Design Structure Matrix PDF
Design Structure Matrix PDF
Design Structure Matrix PDF
Development Process
by
Yonas Nebiyeloul-Kifle
at the
Massachusetts Institute of Technology
May 2005
The author hereby grants to MIT permission to reproduce and to distribute publicly paper
and electronic copies of this thesis document in whole or in part.
Signature of
Author A/ I _ -- -
( 6t an School o'fManagement
May, 2005
Certified
by
, -
cL1
Steven Eppinger, Thesis Advisor
Deputy Dean, Sloan School of Management
Accepted
by.
INSTITUTE
MASSACHUSETTS David Cappupo, Executive Director of A Program
Sloan Schoof Management
MASSACHUSETTS INSTITUTE
OF TECHNOLOGY
SEP 0 1 2005
"4i·SVO
4n-
LIBRARIES
Application of the Design Structure Matrix to Integrated Product
Development Process
by
Yonas Nebiyeloul-Kifle
Abstract
2
Acknowledgements
Special thanks are also in order for my thesis advisor, Professor Steven Eppinger. I have
learned much from his guidance and his pioneering work in DSM research. The fresh
insights and valuable feedback he provided have benefited my analyses and broadened
my understanding of the key management concerns involved in the product development
process. Working with him has been a greatly inspiring and enjoyable experience.
On a personal note, I would like to thank my mother, wife, sister in-law and three
children who have made my time at Sloan exciting and rewarding. Thank you for your
love, affection and unwavering support.
3
Table of Contents
4
1. Introduction
Product development process is the new frontier for achieving competitive advantage in
today's rapidly changing business environment. The product development process itself
particularly true for defense contractors that have to derive significant value from their
product development process is the Design Structure Matrix (DSM). This tool provides
critical capability to model the information dynamics of the development process. This
The thesis starts with a brief background to the product development process and goes on
to discuss the methodology employed to create the baseline DSM model for the project
discusses the analysis performed using DSM partitioning and concludes with a
presentation of the improved project planning and management phase of the integrated
5
2. The product development process
A typical integrated product development process entails six stages with several layers of
hierarchical sub-processes within each phase or stage. Figure 1 presents the typical
Wl
= I r.
E .i
to a, 0~
, C*3
0 to
0
W
a
U
"I
'a4 a0 rjA
a) 0C,) v
'5 P4
0I 3C).I
The 'planning stage', also known as the 'program management and planning phase'
entails seven sub-processes primarily dealing with the project planning activities and start
up gate reviews. This stage is selected for DSM application because it contains a high
degree of coupled activities and can thus benefit significantly from DSM applications.
activities/iterations since they lack the critical insights required. Eppinger (2001) drew
attention to the critical inadequacies of such traditional project management tools. DSM
6
processing unit that receives inputs from previous tasks and transforms these into suitable
information for subsequent tasks downstream. Mapping these dynamics provides the
critical insights required to deal with iterations. DSM affords this capability by modeling
the information flows rather than the work flows. However, modeling the information
exchange is not a trivial task. For example, capturing the planning phase process
information exchange dynamics using network diagrams would require several hundreds
of pages of block diagrams. DSM facilitates this task too, i.e. by providing a matrix
3. DSM fundamentals
A Design Structure Matrix (DSM) is a square matrix with identical number of rows and
tool for decomposition and analysis of complex processes. Most notably, it has proved its
One popular type of DSM is an activity-based DSM. An activity-based DSM shows the
interactions of each activity in a given process with every other activity in that process.
row reveals all those activities upon which the particular activity represented in that row
'There are several types of DSM. Examples include activity-based DSM, parameter-based DSM,
component-based DSM and parameter-based DSM. Additional information on DSM can be found at
http://cipd.mit.edu and http://www.dsm.org
7
depends for information; reading down a column reveals all activities to which the
information feedback. Thus, the DSM provides a snapshot of information flow in the
process.
