Organizational Design Optimization Using Genetic Programming Bijan - - PowerPoint PPT Presentation

SLIDE 1

Organizational Design Optimization Using Genetic Programming

Bijan KHosraviani, Raymond E. Levitt and John R. Koza

S tanford University

GECCO 2004 Conference June 30, 2004

SLIDE 2

June 30, 2004 Organizational Design Optimization 2 Intro Objectives Methodology Results Introduction Conclusions

Presentation Outline

Introduction / Motivation Objectives / Research Questions Research Methodology Results to Date Conclusions

Introduction

SLIDE 3

June 30, 2004 Organizational Design Optimization 3 Intro Objectives Methodology Results Introduction Conclusions

Evolution of Organization Design

Trial-&-Error Adaptation Org’n Analysis: VDT/SimVision Org’n Design: VDT+Optimizer

• 1. Set project objectives.
• 1. Set project objectives.
• 2. Propose alternative
• rganizations.
• 3. Model alternative
• rganizations and

simulate each one to predict outcomes.

• 4. Choose solution that
• ptimizes outcomes.
• 1. Set project objectives.
• 2. Propose organization.
• 2. Propose initial
• rganization as starting

point for optimization.

• 3. Complete project

using proposed

• rganization and
• bserve outcome.
• 3. Evolve many alternative
• rganizations; predict

performance of each

• ne; evaluate “fitness”.
• 4. Succeed or fail.

Try to learn and adapt.

• 4. Evolve optimal organi-

zational configuration by selective reproduction & mutation of alternatives.

Introduction

SLIDE 4

June 30, 2004 Organizational Design Optimization 4 Intro Objectives Methodology Results Introduction Conclusions

Motivation

Project organization design

is a complex, multi- dimensional, optimization problem

Analysis tools exist for

• rganizational design, but

no known automated

• ptimizer exists

Finding an optimal or near-

• ptimal solution is a

challenging task even for an experienced PM

Start Fab Test & Deliver Project Lead Marketing Team Chip Architect Test Engineering St Foundry Lead Logic Design Team 1 Foundry Test Engineer Foundry Layout Engineer Verification Team Design Coordination Develop Specification Insert Scan Partition Chip Gen Test Suite PlaceRoute FullChipSynth Verify RTL FloorPlanning Write B1RTL Sim Gates Assemble RTL PhysVerifn Verify B1RTL Generate Test Vectors Synth_B1RTL 0.8 4 4 4 4 1 1 1 1 1 1 1 1 1 1 Management Meeting Architecture Team Meeting Foundry Team Meeting 1 1 1 1 1 1 1 1 1 1 1 1

Introduction

SLIDE 5

June 30, 2004 Organizational Design Optimization 5 Intro Objectives Methodology Results Introduction Conclusions

Objectives / Research Questions

Objectives

Develop an optimizer for VDT using evolutionary computing

techniques to help project managers find near-optimal designs for their project organizations

Validate the postprocessor against both theory and practice

Research Questions

How can GP help a highly experienced manager in designing

a project organization?

Are “optimal” solutions found by GP in-line with organization

theory and management best practices?

What are the limits of GA/GP for organization design?

Introduction Objectives

SLIDE 6

June 30, 2004 Organizational Design Optimization 6 Intro Objectives Methodology Results Introduction Conclusions

Evolutionary Computing Approach to Project Design Optimization

L I M S S t a r t 5

• 1
• 1

I n it ia t e t h e P r

• je

c t V a li d a t e T e s t T r a in C R P C

• n

f i g u r e S

• f

t w a r e D e s ig n S y s t e m P l a n t h e P r

• j

e c t P h a s V a li d a C

• m

p 1 1 / P h a s e 3 : D e s ig n a n d C

• n

fi g u r a ti

• n

C

• m

p le t e 9 / 1 3 P h a s e 2 : S W S e le c t i

• n

7 /1 2 P h a s e 1 : P la n A p p r

• v

e d 6 /7 P h a s e : P r

• je

c t R e v ie w 5 /1 7 S e l e c t S

• f

t w a r e P r

• j

e c t L e a d L IM S T e s t A p p li c a t i

• n

C

• n

s u lt a n t P r

• je

c t M a n a g e r L IM S C

• r

e W e e k ly P r

• je

c t M e e t i n g 1 W

• r

k D a y s S

• ft

w a r e R e c e iv e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7 L IM S S ta r t 5

