2875 Fall 2004

Assembly System Design Techniques
Goals of this class
Introduce system design methods

Understand the things that must be considered
Look at two ways to approach it
Learn about SelectEquip
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
Assembly System Design Techniques
Assembly system design algorithms exist
They solve the Equipment Selection and Task

Assignment problem
Methods include dynamic programming,
travelling salesman, mixed integer-linear
programming, and a heuristic called ASDP
These algorithms will design an assembly or other
process line to meet average production
requirements, adjusted for a fixed % uptime
Detailed simulation is needed to verify production
rate and study queues and other issues
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
What to Model
The tasks that need to be done

The number of units needed per year
What resources are available or applicable to a given task
What each resource costs to buy

What tool it needs for each task
How long it will take to do the task, change tools, etc
What is its uptime and other operating characteristics
Time for transport from station to station

Reuse of a resource for several tasks
Reuse of tools at one station
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
History
Heuristics by R E Gustavson at Draper and N H

Cook at MIT in 1970s
Solutions based on OR techniques by Prof Graves
and OR Center students
Terry Huttner, 1977 - mixed linear-integer
programming
Bruce Lamar, 1979 - bus routing algorithm
Carol Holmes, 1987 - multiple products, dynamic

programming
Curt Cooprider, 1989 - uncertain demand, dynamic
programming
Holmes-Cooprider method reprogrammed by

Mike Hoag, 2001.
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
System Selection Criteria
Minimize annualized cost

= unit labor cost + annualized cost of capital
Systems can be forced to be all manual, all robot,

or all fixed automation just by removing unwanted
resource classes
A wide variety of preferences can be
accommodated this way
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
Summary of Required Input
Info about assembly resources with cost, operation

time, and rho or installed cost factor
rho relates total cost to equipment cost
Info about assembly tasks with operation time and

tool number for each resource
Annual production volume, labor cost, min
acceptable rate of return, number of shifts available
Rate of return expressed in annualized cost factor
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
Title
Working days/year ________
Shifts available ______
Date
Annualized cost factor
Avg loaded labor rate ($/hr)
Station-station move time (s)

Resource data set name: _________
________
________
________
Task data set name:______
10
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
Data Input (Applicable

Technology Chart)
For each resource:

When a resource can be used:
hardware Cost ($)
C
installed cost/hardware cost
Operation
Tool
rho
e
% uptime expected
time (s)
number
v
operating/maintenance rate ($/hr)
Tool change time (s)
Hardware cost
Tc
Max # stations/worker
Ms
NOTE: SEE FIG 14.8 OF CONCURRENT DESIGN AND PP 434-435
Resource:
__________
__________
__________
__________
__________
......
C_______
C_______
C_______
C_______
C_______
.......
rho_____
rho_____
rho_____
rho_____
rho_____
.........
e_______
e_______
e_______
e_______
e_______
...........
v_______
v_______
v_______
v_______
v_______
........
Tc______
Tc______
Tc______
Tc______
Tc______
.........
Ms______
Ms______
Ms______
Ms______
Ms______
Task:
1
|
|
|
|
|
SEQUENCE
TASK
TASK TYPE
P = PLACE/ORIENT
T=TIGHTEN BOLT, SCREW, ETC
I=INSERT PART(S)
M=MEASURE
S=MODIFY SHAPE
A=ALIGN
Asst Sys Des Tech
11/16/2004
INSPECTION
B=BOLT TORQUE
G=GAUGE DIMENSION
C=COMPARISON
Daniel E Whitney
Assembly Planning Chart
NAME
DATE
PREPARED BY
NOTE: SEE FIG 14.7 OF CONCURRENT DESIGN SHEET
OF
PAGE 433
Applicable Technology Chart
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
Assembly Planning Chart
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
10
Basic Nominal Capacity Equations
# operations/unit * # units/year = # ops/yr

# ops/sec = # ops/yr * (1 shift/28800 sec)*(1 day/n shifts)*(1 yr/280 days)
cycle time = 1/(ops/sec) = required sec/op
equipment capability = actual sec/op
actual sec/op < required sec/op -> happiness
required sec/op < actual sec/op -> misery (or multiple resources)
Typical cycle times: 3-5 sec manual small parts
5-10 sec small robot
1-4 sec small fixed automation
10-60 sec large robot or manual large parts
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
11
How the Holmes-Cooprider Method
Works
The maximum takt or cycle time is calculated

based on annual volume requirement and # shifts
Each resource is tested to see if it can do one task
without running out of time, two tasks, three tasks,
etc.
A network is built where pairs of nodes are tasks,
and arcs are resources
Each arc has a cost based on investment, tools, and
labor (labor cost based on time used)
The shortest path through the network is the string
of selected resources and the tasks they will do
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
12
Network
$20K
$10K
$10K
Etc.
$7K
$7K
$15K
$14K
Shortest path
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
13
Network Models of Assembly Systems
Model of system as flows in a network
Represents equilibrium state

Based on probabilities and costs
1.0, $10
0.9, $20
0.1, $50
Outbound probabilities add to 1.0

Equilibrium solution gives average cost to go
through and average flow on each branch
Asst Sys Des Tech
11/16/2004
Daniel E Whitney
17
Equations
pij=pr of going from node i to node j

cij=cost of going from node i to node j
fij=flow from node i to node j
yi = total flow out of node i
pij
cij
fij = yi pij
where we must have
pij = 1 for each
Node i
Node j
Conservation of flow at node j:
y j = y j p jj +
y k pkj
+ x j
Y
= P T Y + X
k. j
xj=flow into node j from outside
Solution:
Y= IP
cost = fij cij
pjj=0
Asst Sys Des Tech
T 1
11/16/2004
Daniel E Whitney
18
Example System: an assembly with two
subassemblies and several test and rework
stations
Rework
the Assy
$50
Both Fail:
0.002
Subassy #2 Already Done
Build #1
$10
New
Parts
Subassembly #1
Test 0.9 OK
#1
$1
0.1
Fail
Rework
#2
$10
Rework
#1
$40
Rework#1
with #2
Attached
$80
Asst Sys Des Tech
11/16/2004
Build #2
Assemble
#1 To #2
$20
Subassembly #2
Daniel E Whitney
Test
#1 & #2
$2
Done
#2 Fails:
0.1
#1 Fails:
0.02
19
Network Equivalent of Example
0.9
$0
0.002
0.02
$80
$50
New
Parts
$0
$11
0.1
$40
A
B
C
Asst Sys Des Tech
0.9
$0
$20
$2
4
0.1
$10
1
$11
0.1
$40
Build/repair Subassembly #1 and Test it

Build/repair Subassembly #2 and Test Both
Repair/rebuild #1 While Attached to #2
11/16/2004
Daniel E Whitney
20
Done
Matlab Solution
P=zeros(8)
C=zeros(8)
%Arc probabilities:
P(1,2)=1;
P(2,1)=.1;
p(2,3)=.9;
P(2,3)=.9;
P(3,4)=1;
P(4,5)=1;
P(5,3)=.1;
P(5,1)=.002;
P(5,6)=.02;
P(5,8)=1-P(5,6)-P(5,3)-P(5,1);
P(6,7)=1;
P(7,4)=.9;
P(7,6)=.1;
X=[1 0 0 0 0 0 0 0];
X=X'
Asst Sys Des Tech
11/16/2004
Y=inv(eye(8)-P')*X
Y=
1.1136
1.1136
1.1162
1.1390
1.1390
0.0253
0.0253
1.0000
YY=[Y Y Y Y Y Y Y Y]
F=box(YY,P)
Daniel E Whitney
21
Equilibrium Flows
0.0000 1.1136
0.1114 0.0000
0.0000 0.0000
0.0000 0.0000
0.0023 0.0000
0.0000 0.0000
0.0000 0.0000
0.0000 0.0000
Asst Sys Des Tech
11/16/2004
0.0000
1.0023
0.0000
0.0000
0.1139
0.0000
0.0000
0.0000
F=
0.0000 0.0000
0.0000 0.0000
1.1162 0.0000
0.0000 1.1390
0.0000 0.0000
0.0000 0.0000
0.0228 0.0000
0.0000 0.0000
Daniel E Whitney
0.0000 0.0000
0.0000 0.0000
0.0000 0.0000
0.0000 0.0000
0.0228 0.0000
0.0000 0.0253
0.0025 0.0000
0.0000 0.0000
0.0000
0.0000
0.0000
0.0000
1.0000
0.0000
0.0000
0.0000
22
Cost Solution
%Arc costs:
C(1,2)=11;
C(2,1)=40;
C(3,4)=20;
C(4,5)=2;
C(5,1)=50;
C(5,3)=10;
C(5,3)=10;
C(5,6)=80;
C(6,7)=11;
C(7,6)=40;
cost=sum(sum(box(C,F)))
cost =
$44.7608
Cost without rework = $33
Asst Sys Des Tech
11/16/2004
%FF = total flow in system

FF=sum(sum(F))
FF=5.6720
%EX=excess flow
EX=FF/5
EX =
1.1344
Total flow without rework = 5

Capacity devoted to rework = 13.44%
Daniel E Whitney
23
Key Characteristics
Goals of this class

Introduce Key Characteristics (KCs)
Define the notions of KC delivery and KC delivery chain
Understand the relationship between KC delivery chains and partto-part location
Appreciate how many KCs an assembly can have, including the
concept of KC conflict
See some examples
KCs_04.ppt
9/13/2004
Daniel E Whitney
Key Characteristics (KCs)
Key characteristics are product requirements that demand

attention because
they are critical for performance, safety, or regulations
AND
they are at risk of not being achieved due to process variations
Usually, KCs are geometric relationships between features

on non-adjacent parts
Two basic issues for KCs are
priorities
flowdown
KCs_04.ppt
9/13/2004
Daniel E Whitney
Chain of Delivery of Quality
Image removed for copyright reasons.
Source:
Figure 2-1 in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
and Role in Product Development. New York, NY: Oxford University Press, 2004. ISBN: 0195157826.
No single part delivers the KC.

KCs_04.ppt
9/13/2004
Daniel E Whitney
Chains Deliver KCs
KCs are delivered by chains that must operate

repeatibly
Chains are made up of:
physical elements: parts, sub-assemblies, tools, and fixtures

the associated organizations (supply chain)
the capability of the processes (technology)
Each KC is delivered when its chain is complete
KCs_04.ppt
9/13/2004
Daniel E Whitney
KC Priorities
Everything is important to someone
KCs should be confined to things that are not only

important but are at some risk of not being
achieved
Usually, manufacturing or assembly variation are
considered to be the main threat
So there is a direct link between KCs and
assembly tolerances
If there is no systematic process for identifying
KCs, and if priorities are not assigned, then KCs
tend to proliferate
KCs_04.ppt
9/13/2004
Daniel E Whitney
When Can Key Characteristics Be Used?
During concept design, to capture customer reqmts
During system engineering, to flow down reqmts

to lower levels of the design process
During detail design, to deliver reqmts via tolerances
and process planning
During supplier selection and preparation of specs,
to define deliverables
During program management, to track and assure
achievement of requirements
KCs_04.ppt
9/13/2004
Daniel E Whitney
KCs_04.ppt
9/13/2004
Daniel E Whitney
Desktop Stapler
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
Stapler Parts
Source:
Stapler KCs
KCs_04.ppt
9/13/2004
Daniel E Whitney
Each KC is Delivered by a Chain
Chains of Delivery
HANDLE
CARRIER
PIN
PUSHER
RIVET
STAPLES
ANVIL
BASE
KCs_04.ppt
9/13/2004
Daniel E Whitney
10
Key Characteristics and the Liaison
Diagram
Source:
Figure 1-3 in [Whitney 2004] Whitney, D. E.
Mechanical Assemblies: Their Design, Manufacture,

and Role in Product Development.
New York, NY: Oxford University Press, 2004.
ISBN: 0195157826.
Source:

