Thorsten Pawletta & Olaf Hagendorf · 2017. 5. 30. · Hochschule Wismar - University of Applied Sciences RG Computational Engineering & Automation (CEA) Thorsten Pawletta & Olaf

Hochschule Wismar - University of Applied SciencesRG Computational Engineering & Automation (CEA)

Thorsten Pawletta & Olaf Hagendorf

Invited Talk at the Workshop on Trends in Computational Sciense (TCSE), 13th-14th Feb. 2012, in front of MATHMOD Conf., Vienna, 15th-17th Feb.2012

1. Motivation2. Modular, hierarchical systems3. Modeling & simulation4. Manual sim. based system optimization (SSO)5. Semi-automatic SSO6. Full-automatic SSO

1. SES/MB framework2. Mapping of system structures3. Combination to a complete approach

7. Application exampleConclusion

2TCSE Workshop, Vienna, 2012/02

Contents

Engineering systems can be implemented using different system designs and several strategies

⇒Set of system designs Any system design is a composition of systems &

systems are configured using parameters⇒Modular, hierarchical composition of systems (system

structure)⇒Set of system parameters for each system

An engineering objective: find the best system design

⇒Optimal system structure with optimal system parameters


1. Motivation

System types: Atomic system: non-decomposable systems

with dynamic behavior A=(X, Y, S, δext, δint, δcon, λ, ta) [ZPK_2000]

Coupled system: set of systems and relationsC=(X, Y, D, {Mdd∈D}, EIC, EOC, IC) [ZPK_2000]


2. Modular, Hierarchical Systems

B FE CDA

Modeling1. Specification of reusable models for atomic &

coupled systems => model base (libraries)2. Specification of a specific model (one system

structure with configurable system parameters )

Simulation (most simple experiment)3. Execution of a specific model within a simulation

runtime system


3. Modeling & Simulation

One cycle: eval. of one system design(structure, parameters)


4. Manual Simulation Based System Optimization (SSO)

manual changesof parameters and

structures

system

model

executable model

modeling

programming

simulation

result OK?

yes

no

components steps

man

ual c

hang

es

solution

Inner cycle: eval. of (structure , {parameters}) Outer cycle: eval. of ({structure}, {parameters}) 7TCSE Workshop, Vienna, 2012/02

5. Semi-Automatic Simulation Based System Optimization (SSO)

(Classic parameter optimization)

optimizationmethod fitness

function parameter changes

modeling

programming

simulation

result OK?

yes

noparameter optimized model

system

model

executable model

No

Yes

solution

components steps

result OK?

man

ual c

hang

es o

f mod

el s

truct

ur

simulationresults

performance

manual changes ofstructures

automatic changes ofparameters

Current status Modular, hierarchical model of a single system design Simulation based evaluation Configurable system parameters Numerical parameter optimization approach

Additional requirements for full-automatic SSO Formal specification of all system designs

({system.structures}, {system parameters}) Automatic generation of models/executable models Mapping of {system structures} ⇆ {numerical data} for

a structure optimization equivalent to a param. optim.


6. Full-Automatic Simulation Based System Optimization (SSO)

( )

Pruned Entity StructurePruned Entity Structure executable model

Formal specification of all system designs & dynamics Automatic generation of single simulation models



SES/MB

Model BaseSystem Entity Structure

{1,3}

Tree = ({nodes}, {edges}, {attributes}) Entity node: atomic or coupled system◦ node attributes: system parameters

Aspect node: decomposition of a system◦ coupling specification

Multi-aspect node: specific decomposition of a system◦ properties

Specialization node: taxonomy of a system◦ selection rules

Selection constraints: interdependencies of systems





System decomposition/composition

Adec

B CCdec1

F G

A

{(B.out,C.in)}

{(C.in,F.in),(F.out,G.in)}

{p1=42}

A

B

C

F G



System decomposition/composition (2 variants)