• 1

1 In iti a te th e P r

• j

e c t V a lid a te T e s t T r a in C R P C

• n

fig u r e S

• ftw

a re D e s ig n S y s te m P la n th e P r

• je

c t P h a s V a lid a C

• m

p 1 1 / P h a s e 3 : D e s ig n a n d C

• n

fig u r a ti

• n

C

• m

p le te 9 /1 3 P h a s e 2 : S W S e le c tio n 7 / 1 2 P h a s e 1 : P l a n A p p r

• v

e d 6 /7 P h a s e : P r

• je

c t R e v ie w 5 /1 7 S e le c t S

• ftw

a r e P r

• je

c t L e a d L IM S T e s t A p p lic a tio n C

• n

s u l ta n t P r

• je

c t M a n a g e r L IM S C

• r

e W e e k ly P r

• je

c t M e e ti n g 1 W

• r

k D a y s S

• ftw

a r e R e c e i v e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7 L I M S S t a r t 5

• 1
• 1

I n it ia t e t h e P r

• je

c t V a li d a t e T e s t T r a in C R P C

• n

f i g u r e S

• f

t w a r e D e s ig n S y s t e m P l a n t h e P r

• j

e c t P h a s V a li d a C

• m

p 1 1 / P h a s e 3 : D e s ig n a n d C

• n

fi g u r a ti

• n

C

• m

p le t e 9 / 1 3 P h a s e 2 : S W S e le c t i

• n

7 /1 2 P h a s e 1 : P la n A p p r

• v

e d 6 /7 P h a s e : P r

• je

c t R e v ie w 5 /1 7 S e l e c t S

• f

t w a r e P r

• j

e c t L e a d L I M S T e s t A p p lic a t i

• n

C

• P

r

• je

c t M a n a g e r L I M S C

• r

e W e e k ly P r

• je

c t M e e t i n g 1 W

• r

k D a y s S

• ft

w a r e R e c e iv e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7

• Gen. 1

LIMS Start 5-1-0 1 Initiate the Project LIMS Com ple 12-13

• Roll-Out

Validate Approve Validate Test Train CRP Configure Software Design System Plan the Project Go Live Phase 4: Validation Complete 11/1 Phase 3: Design and Configuration Com plete 9/13 Phase 2: SW Selection 7/12 Phase 1: Plan Approved 6/7 Phase 0: Project Review 5/17 Select Software P roject Lead LIMS Test LIMS QA Application Consultant LIMS Selection & I l t ti 5 Day Lag Project Manager LIMS Core Week ly Project Meeting Software Rece ived Core Lead Synced 07- 27 Synced 07- 27 L I M S S t a r t 5

• 1
• 1

I n i t i a t e t h e P r

• j

e c t V a l i d a t e T e s t T r a i n C R P C

• n

f i g u r e S

• f

t w a r e D e s i g n S y s t e m P l a n t h e P r

• j

e c t P h a s V a l i d a C

• m

p 1 1 / P h a s e 3 : D e s i g n a n d C

• n

fi g u r a ti

• n

C

• m

p l e t e 9 / 1 3 P h a s e 2 : S W S e l e c t i

• n

7 / 1 2 P h a s e 1 : P l a n A p p r

• v

e d 6 / 7 P h a s e : P r

• j

e c t R e v i e w 5 / 1 7 S e l e c t S

• f

t w a r e P r

• j

e c t L e a d L I M A p p l i c a t i

• n

C

• n

s u l t a n t P r

• j

e c t M a n a g e r L I M S C

• r

e W e e k l y P r

• j

e c t M e e t i n g S

• ft

w a r e R e c e i v e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7 L I M S S t a r t 5