New York, NY: Oxford University Press, 2004. ISBN: 0195157826.
Liaison Diagram
KCs_04.ppt
9/13/2004
KCs
Daniel E Whitney
11
Only Some Liaisons Matter in KC
Delivery
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
12
The Delivery Path for Each Stapler KC
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
13
Optical Disk Drive KCs
Images removed for copyright reasons.
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
14
KC Flowdown
Product KCs can be defined for customer
requirements and then decomposed into

lower level Assembly and Manufacturing KCs
Achievment of the PKCs requires
achievement of the AKCs and MKCs
Full implementation requires that each AKC
and MKC meet a specific tolerance or Cpk
Suppliers capability may limit AKCs and MKCs,

requiring flow-up and negotiation
KCs_04.ppt
9/13/2004
Daniel E Whitney
15
Some Statistics
A person at GM said
60% of body sheet metal tolerances can be met
40% must be altered to meet shop capabilities
A patent from Boeing on tolerancing says that typically 8

parts are involved in a tolerance chain (probably similar to
the length of a KC chain for us)
A survey of 600 consumer products by Ulrich and Ellison
reveals that about 6 parts are involved in delivering
functions that differentiate the product in the marketplace
You dont get real numbers like this every day
KCs_04.ppt
9/13/2004
Daniel E Whitney
16
How Parts Locate Each Other to Deliver
Quality at the Customer Level
DOOR
CRAFTMANSHIP KC
CAR BODY
KC=Key Characteristic
BODY TO
HINGE FLAP1: 6 DOF
HINGE FLAP 1 TO
HINGE FLAP 2: 5 DOF
HINGE FLAP 2 TO
DOOR: 6 DOF
KCs_04.ppt
9/13/2004
Daniel E Whitney
17
CUSTOMER
PERCEPTION
OF DOOR
DOOR CLOSING
FORCE
WATER LEAKAGE
AND WIND NOISE
SEAL
TIGHTNESS
DOOR-BODY
ALIGNMENT
IN/OUT
DOOR-BODY
ALIGNMENT
UP/DOWN AND
FORE/AFT
DOOR MOUNTING
METHOD AND
EQUIPMENT
DOOR ATTACHMENT
TO BODY
DOOR FRAME
PERIMETER
SHAPE ACCURACY
DOOR PERIMETER
SHAPE ACCURACY
BODY ASSEMBLY
METHOD AND
EQUIPMENT
SEAL ATTCHMENT
TO BODY
SUBASSEMBLY
PROCESS KCS
DOOR ASSEMBLY
METHOD AND
EQUIPMENT
DOOR PARTS
ASSEMBLY
BODY PARTS
FABRICATION
DOOR PARTS
FABRICATION
9/13/2004
DOOR
SUBSYSTEM KCS
SUBSYSTEM
ASSEMBLY
PROCESS KCS
DOOR THICKNESS
ACCURACY
BODY PARTS
ASSEMBLY
KCs_04.ppt
HINGE ATTACHMENT
TO DOOR
}
}
PARTS FAB AND

ASSEMBLY KCS
Daniel E Whitney
18
Car Door KC Flowdown
FLUSHNESS OF
DOOR-BODY
SURFACES
UNIFORMITY OF
DOOR-BODY GAPS
CUSTOMER
REQUIREMENTS:
PRODUCT KCS
KCs_04.ppt
9/13/2004
Daniel E Whitney
Car Door Exterior Gaps
Source:
19
Door Assembly
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
21
How Doors are Built
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
22
Door Hem
KCs_04.ppt
9/13/2004
Daniel E Whitney
23
Car Door Design KCs
Side View
Top View
up-down
A Pillar
B Pillar
Weather seal KC depends

on placement of the
inner panel
A Pillar
B Pillar
Hinges
Outer
Fender
Latch
bar
fore-aft
door tolerances and fit
KCs_04.ppt
9/13/2004
Outer
Fender
Appearance KC depends
on placement of the
outer panel
in-out
Appearance KC
= uniformity of
this gap
Daniel E Whitney
25
Two Door Methods - There Are Many
= 6 DOF LOCATION
DOOR
HINGES
HINGES
DOOR
HINGE-MOUNTING FIXTURE
DOOR MOUNTING FIXTURE
Assembly Step 1a
Assembly Step 2a
LOCATOR
CONES
HINGES
DOOR
HINGES
DOOR
HINGE-MOUNTING FIXTURE
Assembly Step 1b
KCs_04.ppt
9/13/2004
Assembly Step 2b
Daniel E Whitney
26
KC Conflict in Door Assembly
Difficult to achieve both

KCs this way!
Impossible
to assemble
this way!
Attach door outer

to door inner,
aligning parts
Align door inner

to seal, then attach
inner to frame
Align door outer

to frame gaps, then
attach outer to inner
Mount door (inner+outer)

to frame and align seals,
possibly misaligning gaps
Mount door (inner+outer)

to frame and align gaps,
possibly misaligning seals
Not enough independent DOF available

KCs_04.ppt
9/13/2004
Daniel E Whitney
27
Ford Hinge Mounting
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
28
Ford Hinge Mounting Fixture
Photo removed due to copyright restrictions. (Detail of car door front and rear locator pins and holders for hinges.)
KCs_04.ppt
9/13/2004
Daniel E Whitney
29
Door on Hinge-Mounting Fixture
Photo removed due to copyright restrictions. (Detail of front and rear car door mounting locators.)
KCs_04.ppt
9/13/2004
Daniel E Whitney
30
Ford Door Mounting to Car
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
32
Mustang Body in White
Photo removed due to copyright restrictions. (Detail of car door front and rear locators.)
KCs_04.ppt
9/13/2004
Daniel E Whitney
33
34
Door Mounting Tool for Mustang
Source:
An Interesting Wrinkle
Doors are usually installed on a car before
painting and removed for trim installation
you can grab a door rigidly (accurately) when there is

no paint to scratch
it is easier to install stuff on/in the door and in the car if
the doors are separate
The challenge is to get them back on in the right
place without the benefit of assembly tooling
It is done cleverly with the hinges

install door+hinges to car, remove door from hinges
remove a temporary hinge pin, reinstall a final one
check which bolts have paint to see how it was done
KCs_04.ppt
9/13/2004
Daniel E Whitney
35
GM Hinge Mounting
Source:
GM Method
KCs_04.ppt
9/13/2004
Daniel E Whitney
37
GM Take-apart Car Door Hinges
Photo removed due to copyright restrictions. (Detail of car door hinges and locator cone.)
KCs_04.ppt
9/13/2004
Daniel E Whitney
38
GM Door Mounting to Car
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
39
Ford Locator Drawing
Source:
KCs_04.ppt
9/13/2004
Daniel E Whitney
40
Conclusions
KCs are the link between functional customer needs and

physical realizations at the assembly level
KCs are delivered by chains of parts
KCs can be delivered in more than one way
Design of KC delivery requires definition of location

schemes by which parts are related to each other in 6 dof
Assemblies typically have many KCs, and they can
conflict
KCs_04.ppt
9/13/2004
Daniel E Whitney
41
Mechanical Assembly and Its Role in
Product Development
Instructor : Dr Dan Whitney
7/25/2005
Class I Intro _04
Daniel E Whitney 1997-2004
Goals of this Course
Understand a systematic approach to analyzing

assembly problems
Appreciate the many ways assembly influences
product development and manufacturing
See a complete approach that includes technology,
systems engineering, and economic analysis
Get a feeling for what is technologically feasible
Practice the systematic process on a semester-long

group project of your own
7/25/2005
Class I Intro _04
About Me
Mechanical engineering and history at MIT
Taught at MIT 68 - 74
19 years at Draper Lab doing research and consulting in robotics
and assembly
Applied system engineering techniques to product design for
assembly and assembly process design
Came to understand fuzzy boundary between management and
engineering
Returned to MIT in 1993
Teach Sloan-Eng Product Development Class to SDM
Also interested in complex systems

7/25/2005
Class I Intro _04
Assembly in the Context of Product Development

Org, MOT, Strategy
Technology And
Market
Customer Needs
Requirements
One product?
Product Family?
Customization ?
Reuse?
65% outsourced
Flow of influence of decisions

This course
Functions
Architecture
Assemblies
Parts
Evolution of
Market and Technology
7/25/2005
Direct info flow in product dev.
Product design and

development
Class I Intro _04
Fabrication &
Inspection
Assembly
and Test
Follow-on product
reuses arch,
some assemblies,
some parts
Distribution
Use
Upgrade
Repair
Disassembly
Reuse Recycle
4
Assemblies are Systems
Assembly is inherently integrative

Assemblies can be designed top-down
Decomposition and interface management are key
Assemblies exhibit non-colocation of cause and

effect
Assemblies also violate a hidden assumption:
big causes have big effects while small causes

have small effects
7/25/2005
Class I Intro _04
Assemblies are Complex
Hundreds, thousands, tens of thousands of parts
Divided into manageable subassemblies which are
further divided, so that most have 10 - 20 parts
Many versions of the same product based on

different subassemblies or parts:
many opportunities for customization
many chances for error

Im continually amazed that the thing works at
the end of the line.
7/25/2005
Class I Intro _04
Chain of Delivery of Quality

Shows clearly who delivers what and
how long the chains of delivery are
Source:
Figure 1-8 in [Whitney 2004] Whitney, D. E. Mechanical Assemblies:
Their Design, Manufacture, and Role in Product Development .
PART COUNT: 9
PART SOURCES: 7
TOOL COUNT: 5
TOOL SOURCES: 4
CHECK FIXTURE COUNT: 2
CHECK FIXTURE SOURCES: 2
DISPERSAL INDEX: 81%
Oh, we buy
the radiator
support
7/25/2005
Class I Intro _04
Course Mechanics
Class lectures and discussions Mon & Wed 1:00 -2:30
Textbook at the Coop: Mechanical Assemblies: Their Design,

Manufacture, and Role in Product Development, Oxford
University Press, 2004
Reading and homework assignments on MIT Server
A group project to be done in phases during the term
Homework
6 project reports
4 problem sets
A mid-term and a final project presentation
No quizzes or final exam
Grade formula: 1/3 on homework, 1/3 on project reports, 1/3 on
midterm and final presentation
7/25/2005
Class I Intro _04
Project Guidelines
Buy a small assembled product costing no more than ~$35

and having 10 to 20 parts
Be sure you can take it apart and put it back together
together
Save the packaging and instructions

SDM students can use a product from work
You will analyze it in detail technically and economically
and design an assembly line
Wednesday Sept 15 hand in a description of what you
bought and names/e-mails of team members
Examples: hand-held power tools, small clocks and timers,
Luxo lamps, small home appliances, toys
Schedule a time to show it to me for an hour ASAP
Hand in Request for Payment to get reimbursed.

7/25/2005
Class I Intro _04
10
Electric Drill
Source:
7/25/2005
Class I Intro _04
11
Toy
Source:
7/25/2005
Class I Intro _04
12
Staple Gun
Source:
7/25/2005
Class I Intro _04
13
Juicer
Source:
7/25/2005
Class I Intro _04
14
More Mechanics
Sign up on MIT Server.
Join the group called 2.875 Mechanical Assemblyand go

to documents
One page word doc syllabus under Public Files
Class schedule, assignments, and project deliverables

excel file called detailed_schedule under Detailed
Syllabus
Class slides pdf files under Lecture Notes
All class presentations, homework assignments, and other

info will be posted there
Right now, download the word doc under Public Files and
the excel file under Detailed Syllabus
Look at Excel sheets 1, 2, and 3 for detailed outline,
reading assignments and project deliverables
7/25/2005
Class I Intro _04
15
Class Schedule -Typical

Tuesday Sept 3 Registration Day
Wednesday Sept 4 Class 1 Introduction,

Logistics, Context, History
Monday Sept 9 Class 2 Assembly in the

Small - Step-by-step process -Assembly
Motions and Forces
Wednesday Sept 11 Class 3 Assembly in the

Small -Rigid part mating theory & RCC
Monday Sept 16 Class 4 Mathematical

models of assemblies, Feature-based
modeling of assemblies
Wednesday Sept 18 Class 5 Constraint in

Assembly-1
Monday Sept 23 Holiday No Class
Wednesday Sept 25 Class 6 Constraint in

Assembly-2
Monday Sept 30 Class 7 Key

Characteristics
Wednesday Oct 2 Class 8 Variation buildup

in assemblies-1
Monday Oct 7 Class 9 Variation buildup in

assemblies-2
Wednesday Oct 9 Class 10 Assembly

sequence analysis, algorithms, and software
Monday Oct 14 Holiday No Class
Wednesday Oct 16 Class 11 The Datum

Flow Chain-1
Monday Oct 21 Class 12 The Datum Flow

Chain-2
Wednesday Oct 23 more DFC
7/25/2005
Class I Intro _04
Assembly
in the small
Assembly
in the large
17
Class Schedule - Typical
Monday Oct 28 Class 13 Assembly

in the Large - basic issues,
economics, step-by-step process
Wednesday Oct 30 Class 14 Product

