tools for innovation: the design structure matrix

MITSloan

Tools For Innovation:The Design Structure Matrix

Thomas A. Roemerand Steven EppingerSummer 02, 15-761

MITSloan

OutlineOverview

Traditional Project Management Tools and Product Development

Design Structure Matrix (DSM) BasicsHow to createClassification

The Iteration Problem: Increasing Development SpeedSequencing, Partitioning and Simulation

The Integration Problem: DSM ClusteringOrganizational Structures & Product Architectures

MITSloan

Classical Project Management Tools

Gantt Charts

Graph-based: PERT, CPM, IDEF

Act

ivity

Time

MITSloan

CharacteristicsComplex DepictionFocus on Work Flows

DSM focuses on Information FlowsIgnore Iterations & Rework

Test results, Planned design reviews, Design mistakes, Coupled nature of the process

Decomposition & IntegrationAssume optimal Decomposition & StructureIntegration of Tasks not addressed

MITSloan

Design IterationIteration: The repetition of tasks due to new information.

Changes in input information (upstream)Update of shared assumptions (concurrent)Discovery of errors (downstream)

Fundamental in Product developmentOften times hidden

Understanding Iterations requires Visibility of information flows

MITSloan

Decomposition and Integration

Decomposition: Dividing a complex problem into smaller sub-problems.

Integration:Combining sub-problems to achieve set goals.

MITSloan

A Graph and its DSM

A C

D

B

H

F

G

E

I

A B C D E F G H IA A XB B XC X CD D XE EF X F XG GH X X HI X I

MITSloan

Creating a DSMDesign manualsProcess sheetsStructured expert interviews

Interview engineers and managersDetermine list of tasks or parametersAsk about inputs, outputs, strengths of interaction, etcEnter marks in matrixCheck with engineers and managers

Questionnaires

MITSloan

Four Types of DSMsIteration

Activity based DSMParameter based DSM

SequencingPartitioningSimulation

IntegrationTeam based DSMProduct Architecture DSM

Clustering

MITSloan

Iteration Focused Tools

Concepts, Examples, Solution Approaches

MITSloan

Sequencing Tasks in Projects

Possible Relationships between Tasks

A

B

A

B

A B

Dependent(Series)

Independent(Parallel)

Interdependent(Coupled)

MITSloan

Sequencing AlgorithmStep 1: Schedule tasks with empty rows firstStep 2: Delete the row and column for that taskStep 3: Repeat (Go to step 1)Step 4: Schedule tasks with empty columns lastStep 5: Delete the row and column for that taskStep 6: Repeat (Go to step 4)Step 7: All the tasks that are left unscheduled are coupled. Group them into blocks around the diagonal

MITSloan

Task Sequencing Example

Figure 4 at: http://faculty.erau.edu/ericksol/projects/futurspcrft/kristof/SPfinal.html

Space Shuttle Main Engine

MITSloan

Product Decomposition

Engine Components

MITSloan

Turbo Pump

MITSloan

Project Scope

Activity

SM

SW

W

M W

Activity

Activ

ity

II

I II

I I

Information

Activ

ity

M

M

I

II

Verify A-A & A-I Matrices

Verify A-A & A-I Matrices

Verify OA-

Diagram

Verify OA-

Diagram

Organized Activity Diagram

Questionnaire

I

Gate Reviews

II III IV

Stage Overlapping

...

ConsultKey Functions

and Clients


and Clients

Time Cost Trade-Offs


and Executives


and Executives

OpportunityCosts

OpportunityCosts

Final Process Structure

SensitivityAnalysis

SensitivityAnalysis

MITSloan

Corresponding LiteratureStructuring Product Development Processes

Ahmadi, Roemer & Wang, European J. of Operational Research 2001

Structuring Product Development ProcessesAhmadi, Roemer & Wang,

European J. of Operational Research 2001

Managing Development Risk in ProductDesign Processes

Ahmadi & Wang, Oper. Res. 1999`

Managing Development Risk in ProductDesign Processes

Ahmadi & Wang, Oper. Res. 1999`

Time-Cost Trade-Offs in Overlapped ProductDevelopment Processes

Roemer, Ahmadi & Wang, Oper. Res. 2000

Time-Cost Trade-Offs in Overlapped ProductDevelopment Processes

Roemer, Ahmadi & Wang, Oper. Res. 2000

...

Accelerating Product Development Roemer & Ahmadi, In review at Oper. Res.

Accelerating Product Development Roemer & Ahmadi, In review at Oper. Res.

MITSloan

Finding Coupled StagesBlock

Creation

Block Decomposition

Design Time Computation

MITSloan

Dependency Relations in Conceptual Design Block

For this graph, seeTime-Cost Trade-Offs in Overlapped ProductDevelopment Processes by Roemer, Ahmadi & Wang, Operations Research, 2000.

MITSloan

Block Decomposition

∑∈Aij

ijijij ynamin

∑=

∀=M

mim ix

1 ,1

mCxN

iim ,

1∀≤∑

=

mjiyxxM

mhijjhim ,, ,0

1∀≤−− ∑

+=

{ } mj iyx ijim ,, ,1,0, ∀∈

s.t.

i,j = index for activities, i,j = 1,2,…,N;

m = index for stages, m = 1,2,…,M;

A = the set of directed arcs in the design graph;

aij = the level of dependency of activity i on j

⎩⎨⎧ =

=otherwise1

1if number) large (a aWn ij

ij

⎩⎨⎧

=otherwise0

stage toassigned is activity if1 mixim

⎩⎨⎧

=otherwise1

stagesbetween back feed a is arc if0 ijyij

MITSloan

Resulting Structure for Conceptual Design Block

For this graph, seeTime-Cost Trade-Offs in Overlapped ProductDevelopment Processes by Roemer, Ahmadi & Wang, Operations Research, 2000.

