Multidisciplinary Aircraft Design Optimisation Nickolay Jelev (Meng) 20-22 nd November 2017 [email protected]Academic supervisors: Prof. Andy Keane Dr Andrรกs Sรณbester Industrial supervisor: Dr Carren Holden Jointly funded by Airbus UK and the University of Southampton.
19
Embed
Multidisciplinary Aircraft Design OptimisationMDF) โข Individual Discipline Feasible, ... architecture and stands for โSimultaneous Analysis and ... Axial Stress: ๐= ๐ผ ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Academic supervisors: Prof. Andy Keane Dr Andrรกs Sรณbester Industrial supervisor: Dr Carren Holden
Jointly funded by Airbus UK and the University of Southampton.
Overview of the Design Process
Overview of Multidisciplinary Design Optimisation (MDO) Architectures
The Blackboard Architecture:
โข Multidisciplinary Pattern Search
โข Application UAV wing design problem
Future Work:
โข Team Based design activity to test the proposed MDO framework
โข Use data mining to speed up convergence
2
Contents
3
Design Process
Multidisciplinary Design Optimisation (MDO)
4
โข Methods that solve problems
consisting of a number of domains.
They can better exploit the interactions
between the disciplines, thus in theory
arrive at a superior design than by
optimizing each discipline sequentially.
Overview of the Research Field
Monolithic Method
โข Simultaneous Analysis and Design, (SAND)
โข Multiple Discipline Feasible, (MDF)
โข Individual Discipline Feasible, (IDF)
5
Distributed Method
โข Concurrent Subspace Optimisation, (CSSO)
โข Collaborative Optimisation, (CO)
โข Enhanced Collaborative Optimisation, (ECO)
โข Bi-Level Integrated System Synthesis, (BLISS)
โข Analytical Target Cascading, (ATC)
โข Exact and Inexact Penalty Decompositions, (EPD/IPD)
โข Quasi-Separable Decomposition, (QSP)
โข MDO of Independent Subspaces, (MDOIS)
โข Etcโฆ
Abstract Method
โข Bayesian Based Methods
โข Game Theory Methods
โข Blackboard Methods
โข Fuzzy Logic Methods
โข Etcโฆ
Why Multidisciplinary Design Optimisation? โข There is ample evidence and a shared consensus among academics that MDO methods
produce superior results than sequential one at the time domain optimisation.
โข Two accepted categories of MDO approaches in Academia:
6
Monolithic Distributed
All analyses routines are combined under a single optimiser.
Domain level optimisers are coupled with analyses routines to optimise local objectives. A system level optimiser coordinates the disciplines to a single optimal design.
Advantages: Advantages:
Generally faster to converge and more robust Designed to fit the already existing organisational structure in a company
Domains can operate independently of other domains and take advantage of low cost distributed computing
Disadvantages: Disadvantages:
Maintenance difficulties of merging numerous analysis tools under a single optimiser
Human out of the loop process Difficult to implement in an organisational
structure. Non trivial gradient computation
Generally much slower to converge Some require a non trivial problem decomposition
Change in Design Process
7
8
The Blackboard Framework User Interface
Final Design Rule base
Blackboard
Data Mining
Database
To domains: Search Space From domains: Preferred Designs
Aerodynamics Group,
Minimising Drag
Weights Group,
Minimising mass
Structures Group,
Minimising Stresses
Starting Design and
Search Space
Controls Group,
Improving Stability
Manufacturing Group, Ease of Manufacturing
User Interface
Final Design Rule base
Blackboard
Data Mining
Database
To domains: Search Space From domains: Preferred Designs
Aerodynamics Group,
Minimising Drag
Weights Group,
Minimising mass
Starting Design and
Search Space
Manufacturing Group, Ease of Manufacturing
Structures Group,
Minimising Stresses
Controls Group,
Improving Stability
Multidisciplinary Pattern Search
9
The Hooke and Jeeves Pattern
Search
The Multidisciplinary Pattern
Search
Simplified UAV Wing Design
10
System:
Shared Variables: ๐, ๐, ฮ,๐ก
๐ ๐๐๐ฅ
Structures Local Variables: ๐ก๐ ๐๐๐ , ๐๐๐ข๐ก๐๐