Architecture, flexibility
Monday Nov 4 Class 15 Design for

Assembly Theory, Examples and
video
Wednesday Nov 6 Class 16 AITL

System Design Issues: Kinds of
assembly lines and equipment,
production volume, cycle times
Class 18
Monday Nov 11 Holiday No Class
Wednesday Nov 13 Class 17 Midterm presentation of student projects

covering first three reports
Monday Nov 18 Class 19 Assembly

in the large: Workstation design
issues
Wednesday Nov 20 Class 20

Assembly System Design Software
Monday Nov 25 Class 21 Discrete

Event Simulation
Wednesday Nov 27 Class 22

Economic analysis of assembly
systems
Monday Dec 2 Class 23

Outsourcing, & supply chain
management
Wednesday Dec 4 Class 24 767

Wing Case Study
Monday Dec 9 Class 25 Student

project presentations
7/25/2005
Class I Intro _04
18
Wednesday Sept 18
Problem set on rigid part mating due
Wednesday Sept 25
First project report due:
Completely describe the product
Each report
generates info
for later
reports
Wednesday Oct 2
Problem set on 4x4 matrices due
Wednesday Oct 16
Problem set on tolerances and constraint due
Wednesday Oct 23
Problem set on DFCs due
Wednesday Oct 30 Second project report due: DFC
analysis of your product
Wednesday Nov 13
Third project report due:
Choreograph each assembly step & DFA
Wednesday Nov 20
Fourth project report due:
Design a workstation
Wednesday Nov 27
Fifth project report due:
Create a floor layout
7/25/2005
Wednesday Dec 4 Sixth project report due:

Economic analysis of this layout and Discrete event
simulation
Class I Intro _04
Project Reports and Homework
Wednesday Sept 11
Student project descriptions due
19
Why is Assembly Important?
Assembly is inherently integrative

brings parts together
brings people, departments, companies together
can be the glue for concurrent engineering
Assembly is where the product comes to life
there arent many one-part products
Assembly is where quality is delivered

quality is delivered by chains of parts, not by any
single most important part
A paradox: assembly is not a big cost element
7/25/2005
Class I Intro _04
20
Make-Buy Complexity:
Product Development on a Web
A PRODUCT DESIGN
STARTS OUT FROM
ONE POINT
ASSEMBLER/
DESIGNER
ASSEM
S
DESIGN
DE S
IGN
TOOLING
SUPPLIER
PARTS
SUPPLIER
HU
N
7/25/2005
DRE
DS
OR THOU
S AN
Class I Intro _04
ASSEMBLY
SUPPLIER
OLS
BLY TO
PARTS
SUPPLIER
ASSEMBLY
SUPPLIER
NS
SP
IF
C
E
PARTS
TOOLING
SUPPLIER
TI
A
IC
ASS
DE
EM
SIG
BLI
ES
NS
NS
SP
EC
IF
IC
AT
IO
I
BR
A
F
TI
CA
NT
LS
OO
DE
NS
PA
RT
S
FINALASSEMBLY
ASSEMBLY IS THE
FINAL
"MOMENT
MOMENT OF
OF TRUTH"
TRUTH FOR
THE ENTIRE PROCESS
G
SI
IT GETS DISPERSED
OVER THE SUPPLY WEB
PARTS
SUPPLIER
ARS
DS OF MILES AND THREE TO TEN YE
21
Assembly Links Unit Manufacturing
Processes to Business Processes
Business Context
Production volume
Model mix, variety, customizing
Upgrade, update
Reuse, carryover
Outsourcing
System Level
Subassemblies
Assembly sequences
Assembly quality
Line automation
Line layout
People involvement
Technical Details
Individual part joining

Individual part quality
Part prep, logistics, feeding
Manual vs automatic
Speed, cost
7/25/2005
Class I Intro _04
23
Example of This Link
Denso makes many kinds of panel meters for Toyota.

Toyota orders different ones in different amounts every day.
Denso designed an assembly family of meters and can
make any quantity of any kind at any time by selecting the
right parts. Assembly interfaces were standardized for all
parts. The result is assembly-driven manufacturing.
Source:
7/25/2005
Class I Intro _04
24
History and Present Status
Traditional unit processes studied for 150+ years
Assembly studied perhaps 40 years

Most assembly is manual
Most assembly process design is manual
Surge in interest in robot assembly in the 70s
Interest in appropriate technology today
7/25/2005
Class I Intro _04
25
Need for a Systems Approach
We design parts, we dont design assemblies
We spent all day identifying the reasons why certain

features on certain parts relate to features on other
parts
Tolerances are those little numbers that you have to
put on the drawing before the boss will sign off
You cant have both cosmetic quality and functional
quality (car doors)
7/25/2005
Class I Intro _04
26
Outline of Requirements-Driven Step-
by-Step Process
Assess business context

managements objectives and constraints
character of the product
Analyze assembly in the small
understand each part, determine risks
recommend redesigns
Analyze assembly in the large
revisit business context
take system view: technical and economic analysis
design processes: assy sequence, line layout, equipment

make final recommendations
7/25/2005
Class I Intro _04
27
Manual vs Automated Assembly
People just do it
Machines cant just do it
It was hoped that robots could just do it
Early robot research focused on imitating what

people do in general
behave flexibly
use their senses
react to the unexpected
fix mistakes that should not have occurred in

the first place
7/25/2005
Class I Intro _04
28
Robotics as a Driver
Robotics raises a number of generic issues:

flexibility vs efficiency
generality vs specificity
responsiveness or adaptation vs preplanning
absorption of uncertainty vs elimination of uncertainty
lack of structure vs structure
7/25/2005
Class I Intro _04
29
What Happened...
Robots were too slow and too costly
No one knew how to do an economic analysis and

most didnt care at first
People do what they do because of their strengths
and weaknesses - same with robots
The unexpected is not supposed to happen in a
factory - planning for it is not the right attitude
Today there is a place for robots, people, and fixed
automation in assembly
The issue is to decide which is best and how to
prepare the environment
7/25/2005
Class I Intro _04
30
Video
Computer-Controlled Assembly
Made at Draper 25 years ago
First convincing lab demo of robot assembly
Illustrates most of the elements of good assembly

analysis
7/25/2005
Class I Intro _04
31
Sony Video
Compare 20+ years later:

multiple parts feeders at one station
tool changer head
4 - 6 sec operation time per part
It is a complete solution
robot and tool set (VCR and school of VCR)
part tray loader
transport
controllers
Used for cameras, VCRs, Walkmen, disk drives...
7/25/2005
Class I Intro _04
32
Assembly Fine Motions
Goals of this class
outline the basic issues of assembly in the small

how to look at and take apart a product
characterize assembly motions
model motions in 3 dimensions
relate motions and forces using vector-matrix models
Class 2 Assy motions 9/7/2004
Basic Issues in Assembly in the Small
The steps are:
understand each assembly step in detail

do conventional Design for Assembly
identify high risk areas
identify necessary experiments
recommend local design improvements
Greg Blonder Video (17 min of 1:45)
Talk given at Bell Labs Jan 16, 1990
Has been analyzing precision consumer products

for about 5 years
Takes about a day to analyze each one
Takes apart and analyzes a $100 Canon camera
Classifies ~ 400 parts including 60 screws (!)
Tells what it means to get high assembly yield
Tries to explain it to a high level executive

Discusses shutter and flash subassemblies
He later became a VP of Lucent
Understand Each Step
Take the product apart (or use drawings if thats all you have)
Get really familiar with every part and its role in the
assembly (story: Yes, Alex)
Make a structured bill of materials
Draw a picture (2D is OK) of each part
Make an exploded view drawing
Choose any convenient assembly sequence
Study each part mate and draw it, noting each place on
each feature where the parts touch during assembly
Note where the part can be gripped

Note how the part can be fixtured before assembly
Note any problems that could occur

Do Conventional DFA
The issues are: (Boothroyd except where noted)

assembling each part feeding/presenting
handling/carrying/getting into position
inserting without damage, collisions, fumbling
reducing part count (driven by local economic analysis)
two adjacent parts of same material?

do they move wrt each other after assembly
is disassembly needed later (use, repair, inspection, upgrade...)
the part is a main function carrier?(Fujitsu)
if not, consider combining them (but see DFA class)
are there too many fasteners? (but see DFA class)
identifying cost drivers (Denso)

Identify High Risk Areas
Showstoppers, safety, regulatory issues
Do we understand the process?

Vague instructions: use a small amount, wipe off excess,
check for defects, clean fiber
Tasks only one person can do
Person can do it but can a machine??

Look for part damage, find a part that works better
Unavailability of an otherwise attractive sequence
story: disassembly after balancing
Inability to guarantee a constant task time

Calibration
Typical errors: wrong part, incorrect assembly of a good

part, correct assembly of a bad part
Identify Necessary Experiments
Usually address high risk areas

Determine physical feasibility
Determine economic feasibility
Generate metrics for successful assembly and
means for detecting failures on the fly
cycle time
checks for part correctness/presence/placement
avoidance of parts damage

awareness of potential undocumented sources of
trouble
Recommend Local Design Improvements
These address the high risk areas as well as

physical and economic feasibility
There usually is no strategic or system content in
these kinds of improvements
Assembly in the large addresses such issues
Keep an Eye Out for Mysteries
Parts or features that dont seem to do anything
Parts that look greatly over- or under-designed
Pay attention to what the fool says
Hole in Battery-Operated Vac Housing
Source:
10
Laptop Heat Removal
Heat sink
(CPU invisible)
Fan
Heatpipe
Radiator
11
Gateway Laptop, 2002
HEAT PIPE
DVD
CARD
SLOT
HDD
ME
MO
RY
CPU
OTHER
STUFF
BATTERY BAY
12
Over-designed Part
Images removed due to copyright restrictions. (Photos of hinge mounts and heat pipe laptop parts.)
13
Gross and Fine Motions
Assembly alternates between two kinds of
motions:
Gross motions
are fast and do not need high accuracy
basically used for transport
large compared to size of part
Fine motions
need high accuracy
are likely to be slower than gross motions
small compared to size of part

14
Characteristics of Gross and Fine Motions
Parts do not (should not) contact during gross motion
Parts normally contact during fine motion

Fine motion is basically a series of controlled contacts
15
Nature of Gross Motions
Errors:
Preplanning:
could happen
they can be seen but
not felt until too late
people use sensors
machines use
preplanning
- is rewarded
- errors can be
catastrophic
- low cost to avoid
them
- savings are reaped
many times
- characterized by
structure - an open
loop approach
16
Nature of Fine Motions
Errors:
Preplanning:
are unavoidable for

reasonable cost
they can be felt but
not seen
they generate
signals that can be
used to correct them
is not rewarded
even tiny errors can
stop some tasks
cost of avoiding
them grows too fast
as their size shrinks
a closed loop
approach is suitable
17
Essence of Force Feedback Fine Motions

Make a model of the process, giving forces that
result from various classes of geometric errors
Invert the model so that sensed forces can be used
to estimate the errors
Make up a strategy that generates corrective
motions in response to specific errors
Design a controller that will execute the response
motions without going unstable
motion force motion force etc!
19
Multi-axis Force-Fine Motion Models

Source:
and Role in Product Development .
Typically requires a
matrix relation between
task coordinates X and
command coordinates
x= J
= J 1 x
Typically requires a
matrix relation between
sensed parameters and
responses
x = K fF
20
Differential Motion Analysis

2 L
V
1.414 1 L
1
L
2
X direction: = - 0.707 2 L
Y direction: = 1.414 1 L + 0.707 2 L
0
[
[
-0.707 L
J=
]
]
1.414 L 0.707 L
0.707/L 0.707/L
J-1 =
-1.414/L
If V=[1,0]T, then =[.707,-1.414]T

2=-2 1 for all joint angles
21
Typical Motion-Force Process with
Geometric Errors
Lateral error requires

lateral motion to
correct it
Angular error requires

angular motion to
correct it
22
Active Force-Motion Strategy
Source:
23
Closed Loop Force-Motion System

.
x
Original Motion
Modified Motion
J-1
Command
Command
Robot dt
Object Position
KE
KF
Sensor
Actually modeled and analyzed as a sampled time system.