• 1
• 1

I n i t i a t e t h e P r

• j

e c t V a l i d a t e T e s t T r a i n C R P C

• n

f i g u r e S

• f

t w a r e D e s i g n S y s t e m P l a n t h e P r

• j

e c t P h a s V a l i d a C

• m

p 1 1 / P h a s e 3 : D e s i g n a n d C

• n

fi g u r a ti

• n

C

• m

p l e t e 9 / 1 3 P h a s e 2 : S W S e l e c t i

• n

7 / 1 2 P h a s e 1 : P l a n A p p r

• v

e d 6 / 7 P h a s e : P r

• j

e c t R e v i e w 5 / 1 7 S e l e c t S

• f

t w a r e P r

• j

e c t L e a d L I M A p p l i c a t i

• n

C

• n

s u l t a n t P r

• j

e c t M a n a g e r L I M S C

• r

e W e e k l y P r

• j

e c t M e e t i n g S

• ft

w a r e R e c e i v e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7

• Gen. 2

L I M S S t a r t 5

• 1
• 1

I n it ia t e t h e P r

• je

c t V a li d a t e T e s t T r a in C R P C

• n

f i g u r e S

• f

t w a r e D e s ig n S y s t e m P l a n t h e P r

• j

e c t P h a s V a lid a C

• m

p 1 1 / P h a s e 3 : D e s ig n a n d C

• n

fi g u r a ti

• n

C

• m

p le t e 9 / 1 3 P h a s e 2 : S W S e le c t i

• n

7 /1 2 P h a s e 1 : P la n A p p r

• v

e d 6 /7 P h a s e : P r

• je

c t R e v ie w 5 /1 7 S e l e c t S

• f

t w a r e P r

• je

c t L e a d L I M S T e s t A p p lic a t io n C

• P

r

• je

c t M a n a g e r L IM S C

• r

e W e e k ly P r

• je

c t M e e t in g 1 W

• r

k D a y s S

• ft

w a r e R e c e iv e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7 L IM S S t a r t 5

• 1
• 1

In i t ia t e t h e P r

• je

c t V a l id a t e T e s t T r a i n C R P C

• n

f i g u r e S

• f

t w a r e D e s i g n S y s t e m P l a n t h e P r

• je

c t P h a s V a l id a C

• m

p 1 1 / P h a s e 3 : D e s ig n a n d C

• n

f ig u r a t io n C

• m

p l e t e 9 / 1 3 P h a s e 2 : S W S e le c t i

• n

7 /1 2 P h a s e 1 : P l a n A p p r

• v

e d 6 / 7 P h a s e : P r

• j

e c t R e v ie w 5 / 1 7 S e le c t S

• f

t w a r e P r

• j

e c t L e a d L I M S T e s t A p p lic a t io n C

• P

r

• j

e c t M a n a g e r L IM S C

• r

e W e e k ly P r

• j

e c t M e e t in g 1 W

• r

k D a y s S

• f

t w a r e R e c e iv e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7 M e e t i n g 1 L IM S S t a r t 5

• 1
• 1

In i t ia t e t h e P r

• je

c t V a l id a t e T e s t T r a i n C R P C

• n

f i g u r e S

• f

t w a r e D e s i g n S y s t e m P l a n t h e P r

• je

c t P h a s V a l id a C

• m

p 1 1 / P h a s e 3 : D e s ig n a n d C

• n

f ig u r a t io n C

• m

p l e t e 9 / 1 3 P h a s e 2 : S W S e le c t i

• n

7 /1 2 P h a s e 1 : P l a n A p p r

• v

e d 6 / 7 P h a s e : P r

• j

e c t R e v ie w 5 / 1 7 S e le c t S

• f

t w a r e P r

• j

e c t L e a d L IM S T e s t L IM S A p p l ic a t io n C

• n

s u lt a n t P r

• j

e c t M a n a g e r L IM S C

• r

e W e e k ly P r

• j

e c t M e e t in g 1 W

• r

k D a y s S

• f

t w a r e R e c e iv e d C

• r

e L e a d S y n c e d 7

• 2

7 S y n c e d 7

• 2

7 M e e t i n g 1

• Gen. 0

Introduction Objectives Methodology

SLIDE 7

June 30, 2004 Organizational Design Optimization 7 Intro Objectives Methodology Results Introduction Conclusions

VDT Case Study: VDT Case Study: Design-Build Biotech Plant Case Design-Build Biotech Plant Case

Objective

Shorten the simulation

duration while maintaining acceptable quality risk

Acceptable interventions:

Increase the skill level (from

low to medium, or medium to high) for any one skill for any

• ne actor.

Add a total of up to 3 FTE’s in

increments of not less than 0.5 FTE to any combination of actors.