A 18 IC ID IE F IG
Activity A _ X
Activity BX X X X
Activity
Activity
Activity E X I
Activity FX X
Activity GXXX
The key to understanding activity-based DSM lies in grasping the concept of iteration
and the difference between serial, parallel and coupled activities. In many design
processes, much re-work results from poorly sequenced activities where outputs of
activities are not available or coordinated in a timely fashion to other activities in the
process. Still more re-work stems from highly coupled activities where mutually
dependent teams could converge to an acceptable solution only through iteration. The
first step towards reducing development time and variation is to minimize or eliminate
iterations in the process. This can be achieved by improving the activity sequencing such
that the information required by an activity is available when it is time to perform that
particular activity. Reordering the rows and columns of the descriptive DSM reveals a
prescriptive DSM that minimizes feedback points in the process. Therefore, the goal of
DSM partitioning or sequencing becomes an effort to get as many of the interfaces below
8
the diagonal of the DSM as possible. The activity-based DSM techniques are discussed
The use of I)SM in both research and industrial practice increased greatly in the 1990s.
Kusiak and Wang (1993) demonstrate the use of an activity-based DSM to restructure the
automobile design process. Browning (1998) applied DSM at Boeing Information, Space
and Defense Systems Group to model and analyze cost, schedule and performance for the
Uninhabited Combat Air Vehicle program. Still others have applied and expanded the
(1998) at NASA have developed a web-based DSM tool for system monitoring and
the alignment of the interaction patterns between the product, process and the
Modeling the information dynamics of the product development process is thus essential
for managing the complexity involved. The first task in studying complexity is to
requires efficient techniques to capture the dependencies involved. DSM provides this
partitioning.
9
Construction of the baseline DSM model for the planning stage was performed using a
hierarchical DSM simulator. The hierarchical DSM simulator provided the capability
required to capture the structure of the integrated product development process. The data
and the input/output interdependencies for the planning stage of the integrated product
development process were gathered from process experts through Raytheon's integrated
format the information for the simulator application and some activities were re-labeled
The baseline DSM model depicted the activities, information exchange or flow using an
"X" mark. Figure 3 presents the baseline DSM model for the planning phase of the
integrated product development process. The model captures the information exchanges
including the iterative dependencies. Reading across a row shows the information inputs
required to complete the corresponding activity, and reading down the column shows the
requires recognition of the iterative information flows and an appreciation of the inherent
according to the particular temporal role played by information exchange in the execution
10
C) 2 =Requires the input information but the activity could be executed with a
The baseline DSM model with the I/O characterization is presented in Figure 5. The
baseline DSM also captured the first level hierarchy of the planning phase to accurately
model the process structure in the integrated product development process. For
The modeling activity could easily be expanded to analyze the development process
Identifying iterations or feedback points and appropriately planning for their management
provides a crucial means for improving quality in product development. Iterations could
techniques or decreasing the coupling by making fewer feedback loops. Analysis of the
baseline DSM model revealed the information exchange points that involve critical
iterations or feedback loops. These could be easily identified from the DSM model
(presented in Fig. 4) because they lie above the diagonal. The diagonal separates the
feed-forward information exchange (below the diagonal) from the feedback information
that lies above the diagonal. The feedback points denote that information from the
subsequent activity may force a re-work of the preceding activity or may trigger delays.
One can also ascertain the sequential, parallel and coupled activities or feedback. Figure
4 presents these observations and the major feedback loops that are the source of
11
The baseline model also provides insights on activity dependencies across sub-processes
is difficult to determine activity duration and costs. Consider for example, the Hardware
Engineering activity. Since this activity depends on information from various sources,
difficult to estimate its exact duration or cost. This complexity creates difficulty for most
mangers to adequately comprehend and plan for the actual efforts and resources required.
resource allocation. This effect is clearly demonstrated in the project planning and
management phase where full resource loading is required for the entire duration of the
planning phase. Figure 5 presents the three activities and their overall duration in the
process.
The insights gained from the baseline DSM model are also used to identify the activities
that have the most impact on activities downstream. This is achieved by simply reading
down the columns of the activities. Accordingly, activities are classified as follows:
A) Group A: activities with output information that feed other activities in more than
5 sub-processes.
B) Group B: activities with output information that feed other activities in more than
2 sub-processes.
12
C) Group C: activities with output information that feed other activities in less than
three sub-processes.