Adec

B CCdec1 Cdec2

F G H I

A

{(B.out,C.in)}

{(C.in,H.in1),(H.out,I.in),(I.out,H.in2)}

{(C.in,F.in),(F.out,G.in)}

A

B

C

F G

A

B

C

H I



Specific decomposition/composition

Adec

CDmaspec

{1,2,3}

D

A

{(..),..}

B

AD1 C

AD1

CD2

D3

A D1C

D2



Specialization

Adec

CEspec

E1 E2 E3

A

{(..),..}

{selection rules}

E

AE1 C

AE2 C

AE3 C



Selection constraints/Structure rules

Adec

CCdec1 Cdec2

F G H I

A

{(..),..}

{(..),..}{(..),..}Dmaspec

{1,2,3}

D

BEspec

E1 E2 E3

{selection rules}

E

without constraints/rules:18 variants

with constraints/rules:12 variants

constraints rules

≡ {((E2 ∩ H) ∪(E3 ∩ F) )}

Adec

CCdec1 Cdec2

F G H I

A

{(..),..}

{(..),..}{(..),..}Dmaspec

{1,2,3}

D

BEspec

E1 E2 E3

{selection rules}

E



Adec

CE3Cdec1

D1

F G

A

D2

SES

PES

A

E3

C

F GD1

D2

≡ {((E2 ∩ H) ∪(E3 ∩ F) )}

x

xx

x

x

Current state: Modular, hierarchical model of a single system design Simulation based evaluation Configurable system parameters Numerical parameter optimization approach

Additional requirements Formal specification of all system designs

({system structures}, {system parameters}) Automatic generation of models/executable models Mapping of {system structures} ⇆ {numerical data} for

a structure optimization equivalent to a parameter opt.



?

( )

Adec

CCdec1 Cdec2

F G H I

A

{(..),..}

{(..),..}{(..),..}Dmaspec

{1,2,3}

D

BEspec

E1 E2 E3

{selection rules}

E



xS= (xS1,xS2,xS3)Dmaspec => xS1 ϵ {1,2,3}Espec => xS2 ϵ {1,2,3}Cdec => xS3 ϵ {1,2}

⇒ n decision nodes → n variables

Adec

CCdec1 Cdec2

F G H I

A

{(..),..}

{(..),..}{(..),..}Dmaspec

{1,2,3}

D

BEspec

E1 E2 E3

{selection rules}

E



xsi= (3,3,1)

xS2=3 => D1,D2,D3xS3=3 => E3xS1=1 => C1

checking structure rules:{((E2 ∩ H) ∪ (E3 ∩ F) )} → OK

Adec

CCdec1

F G

A

D1 D2 D3 E3



xsi= (3,3,1)

A

E3

C

F G

D1

D2

D3

Current state: Modular, hierarchical model of a single system design Simulation based evaluation Configurable system parameters Numerical parameter optimization approach

Additional requirements Formal specification of all system designs

({system structures}, {system parameters}) Automatic generation of models/executable models Mapping of {system structures} ⇆ {numerical data} for

a structure optimization equivalent to a parameter opt.



( )

cycle: eval. of ({structure} , {parameters})

