options and utp rsm code plug-ins to isight and options ... · rsm doe and rsmmethods including...
TRANSCRIPT
CEDC
Options and UTP RSM codeplug-ins to iSIGHT and OptionsOptimiser Plug-In to Model Center
Objectives:• To incorporate the OPTIONS opti-
misation, DOE and RSM algo-rithms into iSIGHT as a plug-in.
• To incorporate the UTP developed(Stephen Leary) DOE and RSMalgorithms into iSIGHT as a plug-in.
iSIGHT Plug-ins:The following four things need to bedone to add an optimisation tech-nique (i.e., couple a code) to iSIGHT:1. Provide two C routines (one
through a .tgen file) to create anew Tcl package and Tcl com-mand for the optimisation tech-nique
2. Use ESI C API calls to communi-cate between the optimisationtechnique and iSIGHT
3. Use the Technique Generator(Techgen) utility to automaticallycreate support code needed toadd your technique to iSIGHT
4. Any GUI interfaces need to beprogrammed (in Tcl/Tk) & incor-porated into the overall controlstructure.
Basis for iSIGHT plug-in development – theTcl language:• Tcl language consists of an inter-
preter that understands a base setof commands implemented in a Clibrary.
• Tcl is powerful because it suppliesa mechanism for extending thebase command set with addition-al commands written in C or Tcl asdynamically loadable, sharedlibraries.
• This is the way iSIGHT itself isimplemented.
strained objective functions and usetheir own internal SimulatedAnnealing (SA) based optimiser formodel training and optimisation.
Options Plug-In forModelCenter:• Objective: Incorporate the Options
optimisation package with its ownGUI into ModelCenter, so that aproblem defined in MC can be opti-mised using one or more Optionsmethods and the results stored in anMC Data Collector for further dis-play/processing by MC facilities.
• Phoenix supply a Developer’s Guidefor Plug-Ins to ModelCenter. Aplug-In can be written as a COM compo-nent (using C++, VB, etc) or Javacomponent (with Java Swing GUI)
ModelCenter Plug-InInterfaces:• Tools-Trade Study Plug-In: a Trade
Study can be any algorithm thatrepeatedly runs an analysis code toreach some objective (e.g. an opti-mization, DOE or parametric study).
• Component Plug-In: incorporateplug-in as a component in the ServerBrowser so that it can be includedinto Workflows. On the Workflow,click on component icon to bring upits GUI.
• Data Collector Plug-In: appears as atoolbar item in the Data Collector,and is run from there using data fromthe Data Collector. It is used for post-processing operations, such as build-ing a response surface model.
Some combinations of the above arevalid, such as Trade Study +Component plug-in or Data Collector +Component plug-in.
Options Trade Study/Component Plug-In:What’s involved:• Develop a Java plug-in program
according to the Trade Study andComponent interface requirementsand the JavaBeans specification.
• Develop a series of table driven(JTable/JTableX) Java Swing basedGUI interfaces to drive Options.
• Integrate the Options package(Windows version) as a DLLthrough the JavaOptions (JNIbased) interface developed for theGeodise project.
• Package the plug-in as a JAR fileand register the plug-in withModelCenter.
Options Plug-In ModelSet-up:
The main Options Plug-In GUI show-ing problem data and selection buttonsto drive the optimisation throughlower level GUIs.
Options MethodSelection (Scheduling)GUI:
Sections:• Optimisation, DOE and RSM meth-
ods available for selection (45+ DOE,RSM and optimisation algorithms).
• Run control parameters and but-tons:✷ Optimisation mode: Minimise✷ Maximum iterations for next step:
500✷ Maximum major iterations
(updates) for RSM models: 5✷ Call method selected parameter
setting GUI✷ Display Options Debug output
file ✷ Display Options optimisation log
file • Run schedule selected:
✷ Build Kriging model (usingExpected Improvement Function)of the Objective Function andmain constraint. MinimiseKriging predictor on objectivefunction using GA followed bylocal search.
Options MethodParameter Setting:
Change running parameters for GAalgorithm.
Use Model Center RSMToolkit to gen-erate a plot of the response surfaceusing data generated by the ‘GenerateRSM surface points’ option in the mainGUI.
Tony Scurr ([email protected])Computational Engineering Design GroupSchool of Engineering SciencesUniversity of SouthamptonHighfield, SouthamptonSO17 1BJ
Options iSIGHT Plug-In GUI:
Run control parameters:Maximum iterations: 500Optimisation mode: MinimiseLast method selected: dynamic hill climbing
Optimisation, DOE and RSM methods availablefor selection (45+ DOE, RSM and optimisation algo-rithms).
Run schedule selected:Build Kriging model of Objective Function and mainconstraint. Minimise Kriging predictor on objectivefunction using GA followed by local search.
Run Task Log: showing building of objective func-tion and constraint surfaces using Kriging, followedby a Genetic Algorithm (GA) optimisation.
UTP RSM Plug-In GUI:
Run control parameters:Maximum iterations: 100Optimisation mode: MinimiseLast method selected: RBF Optimise
RSM
DOE and RSM methods including Polynomial,Radial Basis Function (RBF) and Kriging algorithms,the last two with gradient enhanced algorithmic ver-sions.
Run schedule selected:Train RBF to derive optimal sqsig parameter fol-lowed by building and optimisation of the RBF RSMsurface.
The plug-in is implemented in thesame way as the Options plug-in, i.e.using the iSIGHT Techgen utility. Theplug-in contains a number of RSMmethods, including Polynomial, RBF,Kriging, the latter two with addition-al gradient enhanced methods. TheRSM methods only handle uncon-
This article may be found athttp://www.soton.ac.uk/~cedc/posters.html