This allows us to single step through history, observing
the dynamics as carefully as we want.
24
Force-Motion Stability
Stability Criterion:
KF KE T < 1
Essentially means that not all the accumulated

contact force can be removed during the next T
see next slide
Problem is made worse by stiff coupling to

environment
Problem is made better by faster sampling, up to a
point
25
The Stability Criterion in Words

K FK E T < 1 (1)
Xi + 1 = Vi T
Fi + 1 = K E Xi + 1
Vi = KF Fi
multiply both sides of (1) by V :
K F(K E ViT ) < Vi
K F Fi + 1 < K FFi
Fi + 1 < Fi
26
Motions Made by Choice of KF
Diagonal KF creates damping, nulling feedback
Cross terms in KF turn sensed force into rotation or sensed

torque into translation
Box packing, putting records on turntables, crank-turning
27
Some Force Feedback Matrices
R
S
L
Tw
Tn
Tl
Command Reach
Command -Lift
Lift
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
100 0
0
0
0
0 100 0
0
0
0
0 100 0
0
0
0
0 100 0
0
0
0
0 100
0
0
0 100
200 100
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Sweep
R
(L)
Command -Lift
moment about reach =

-Flift R sin
moment about sweep =
Flift R cos
0
0
0
0
0
0
0
0
0
0
0
0
0 VR
0 0
0 0
0 0
0 0
0 0
0
VR
0
0
0
0
0
0
0
0
0
0
28
Lateral Error Can Become Angular Error
with Disastrous Results
29
DFA
Goals of this class:
DFA03.ppt
Place DFA in context

Learn basic principles of Design for Assembly
Understand background and history
Understand its strong and weak points
11/2/2004
Daniel E Whitney
The Multiplier According to Ford and GM
or: Why Is DFM/DFA Important?
For every product part, there are about 1000
manufacturing equipment parts*
Or, for every toleranced dimension or feature on a

product part, there are about 1000 toleranced
dimensions or features on manufacturing
equipment
Such equipment includes fixtures, transporters,
dies, clamps, robots, machine tool elements, etc
*Note: Fords estimate is 1000, GMs is 1800. Both are informal estimates.
DFA03.ppt
11/2/2004
Daniel E Whitney
A Few Quotes
Just because you can make something doesnt

mean you can manufacture it.
Its very hard to make cheap [low cost] stuff - you
get buried by your mistakes.
I dont understand why it wont assemble. It
passed inspection.
Word came down that we couldnt use screws. So
we used snap fits. Then word came down that it
had to pass a drop test. So we dropped it and it
fell apart...
DFA03.ppt
11/2/2004
Daniel E Whitney
Goals of DFM/DFA
Historically, conventionally
reduce costs, simplify processes
improve awareness of manufacturing issues during design
More broadly (a goal of this course)

align fabrication and assembly methods to larger goals
ability to automate, systematize, raise quality, be flexible

access to assembly-driven business methods like delayed
commitment
innovative designs, outsourcing (Siemens intake manifold)
Broad view inevitably pushes DFM/DFA earlier into

the product development process where it blends with
architecture (see AITL Basic Issues and Product Architecture
classes)
DFA03.ppt
11/2/2004
Daniel E Whitney
The Assembly from Heaven*
Can be assembled one-handed by a blind person

wearing a boxing glove
Is stable and self-aligning
Tolerances are loose and forgiving
Few fasteners
Few tools and fixtures
Parts presented in the right orientation
Parts asymmetric for easy feeding

Parts easy to grasp and insert
*Dr Peter Will, ISI
DFA03.ppt
11/2/2004
Daniel E Whitney
The Assembly from Hell
The opposite in each case from the previous slide
DFA03.ppt
11/2/2004
Daniel E Whitney
History of DFA
Deep background in Group Technology

Coding and classification schemes
European design tradition

Value Engineering
each part must be justified
Boothroyd
part feeding physics - 1960s

part handling and insertion experiments- 1970s
assertion that assembly cost = 30 - 50% of manufacturing cost
DFA methodology and software - 1970s-80s
switch to assertion that parts are the main cost and fewer parts
= less cost, even if those parts are more complex
DFA03.ppt
11/2/2004
Daniel E Whitney
Sample Cost Breakdowns
VCRs: 90% parts, 5% labor
Car engines: 75% parts, 7% labor, 7% capital, 7%

consumables*
Mini computers: 65% parts, 25% labor
Fighter planes: 50% parts and tooling, 40% labor
Most of the above are pretty crude estimates because, for
the most part, companies do not really know their costs
Also, data look different depending on whether labor

component of purchased items is visible or not (See class
on Economic Analysis)
*Data from 27 engine lines, International Motor Vehicle Program
DFA03.ppt
11/2/2004
Daniel E Whitney
10
Characteristics of Traditional DFA
Uses an easy to understand metrics-driven approach (metrics

story about demurrage)
Uses a relative cost and time metric

DFA in the small simplifies each assembly step

single parts
manual assembly
small parts
uses many context-free metrics to assess difficulty levels of
feeding and handling
DFA in the large emphasizes part count reduction

It is essentially another force in product architecture
Advanced plastics make part count reduction more attractive
DFA03.ppt
11/2/2004
Daniel E Whitney
11
Traditional DFA
The issues are: (Boothroyd except where noted)

assembling each part -estimating and reducing time
feeding/presenting
handling/carrying/getting into position (Sony exploded views)
inserting without damage, collisions, fumbling
reducing part count (originally driven by local economic

analysis, now driven by part cost itself)
two adjacent parts of same material?

do they move wrt each other after assembly
is disassembly needed later (use, repair, inspection, upgrade...)
is the part a main function carrier?(Fujitsu, Lucas, (Pahl & Beitz))
if not, consider combining them (but see Architecture class)
are there too many fasteners?
- identifying cost drivers (Denso)
DFA03.ppt
11/2/2004
Daniel E Whitney
12
How to Do Traditional DFA
Make a structured bill of materials

Identify every part mate and understand it
Choose a reasonable assembly sequence
Use the tables to estimate handling and mating times
Label theoretically necessary parts, excluding all fasteners
Calculate
assembly efficiency =
3* # of theoretically needed parts

total predicted assembly time
This ranges from 5% for kludges to 30% for good designs
DFA03.ppt
11/2/2004
Daniel E Whitney
13
DFA Spreadsheet
On SoanSpace there is a folder called DFA

Software
In it is DFA.xls with the handling and insertion
data from the previous two slides
Enter your code numbers and labor rate ($/sec)
and the sheet will calculate times and costs
DFA03.ppt
11/2/2004
Daniel E Whitney
16
DFA Spreadsheet
Cost per sec

Type cost
$0.04
Boothroyd-Dewhurst Data
Used in book by perm ission
Handling Code
Time, sec
1.13
0
1.43
1
1.69
2
1.84
3
2.17
4
2.45
5
1.5
10
1.8
11
2.06
12
DFA03.ppt
per second in cell B2

Type handling codes in column E
If you type a non-ex istent code, the value will be #N/A
Don't change values in yellow cells
Product handling codes
Product handling time values
1.13
0
71
#N/A
11
1.8
15
3
11
1.8
1.13
0
1.13
1.13
1.13
11/2/2004
Type insertion codes in column K
Handling Cost Product Insertion codes

$0.05
1
#N/A
11
$0.07
10
$0.12
11
$0.07
11
$0.05
33
Daniel E Whitney
Product insertion times

3
5.2
3.7
5.2
5.2
#N/A
1.5
1.5
1.5
Insertion Cost Total time

$0.12
2.13
$0.21
#N/A
$0.15
11.8
$0.21
14
$0.21
12.8
#N/A
34.13
17
Make Each Step Easier
Add chamfers and lead-ins

Make the assembly point visible and reachable
Design parts so that they do not tangle

Make assembly happen from above
Design the product to assemble in layers
Make the parts easy to assemble the right way
Symmetric if orientation does not matter

Obviously asymmetric if orientation matters
DFA03.ppt
11/2/2004
Daniel E Whitney
18
Heavy Duty Staple Gun
Source:
Assembly efficiency = 17% before improvements

= 25% after improvements
= 30% with some functional risk
DFA03.ppt
11/2/2004
Daniel E Whitney
19
Low Cost Staple Gun
Source:
Assembly efficiency = 31%
Contains many of the suggested improvements
But is it a better staple gun?
DFA03.ppt
11/2/2004
Daniel E Whitney
20
Part Count Tradeoffs
PARTS CONSOLIDATION:
FEWER PARTS AND LESS
FASTENING
ONE PART PER

FUNCTION
MANY
SIMPLE
PARTS
FEWER BUT
MORE COMPLEX
PARTS
LOTS OF
INTERFACES
IN ASSEMBLY
MORE FUNCTION
SHARING
EXTRA WEIGHT,
EXTRA FAULT
OPPORTUNITIES
PARTS TAKE
LONGER TO
DESIGN AND
PROTOTYPE
EXTRA CHANCES
FOR ERRORS
LOTS OF
LOGISTICS,
FAB ACTIVITY,
& ASSY ACTIVITY
EXTRA
"SUPPORT"
COST
MORE ACTIVITY
DURING FAB,
LESS DURING
ASSEMBLY
PARTS COST
MORE
FLEXIBILITY
IS POSSIBLE
DURING
ASSEMBLY
FEWER OPPORTUNITIES
FOR ON-LINE FLEXIBILITY
QUALITY IS
CREATED
DURING
ASSEMBLY
QUALITY CREATED
DURING FAB
PART COUNT TRADEOFFS
DFA03.ppt
11/2/2004
Daniel E Whitney
21
A Few Conceptual Questions
Whats a base part?
remember the alternator - the nut is the base in the

only assembly sequence family that achieves assembly
without reorientation
different types of product structure exist (Arch. class)
What do you mean difficult?

for manual assembly, Boothroyd has some time-based
data (originally derived from grad students)
Why avoid screws?

25 years agos reasons may not apply any more
see Blonder video
DFA03.ppt
11/2/2004
Daniel E Whitney
22
Manual vs Automatic Assembly DFA
Whats easy for a person

reorienting the assembly
quickly eyeballing the part (story about bad filament)
Whats easy for a machine

picking up little parts
using tools that are like tweezers
Part jams occur most often in feeder tracks

Denso: perfect parts dont jam!
A different balance between gross motion and fine

motion times
Different ways of inspecting
DFA03.ppt
11/2/2004
Daniel E Whitney
23
Complex Molded Part
Source:
DFA03.ppt
11/2/2004
Daniel E Whitney
24
Home Hot Water System Family Parts
Source:
DFA03.ppt
11/2/2004
Daniel E Whitney
Poschmann AG
25
Melt-Core Technology for Water Heater Parts
Source:
DFA03.ppt
11/2/2004
Daniel E Whitney
Poschmann AG
26
Regression Model for Mold Cost and
Time
$Cost = 22500 + .82*Size, cc + 30*#Dimensions
+ 2940*#Actuators + 7630*High finish + 5470*

High tolerance (R2 = 0.911)
Time (weeks) = 13 +0.000055*Size +
0.007*#Dimensions (R2 = 0.7)

Ref A. Fagade and D. Kazmer, Optimal Component Consolidation in Plastic Product Design,
Boothroyds model includes a complexity factor

1. 24
that drives cost C = (number of features)
ASME DETC/DFM-8921, 1999.
DFA03.ppt
11/2/2004
Daniel E Whitney
27
Questions of Scope
When can DFA be applied?
When should DFA be applied? When is DFA not

the right approach?
What information is needed before DFA can be
applied?
What should the designers priorities be?
Can/should DFA be separated from the rest of

product design?
DFA03.ppt
11/2/2004
Daniel E Whitney
28
Development Time Not Critical
Development Time Critical
DFM-DFA Strategies
Low Lifetime Production Volume
High Lifetime Production Volume
Example Products:
High performance computers
Telecommunications equipment
DFM Strategy:
Avoid long lead time tooling
Use standard components
Minimize production risk
Example Products:
Notebook computers, Toys
DFM Strategy:
Minimize complexity of most
complex part
For complex parts, use processes
with fast tool fab
Apply traditional DFM to less
time-critical parts
Example Products:
Machine tools
Electrical distribution equipment
DFM Strategy:
Avoid expensive tooling
Use standard components
Other issues likely to dominate
Example Products:
Blank videocassettes
Circuit breakers
DFM Strategy:
Use traditional DFM-DFA
Combine and integrate parts
Consider automatic assembly
Source: Ulrich, Sartorius, Pearson, Jakiela, DFM Decision-making, Mgt Sci, v 39 no 4, Apr 1993.
DFA03.ppt
11/2/2004
Daniel E Whitney
29
The Pneumatic Piston Redesign*
Was the original function completely understood?
Was it preserved in the redesign?
*Product Design for Assembly by Boothroyd and Dewhurst, workbook 1991
DFA03.ppt
11/2/2004
Daniel E Whitney
30
The Water Pump Redesign
What are the differences between the old and new

designs?
from the POV of product function
from the POV of assembly
What are we looking at in this example?
DFA03.ppt
11/2/2004
Daniel E Whitney
32
Redford-Chal Pump Redesign
Source:
DFA03.ppt
11/2/2004
Daniel E Whitney
33
DFA at Sony
Applied to products like Handicams