The letters A, B and C are used to denote these group classifications and are presented in
Improving the project planning phase of the integrated product development process
triangular form provides an optimal architecture for the process since it eliminates and
While product development is essentially iterative, unplanned iterations cost money and
time to market. Iterations occur primarily because of new information from activities
Consider for example, the dependency of 'logistics support' and 'scope management'
activities, labeled "0" and the dependency of 'process tailoring' and 'metrics planning'
activities dependency labeled "1". The schematic of the baseline model presented in
Figure 6A highlights these feedback points. It becomes evident from studying these
input/output dynamics that the dependency assigned a "O" value needs to be below the
13
diagonal and the dependency with "1" needs to be close to the diagonal. This modified
strategy can provide the capability needed to execute logistics support without having to
wait on 'scope management' activity to complete its work. This speeds the process and
also allows 'scope management' to complete its work earlier. 'Metrics planning' and
'process tailoring' are grouped together enabling concurrent execution. This improves
coordination and resource efficiency. The results are presented in Figure 6B as part of the
The insights gained from the grouping strategy based on output information also helped
in determining the sequencing approaches. For example, the activity 'gate review' and
'contract baseline' presented in Figure 8 are classified as group A activities with impacts
on activities in more than five sub-processes. These activities are therefore, architected to
be performed earlier in the process. As the name suggests, the 'gate review' activity
represents the check points within the project management and planning process. There is
activities in the process. That is, for example, 'Hardware Engineering' under the baseline
model has to wait until the 'gate review' activity reaches completion. In the improved
earlier since it could independently coordinate its needs. This saves money by speeding
up the development process and most importantly, also affords the flexibility for
14
'Subcontract Material' and 'Systems Engineering' activities are grouped together. It is
crucial to start these coupled activities as early as possible because they span the entire
project management and planning phase. This also provides flexibility to execute and
this thesis holds that these activities warrant special attention as they provide additional
opportunity for improvement. The critical information they demand is from 'financial'
planning activity. Prudent project management thus demands intimate attention to this
grouped activity. Figure 10 presents the iteration points that need special attention after
The proposed improved process contains less iterative dynamics, hence is faster and more
reliable. This could also be demonstrated using Browning's 1998 model that presents
DSM based methodology for analyzing cost, schedule and performance in complex
and Monte Carlo simulation to predict the distribution of possible project duration. The
application involves assigning estimates to both re-work probabilities due to iteration and
demonstrate the performance of the improved proposed process vis a vis the baseline
process, the analysis for this thesis identified coupled activities in the improved process2 .
2This was done merely to show the 'worst case scenarios' in the improved process and for simulation and
parameter estimation convenience. It should be noted that the proposed improvement has eliminated a
15
The same parameter estimates were subsequently applied for the probability, impact and
overlap matrix for both the baseline DSM model and improved DSM model. The
activities selected are presented in Figures 12A and 12B which highlight the coupling
effects which made them ideal candidates for performance evaluation. For ease of
reference, the selected activities and their I/O dynamics DSM models are presented
Baseline OSM
Activity 1/0 dynamics Tag 17 18202223
Systems Engineering - Process ID 2-5.01 17 1 1 1
Hardware Engineering - Process ID 2-5.02 18 0 1
Specialty Engineering - Process ID 2-5.04 20 0 1
Product / System Integration - Process ID 2-5.06
Verification and Validation - Process ID 2-5.07
22
23
0
0
0
0
*
0
0
The next step involved crafting the DSM probability matrix. That is, for each
significant number of feedback points. This significant improvement in overall performance can be readily
appreciated by selecting activities with no feedback in the improved process and comparing them with the
baseline process where they previously exhibited considerable feedback.
16
activity in the process. Figure 14A and B present the probability estimates applied to
characterize the DSM probability matrix. For example, the circled 0.1 estimate in Figure
14A indicates that there is an estimated 10% chance of completion of 'Verification and
Validation' planning or that uncoordinated information exchange will cause at least some
re-work of 'Hardware Engineering' planning. In general, numbers above the diagonal are
estimates of the likelihood or the odds of feedback from one activity to another. The
numbers below the diagonal have similar characteristics, but represent the chances of a
be re-worked there would be a 10% chance that 'Specialty Engineering' planning would
Baseline DSM
Probability Matrix Ta 17 18 20 22 23 eedback
Systems Engineering - Process ID 2-5.01 17 0.10 o.o1 0.10
Hardware Engineering - Process ID 2-5.02 18
Specialty Engineering - Process ID 2-5.04 20 0.00 0.10 0.05
Product / System Integration - Process ID 2-5.06 22 0.10 0.10 0.20
Verification and Validation - Process ID 2-5.07 23 0.20 0.20 0.20
17
The extent of re-planning caused or that had to be performed for each probability is also
captured using DSM impact structure as presented in Figures 15A and B. For example,
an estimated 20% of the planning for 'Hardware Engineering' would have to be re-
worked or re-done if the projected 10% likelihood of the iteration caused by the
Baseline DSM
Impact Matrix Tag 17 18 201 22 2 3
Systems Engineering - Process ID 2-5.01 17 0.1 0.0 0.2
Hardware Engineering - Process ID 2-5.02 18 0.2 )
Specialty Engineering - Process ID 2-5.04 20 0.0 0.2 0.0
Product / System Integration - Process ID 2-5.06 22 0.2 0.2 0.4
Verification and Validation - Process ID 2-5.07 23 o.i o.1 o.1
In addition, it is important to note that this thesis also considered the effects of learning
that come with the implementation of improved processes. The assumption made for the
purposes of this analysis, is that the probability of re-working will fall by a constant
fraction with each iteration3 . To adequately simulate the dynamic nature of learning this
3 Although it should be noted that Smith and Eppinger (1997) hold that it is not always readily apparent
whether re-work probability would increase or decrease with iterations of a task.