Change levels of

centralization, formalization, or matrix strength

Introduction Objectives Methodology Results

S tru ctu ral D esig n S u b team

Start Long Lead Purchas ing A pply for E xc Perm it Seek Z

• ning

Variance Provide G M P Select K ey Subs Project C

• ordination

E s tim ate C

• s

t G M P A ccepted

Projec t C

• ord

ination M eeting

C hoos e C

• ns

truction M ethods Select Subcons ultants A rch Program E s tim ate Tim e D efine Scope C hoos e facade m aterials C hoos e Struct. Sys tem D es ign C

• ordination

A rch itectu ral D esig n S ub tea m D esig n P M C lien t P M P ro ject E ng in eers C

• n

struc tio n P M P ro cu rem en t S u b team R ead y to E xcavate

SLIDE 8

June 30, 2004 Organizational Design Optimization 8 Intro Objectives Methodology Results Introduction Conclusions

A Plausible Fitness Function for this Problem = SPD + TFTE * FTEW + (FRI i * FRIWi + PRI i * PRIWi + CRi * CRWi) Where

SPD = Simulated Project Duration TFTE = the Total FTE added FTEW = FTE Weight ( if TFTE > 3.0 = > equals 1000 otherwise 1) FRI(i) = Functional Risk Index for activity i FRIW(i) = FRI weight for activity i (if FRI(i) > 0.5 = > equals 1000 otherwise 1) PRI(i) = Project Risk Index for activity i PRIW(i) = PRI weight for activity i (if PRI(i) > 0.5 = > equals 1000 otherwise 1) CR(i) = Communication Risk for activity i CRW(i) = CR weight for activity i (if CR(i) > 0.5 = > equals 1000 otherwise 1) M = maximum number of activities

Fitness Function

∑

= M i 1 Introduction Objectives Methodology Results

SLIDE 9

June 30, 2004 Organizational Design Optimization 9 Intro Objectives Methodology Results Introduction Conclusions

Transforming Genetic Tree

• 1st Skill of Actor P1 is increased by
• ne level
• 2nd skill of P1 is decreased by one

level

• 3rd skill of P1 is increase by one

level Activities of P3 and P5 are swapped

• Centralization is increased by one level
• Formalization is decreased by one level
• Matrix Strength is increased by one level

FTE is increased by 0.5 Introduction Objectives Methodology Results

SLIDE 10

June 30, 2004 Organizational Design Optimization 10 Intro Objectives Methodology Results Introduction Conclusions

Actors Skill Levels

S truc tura l De s ign S ubte a m

Start Lo n g Lead Pu rch as in g A p p ly Exc Permit Seek Z o n in g Varian ce Pro v id e GM P Select Key Su b s Pro ject Co o rd in atio n Es timate Co s t GM P A ccep ted Pro ject Co o rd in atio n M eetin g Ch o o s e Co n s tru ctio n M eth o d s Select Su b co n s u ltan ts A rch Pro g ram Es timate T ime Defin e Sco p e Ch o o s e facad e materials Ch o o s e Stru ct. Sy s tem Des ig n Co o rd in atio n

Arc hite c tura l De s ign S ubte a m De s ign P M Clie nt P M P roje c t Engine e rs Cons truc tion P M P roc ure m e nt S ubte a m Re a dy to Ex c a va te

• Generic
• Structural
• Generic
• Design

Coordination

• Mechanical
• Architectural
• Biotechnology
• Generic
• Architectural
• Management
• Biotechnology
• Generic
• Constructability
• Mechanical
• Geotechnical
• Cost Engineering
• Generic
• Biotechnology
• Local Politics
• Management
• Mechanical
• Generic
• Purchasing
• Negotiation
• Electrical
• Mechanical
• Generic
• Management
• Negotiation
• Local Politics
• Constructability
• Scheduling
• Mechanical
• Cost Engineering
• Total of 29 skills for 7 Actors
• Each skill can be set to 3 levels of low, medium, high
• Total Number of combinations = 329 = 6.8 X 1013

Introduction Objectives Methodology Results

SLIDE 11

June 30, 2004 Organizational Design Optimization 11 Intro Objectives Methodology Results Introduction Conclusions

Optimizing Actors Skill Levels

Genetic Tree Set up

• Terminal Sets = { Up, Down, Same}
• Function Sets = { P1..P7}
• Population size M = 100
• Maximum number of generations = 50
• Crossover = 90% Mutation = 3% Reproduction = 7%

Introduction Objectives Results Methodology

Best I ndividual found after 16 generations Was it the optimal solution?!

