Multidisciplinary Optimization

Slides:



Advertisements
Similar presentations
Linear Programming Problem
Advertisements

GRAPH BALANCING. Scheduling on Unrelated Machines J1 J2 J3 J4 J5 M1 M2 M3.
Global Approximate Inference Eran Segal Weizmann Institute.
1 Chapter 8: Linearization Methods for Constrained Problems Book Review Presented by Kartik Pandit July 23, 2010 ENGINEERING OPTIMIZATION Methods and Applications.
1 Bandwidth Allocation Planning in Communication Networks Christian Frei & Boi Faltings Globecom 1999 Ashok Janardhanan.
Stevenson and Ozgur First Edition Introduction to Management Science with Spreadsheets McGraw-Hill/Irwin Copyright © 2007 by The McGraw-Hill Companies,
Stevenson and Ozgur First Edition Introduction to Management Science with Spreadsheets McGraw-Hill/Irwin Copyright © 2007 by The McGraw-Hill Companies,
Operations Research Assistant Professor Dr. Sana’a Wafa Al-Sayegh 2 nd Semester ITGD4207 University of Palestine.
1 1 Slide © 2005 Thomson/South-Western Chapter 2 Introduction to Linear Programming n Linear Programming Problem n Problem Formulation n A Maximization.
Chapter 7 Handling Constraints
Distributed Lagrangean Relaxation Protocol for the Generalized Mutual Assignment Problem Katsutoshi Hirayama (平山 勝敏) Faculty of Maritime Sciences (海事科学部)
Framework for MDO Studies Amitay Isaacs Center for Aerospace System Design and Engineering IIT Bombay.
Low-Rank Kernel Learning with Bregman Matrix Divergences Brian Kulis, Matyas A. Sustik and Inderjit S. Dhillon Journal of Machine Learning Research 10.
Linear Programming Erasmus Mobility Program (24Apr2012) Pollack Mihály Engineering Faculty (PMMK) University of Pécs João Miranda
Conformant Probabilistic Planning via CSPs ICAPS-2003 Nathanael Hyafil & Fahiem Bacchus University of Toronto.
Jan. 28, 2004MDO Algorithms - 1 Colloquium on MDO, VSSC Thiruvananthapuram Part II: MDO Architechtures Prof. P.M. Mujumdar, Prof. K. Sudhakar Dept. of.
Business Mathematics MTH-367 Lecture 14. Last Lecture Summary: Finished Sec and Sec.10.3 Alternative Optimal Solutions No Feasible Solution and.
1 1 Slide Graphical solution A Graphical Solution Procedure (LPs with 2 decision variables can be solved/viewed this way.) 1. Plot each constraint as an.
1 2 Linear Programming Chapter 3 3 Chapter Objectives –Requirements for a linear programming model. –Graphical representation of linear models. –Linear.
The Transportation and Assignment Problems
Automatic Test Generation
Auburn University
An Introduction to Linear Programming
Decision Support Systems
Memory Hierarchy Ideal memory is fast, large, and inexpensive
Virtual memory.
Chapter 7. Classification and Prediction
An Introduction to Linear Programming Pertemuan 4
Chap 10. Sensitivity Analysis
CSPs: Search and Arc Consistency Computer Science cpsc322, Lecture 12
Author: Vikas Sindhwani and Amol Ghoting Presenter: Jinze Li
SOFTWARE DESIGN AND ARCHITECTURE
Lecture 3.
Efficient Join Query Evaluation in a Parallel Database System
Solving Systems of Linear Equations: Iterative Methods
C.-S. Shieh, EC, KUAS, Taiwan
CSPs: Search and Arc Consistency Computer Science cpsc322, Lecture 12
The minimum cost flow problem
Auburn University COMP7330/7336 Advanced Parallel and Distributed Computing Mapping Techniques Dr. Xiao Qin Auburn University.
Chapter 3 Brute Force Copyright © 2007 Pearson Addison-Wesley. All rights reserved.
ME 521 Computer Aided Design 15-Optimization
Database Performance Tuning and Query Optimization
CSPs: Search and Arc Consistency Computer Science cpsc322, Lecture 12
Chapter 6. Large Scale Optimization
Chapter 3 The Simplex Method and Sensitivity Analysis
1.206J/16.77J/ESD.215J Airline Schedule Planning
CIS 488/588 Bruce R. Maxim UM-Dearborn
Chapter 3 Brute Force Copyright © 2007 Pearson Addison-Wesley. All rights reserved.
Resolution Proofs for Combinational Equivalence
Chapter 8. General LP Problems
3. Brute Force Selection sort Brute-Force string matching
Algorithms and Problem Solving
Authors: Barry Smyth, Mark T. Keane, Padraig Cunningham
UNIVERSITY OF MASSACHUSETTS Dept
Chapter 11 Database Performance Tuning and Query Optimization
Incremental Problem Solving for CS1100
Linear Programming Problem
UNIVERSITY OF MASSACHUSETTS Dept
Chapter 8. General LP Problems
Chapter 4 The Simplex Algorithm
Md. Tanveer Anwar University of Arkansas
Part 4 Nonlinear Programming
Graphical solution A Graphical Solution Procedure (LPs with 2 decision variables can be solved/viewed this way.) 1. Plot each constraint as an equation.
Simplex method (algebraic interpretation)
Chapter 8. General LP Problems
Normal Form (Matrix) Games
Chapter 6. Large Scale Optimization
Constraints.
Multiobjective Optimization
Presentation transcript:

Multidisciplinary Optimization

Multidisciplinary Optimization Multidisciplinary Optimization (MDO) focuses on solving optimization problems spanning across multiple interacting disciplines

Disciplinary Analyses As with all previous optimization problems, we optimize a design point, but for each discipline, a disciplinary analysis is performed on that design point to generate a set of outputs called response variables for each disciplinary analysis where x is the design point, y is a vector of response variables, m is the number of disciplines, and F is some function used to compute the ith response variable vector Note the coupling of response variables between disciplines

Disciplinary Analyses An assignment is an association of a set of variable names with corresponding values related to the problem A disciplinary analysis can be thought of as a way of updating the response variables for the discipline using an assignment An assignment can be represented in code as a dictionary or associative array data structure Using these structures, variables are not restricted to numbers, but can be any object such as a vector

Interdisciplinary Compatibility Interdisciplinary compatibility means response variables are compatible with all disciplinary analysis Interdisciplinary compatibility holds for a particular assignment if the assignment is unchanged under all disciplinary analyses Finding an assignment that satisfies interdisciplinary compatibility is called a multidisciplinary analysis

Interdisciplinary Compatibility System optimization for a single discipline requires an optimizer to select design variables and query the disciplinary analysis in order to evaluate the constraints and objective function Only one discipline, so no coupling between disciplines

Interdisciplinary Compatibility If there are multiple disciplines, then dependencies have to be considered

Interdisciplinary Compatibility Dependencies can be mapped to form a dependency graph If disciplines are mutually dependent on each other, then they have a dependency cycle

Interdisciplinary Compatibility If a dependency graph has no cycles, then there always exists an evaluation ordering such that dependencies are satisfied called a topological ordering Kahn’s algorithm can find such orderings

Interdisciplinary Compatibility If there are dependency cycles, then no topological ordering exists A general approach is iterative techniques such as the Gauss- Seidel method Depending on the nature of the problem, iterative methods can converge slowly

Architectures General MDO problem formulation where f and depend on both the design and response variables

Multidisciplinary Design Feasible The multidisciplinary design feasible architecture structures the MDO problem such that standard optimization algorithms can be directly applied to optimize the design variables

Multidisciplinary Design Feasible Primary advantage of the multidisciplinary design feasible architecture is that the procedure is simple The primary disadvantage is that it can be expensive, requiring multiple discipline analyses Problems with complex dependencies may not converge

Sequential Optimization The sequential optimization architecture attempts to optimize the overall MDO problem by optimizing over multiple subproblems A subproblem is an optimization procedure performed over a subset of the design variables while focusing on a particular discipline The system-level optimizer takes the results from each subproblem and optimizes over the remaining variables This architecture often leads to suboptimal solutions and is often used as a baseline against which other optimization architectures are measured

Sequential Optimization

Individual Discipline Feasible The Individual Discipline Feasible architecture allows disciplinary analyses to be conducted in parallel A set of coupling variables c are introduced which serve as proxies for the response variables The optimizer modifies the design point at each iteration until c has converged to a feasible equivalent to y

Individual Discipline Feasible

Collaborative Optimization The Collaborative Optimization architecture breaks down a problem into disciplinary subproblems that have full control over local design variables and discipline-specific constraints and can be analyzed in parallel Coupling variables are defined to enable iterative convergence to a feasible optimal solution Collaborative Optimization is an example of a distributed architecture

Collaborative Optimization

Simultaneous Analysis and Design The simultaneous analysis and design (SAND) architecture avoids coordinating between multiple design analyses and combines the entire optimization process into a single optimizer Any disciplinary analysis can be transformed into residual form The MDO problem can then be stated simply as

Simultaneous Analysis and Design An advantage of SAND is that the solution process is simpler, provided the problem supports transformation into residual form SAND can also traverse infeasible regions of the search space to find an optimal, feasible solution A disadvantage is that not all problems support transformation to residual form

Summary Multidisciplinary design optimization requires reasoning about multiple disciplines and achieving agreement between coupled variables Disciplinary analyses can often be ordered to minimize dependency cycles Multidisciplinary design problems can be structured in different architectures that take advantage of problem features to improve the optimization process The multidisciplinary design feasible architecture maintains feasibility and compatibility through the use of slow and potentially nonconvergent multidisciplinary design analyses

Summary Sequential optimization allows each discipline to optimize its discipline-specific variables but does not always yield optimal designs The individual discipline feasible architecture allows parallel execution of analyses at the expense of adding coupling variables to the global optimizer Collaborative optimization incorporates suboptimizers that can leverage domain specialization to optimize some variables locally

Summary The simultaneous analysis and design architecture replaces design analyses with residuals, allowing the optimizer to find compatible solutions but cannot directly use disciplinary solution techniques