automatic changes ofstructures and

parameters




7. Application Example

System Design

Splicer URS DigiURS DigiSplicer Software App

Development

AnalogPrinter

Scanner

CD Production

Digital Printer

Development

Cutter DigiCutter

LoginIn-sorter

Out SorterShipping

analog material

digital data

paper/picture/others

analog machine

digital machine



controller_lsspec

ctrl1 ctrl2DEP_LOGINdec1 DEP_LOGINdec3

queue_box2

queue_batchsplicermaspec

splicer

{#_of_splicers={1,…,6}}

DEP_LOGINdec2ctrl3

DEP_SPLICERdec

MODELdec

queue_order

queue_box1

sorter_manu

sorter_manu

queue_order

queue_box1

sorter_auto

queue_order

queue_box1

sorter_auto

MODEL

DEP_SPLICERCONTROLLER_LSDEP_LOGIN

model parameter

#_of_operators_ls={1,6}#_of_operators_pc={1,6}filter={0, 0.2, … 0.8, 1}

structure rules:{max(manu_login+auto_login,#_of_splicers)=#_of_operators}

{auto_login=1}

{manu_login=1}

{auto_login=1}

{manu_login=1}

SES

controller_pcspec

ctrl1 ctrl2 ctrl3

CONTROLLER_PC

queue_batch1

queue_batch2printer_analog

DEP_ANALOGdec

DEP_ANALOG

printer_analog cutter_analog

queue_batch1

queue_batch2printer_digi

DEP_DIGITALdec

DEP_DIGITAL

printer_digi cutter_digi

filter

162 system structures 3 system parameters 34992 system designs



controller_lsspec


queue_box2


splicer


DEP_LOGINdec2ctrl3

DEP_SPLICERdec

MODELdec

queue_order

queue_box1

sorter_manu

sorter_manu

queue_order

queue_box1

sorter_auto

queue_order

queue_box1

sorter_auto

MODEL


Model Parameter



{auto_login=1}

{manu_login=1}

{auto_login=1}

{manu_login=1}

SES

controller_pcspec

ctrl1 ctrl2 ctrl3

CONTROLLER_PC

queue_batch1


DEP_ANALOGdec

DEP_ANALOG


queue_batch1


DEP_DIGITALdec

DEP_DIGITAL


filter

162 system structures 3 system parameters 18145 system designs



xS=(xDEP_LOGIN, xcontroller_ls_spec, xsplicermaspec, xcontroller_pc_spec )xP=(x#_of_operators_ls, x#_of_operators_pc, xfilter)S=(xS×xP) ⇒ 7 dimensional search room

controller_lsspec


queue_box2


splicer


DEP_LOGINdec2ctrl3

DEP_SPLICERdec

MODELdec

queue_order

queue_box1

sorter_manu

sorter_manu

queue_order

queue_box1

sorter_auto

queue_order

queue_box1

sorter_auto

MODEL


Model Parameter



{auto_login=1}

{manu_login=1}

{auto_login=1}

{manu_login=1}

SES

controller_pcspec

ctrl1 ctrl2 ctrl3

CONTROLLER_PC

queue_batch1


DEP_ANALOGdec

DEP_ANALOG


queue_batch1


DEP_DIGITALdec

DEP_DIGITAL


filter



26 system designs with Fmin=0.26



average number of investigated individuals to find a global optimum 226,4

global optimum 47x

near optimal results with max 1% error 26x

results with 1 … 5% error 9x

results with 5 … 10% error 18x

numerical optimisation method: GA

Average results of 100 optimization experiments:



complete enumeration 18145 simulation runs Finding of global

optimum guaranteed

SSO ca. 226 simulation runs Finding of global

optimum not guaranteedBut with 73% probability

finding of a solution with error

manual simulation based system optimization semi-automatic simulation based system

optimization⇒full-automatic simulation based system

optimization◦ formal description of {system designs}◦ automatic model generation◦ mapping {system structures} → {numerical parameter}⇒using of existing numerical parameter optimization

method possible◦ integration into traditional optimization algorithm


8. Conclusion


• O. Hagendorf, T. Pawletta: A Framework for Simulation-Based Structure and Parameter Optimization of Discrete Event Systems. In: Discrete-Event Modeling And Simulation, Ed. G.A. Wainer and P. J. Mosterman, CRC Press, 2011, 199-222

• [ZPK_2000] B.P. Zeigler, H. Prähofer, T.G. Kim: Theory of Modelingand Simulation (2nd Ed.), Academic Press, 2000

Simulation Based�Evaluation and Optimization of Modular, Hierarchical System Designs Using A Graph Based SpecificationContents1. Motivation2. Modular, Hierarchical Systems3. Modeling & Simulation4. Manual Simulation Based� System Optimization(SSO)5. Semi-Automatic SSO6. Full-Automatic SSO6.1 SES/MB Framework (Zeigler et al.)Characteristics of SESCharacteristics of SESCharacteristics of SESCharacteristics of SESCharacteristics of SESCharacteristics of SESSES/MB Based Model GenerationFull-Automatic SSO6.2 Mapping of �{system structures} → {numerical data}6.2 Mapping of �{system structures} ← {numerical data}6.2 Mapping of �{system structures} ← {numerical data}Full-Automatic SSO6.3 Full Approach7. Application: Production Planning of a Photofinishing LabSES of the exampleSES of the exampleSES of the exampleFitness functionResults: Complete EnumerationResults: SSOResults: comparisonConclusionFoliennummer 32

Thorsten Pawletta & Olaf Hagendorf · 2017. 5. 30. · Hochschule Wismar - University of Applied Sciences RG Computational Engineering & Automation (CEA) Thorsten Pawletta & Olaf

Documents