18
thesis employed Cory's (2001) modified version of Browning's (1998) model4 . For
example, for learning curve (LC) parameter 0.2 and initial iteration probability 0.1, the
Probabilityof iteration/rework
0.12 -
01
p - PfR(tn)= 1.n
X .8'
0.08
0
0.
.2
CL
0
0.06.
a,
ti
0 0 04
a,
002
0
0 1 2 3 4 5 6 7 8
Numberof iteration
I-- Probability
of iteration/rework]
The learning curve estimates utilized for performance analysis are presented in Figure 19
with the duration parameters simulated. The last step in characterizing the DSM structure
for performance analysis involved capturing the amount of overlap between the activities.
To simulate the effects of concurrency between the activities, all elements of the DSM
structure for the overlap matrix are assigned a zero value. That is, for example, 0% of
19
Engineering'. Figures 17A and B present the DSM overlap matrix for the baseline
Baseline DSM
Overlap Matrix Tag 18jl 201 221 231
Systems Engineering - Process ID 2-5.01 _l
_
17 .l
., ._
_ .
The amount of impact on the overlapped portion of the activities is assumed to be the
same as the impact matrix parameters since the activities completely overlap. Browning
(1998) showed that activity durations could be modeled as random variables with
triangular distributions. Estimates of the most likely duration value (MLV), best case
duration value (BCV), and worst case duration value (WCV) provide the endpoints of
20
_
Ec
Figure 19 presents the duration and the learning curve parameters applied. The sum of
these randomly chosen activity durations will constitute one possible outcome for the
total duration for these activities. Numerous repeats will yield the probability distribution
fimction (PDF) for the duration. 5 Accordingly, Mote Carlo simulation 6 was adopted by
this thesis, with a view to demonstrating the performance of the selected activity
durations in the proposed improved process, in comparison with the baseline process.
A total of 1000 simulation runs were performed to compare the distribution of the
5 Total process duration can also be shown using cumulative distribution function (CDF).
6 This thesis employed Soo-Haeng Cho's (2001) Microsoft implementation that drew on much of
Browning's work, including application of Latin Hypercube Mote Carlo simulation to capture the skewed
right hand shape of the durations.
21
demonstrated that the proposed improved process has slightly higher duration mean but it
140 - - --- - 1
0.9
120
0.8
2 100
AR age 27.6 0.7
Idad Diation 1.8 --
80 '--' -A(
"Is------ , ar 0.6
0.5 R
60 --------
0.4
* 40 . . ..
o0.3
=iEI 0.2
20 . · .- - ·
.-- · - · .-
0 - -'- I I I I -
.ft
J#HHIIII
I i :. - '- ' '- ' ' - · - · f ,, , I I
0.1
0
19.1 21.2 23.4 25.5 27.7 29.8 320 34.1 36.3 38.4
Duration in person moh
IaJ - 7~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.
iz~, Iy-
0.9
100
0.8
2 80 0.7 1
0.6
pe27.8
a -0.5
id DeVation0.7
0.4
* 40 I I . I. I
- 0.3 3
20 0.2
- 0.1
0 ,i i- - !Ii-i-; ; i 1W . 0
19.4 21.3 23.3 25.2 27.1 29.0 3Q9 328 34.7 36.6
Duartion in personmoth
22
Although the improved process has a higher mean it resulted in significant reduction in
activity duration variability. It is worth reiterating here that the performance evaluation
was done with a select few activities that exhibit feedback characteristics in the improved
process. Therefore, it is evident the overall improved process will exhibit a much reduced
average duration and variability, given that the improved process delivered significantly
7. Conclusions
competitive technology firms. In the future, it is unlikely that complex system products
will compete solely on the basis of technical performance (Browning, 1998). The
required amount of sophistication of project planning and control for complex process
complexity management is both essential and strategic. Balancing all aspects of the
analytical tool. This thesis presented DSM methodology and its application for process
improvement.
