simulating multibody applications with openmodelica · multibody simulation with openmodelica slide...

Fakultät Maschinenwesen, Professur für Baumaschinen und Fördertechnik

Linköping, 04.02.2013

Simulating MultiBody Applications with

OpenModelica

– Current Status -

Christian Schubert, Dresden University

Multibody simulation with OpenModelica Slide 2

Outline

1. Introduction

2. Feature Status

3. Performance by looking at Examples

a. Examples

b. Translation

c. Compilation

d. Simulation

4. Possible Future Directions

04/02/2013


Applications

Models

04.02.2013


Applications

Simulator

04.02.2013


Applications

Toolchain

04.02.2013

=> Performance is most important to us

Modelica Model

OMC

FMU

SARTURIS

Motion Platform

Driver Cabin

Visuals


Required Features

Handling of Records, Vectors, Matrices

Overconstrained Connection Graph

Annotations: Inline, (__Dymola_)InlineAfterIndexReduction, derivative, noDerivative, Evaluate

Dynamic State Selection

Handling of replaceable gravityFunction

Current Status

04.02.2013


Peformance/Comfort Features

Tearing Reducing size of sparse blocks /

Nonlinear Solver (Kinematic Loops)

Analytic Jacobians

Robust and fast Event Handling

Support for large models

Visualization

/

Current Status

04.02.2013


Current Status

Necessary features implemented

Workaround for gravityFunction

Multibody should work

One test from MSL 3.2.1 fails: Elementary.UserDefinedGravityField

Performance/Comfort functions not finished

Let us see how they perform

04.02.2013

https://test.openmodelica.org/~marsj/MSL32/Modelica.Mechanics.MultiBody.Examples.Elementary.UserDefinedGravityField.err


Examples

RobotR3 EngineV6 Pendulum N

Picture

#Vars 4921 12491 ≈ 242 N

#States 36 4 2 N

#Nls 0 6 0

Linear 137 -> 6 322 -> 31 ≈ 12 N -> N

04.02.2013


Examples

Pendulum with N bodies

04.02.2013

…

model Pendulum_N constant Integer N = 2; inner World world; Revolute revolute[N]; BodyCylinder bodyCylinder[N]; Damper damper[N]; equation connect(world.frame_b, revolute[1].frame_a); connect(revolute.frame_b, bodyCylinder.frame_a); connect(damper.flange_a, revolute.support); connect(damper.flange_b, revolute.axis); connect(bodyCylinder[1:N-1].frame_b, revolute[2:N].frame_a); end Pendulum_N;


Examples

Division into three steps

04.02.2013

Translation Compilation Simulation

Modelica Model

C Source Code

Executable Resultfile


Translation

RobotR3 EngineV6

04.02.2013

Cra

she

d 85

90

95

100

OMC 1.8.1 OMC 1.9.0.b3 Dymola

0

5

10

OMC 1.8.1 OMC1.9.0.b3

Dymola

Software

0

5

10

15

20

25

30

35

OMC 1.8.1 OMC1.9.0.b3

Dymola

Software

translateModel

Templates

Simcode

Backend

Frontend≈ C

rash

ed


Translation

04.02.2013

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

1,8

10 30 50 70 90 110 130 150 170 190 210 230 250

Tim

e p

er

bo

dy

Number of Bodies

Normalized Time for Translation (with Tearing)

Templates

Simcode

Backend

Frontend


Translation

• Translation lies within factor 10 of Dymola

• Translation scales with O(N²)

• Problems have been identified and are being fixed

04.02.2013


Compilation

04.02.2013

1

10

100

1000

0 20 40 60 80 100

Re

qu

ire

d T

ime

/s

Number of bodies

Compilation - Pendulum N

OMC - Mingw

OMC - Visual Studio

Dymola - Visual Studio


Compilation

• Compilation as long as translation

• Compilation limits model size

• gcc faster than Visual Studio

• much slower than Dymola

04.02.2013


Simulation

Tearing

• Deactivated by default for linear systems

• +d=doLinearTearing

• Always treated as nonlinear system

• => Numerical Jacobian

• Much faster/more robust with symbolic Jacobian

• +d=doLinearTearing,NLSanalyticJacobian

04.02.2013


Simulation

04.02.2013

0

0,05

0,1

0,15

0,2

0,25

0,3

t RH

S/m

s

RobotR3

Erro

r

0

0,2

0,4

0,6

0,8

1

1,2

1,4

1,6

t RH

S/m

s

EngineV6


Simulation

04.02.2013

0

1

2

3

4

5

6

7

8

9

10

0 20 40 60 80 100

t RH

S/m

s

Number of bodies

Evaluating x_dot=f(x,t)

OMC

OMC - Tearing

OMC - Tearing Analytical

Dymola


Simulation

EngineV6 (𝐟 t, 𝐱 equally fast)

Not everything is counted! Event detection has to be improved

04.02.2013

OMC Dymola

Simulation time[s] 9,72 5,18

steps 8081 7912

events 544 335

F-Evaluations 12273 24548

Jacobians 3338 3215


Simulation

Pendulum with 40 bodies (𝐟 t, 𝐱 3x slower)

04.02.2013

OMC Dymola

Simulation time[s] 0,951 0,421

steps 257 240

events 0 0

F-Evaluations 364 1837

Jacobians 14 17


Possible Future Directions

Short term goals: Finishing • Dynamic State Selection • Tearing • Analytic Jacobians Mid term goal: • Visualization • Clean up OpenModelica-Trac Long term goal:

A complete redesign of the generated code should be considered

04.02.2013


Thank you

04/02/2013

Christian Schubert, Dipl.-Ing, M.Sc. Dresden University of Technology Institute of Processing Machines and Mobile Machines E-Mail: [email protected] Tel.: +49 351 463-39278

simulating multibody applications with openmodelica · multibody simulation with openmodelica slide...

Documents