No – But it was pretty close: (Both reduced schedule by 69 days)

⌧Optimal solution Simulation Project End = 1/17/2001 8:29AM ⌧GP near-optimal Simulation Project End = 1/17/2001 2:25PM ⌧ Difference:

• Skill 4 (Geotechnical) of Project Engineer increased from medium to high

SLIDE 12

June 30, 2004 Organizational Design Optimization 12 Intro Objectives Methodology Results Introduction Conclusions

Best Individual of Generation 16

(Up P4 (Same (Same P1 P2) (Up (Up P5 (Same (Same (Up (Same P3 P1) (Up P6 (Up P4 (Up (Down P3 P2) P0)))) (Up (Down P3 P2) P0)) (Same (Up P0 (Up P0 (Down P3 P2))) (Up (Down P5 P2) P0)))) P0))) Introduction Objectives Methodology Results

SLIDE 13

June 30, 2004 Organizational Design Optimization 13 Intro Objectives Methodology Results Introduction Conclusions

Optimization Using Reassignment, Attention Allocation and Actors’ FTEs

Genetic Tree Set up

Terminal Sets = { Up, Down, Same, FTE, Assign, Aloc} Function Sets = { P1..P7} Population size M = 3000 Maximum number of generations = 100 Crossover = 90% Mutation = 3% Reproduction = 7%

Introduction Objectives Methodology Results

10 20 30 40 50 60 70 Human GP # o f d ays

75 77

Best I ndividual found after 21 generations Found Best Solution Ever!

Student/Manager Simulated End date= Dec 7, 2000 GP Solution Simulated End date = Dec 5, 2000

GP Solution:

Matched FTE additions additions in same location & same quantity Found additional Reassignment + changes in Attention Allocation

SLIDE 14

June 30, 2004 Organizational Design Optimization 14 Intro Objectives Methodology Results Introduction Conclusions

Comparing Assignments Comparing Assignments and FTEs Increments and FTEs Increments (GP vs. Student Sol (GP vs. Student Solution) tion)

S truc tura l De s ign S ubte a m

Start Lo n g Lead Pu rch as in g A p p ly Exc Permit Seek Z o n in g Varian ce Pro v id e GM P Select Key Su b s Pro ject Co o rd in atio n Es timate Co s t GM P A ccep ted Pro ject Co o rd in atio n M eetin g Ch o o s e Co n s tru ctio n M eth o d s Select Su b co n s u ltan ts A rch Pro g ram Es timate T ime Defin e Sco p e Ch o o s e facad e materials Ch o o s e Stru ct. Sy s tem Des ig n Co o rd in atio n

Arc hite c tura l De s ign S ubte a m De s ign P M Clie nt P M P roje c t Engine e rs Cons truc tion P M P roc ure m e nt S ubte a m Re a dy to Ex c a va te

0.5 1.0 0.5 0.5 0.5 0.5 1.0 0.5 0.5 0.5 GP Students 1.0 1.4 .75 1.2 Introduction Methodology Results Objectives

SLIDE 15

June 30, 2004 Organizational Design Optimization 15 Intro Objectives Methodology Results Introduction Conclusions

Project Duration Improvement Project Duration Improvement Before and After Intervention Before and After Intervention

Introduction Methodology Results Objectives

SLIDE 16

June 30, 2004 Organizational Design Optimization 16 Intro Objectives Methodology Results Introduction Conclusions

Quality Risk Improvement Quality Risk Improvement Before and After Intervention Before and After Intervention

Introduction Methodology Results Objectives

SLIDE 17

June 30, 2004 Organizational Design Optimization 17 Intro Objectives Methodology Results Introduction Conclusions

Fitness Improvement thru Generations Fitness Improvement thru Generations

27000 32000 37000 42000 47000 52000 57000 62000 1 6 11 16 21 26 31 Generation Numbers Fitness Value

Best Fitness Value Mean Fitness

Introduction Objectives Methodology Results

SLIDE 18

June 30, 2004 Organizational Design Optimization 18 Intro Objectives Methodology Results Introduction Conclusions

Conclusion & Future Research

Methodology

Conclusion

GP post processor for VDT beats the best human

trial-and-error performance of > 40 graduate student & practitioner teams over the past 8 years

Future Research

Develop new “Micro-Contingency Organization

Theories Using GP Optimizer

Integrate GP Optimizer into VDT Add additional variables to optimize as VDT is

extended to model impacts of cultural differences in global projects

Introduction Objectives Conclusions Results

SLIDE 19

June 30, 2004 Organizational Design Optimization 19 Intro Objectives Methodology Results Introduction Conclusions