The activity-based DSM applied for the project planning and management phase
feedbacks and potential iterations they can cause. The thesis performed deeper analysis of
the interdependencies by classifying the I/O dynamics and the output characteristics of
the activities., The insights gained and the partitioned performed rendered an improved
process for the project management and planning phase of the integrated product
23
development process. For example, under the current or the baseline process the
'hardware engineering' planning team had to wait for the 'metrics planning' team to
define their measurement and analysis strategy for their project. The proposed improved
process eliminates this delay in time and the possible iterations because it proposes to
execute the 'metrics planning' earlier in the process. This results in the delivery or
exchange of the program's measurement and analysis plan to the 'Hardware Engineering'
planning activity in a timely fashion. The proposed improved process also eliminates the
Development' planning. Under the proposed improved process, the 'Data Management'
planning team defines the critical, often contractual, data exchange format for software
code archiving and exchange earlier on in the project planning and management phase,
thus enabling the 'Software Development' planning team to plan appropriately for
specifications without delay or iteration. Furthermore, the improved process proposes that
critical 'Risk and Opportunity (R&O) planning' activity be executed before 'Financial
Planning' activity. That is, when the 'Risk and Opportunity planning' team establishes
and delivers the methods for classifying and tracking R&O the 'Financial planning' team
is able to define the accounting procedures for the R&O without delay or iteration.
Thus, the improved process delivered through the application of DSM techniques saves
both money and time. In addition, it addresses a number of typical weaknesses in the
current project management and planning process. It eliminated several iterative loops by
partitioning the matrix and sequencing the activities. For example, activities within the
24
'program detail' and planning sub processes could now be executed without any iterative
loops. The application identified key control points and efficient coordination structures.
The improved process is more robust and allows planners to adjust the process to meet
customer expectations.
If appropriately aligned with the organizational structure of the team supporting the
project management and planning phase, the optimized process architecture could
potentially result in further improvements. Indeed, Eppinger and Salminen (2001) have
shown the value of DSM application in three areas of the development process - the
The findings of this thesis also helped highlight that the contemporary emphasis on
improvement could result in increased coupling and iterations. Research has shown that
concurrency becomes detrimental (AitSahlia et al., 1995). Indeed, some models have
shown that the goal of process improvement should not be focused on concurrency alone.
The objective should be to achieve optimization in all aspects of the development process
25
8. Analyses
26
(3;
a)
:
bju
.5
a
9
c)
o
(N
aa
k
m
a)
7:$
-o
Cd
a
CLI
c)
0
0o
o(
ra
0
.N
c)
c)
.a ri
3
C
L;)
ca
0
vt:
0a)
ao
0
O
C#2
ous
S
a) .)
cn
cn
CI)
cT +1
acn
©
o
o1)
0
rd
a) a)
CZ)
C.)
0
a)
.- 4
0a),
C
blo
ct
vc)
Q
.E
c)
.bb
o0ci
C
u
ca)
a)
a
o
a)
0
o a)
a.)
c)
a.
L;
v)
ua)
*E)
t~o
Cd
0
so
C3
C
E
a
0
0
c)
0
03
Q
a)
c)
'e
To
.-0
Q
a)
;t4
o
cot
(V
N1
- ' -
- - 0
0
o00 m
°)0~~ ' . S° C)
CO - 0 . 0
-
(0
e- - ALi , 04 0 0 0 o
* lo
.- -' 0
,:ta
a- .
·
. o mF =
- -
0.
CD 0 o
N " o 0 L;
0
(0 - 0 0000000 00 0
eO 0 0 0 0 0 0
0
ia ./'
cn
U 00-
-
-00 0 -
CJ, -
r-o 0
04a.-_-
a a aa0CD0 CC
UC 0 0
o800 0ic
~~r~pnio~o
_-o g, - -.-a -(
P 0, d~;lgl~J~l
00 o-~ 00Cm, ..
