+ All Categories
Home > Documents > Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Date post: 02-Jan-2016
Category:
Upload: angelina-wilkins
View: 220 times
Download: 1 times
Share this document with a friend
Popular Tags:
32
Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1
Transcript
Page 1: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Robin McDougallScott Nokleby

Mechatronic and Robotic Systems Laboratory

1

Page 2: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Outline

Context and BackgroundMulti-Objective PSO (MOPSO) via Pareto

DominanceParallelization of PSO (PAPSO)MOPAPSOResults

2

Page 3: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

IntroductionIncreasing role for global optimization

techniques in engineering designAmbition in design leads to more highly-

parameterized systemsMore parameters lead to increasingly non-

linear objective function surfaces

3

Page 4: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

IntroductionParticle Swarm Optimization (PSO) gaining

increasing attention in both research and applications

Over time a number of variants have been utilized with great success

Many on the “Particle” level: Particle Accelerations Transient Social and Personal Weights Dynamic Forms … and many more

4

Page 5: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MotivationOptimization-Based Mechanism Synthesis

(OBMS)Highly parameterized systemUse optimization techniques to choose

parametersMore parameters typically lead to more non-

linear objective function surfacesEffects of which can confound

traditional optimization techniques

5

Page 6: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MotivationReplace previously used deterministic

techniques with a global optimization technique

No need for parameter transforms No need for “pseudo-global” techniques

Prevent artificial constriction of search space

6

Page 7: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MotivationA typical OBMS objective could be to design

a system to follow a given path…

7

Page 8: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MotivationOBMS with PSO synthesized mechanisms

which could fulfill this task better than deterministic algorithms

8

Page 9: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MotivationWith one major caveat….PSO took hours instead of minutes

9

Page 10: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Objectives

Use Multi-Objective PSO (MOPSO) to handle multi-objective problem specifications

Use Parallel Asynchronous PSO (PAPSO) to speed things up

Both topics well covered in the literature individually

Little mention of combining the two

10

Page 11: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Multi-Objective Optimization

Engineering design choices often involve balancing competitive objectives:

Cost vs. Performance Size vs. Strength Effectiveness vs. Efficiency

What options are available to us to deal with these competing objectives?

11

Page 12: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Multi-Objective OptimizationCould use a weighting scheme

Concerns: With no prior knowledge, how do you select the

weights? Potential to unfairly influence optimization before

execution begins

12

Page 13: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Multi-Objective OptimizationWhat we would like to do is change:

to

Shift the decision on when to decide how influential each objective will be to after the optimization effort instead of before hand

13

Page 14: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Multi-Objective OptimizationMOPSO uses Pareto Dominance to determine

set of solutions for one or more competing objectives

Each point in the optimal set constitutes a non-dominated solution

Two objective function systems form a front, more, a hyper-surface

14

Page 15: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Multi-Objective OptimizationImagine a two objective function system:

15

•Instead of a single optimum solution, MOPSO delivers a front of non-dominated solutions•Each point on the front represents the best possible solution for a given objective function with respect to the other objective functions

Page 16: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPSOMOPSO requires two significant changes to

the basic form of PSO:Creation and active maintenance of a

repository to collect the non-dominated candidate solutions

Modification of the basic form of the velocity equation to choose a social leader form this repository instead of of a global best

16

Page 17: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPSONo longer a single social leader (SL)

available in MOPSO

Instead, need to choose a particle from the repository to serve as the SL

Use weighted Roulette Wheel procedure to select SL

Biased towards sparsely populated regions of the emerging front

17

Page 18: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

PAPSOReduce runtime by performing swarm

activities simultaneouslyPSO lends itself well to parallelizationFitness, velocity, position updates

independent per swarmProcessors:

Master Processor to administrate swarm Slave Processors perform Objective Function

Evaluations and Particle Updates

18

Page 19: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

PAPSONotes:If (# Particles > Number of Processors)

FIFO queue for particles

Asynchronous nature mitigates negative performance effects caused by runtime variability

Runtime improvement proportional to ratio of Objective Function Calculation time to Network Transmission Time

19

Page 20: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSOThe idea is to combine these variants:

MOPSO to provide formal multi-objective support PAPSO to speed things up

Requirements: Should match MOPSO results Should reduce overall runtime

20

Page 21: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSOTwo Roles for Processors…

21

One Master N# of Slaves

Initializes the swarm

Creates a FIFO particle queue

Dispatches the first “N” jobs

Catch GBEST, PPOS, PVEL

Update Velocity

Calculate OFs for each Object

Return OFs, PPOS and PVELCatch updated particle specs.

Dispatch next particle job

Update the repositoryEvery “m” iterations

Page 22: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSO – Benchmark TestsTo test effectiveness of MOPAPSO implementation:

Used two MOPSO benchmarks from the literature before applying to OBMS

Configuration:Nine-node grid running Rocks Cluster Distribution5 dual core 2.0 GHz processors with 1GB RAM eachacslX Interpconsole v2.4.1 using MPICH240 Particles, 100 “Iterations” (100 x 40 = Updates)

22

Page 23: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSO – Benchmark One

23

Page 24: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSO – Benchmark One

24

Page 25: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSO – Benchmark Two

25

Page 26: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

MOPAPSO – Benchmark Two

26

Page 27: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

27

Page 28: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Objective Functions

28

Page 29: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Results

29

Page 30: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Runtime Improvements

30

Page 31: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Conclusions

31

A high-level implementation of MOPAPSO has been developed

Testing of the algorithm on two benchmark problems showed that MOPAPSO can easily locate Pareto-fronts for multi-objective problems

MOPAPSO effectively solved OBMS featuring multiple objectives

Page 32: Robin McDougall Scott Nokleby Mechatronic and Robotic Systems Laboratory 1.

Acknowledgements

32

Ontario Research Fund


Recommended