Our designers take assembly into account early.
Method:
concept designs are sketched in exploded views
each concept is subjected to DFA analysis and scored
concept selection criteria include DFA score
A Sony engineer made a complete exploded view

drawing of a Polaroid camera in 20 minutes!
DFA03.ppt
11/2/2004
Daniel E Whitney
34
Sony Walkman II Mechanism
Source:
DFA03.ppt
11/2/2004
Daniel E Whitney
35
Sony Exploded View
Source:
DFA03.ppt
11/2/2004
Daniel E Whitney
36
Hitachi Assembly Reliability Evaluation
Method
Source:
Source: Hitachi; Suzuki, Ohashi, Asano, and Miyakawa

DFA03.ppt
11/2/2004
Daniel E Whitney
38
Design for Recycling and Reuse
Source:
Source: Kanai, Sasaki, and Kishinami

DFA03.ppt
11/2/2004
Daniel E Whitney
39
Web Sites from Google
http://www.intel.com/design/quality/pcdesign/assembly.htm
http://www.engineer.gvsu.edu/vac/ (class notes)
http://www.dfma.com/ (Boothroyd-Dewhurst company)
http://www.johnstark.com/pb18.html (a list of books)
http://www.munroassoc.com/design.htm (consulting, training)
DFA03.ppt
11/2/2004
Daniel E Whitney
40
Flexible Manufacturing Systems
Goals of this class:

Understand goals of FMS
Place FMS in context of manufacturing
Understand the history
Take some lessons about appropriate technology
11/24/2004 FMS
Background
Batch production - since the Egyptians?

Mass production - 1880-1960
Flexible production - ?
Lean production - since 1970?
11/24/2004 FMS
More Background
Manually operated machine tools since 1700s
Roger Woodbury History of the Milling Machine,

1960
Steam and electric powered machines since 1820s
Computer-controlled machines since 1960s

Manufacturing systems awareness since Henry
Ford or arguably much earlier
11/24/2004 FMS
Computers and Manufacturing
Numerical control of machine tools R&D at MIT,

1950s - see photo gallery along corridor
From WW II gun servos
Early 1950s Air Force SAGE system
Computer-aided design R&D at MIT in the 1960s
If the computer can guide the tool, then it can hold part
shape in its memory
11/24/2004 FMS
Results of MIT NC Project
Air Force funding aimed the project at complex parts

requiring 5 axis machining
MITs response included complex implementation and
abstract programming language
Simple record playback solution rejected
Useful output mainly benefited the defense industry and
had little to offer small business with 2D or 2.5D needs
Story documented (with exaggerated Marxist

interpretation) by David Noble in Forces of Production,
Oxford Univ Press, 1986
Market gap in small business making simple parts not
filled for 2 decades
11/24/2004 FMS
Numerical Control Technology
Initially one computer for each machine
Computer programmed in APT (Automatically
Programmed Tool), a language like LOGO
By the 1970s, a central computer controlled many

machines - DNC (direct numerical control)
By the 1980s each machine had its own computer,
possibly loaded with instructions from a central
computer - CNC (computer numerical control)
11/24/2004 FMS
Job Shops and Flow Lines
Ford style flow lines utilize equipment at a high level but

are inflexible and costly

Big initial investment requires years to pay back

Dedicated to one part or a very limited family
At risk if the part is no longer needed
One failure stops the whole line
Job shops are flexible but utilization is low

Some asserted that utilization is as low as 5%

Machines time is lost due to setups made on the machine
Parts time is lost due to complex routing and queuing
Big WIP
Flexibility can be defined several ways, including

Different part mix
Different production rate of existing parts
Different machines or routing if one breaks
11/24/2004 FMS
Past Approaches to Utilization
Improvement
Faster changeover AKA SMED
Reduction of setups
Standardization
Use of same setup for several parts
Same setup: Group Technology
Classify parts and code them

Design generic tooling, fixtures, and processes for each
class of part
Ignore the differences that do not matter
11/24/2004 FMS
Ungrouped and Grouped Parts
Images removed due to copyright restrictions.
www.strategosinc.com/ group_technology.htm
11/24/2004 FMS
A Misplaced Effort: Adaptive Control
Adaptive control speeds up a cutting process by

adjusting the feed and speed corresponding to
material hardness and cutter sharpness
Without adaptive control the feed and speed have
to be reduced to avoid random hard spots breaking
the cutting tool
But speeding up the cutting process just makes the
machine finish sooner and makes the utilization
gap even more obvious
11/24/2004 FMS
10
The Flexible Manufacturing System Idea
This idea sprang up in several places at once in the mid 1960s
The basic idea was a computer-controlled job shop with flow

line characteristics
Group technology still important - system aimed at one kind of
part, such as prismatic < 2 ft sq, or rotational < 6 diameter
Computers perform scheduling, routing, and detailed cutter

path control
Pioneering developments by Molins (UK), Cincinnati
Milacron and Kearney&Trecker (US), Gildemeister in W.
Germany, Fritz Heckert Werkzeugmachinenkombinat in E.
Germany
Dueling patents between Molins and Milacron (Molins won)
11/24/2004 FMS
11
Volume and Variety - The Claimed Niche
Volume (parts/hour)
Fixed automation
Sets of special machines
FMS
Cells
Job shop
Variety (number of kinds of parts)

11/24/2004 FMS
12
Early Customers and Partners
Molins made cigarette-making machines
Milacron partnered with Ford to make engine
blocks in small quantities and many variants
Gildemeister partnered with Heidelberg

Druckmachinen to make printing presses
Fritz Heckert made machine tools and partnered
with its own internal business to make simple
Bridgeport-style milling machines
11/24/2004 FMS
13
Typical Big NC Machine
http://www.hildebrandmachinery.com
11/24/2004 FMS
16
Political/Historical Context
Context overlays the technological revolution
Challenge to US manufacturing from overseas, particularly

Japan - several national big projects in IT and
manufacturing in the 70s and 80s
Defense mentality in politics and government-funded
research
Crisis approach to introducing FMS technology to get
government and industry involved in supporting
development
Some hype
75% of all US manufacturing occurs in batches of 50 or

less, a fact still quoted 40 years later
11/24/2004 FMS
17
Claimed FMS Capabilities
Efficiency (high machine utilization based on offline setup using optical comparators)
Flexibility (could be reprogrammed for different
parts)
Capability (could process parts requiring many
operations from many machines)
Scope (could make many different kinds of parts)
Automation (could be programmed remotely and

operated without people)
11/24/2004 FMS
18
Requirements to Support Claims
Rapid programming
Ability to set up tools and parts off line
Ability to place parts and tools on machines accurately

with respect to machines coordinate system so that parts,
tools, machine and NC program all align
In general, these were achieved
Effective scheduling and sequencing of work
High reliability and uptime

In general, these turned out to be unanticipated and proved
to be serious impediments
11/24/2004 FMS
20
Early FMS Implementations - 1970s
These were big systems with big machines
Several architectures were tried

Vendors did not understand system architecture
implications or control issues
Only Milacron had both hardware and software
capability
Technical University of Stuttgart did software and
system integration for Gildemeister - observed by
Whitney in April, 1976
11/24/2004 FMS
21
11/24/2004 FMS
Molins FMS Patent
Content removed due to copyright restrictions.

Please see:
Williamson. Automated machine tool installation with storage means.
US Patent #4,369,563. Filed: October 29, 1970. Issued: January 25, 1983.
22
Please see:
Perry, et al. Machine tool.
US Patent #4,309,600. Filed July 5, 1979. Issued January 5, 1982.
11/24/2004 FMS
Milacron FMS
Patent
23
Milacron Circumferential System

Please see:
Fig 1. Perry, et al. Machine tool.
11/24/2004 FMS
24
Tool Changing and Accurate Location of
Tools on Machine

Please see:
11/24/2004 FMS
25
Accurate Coupling of Pallet to Machine

Please see:
11/24/2004 FMS
26
Control Architecture

Please see:
11/24/2004 FMS
27
Elements of a Process Plan for a Part
Features to be machined
Approach directions needed
Rough and fine cuts needed to achieve required

tolerances and surface finishes
Sequence of cuts
Cutting time (feeds and speeds)
Required tools (kind, shape, size)
Required machine(s) (dof, strength or stiffness,

range of motion)
11/24/2004 FMS
28
Elements of a Shifts Work
Get all the parts made

Keep all the machines busy
Get the needed tools to the machines
Get finished parts out and waiting parts into the

machines quickly
Plan the allocation of parts to machines over time
Replan when a machine breaks or someone wants

a special part made
We installed the FMS to stop the red telephone
11/24/2004 FMS
29
Successful Architecture
Ingersoll-Rand system build by Sunstrand; their

first FMS
A loop architecture with traveling pallets
One piece in-queue and one piece out-queue at

each machine
The system basically ran itself
11/24/2004 FMS
30
I-R System
Machines (6 total)
Conveyor Loop
Load/unload area
Palletizing
Tool setting
11/24/2004 FMS
31
An Unsuccessful Architecture
In-line system for Caterpillar built by Sunstrand,

their next one after the I-R system
12 machines in a line
Two handling carriers on a single rail
Each carrier could hold one part

No in- or out-queues, eliminated (@$75K each) to
save money
No idea what operational problems this would
cause
Gave Prof Richard Wysk his PhD in 1977
11/24/2004 FMS
34
Caterpillar FMS ~ 1976
Machines
Loa d/Unloa d
Area
Ma chine s
Workpieces
Tra ns porters
Two one-arm paper hangers sharing the same crutch
11/24/2004 FMS
35
What Happened
Early systems were too complex and too flexible
Too many kinds of parts were tried on one system
Too many operations were tried on one system

Too many tools were needed (approx 10 per part
at any one station)
Problem of scheduling and dispatching tools was
not anticipated
Parts could not be inserted randomly but had to be
batched - required complex software and
optimization algorithms - called production
smoothing or load leveling today
11/24/2004 FMS
36
PRISMA
East German system built between 1969 and 1974
Highly touted by Milacrons chief marketer

Visited by Nevins and Whitney April, 1976
Porous partly machined parts on the floor
Almost no raw castings at the input
Stacks of finished parts at the output
General Mgr: What do you think?
Nevins: Very impressive. Do you plan to make
any more?
G. M: No!
11/24/2004 FMS
37
What Happened - 2
Systems were too expensive

Systems did not achieve claimed productivity
Sufficient reliability was not achieved until

Japanese applied their methods in the 1980s and
90s
High reliability -> lights out operation -> high
productivity
Typical FMS applications today are simple and
have 3 to 5 machines doing a few operations on a
few kinds of parts
11/24/2004 FMS
43
Sheet Metal Bending System
www.mt-muratec.com/ eg/p/fms/fms_yuatu.html
11/24/2004 FMS
45
Yamazaki Mazak
Built lights-out factory in mid 1980s to make its products

(machine tools) - visited by Whitney in1991
Addressed tool proliferation with given tool method
Addressed system complexity by breaking up factory into

many simple cells having identical tasks, identical
machines, and identical tool sets
Addressed reliability, in part, by reducing cutter depth and
speed at night, eliminating tool breakage, the main failure
preventing lights-out operation
American customers want 120-tool capacity in their tool
carousels - ha ha. Japanese companies are happy with 60.
Some of this documented by the late Prof Jai Jaikumar of
HBS in cases on Yamazaki
11/24/2004 FMS
http://www.mazak.jp/english/
46
Fanuc
Originally a motor company

Built NC machine in 1956!
Developed NC technology in 1960s and 70s
Started building robots in the 1970s