D rja~g
a_
EirjlpN18m
2
I
o
o
E
o
i '
I,-Z
5 (]1- ~ a CC 1 e
0o
0:L
.--e
E
z
a)
.e
.'
a)
a)
a,
'5
V4
0
0)
ig
cd
u
a
a)
cn
03
a)
r.
a)
Cr,
a)
a
's
a.
CO
i
a)
172
cf
a
1)
To
a>
C
E
.-|
a.)
V)
CA
rV)
a)
E
c,
411
av
U
N-
*3
U)
a)
au
421
.=
u
U
C
.9
rc
U
U)
Cd
U
(14
I
r--
I
--.---- *---.--
...
I
.I
O
I- i ~- I~~~~~~~~~~~~~~~~~~~ I
-
I
I
l I
I
I
x I
I
I
r-·
*X I
N'
X X xl
I
X I
X.
o I- - x l
X XX XXX c.)
I- · XXX XXX
I
I .X X
O
.XXXXXX XXX XX
o I-C.Z
II
iI o
> 0 ,-4
CI.
I X X X XX
x XXX XXX X
.XXXXXXXXXX XXXXX
o I'- XX XXX X
XX XX _X___X __
C.
II
.XI x XX X XXX X 1 5
r. 00
o . X X Ccn
5 o
. I .)
I;XX
x I XXX X XXX XX XX C
I
X I
. XX I
I
.X xxxxxxxlx xx XXXXXXX XXX X
X X
--
· XXX
(
. . .
XXX
-
1 ) co I-
- - -
Go 0)
3XXXX!XXXX
-·1
.-
O ,,
'I ' ' -
. -W . . . . . XXXXXXXXXX
X XX .XXXX . (0 -
-
2
I- C IIICI
a)
I ,-4
CO rI
Cb
NI
c)
Cr
0 i
0
i a~ 0
P1 (I
0 !
a- ci) 0c¢:
Cv oO
(U 3.
nr C.) ?1
T a-C 0
'E 0 CU
0
L9.T
9. References
AitSahlia, Farid, E. Johnson, and P. Will. 1995. "Is Concurrent Engineering Always a
Sensible Proposition?" IEEE Transactions on Engineering Management.
Browning, Tyson and S.D. Eppinger. 1998 "A Model for Development Project Cost and
Schedule Planning," M.I.T. Sloan School of Management, Cambridge, MA, Working Paper
no. 4050
Browning, Tyson R., 1998. "Modeling and Analyzing Cost, Schedule, and Performance in
Complex System Product Development", Doctoral Thesis (TMP), Massachusetts Institute of
Technology.
Carrascosa, M., S. Eppinger and D.E. Whitney. 1998 "Using the design structure matrix to
estimate product development time" ASME Design Engineering Technical Conferences
Eppinger, Steven. 2001. "Innovation at the speed of information" Harvard Business Review.
Kusiak, Andrew and J. Wang. 1993. "Decomposition of the Design Process," Journal of
Mechanical Design, Vol. 115
Kusiak, Andrew and J. Wang. 1993 "Efficient Organizing of Design Activities," International
Journal of' Production Research, Vol. 31.
Rogers, James L. 1997. "Reducing Design Cycle Time and Cost Through Process
Resequencing", Proceedings of the International Conference on Engineering Design,
Tampere, Finland.
Rogers, James L. and C. Bloebaum. 1994 "Ordering Design Tasks Based on Coupling
Strengths", 5th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and
Optimization, Panama City, Florida.
39
Rogers, James L., A.O. Salas and R.P. Weston. 1998. "A Web-Based Monitoring System for
Multidisciplinary Design Projects", in Proceedings of the Seventh
AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization,
St. Louis, MO.
Smith, R.P. and S.D. Eppinger. 1998. "Deciding between Sequential and Parallel tasks in
engineering Design," Concurrent Engineering: Research applications
Soo-Haeng Cho. 2001. "An Integrated Method for Managing Complex Engineering Projects
using the Design Structure Matrix advance Simulation" MS Thesis, Massachusetts Institute
of Technology
Welch, Cory. 2001. "Application of the Design Structure Matrix: A case study in process
improvement dynamics", MS. Thesis, Massachusetts Institute of Technology
40