MITSloan

Lead Time Estimation

( )[ ]∑−

=

−− −++=1

1

00/11i

ii

n

niiiii etttetT

µλβ

( ) 1−−= PIjiµ

∑ ==

k

i iTT1

Exp. Iterations

Activity Duration

Stage Duration

Markov Chain

.3

S

21 3

.2.5

.51

T

1 .5

MITSloan

Decomposition and Duration

0%

20%

40%

60%

80%

100%

0 5 10 15 20 25 30

Decomposition Degree (Number of Process Stages)

Blo

ck D

urat

ion

MITSloan

Stage Size and Duration

Number of Activities

1 2 3 4 5 6 7 8 9 10 11

Number of IterationsDuration

0%

200%

400%

600%

800%

1000%

05101520253035

Dur

atio

n In

crea

se

Num

ber o

f Ite

ratio

ns

MITSloan

Task A

Task B

Task C

X

X

X

DSM Simulation

Task A requires input from task CPerform A by assuming a value for C’s outputDeliver A’s output to BDeliver B’s output to CFeed C’s output back to A

Check initial assumption (made by A)

Update assumption and repeat task A.

MITSloan

Simulating Rework

Task A

Task B

Task C

X

X

R

R is the probability that Task A will be repeated once task C has finished its work.

R = 0.0 : There is 0 chance that A will be repeated based on results of task C.R = 1.0 : There is 100% probability that A will be repeated based on results of task C.

MITSloan

Simulating 2nd Order Rework

Task A

Task B

Task C

R2

X

X

Second Order rework is the rework associated with forward information flows that is triggered by feedback marks.

First order rework: Output of task C causes task A to do some rework2nd order rework: Consequently there is a chance tasks depending onA (e.g. task B) will also be repeated.

MITSloan

Simulating Rework Impact

Task A

Task B

Task C

X

X

I

I = 0.0 : If task A is reworked due to task C results, then 0% of task A’s initial duration will be repeated

I = 1.0 : If task A is reworked due to task C results, then 100% of task A’s initial duration will be repeated

MITSloan

Simulation Results

DSM contains rework probabilities and impactsCost and time add upMany runs produce a distribution of total time and costDifferent task sequences can be tried

Related reading: “Modeling and Analyzing Complex System Development Cost, Schedule, and Performance” Tyson R. Browning PhD Thesis, MIT A&A Dept., Dec 99

MITSloan

Gantt Chart with Iteration

Typical Gantt chart shows monotone progressActual project behavior includes tasks stopping, restarting, repeating and impacting other tasks

Related Reading: “Modeling and Analyzing Complex System Development Cost, Schedule,and Performance” Tyson R. Browning PhD Thesis, MIT A&A Dept., Dec 99

MITSloan

Lessons Learned: IterationDevelopment is inherently iterative.Understanding of coupling is essential.Iterations improve quality but consumes timeIteration can be accelerated through:

Information technology (faster iterations)Coordination techniques (faster iterations)Decreased coupling (fewer iterations)

Two Types of Iteration:Planned Iterations (getting it right the first time)Unplanned iterations (fixing it when it’s not right)

MITSloan

Integration Focused Tools

Concepts, Examples, Solution Approaches

MITSloan

Team SelectionTeam assignment is often opportunistic

“We just grab whoever is available.”

Not easy to tell who should be on a teamTradition groups people by functionInfo flow suggests different groupingsInfo gathered by asking people to record their interaction frequency with others

MITSloan

Clustering a DSM

A B C D E F GA AB BC CD DE EF FG G

A F E D B C GA AF FE ED DB BC CG G

HiNo Dependency Low

MITSloan

Alternative ArrangementOverlapped Teams



HiNo Dependency Low

MITSloan

GM’s Powertrain Division22 Development Teams into four System Teams

Short block: block, crankshaft, pistons, conn. rods, flywheel, lubricationValve train: cylinder head, camshaft and valve mechanism, water pump and coolingInduction: intake manifold, accessory drive, air cleaner, throttle body, fuel systemEmissions & electrical: Exhaust, EGR, EVAP, electrical system, electronics, ignition

MITSloan

Lessons Learned: IntegrationLarge development efforts require multiple activities to be performed in parallel.The many subsystems must be integrated to achieve an overall system solution.Mapping the information dependence reveals an underlying structure for system engineering.Organizations and architectures can be designed based upon this structure.

MITSloan

Conclusions

The DSM supports a major need in product development:

documenting information that is exchanged

It provides visually powerful means for designing, upgrading, and communicating product development activitiesIt has been used in industry successfully

MITSloan

DSM Web Pagehttp://web.mit.edu/dsm/It contains

A tutorialLinks to other DSM sitesOver 100 references of DSM literatureAnalysis tools

MITSloan

Suggested Reading

“Innovation at the Speed Of Information”

By Steven D. Eppinger

HBR January 2001

tools for innovation: the design structure matrix

Documents