Applied robot controllers to simple CNC machines in late
1970s with low cost bubble memory and simple graphical
controls for programming and simulating and monitoring
operations
Drove US NC controls makers (GE, Honeywell, A-B) out
of the market
Addressed needs of small manufacturers and simple
machines for the first time
Fanuc is still important in the controller and robot markets
11/24/2004 FMS
http://www.fanuc.co.jp/en/profile/index.htm
48
Reconfigurable Manufacturing Systems
Japanese demonstrator system in the 1970s

included reconfigurable machine tools
Current research looks at entirely reconfigurable
systems consisting of reconfigurable machines and
transport systems (see U of MI RFMS Center)
Advances in machine design techniques are
included
Economic analysis includes system life cycle(s)
11/24/2004 FMS
49
Current Status
FMS is a niche technology, not the savior of US

manufacturing
It is effective when applied judiciously with
limited aims, complexity, and scope
This is in spite of Jaikumars paper PostIndustrial Manufacturing, HBR NovemberDecember 1986, which claimed that US firms
made less flexible use of FMS than Japanese
firms, and that this was bad for US manufacturing
11/24/2004 FMS
50
Rigid Part Mating
Goals of this class

understand the phases of a typical part mate
determine the basic scaling laws
understand basic physics of part mating for simple
geometries
relate forces and motions arising from geometric errors
compare logic branching and direct error-feedback part

mating strategies
rigid part mating
9/13/2004
Daniel E Whitney 2000
Basic Bandwidth Issues and Time-Mass-
Distance Scaling Laws
Torque required to move a mass M at the end of an
arm of length L in time T is proportional to
M L2/T2
This implies that really fast motions must be really

small or use a small arm with small mass
I estimated
my hands mass = 250g, effective length = 10cm
my arm + hands mass = 1700g, effective length = 35 cm
ratio arm:hand of ML2 = T2 = 85
Dont forget: arm mass+payload mass=M

rigid part mating
9/13/2004
Main Phases of a Part Mating Event
contact
force
Approach
rigid part mating
Chamfer
Crossing
9/13/2004
One-point
Contact
Two-point
Contact
Line
Contact
Required Bandwidth for Chamfer Crossing

lateral motion
E
0.5E
E/2
time
E/2V
0.5E/V
10E/V
20E/V
T
Fourier coefficient = 2 T / (n2 2 ) sin (2 n /T)

T = 20 E / V; = E / 2 V; T / = 40
Period = 2 = T=20E/V
= V /10 E
f = V / 20 E
If V = 10 in/s and E = 0.05", f = 10 Hz
If 5th harmonic must be adhered to, bandwidth needed = 50Hz
rigid part mating
9/13/2004
Trapezoidal Wave Harmonics
Source:
rigid part mating
9/13/2004
Conclusions
Gross motions can be (must be) done by large

arms that necessarily will move slowly
No robot arm with practical reach can make fine
motion error removal adjustments at 50 Hz
Fine motions can be fast if they are done by small
arms, and must be fast to absorb typical errors at
economical speeds
Big tasks with big parts will take a long time
compared to small tasks with small parts
What we see: small parts cycle times are ~5s while

big parts cycle times are ~ 60s.
rigid part mating
9/13/2004
Essentials of Part Mating Theory for
Fine Motions
Quasi-static assumption
Geometry of pegs and holes
Applied forces
Normal reaction forces and friction reaction forces
Entry geometry limits

Wedging conditions
Jamming conditions
Alternate strategies for accomplishing fine motion
rigid part mating
9/13/2004
The Basic Idea
In gross motions, it pays to pre-plan to prevent

errors
In fine motion, it does not pay to try to prevent
errors
So the principle is to anticipate errors and figure
out how to make assembly happen anyway
This requires us to understand three factors:
Geometry
Compliance
Friction
rigid part mating
9/13/2004
Geometry of Peg-Hole Mates

10
l/ D
=
c/
l /d = c /
C = 0.1
C = 0.01
ll /D
2 C
TYPICAL MACHINED
PARTS
C = 0.001
m = 2c
EA
CH
0.1
C = 0.0001
Source:
0.01
0.001

rigid part mating
9/13/2004
FO
PRECISION
PARTS
0.01
0.1
1.0
RADIANS
Dimensioning Practice
Source:
rigid part mating
9/13/2004
10
Geometry Definitions
Insertion Direction
c = (D-d)/D
R
D
rigid part mating
9/13/2004
11
Insertion History
rigid part mating
9/13/2004
12
Insertion History
rigid part mating
9/13/2004
13
Insertion History
rigid part mating
9/13/2004
14
Insertion History
rigid part mating
9/13/2004
15
Insertion History
rigid part mating
9/13/2004
16
Insertion History
rigid part mating
9/13/2004
17
Insertion History
rigid part mating
9/13/2004
18
Insertion History
rigid part mating
9/13/2004
19
Life Cycle of a Part Mate
Source:
rigid part mating
9/13/2004
20
Model of a Compliant Support

Source:
rigid part mating
9/13/2004
All support is assumed

concentrated at one point
and consists of one
lateral stiffness and one
angular stiffness
21
How Compliance Center Reacts to Force
Source:
Force away from

C. C. causes rotation
and translation
rigid part mating
9/13/2004
Force on C. C. causes only translation
22
Forces and Moments - Two Point
Contact Case
Source:
All applied and reaction

forces are expressed
in coordinates at pegs tip
rigid part mating
9/13/2004
23
Forces Applied During Two-point Contact

K
When Lg >> 0
When Lg ~ 0
Lg
Big
Small
Big
rigid part mating
9/13/2004
24
Making Lg Small is Good
How to do it?
Active Robot Force Feedback
Costly
Slow
Some way that acts by itself

It was invented almost 30 years ago
Called Remote Center Compliance
Reduces assembly force
Avoids one of two main failure modes
rigid part mating
9/13/2004
25
INSERTION FORCE Fz, NEWTONS
TWO-POINT
CONTACT
ONE-POINT
CONTACT
l1
l2
Insertion Force History
CHAMFER CROSSING
l*
INSERTION DEPTH l, mm
rigid part mating
9/13/2004
26
Assembly Failure Modes
Both occur during two-point contact
Wedging sets up compressive forces inside the

parts
Jamming results from incorrect insertion forces
We can derive the requirements to avoid both of

these failure modes
rigid part mating
9/13/2004
27
Wedging: Compressive Friction Forces
Prevent Insertion Regardless of Insertion
Force
== t an
-1
== t an
-1
f2
f
rigid part mating
l - d
Wedging can happen if > c/

when two-point contact occurs
9/13/2004
f2
1
l - d
Wedging can be avoided if is small

enough or if two-point contact occurs
deep enough in the hole
28
Whats a Friction Cone?

F
Friction cone
= tan-1
FN
FT
FN
Sliding will occur if FT > FN
FT /FN = tan
So, sliding will occur if tan >
and F will lie on the boundary
of the cone
rigid part mating
9/13/2004
If F is inside the cone

then sliding will not happen
because FT < FN
and F can be any value
29
Conditions for Avoiding Wedging
S =
Lg
2
L g + K / Kx
Source:
rigid part mating
9/13/2004
30
Jamming: Insertion Force Directed the
Wrong Way - Cant Overcome Friction
COMPONENT NORMAL TO PEG AXIS

COMPONENT
PARALLEL TO
PEG AXIS
INSERTION
FORCE
FRICTION
CORRESP.
TO NORMAL
COMPONENT
COMPONENT
PARALLEL TO
PEG AXIS
INSERTION
FORCE
REACTION
TO NORMAL
COMPONENT
Component of
Insertion force
Along insertion direction
Not big enough:
Peg Is Jammed
rigid part mating
9/13/2004
Component of
Insertion force
Along insertion direction
Is big enough:
Peg Goes In
31
Conditions for Avoiding Jamming
Source:
rigid part mating
9/13/2004
32
Jamming Examples

COMPONENT
PARALLEL TO
PEG AXIS
INSERTION
FORCE
FRICTION
CORRESP.
TO NORMAL
COMPONENT
COMPONENT
PARALLEL TO
PEG AXIS
REACTION
TO NORMAL
COMPONENT
Fz
Fz
Fx/Fz is small.
M/rFz is small.
Fx/Fz is big.
M/rFz is big.
9/13/2004
M
Fx
Fx
rigid part mating
INSERTION
FORCE
33
Target Expands as Depth Increases
M
rF
This point moves

As increases
= 2r
2+1
1/
-1
1/
x
z
-( 2+1 )
rigid part mating
9/13/2004
This point moves

As increases
34
Experimental Data
rigid part mating
9/13/2004
35
Experimental Data -2
rigid part mating
9/13/2004
36
Test Your Understanding
Why does insertion force rise and then fall during

two-point contact?
rigid part mating
9/13/2004
37
Experimental Data - 3
When
WhenLLgg==00
there
thereisisbarely
barely
any
anyinsertion
insertionforce.
force.
All
Allthats
thatsleft
leftisis
chamfer
chamfercrossing
crossing
force.
force.
rigid part mating
9/13/2004
38
Test Your Understanding Again
Why does the insertion force not rise after

chamfer-crossing is finished?
rigid part mating
9/13/2004
39
Review of Force Feedback Strategy
Create a coordinate frame at the working point

of the part or tool
Separate lateral and angular sensing and response
motions in that frame
Devise a response strategy
The Remote Center Compliance is a purely

passive implementation of one such strategy
rigid part mating
9/13/2004
40
Simplified Explanation of the Remote
Center Compliance (RCC)
Source:
rigid part mating
9/13/2004
41
RCC Response to External Loads
(d) RCC UNDER

LATERAL LOAD
rigid part mating
(e) RCC UNDER

ANGULAR LOAD
9/13/2004
(f) LINKAGE
RCC UNDER LATERAL
AND ANGULAR
DEFORMATION
42
(d) RCC UNDER

LATERAL LOAD
rigid part mating
9/13/2004
43
(d) RCC UNDER

LATERAL LOAD
rigid part mating
9/13/2004
44
(e) RCC UNDER

ANGULAR LOAD
rigid part mating
9/13/2004
45
(e) RCC UNDER

ANGULAR LOAD
rigid part mating
9/13/2004
46
rigid part mating
9/13/2004
First RCC Experiment - 1
r = 3
Angular Error
47
First RCC Experiment - 2
rigid part mating
9/13/2004
48
Commercial Remote Center Compliances
Source:
rigid part mating
9/13/2004
49
rigid part mating
9/13/2004
50
Assembly is a matter of geometry

Geometry contains errors
in parts
in equipment
But they are small

and usually obscured
Collisions
contain info
Collisions +
stiffness matrix
= force signals
Small stiffness
small signals and
oscillations and
position uncertainty
But: friction limits

the available information
in force data
Large stiffness
large signals but
possible instability
Active force feedback
Passive accommodation (incl small stiffness)

Will be stable
Can be smart
can be unstable
can be "strange"
can be slow
Engineered
Will avoid jamming
Accidental,
contextual
May not avoid jamming
With sensors IRCC Without sensors RCC
basically Lyapunov stable
taxonomy of fine motions
rigid part mating
Costs too much
Detect errors directly
by sensing them
geometrically
Detect errors indirectly

by sensing collisions
Collisions
are bad
Taxonomy
of Fine
Motions
Seek perfection
Accept errors
Ignore errors
(Matt Mason
approach)
9/13/2004
51
Math and Feature Models of Assemblies
Start of series of 5 classes on math/CAD models
basic matrix representations and Feature-based Design

constraint
variation
assembly sequence analysis

Datum Flow Chain
Assy Models
9/16/2004
Daniel E Whitney
Objectives of Assembly Modeling
Provide a computer environment that permits topdown design of assemblies with a persistent database
that captures the assembly as an assembly
Should link to geometry creation (CAD, rulegenerated)
Should permit specification of Key Characteristics,
constraints on location, datums and locators, and
variation analysis for KCs using the assembly model
Should permit assembly planning, vendor interfaces,
ramp-up, and production support
Assy Models
9/16/2004
Daniel E Whitney
Top-down and Bottom-up Design
Top-down defines an assembly in this order:
major customer deliverables

chains of delivery through possible parts
main part mates and necessary features
detailed part geometry
Bottom-up defines
the same things but in the reverse order
requires having some idea of final assembly layout first
Top-down used to be the only way before CAD

CAD seems to encourage bottom-up
Assy Models
9/16/2004
Daniel E Whitney
Sketch of Top-Down Assembly Design
PKC
Pylon
6
PKC
Door
DFC
5
Inlet
Relating KCs to Chains
Concept
Design
Engine
Selected concept
Architecture
Integration risk
Key dimensions
Datum Flow Chain
Featurized
DFC
Location
Responsibility
Dimensional
Control
Constraint (6 DoF)
Parts placed
in DFC with
features
Detail
design
Variation
Tooling
Constraints
Assy Models
top-down assy process
9/16/2004
Assembly
Sequences
Daniel E Whitney
Assembly
Processes
Goals of this Class
Review basic math that relates adjacent coordinate

frames
Model assemblies as chains of frames
Attach these frames to mating features
Introduce feature based design for assembly
Assy Models
9/16/2004
Daniel E Whitney
Basic Math
Uses 4x4 matrices to relate adjacent frames
Permits chaining together of parts

same math is used to describe robots
The matrix contains a rotational part and a

translational part
The matrix is designed to translate first and then
rotate so that rotation does not change position of
new frame
This matrix is a subset of a more general
projection matrix that includes perspective
Assy Models
9/16/2004
Daniel E Whitney
History of this Representation
Basic to Kinematics (Denavit & Hartenberg)

