-Function Interface
SIMULINK has two interfaces:
S-Function for offline simulation
Realtime Workshop for embedded Systems
FMI is targeted for both offline simulation and embedded systems (one interface)
Simulink and FMI have different goals and therefore have different solutions:
S-function targeted to import models in SIMULINK:
Main interface to implement an S-Functions.
(many macros to fill S-Function data structure)
Incomplete interface to call S-Functions.
Whenever a small detail of the S-Function data structure is changed,
existing DLLs must be newly build.
If used in 3 simulation environments → 3 DLLs of the same model
FMI targeted to import models in many simulation environments
Only interface to call model in simulation environment.
Model data structure is secret of model generation environment.
If used in 3 simulation environments on same platform → 1 DLL
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 27
S-function not suited for embedded systems, due to large memory overhead
since all information of a model is stored in the Model DLL (therefore
separate code generation for embedded systems via Realtime Workshop)
FMI: Only the minimum necessary part is stored in C source code or in Model
DLL. All information not needed for execution, is provided in an XML file
(which is needed on host, but not on target microprocessor)
S-function has very complex definition (> 100 C-functions/macros)
Generating S-function is fine. However, there is no simulator that can import
all S-function models (with exception of SIMULINK).
FMI: Simple definition (20 C-functions, no macros, XML schema file)
S-function proprietary format, gives legal problems if used in other simulators
FMI: Wikipedia license for specification, BSD license for schema/header
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 28
Technical issues that are missing in S-Function interface and are available in FMI:
Reliable state event handling
Event iteration over simulation model (not only component model)
Request from submodel to reduce step-size
(for non-linear equations in model that do not converge)
Dynamic selection of states
(as needed for order-reduced higer index systems)
Alias variables
(FMI: alias variables are marked; need to be stored only once, not several
times; important for Modelica models, since many alias variables)
Caching of computed results
(FMI: more efficient solution)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 29
9. Outlook
"FMI for Model Exchange" released at end of January 2010
(technical specification finalized; some discussion about precise license text)
"FMI for Co-Simulation" in a good stage. Will be released in first half year:
Support for: extrapolation/interpolation of interface variables,
variable communication step-size, re-doing a step
→ step-size control possible).
"FMI for Model Exchange" will be further developed. A lot of requirements
available, such as:
Sparse Jacobian
Direct support for arrays and records in xml schema
Improved sample time definition (for embedded systems)
Online changeable parameters
Saving/restoring model state
Other tool vendors are encouraged to support FMI, especially:
VHDL-AMS simulators
Multi-Body simulators
Note: Much simpler as SIMULINK S-Function interface and more powerful.
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 30
10. Acknowledgments
FMI initiated and organized
: Daimler AG (Bernd Relovsky, ....)
Head of FMI development
: Dietmar Neumerkel (Daimler AG)
Head of FMI-for-Model-Exchange: Martin Otter (DLR-RM)
FMI-for-Model-Exchange
Core-Design by:
Torsten Blochwitz (ITI)
Hilding Elmqvist (Dassault Systèmes -Dynasim)
Andreas Junghanns (QTronic)
Jakob Mauss (QTronic)
Hans Olsson (Dassault Systèmes -Dynasim)
Martin Otter (DLR-RM)
Other MODELISAR contributors: Ingrid Bausch-Gall, Bausch-Gall GmbH
Alex Eichberger, SIMPACK AG
Rainer Keppler, SIMPACK AG
Gerd Kurzbach, ITI GmbH
Carsten Kübler, TWT
Johannes Mezger, TWT
Thomas Neidhold, ITI GmbH
Dietmar Neumerkel, Daimler AG
Peter Nilsson, Dassault Systèmes-Dynasim
Antoine Viel, LMS International
Daniel Weil, Dassault Systèmes
Other contributors:
Johan Akesson, Lund University
Joel Andersson, KU Leuven
Roberto Parrotto, Politecnico di Milano
Partially funded by:
BMBF, VINNOVA, DGCIS, organized by ITEA2
Prototypes for FMI evaluation:
Dymola by Peter Nilsson, Sven Erik Mattsson,
Carl Fredrik Abelson, Dan Henriksson
(Dassault Systèmes, Dynasim)
JModelica.org by Tove Bergdahl (Modelon)
Silver by Andreas Junghanns, Jakob Mauss
(QTronic)
Functional Mockup Interface > ASIM Workshop, 4.-5. März 2010 >
Slide 31