Systems Design - New Paradigm K Sudhakar Centre for Aerospace Systems Design & Engineering January 28, 2004.

Systems Design - New Paradigm

K SudhakarCentre for Aerospace Systems Design &

Engineering

http://www.casde.iitb.ac.in/January 28, 2004

Systems Design

RequirementsCapture

DesignProcess

SystemSpecificatio

n

Systems Engineering

Discipline-1 Discipline-2 Discipline-3

System

Systems Engineering?

Need to view things from one level

higher than your work requires

Designers Design Process

Meta DesignMeta Design

Meta-Design

• Increase breadth of knowledge used in decisions

• Increase depth of knowledge used in decisions

• Shorten design cycle time

• Ability to systematically explore design space

• - -

RequirementsCapture

Design theDesign Process

SpecifyDesign Process

Meta Design MDO

Meta Design

ElementsMDO Elements

Researcher’s Perception

• Multi-disciplinary : Increased breadth

• Design – process of translating requirements into product specifications.

• Optimization – Formal method of locating the ‘best’ under ‘constraints’

• Implies use of high fidelity tools. Increase depth.

Industry Perception

• Not a turnkey solution to design!

• Only a tool in the hands of designer to – State design problems formally

– Integrate appropriate fidelity analysis

– Explore design space

– Improve design starting from a baseline

If we can find an optima we will be happy!

If we find global optima we will celebrate!

Optimization

Systems Design

Parameters

Requirements asConstraints

Objective

Analysis

Aerodynamics

Structures

Controls

-ilities

Trajectory

An Example – HSCT (1991-’99)!

• HSCT-2– 5 design variables, 6 constraints – WINGDES, ELAPS, Range equation, engine deck – Time for one cycle = 10 minutes

• HSCT-3– 7 design variables, 6 constraints– ISAAC, COMET, Range equation. Engine deck– Time for one cycle = 3 hours

• HSCT-4– 271 design variables, 31,868 constraints– CFL3D, USSAERO, GENESIS, FLOPS, ENG10 – Time for one cycle = 3 days

HSCT - 4

• Detailed problem definition took more than 1 year to extract from people

• Requirements document touched 100 pages merely to define analysis process, tools used and data flow

• 90% of work went into preparing analysis codes for MDA and integrating them in a proper sequence

Where are we?

• Strengths exist in disciplinary analysis

• No focus on Analysis for Design

• No focus on verification / validation to

characterize uncertainties

• No attempt to capture knowledge with

traceability

CASDE @ Workshop on Framework for System Analysis, ISSA, New Delhi, October 13, 2003

Need for groups to

• Define design problem• Define needs for Analysis for Design• Extract / Establish traceability • Perform Verification / Validation to

characterize uncertainty• Explore design methodologies

New Paradigms• MDO – the process• Frame Works – to deploy the process• Multi-criteria decision making• Design under uncertainty

Components• Surrogate Modeling (DOE, RSM, DACE)• Sensitivity Analysis

Design Under Uncertainty

AnalysisXnom Ynom

Ynom Y

p

• V&V, levels of fidelity• How to fuse• Characterisation

XnomX

p

• Characterisation• How to propagate

• How to assemble System Analysis• How to state design problem?

Frame Work

• Essential infrastructure• Disciplinary autonomy, but system level

integration. (Distributed, heterogeneous environment)

• Tools availability• Requirement Capture for Frame Work?• Commercial Frame Works – iSIGHT,

Phoenix Integration, . . .• CASDE MDO FrameWork Version-II

(March 2004)

MDO Framework

Database

ConfigurationServer

ExecutionManager

MDOController

NameServer

DataServer

OPT1

Optimizer Manager

OPT2 OPT3

AM1

AnalysisManager

AM2 AM3

GUI

Control

Data

A1

A2

A3A4

A5

Execution sequence

A13

Execution Unit

A12 A13 A14

Execution sequence of execution units

A22A1

A2

A3

A4

A5

Parallel Execution

• Architectural design - Intuitive GUI, OO principles, standards based

• Problem formulation - Iterative & branching formulations, legacy codes, multiple optimizers

• Problem execution - Automatic execution, parallel & distributed

• Information access – DB management visualization, monitoring

3D-Duct : An Example

• Duct design in the past?• Is improvements in breadth, depth

possible?• Statement of design problem?• Analysis Tools - Identification, V&V and

Integration• Focus on shrinking design cycle time• Design process?

3D-Duct : Problem Formulation

Entry Exit Location and shape (Given)

Optimum geometry of duct from Entry to Exit ?

Objective/Constraints

• Pressure Recovery• Distortion• Swirl

3D-Duct : Automation for CFD

Generation of entry and exit sections using GAMBIT

Clustering Parameters

Conversion of file format to CGNS using FLUENT

Mesh file

Generation of structured volume grid using parametrization

Duct Parameters(β1, β2, αy, αz)

Entry & Exit sections

Conversion of structured grid to unstructured format

Unstructured CGNS file

CFD Solution using FLUENT

End-to-end (Parameters to DC60)

automated CFD Cycle. Objective/Constraints evaluationUsing UDFs (FLUENT)

DC60

CFD Solution

ContinuationSolution

3D-Duct : Automation for Design

Generation of structured volume grid using parametrizationEntry & Exit

sections

Conversion of structured grid to unstructured format

CFD Solution using FLUENT

Objective/Constraints evaluationUsing UDFs (FLUENT)

DC60

Optimization

Duct Parameters(β1, β2, αy, αz)

ContinuationSolution

Unstructured CGNS file

CFD Solution

3D-Duct : Design Space Reduction

6.19

1.42

(0.61, 0.31, 1.0, 1.0)

Optimized duct from low fidelity rules

24.2116.28DC60

3.532.0PLOSS

(-0.4, 1.5, 0.3, 0.6)

(0.1, 0.31, 0.2, 0.6)

P

Highly infeasibleFrom low fidelity rules

Marginally infeasible from low fidelity rules

P – Parameters; PLOSS – Total Pressure Loss

3D-Duct : Simulation Time

• Strategies– Continuation Method– Parallel execution of FLUENT on a 4-noded

Linux cluster

Time for simulation has been reduced to around 20%.

0 20 40 60 80 100

Time (hrs)

Time per CFD Run

Serial

Parallel

Slapping

3D-Duct : Design Process

Parametrization

Low fidelity Analysis

DOE in reduced space

CFD analysis at DOE points

RS for PR & DC60

OptimizationConstraint

s

LFA Optima

CONCURRENT ENGINEERING Vs MDO

Time into the process

Source: AIAA MDO White Paper, 1991

Life Cycle Emphasis

Design

Manufacturing

Supportability

CE

Systems Design Emphasis Aerodynamics

Propulsion

Structures

Controls

MDO

Visithttp://www.casde.iitb.ac.in/MDO/

Thank You