Used to model assemblies in 1970s:
S N Simunovic Masters Thesis, MIT, 1972
Edinburgh University AI Lab robot assembly 1976
Used by CAD researchers

Steve Coons, 1960s
Gossard and others, 1980s
Used by CAD systems to locate surfaces wrt each

other
Assy Models
9/16/2004
Daniel E Whitney
AXIS "A"
AXIS "B"
HAMMER
HANDLE
CARRIER
PIN
CRIMPER
RIVET
ANVIL
HAMMER
HANDLE
STAPLE
CARRIER
"Y" DIRECTION
PIN
Stapler
TOP VIEW
"Z" DIRECTION
STAPLES
SIDE VIEW
CRIMPER
RIVET
ANVIL
BASE
"X" DIRECTION
Assy Models
9/16/2004
Daniel E Whitney
Stapler Frames and KCs
Source:
Assy Models
9/16/2004
Daniel E Whitney
Frames and Chains
By following the arrows, you can travel from

frame to frame
On the previous slide, the anvil was chosen as the
origin part, and the anvil-pin joint on the anvil was
chosen as the location of the origin frame.
All arrows go out from the origin frame
You can travel from one end of a KC to the other

by moving from frame to frame along the arrows,
sometimes in arrow direction and sometimes in
reverse
Assy Models
9/16/2004
Daniel E Whitney
10
Basic Translation and Rotation
Source:
Translate first, then rotate

Assy Models
9/16/2004
Daniel E Whitney
11
Basic 4x4 Transform
R p
1
R
p
R
T= T
T = T
0 1
0
1
r11 r 12 r13
r21 r 22 r23
T =
r31 r 32 r33
0 0 0
Assy Models
9/16/2004
p
x
py
p
z
1
All the information about

location (position and
orientation) is inside the
matrix
Daniel E Whitney
12
Basic Rotation Matrices

0
1
0 cos.
Rot(x,. ) =
0 sin .
0
0
cos
Rot(y, ) =
sin
0
sin.
cos.
0
9/16/2004
Z1
Zo
.
Xo X 1
0 sin 0
1
0
0
0 cos 0
0
0
1
cos sin
sin cos
Rot(z, ) =
0
0
0
0
Assy Models
0
0
Z1
Y1
Yo
. is in radians
Zo
Yo
Y1
Xo X
1
0 0
0 0
1 0
0 1
Daniel E Whitney
Zo
Z1
Y1
Yo
Xo
X1
No translation 13
Basic Translation Matrix

No rotation
1
trans(x, y,z) =
0
1 0 y
0 1 z
0 0 1
0 0
This and the three basic rotation matrices

are matlab .m files on MIT Server that you can use
Assy Models
9/16/2004
Daniel E Whitney
14
Composite Transforms
T02 = T01 T12

2
R01 p01 R12 p12

=
T02 = T
T
0 1
0 1
R01 R12 R01 p12+p01

T
0
1
Assy Models
T02
p01
p12
1
T01
TT0101locates
locatesframe
frame11in
inframe
frame00coordinates
coordinates
TT12 locates
frame 2 in frame 1 coordinates
12 locates frame 2 in frame 1 coordinates
TT
locates frame
22in
frame
00coordinates
02
frame
in
frame
coordinates
9/16/2004
02 locates
Daniel E Whitney
R12
T12
R01
15
Transform Order is Important
TAB = TA TB . TBA = TB TA
Assy Models
9/16/2004
Daniel E Whitney
16
More About Transform Order
Ti +1 works on new axes created by Ti
T1 n = T01T12 ...T(n 1) n
Assy Models
9/16/2004
Daniel E Whitney
17
Nominal Mating of Parts
Source:
Assy Models
9/16/2004
Daniel E Whitney
18
Example
Source:

and Role in Product Development. New York, NY:
Oxford University Press, 2004. ISBN: 0195157826.
Coordinate Frames
Assy Models
9/16/2004
Daniel E Whitney
>> TAB = trans (3,0,4)
TAB
1
0
=
0
0 0 3
1 0 0
0
1 4
0 0 1
MATLAB(TM) Code
19
Front, Top, and Side Views
Top
Other
Side
Front
Side
Bottom
Assy Models
9/16/2004
Daniel E Whitney
20
Another Example
>> TAC = TAB roty ( dtr (90))

Source:

TAC
0
0
=
1
0 1 3
1 0 0
0 0 4
0 0 1
function degtorad = dtr(theta)

% Converts degrees to radians
degtorad=theta*pi/180;
Assy Models
9/16/2004
Daniel E Whitney
21
Example Feature on Part

Source:
>> TAD = trans (3, 2,4)roty (dtr (90))
TAD
0
0
=
1
0 1
1 0
0 0
0 0
3
2
4
Assy Models
9/16/2004
Daniel E Whitney
22
Feature on Second Part
Source:

Assy Models
9/16/2004
Daniel E Whitney
>> TEF = trans (6,0,1)

1 0 0 6
0 1 0 0
TEF =
0 0 1 1
0 0 0 1
23
Assembling These Parts
>> TDE = rotz (dtr (180))

1 0 0
0 1 0
TDE =
0 0 1
0 0 0
>> TAF = TAD TDE TEF

Source:
TAF
0
0
0
0 0 1 4
0 1 0 2
=
1 0 0 10
0 0 0 1
4x4_examples copy
Assy Models
9/16/2004
Daniel E Whitney
24
Varied Part Location Due to Tolerances
AF
TFB'
T AB'
B'
TBB'
The varied location of Part B can be calculated

from the nominal location of Part A. This process
can be chained to Part C, etc., including errors on
Part B. It uses the same math as the nominal model.
Assy Models
9/16/2004
Daniel E Whitney
25
Equations for Connective Models

Nominal
Source:
Varied
Source:
Assy Models
9/16/2004
Daniel E Whitney
26
A Hierarchy of Assembly Models
Bill of materials - lists the parts in no particular order
Structured BOM - aka drawing tree - groups parts by subassembly

Liaison graph - (Bourjault) parts are dots, joints are lines
Ordered liaison graph - the lines have arrows

Attributed liaison graph - the lines have
constraint or feature information
Ordered-attributed liaison graph (Datum Flow Chain)
Assy Models
9/16/2004
Daniel E Whitney
27
Assembly Types Classified Technically
Source:
Assy Models
9/16/2004
Daniel E Whitney
28
Assembly Types Classified by Liaison
Diagram Form
Hub and spokes

Loop
valve
head
cmsf
Network
crsf
block
rod
piston
Stack
Assy Models
9/16/2004
Daniel E Whitney
29
Inside a Car Engine
Source:
Assy Models
9/16/2004
Daniel E Whitney
30
Feature-based Design
Seeks to rise above geometry and capture intent

First efforts in machined features
slots
pockets
holes
Features look different depending on how they are
made
drafted walls if cast
pockets if cut
pocket
drafted
wall
Feature recognition may be needed

Assy Models
9/16/2004
Daniel E Whitney
31
Assembly Features
Each feature has nominal geometry and a reference
coordinate frame expressed as a 4x4 matrix. It also has
a variety of other attributes as needed for its type.

Source:

.
ISBN: 0195157826.
Assy Models
9/16/2004
Daniel E Whitney
Feature recognition
may not be needed.
Story: Feature recognition of Philips

Head screws
32
Assembly Features and Assembly Data
Models
An assembly feature
A
Origin frame
for a part
Validity of this
model does not
depend on part
shapes being
correct
Typical CAD Model

Based on World Coordinates
9/16/2004
B
CC
C
An Assembly Model
Based on Connecting
Assembly Features
Assy Models
Validity of this
model depends
on shapes of parts
being correct
Assembly model with

variation based on error
in feature location in
part B
Daniel E Whitney
33
A Really Bad Example
Source:
Assy Models
9/16/2004
Daniel E Whitney
34
A Better Way, Based on Features and
Frames
R
2
p12
y
y
O
Assy Models
p01
x
9/16/2004
Daniel E Whitney
35
What The Different Models Do
World coordinate model is good for drawing

pictures of the nominal arrangement; can find
interferences based on errors in the nominal but
cant help you find out why they happened
Chained model is good for capturing relational
information and design intent, and can trace
effects of variation from the nominal; wont
necessarily find interferences because it forces
things to be assembled; can help you find out
why things dont fit
Assy Models
9/16/2004
Daniel E Whitney
36
Assembly Modeling in CAD
Parts can be defined with mating features on them.
Features can be mated directly.
An assembly database builds up automatically.
Assembly knowledge can be accessed.
PEG
HOLE
PART B
A
RT
PA
FEATURE 8
FEATURE 2
8
14
PART A
PART B
FIXT
PART A
FIXTURING
OPTIONS, INCL
TOLERANCES
Assy Models
9/16/2004
ASSEMBLY
CONDITIONS,
EFFECT OF
CLEARANCES
AND FRICTION
PART B
ASSY OF FEATURE
PAIRS
Daniel E Whitney
37
Information in Assembly Models
What parts mate to what parts
What features define the mates and where they are on

the parts
What interfaces must be controlled, plus a formal way
of describing them
Constraints and rule-checking
about assembly in the small
about assembly intent in terms of features
about assembly in the large, including alternate parts
It is a completely abstract and general model based on

connectivity
Geometry is an attribute of the parts
Assy Models
9/16/2004
Daniel E Whitney
38
Example Assembly Data Model
DECLARED ASSEMBLY FEATURE

ON PART_____ (text)
TYPE NAME_____SPECIFIC NAME_____
LOCATION ON PART _____ (4x4)
LOCAL ESCAPE DIRECTION____
(DEFAULT: Z AXIS)
TOLERANCES
GEOMETRY_____
PARAMETERS________
TOLERANCES_____
OPTIONAL: FEATURES IT CAN MATE TO

CONSTRAINTS
MATED TO FEATURE___ ON PART ___
CASE... (other parts in other circumstances)
Assy Models
9/16/2004
Daniel E Whitney
39
Seeker Head
Source:
Assy Models
9/16/2004
Daniel E Whitney
40
Seeker Liaison Diagram
Source:
Assy Models
9/16/2004
Daniel E Whitney
41
PART
PART NAME
BASE
A
Assy Models
OUTER GIMBAL
INNER GIMBAL
OUTER BEARING
RETAINING
SCREW
OUTER BEARING
RETAINING
SCREW
INNER BEARING
RETAINING
SCREW
INNER BEARING
RETAINING
SCREW
9/16/2004
8
1
2
3
4
1
2
3
1
2
1
2
3
1
2
1
2
3
1
2
1
2
3
1
2
FEATURE NAME
BEARING BORE
TRUNNION BORE
TRUNNION BORE
BEARING BORE
BEARING BORE
TRUNNION BORE
TRUNNION BORE
BEARING BORE
RET. SCREW HOLE
TRUNNION
TRUNNION
RET. SCREW HOLE
RET. SCREW HOLE
TRUNNION
TRUNNION
RET. SCREW HOLE
BORE
OUTER DIAMETER
INNER RACE FACE
THREAD
HEAD
BORE
OUTER DIAMETER
INNER RACE FACE
THREAD
HEAD
BORE
OUTER DIAMETER
INNER RACE FACE
THREAD
HEAD
BORE
OUTER DIAMETER
INNER RACE FACE
THREAD
HEAD
Daniel E Whitney
FEATURE CLASS
(CHAMFERED) BORE
BORE
BORE
(CHAMFERED) BORE
(CHAMFERED) BORE
BORE
BORE
(CHAMFERED) BORE
THREADED BORE
(CHAMFERED) PIN
(CHAMFERED) PIN
THREADED BORE
THREADED BORE
(CHAMFERED) PIN
(CHAMFERED) PIN
THREADED BORE
(CHAMFERED) BORE
(CHAMFERED) PIN
PLANE
THREADED PIN
PLANE
(CHAMFERED) BORE
(CHAMFERED) PIN
PLANE
THREADED PIN
PLANE
(CHAMFERED) BORE
(CHAMFERED) PIN
PLANE
THREADED PIN
PLANE
(CHAMFERED) BORE
(CHAMFERED) PIN
PLANE
THREADED PIN
PLANE
Parts and Features List
FEATURE
1
2
3
4
1
2
3
4
5
6
7
42
Assy Models
9/16/2004
Daniel E Whitney
Liaison Diagram with
Features
Source:
43
Feature-based Design Video
Made at Draper in 1990

Illustrates a bottom-up approach
First demo of integrated design of assembly tools

hooked to a CAD system
parts designed with mating features
parts joined by connecting the features
liaison diagram constructed automatically
assembly data model passed to CAE routines for

assembly sequence, assembly system design, and
economic analysis
Assy Models
9/16/2004
Daniel E Whitney
44
Kinematic Constraint in Assembly
Topics
Assembly as zero-stress location
AKA Exact Constraint, Proper Constraint, or
Kinematic Design
AKA 3-2-1 assembly
Assembly features as carriers of constraint,
operationalizing the coordinate frames
Non-zero-stress assemblies
Mathematical analysis of constraint
Class 6-7 Constraint
9/21/2004
Daniel E Whitney
Assembly = Constraint
Assembly = removal of dof = application of

constraint
As constraint is applied, degrees of freedom are
taken away so that a part gets to where it is
supposed to be.
When parts are where they are supposed to be, the
KCs can be delivered, assuming no variation
This is called the nominal design
9/21/2004
Daniel E Whitney
Parts Locate Each Other to Deliver Quality at
the Customer Level
DOOR
CRAFTMANSHIP KC
CAR BODY
KC=Key Characteristic
BODY TO
HINGE FLAP1: 6 DOF
HINGE FLAP 1 TO
HINGE FLAP 2: 5 DOF
HINGE FLAP 2 TO
DOOR: 6 DOF
9/21/2004
Daniel E Whitney
Definitions of Assemblies
Whitehead: An instrument can be regarded as a

chain of related parts any mechanism whose
function is directly dependent on the accuracy
with which the component parts achieve their
required relationships.The Design and Use of Instruments and
Accurate Mechanism, by Thomas North Whitehead, 1934
Whitney: An assembly is a chain of coordinate

frames on parts designed to achieve certain
dimensional relationships called Key
Characteristics between some of the parts or
between features on those parts. Designing Assemblies,
Research in Engineering Design, (1999) 11:229-253.
9/21/2004
Daniel E Whitney
The Three Principles of Statics
Geometric compatibility
Force and moment balance
Stress-strain-temperature relations
We assume rigid parts, so the 3rd principle

does not apply to our work
9/21/2004
Daniel E Whitney
Properly Constrained and Over-
constrained Assemblies
Assemblies that function by geometric compatibility
and force equilibrium alone are called
statically determinate
properly constrained
kinematic or semi-kinematic ~ 3-2-1
You just put them together
Assemblies that require stress analysis are called
statically indeterminate
over-constrained
You cant just put them together

Constraint is a property of the nominal design
9/21/2004
Daniel E Whitney
Constraint is Accomplished by Surfaces
in Contact
The contact permits

some dof to move
with respect to
each other and
prevents motion of
other dof.
The black ones
can move.
The red ones cant.
9/21/2004
Z
Y
X
Daniel E Whitney
Different kinds of
surface pairs
permit and prevent
motion of
different dof
Degrees of Freedom
An objects location in space is completely
specified when three translations (X, Y, Z) and
three rotations (. X ,. Y ,. Z) are specified
How many DOFs are constrained?
z
cube on table (x-y plane)
.z
cube at floor-wall interface

cube at floor-two walls interface
ball on table
ball at floor-wall interface
round peg in blind round hole
.y
.x
Think about the constrained ones

9/21/2004
Daniel E Whitney
Constraint - 1
Proper constraint provides a single value for each of a

bodys 6 degrees of freedom
This is done by establishing surface contacts with
surfaces on another part or parts
If less than 6 dof have definite values, the body is
under-constrained
If an attempt is made to provide 2 or more values for
a dof, then the body is over-constrained because rigid
bodies have only 6 dof
Any extra needed dof must be obtained by deforming
the object
9/21/2004
Daniel E Whitney
Example of Proper and Over Constraint
Two pins in holes
One pin in hole, one pin in slot
This is over-constrained
in the X direction
9/21/2004
Daniel E Whitney
This is properly constrained in X
10
Constraint - 2
Proper constraint permits an assembly to have

unambiguous chains of delivery of KCs
Two pins in holes
One pin in hole, one pin in slot
Which pin determines X

of the blue plate? Cant tell!
The left pin determines X

of the blue plate
If the X location of the left pin changes, where

will the blue plate go?
9/21/2004
Daniel E Whitney
11
Cylinder Head Mate to Block
LOCATING PIN
LOCATING PIN
9/21/2004
Daniel E Whitney
12
When Parts are Joined, Degrees of
Freedom are Fixed
Parts join at places called assembly features
Different features constrain different numbers and

kinds of degrees of freedom of the respective parts
(symmetrically)
square peg to square
6
Parts may join by
one pair of features

multiple features
several parts working together,
each with its own features
5
1
6
5
1
blind hole
pin to hole
pin to slot
pin to hole
pin to slot
When parts mate to fixtures, dofs are constrained
9/21/2004
Daniel E Whitney
13
Assembly Features Carry Constraint
Kinematic constraint passes from part to part

across the feature joints
The degree of constraint can be calculated using
Screw Theory
Proposed feature designs and KC chains can be
examined using Screw Theory to see if they
convey the desired amount of constraint and avoid
constraint errors
9/21/2004
Daniel E Whitney
14
Constraint - 3
CAD systems analyze constraint
But CAD systems, developers, and researchers do
not mean mechanical constraint as we define it
They mean geometric location consistency
Many designs called properly constrained by CAD

systems are actually over-constrained
Different CAD systems do this analysis different
ways and can disagree about the same assembly
9/21/2004
Daniel E Whitney
15
How CAD Systems Test Constraint
A closed chain of frames is set up

A numerical test is done to see if the chain closes
If, so, the assembly is called fully constrained

Actually, it should be called consistent
Detailed tests for constraint/consistency problems

are done by making small shifts and testing for
interference
Tolerance studies are done the same way
Analysis requires detailed geometry
Results depend on how the model was built
9/21/2004
Daniel E Whitney
16
Some Examples
C
9/21/2004
Daniel E Whitney
17
Proper Constraint Zero Stress Assembly
Kinematic design seeks to determine locations of

parts solely or almost solely by means of
geometric compatibility
Locked-in stresses are kept so low that they are
negligible
Exact constraint design is equivalent to 3-2-1
assembly
In effect, in a kinematic assembly, the parts act as
fixtures for each other
You just put them together
9/21/2004
Daniel E Whitney
18
Non-zero Stress Assemblies Can Be OK
Three-leg stool rests firmly and is fully

constrained
Four-leg stool gives the security of an extra leg
and wider footprint but will not rest firmly unless
the legs are elastic enough to deform until all four
are in contact, or there are screw adjustments
Other examples
shrink fit of wheel on shaft

preloaded angular contact ball bearing pairs
pre-stressed concrete and case-hardened armor plate
planetary gear trains
9/21/2004
Daniel E Whitney
19
Constraint Mistakes Happen
Designers make constraint mistakes

Mostly they make over-constraint mistakes
You can see under-constraint mistakes because they
permit unwanted motions
You cant see the stresses caused by over-constraints
It would be nice to have a test that looks for

over/under constraint
9/21/2004
Daniel E Whitney
20
How Airplanes are Built
Boeing:
Ensure that there is open space at max material condition

Fill the gap with shims, reducing gap to XXX
Report remaining gap to Engineering
Lately: use better process control to predict gaps and prepare
standard shims in as many cases as possible
Airbus:
Make parts from 3D CAD/NC
Join them directly
Look, Dr Whitney, no shims!
Both attempt to limit locked-in stress

9/21/2004
Daniel E Whitney
24
Good Over-constrained Assemblies
Preloaded angular contact bearing systems
Preload increases contact stress, creating a stiff bearing

system (see next page)
Planetary gears - redundant locators, no stress
Shrink fit
Heated wheel slips on over shaft, shrinks upon cooling to
make a super-tight joint
Beam built in at both ends

Its stiffer for the same cross section than a simplysupported beam because the ends can support a moment
A good design permits longitudinal motion at the ends
In each case there is an underlying properly

constrained
system!
9/21/2004
Daniel E Whitney
25
Preload
nut
Rotating
shaft
9/21/2004
Daniel E Whitney
Preloaded Angular Contact

Bearings
Fixed housing
26
Planetary Gear Systems
Source:
9/21/2004
Daniel E Whitney
27
Why Does Over-Constraint Occur?
Forces or torques are deliberately inserted, e.g.

Shrinking
Tightening a lock nut
The design attempts to fix more than 6 degrees of
freedom of a part, e.g.
The x position is determined by the parts left end
The parts x position is determined by the parts right
end
There is a fight whose outcome is compression in the x
direction and no easy way to calculate the x position
9/21/2004
Daniel E Whitney
28
How Different Approaches Deal with
Constraint
B
C
Use consistency
check to find
joining errors
Cant detect most
over- or
under-constraints
Local
Feature
Model
World
Model
9/21/2004
Use
interference
check to find
shape errors
Daniel E Whitney
Use screw
theory to
find joining
errors: overor underconstraint
29
Whats the Solution to Over-constraint?
Increase the diameter of one hole

Increase the diameter of both holes
Use one hole and one slot
Match drill one hole pair and use a slot
Accept a little locked-in stress

We will use these ideas later when we use features
to transfer dimensional constraint and location
from part to part using the Datum Flow Chain
over-constraint = ambiguity about dimensional transfer
requires a stress analysis to decide where the parts are
9/21/2004
Daniel E Whitney
30
Tipoffs for Over-constraint
It takes skill to put the parts together and get them

just right
The assembly task is operator-dependent
Fasteners have to be tightened in a particular

sequence
It is hard to get welded parts out of the fixture
Some parts will assemble easily but other

identical ones will not
You can never get everything to line up the way
you want it to
Results are inconsistent
9/21/2004
Daniel E Whitney
32
Location, Constraint, Stability
Constraint determines location

this is done using mates (Whiteheads locators)
Stability keeps or effects location

this is done with contacts (Whiteheads effectors)
screws, wave washers, clamps, welds,
glue, chewing gum, etc
Locating gets it there

Stabilizing keeps it there
This area is a common cause of confusion
9/21/2004
Daniel E Whitney
33
Location and Stability
You know where it is,

even if it might not stay there
It will stay there (its welded)
but you dont know where it is
9/21/2004
Daniel E Whitney
34
Whiteheads Definitions
(examples use less than 6 dof)
Effected by small force
Enough Joints
Effected
by
a large
force
Too Many Joints
Pure kinematic design

3 legged stool with pointtipped legs
Semi-kinematic design with

redundant constraint in the
small contact area of each
locator
Redundant constraint
4 legged stool with point-
tipped legs or non-zero
contact area at each tip
This is really two 3 legged

stools - your choice which
one!
3 legged stool with nonzero contact area at each

leg tip
Semi-kinematic design
3 legged stool with nonzero contact area and each

leg bolted down tight
9/21/2004
Daniel E Whitney
Over-constraint
4 legged stool with each
leg bolted down tight
35
Force Closures and Form Closures
Force closures are one-sided

They support force in one direction at a definite location
They can provide proper constraint
Form closures are two-sided

They can support unlimited force
They will generate over-constraint unless some clearance
is provided
If clearance is provided, then the location is no longer
definite
9/21/2004
Daniel E Whitney
36
One-Side and Two-Side Constraints
One-side (AKA force closure)

Needs an effector
Gives perfect knowledge of location but cant support
an arbitrary force in all directions
Two- or multi-side constraint (AKA form closure)
Needs no effector and can support arbitrary force

Contains its own stabilizer
Actually contains over-constraint
If we relax this over-constraint with a little clearance
then we lose perfect knowledge of location
See Exact Constraint Design Using Kinematic Principles by Douglass Blanding

9/21/2004
Daniel E Whitney
37
Taxonomy of Assemblies
Underconstrained
Mechanisms
mistakesattempts to
achieve location
that lock in
stress or fail
to locate
Over-constrained
Properly
Constrained
zero-stress
(or almost
zero stress)
assemblies
that deliver
Interference
KCs by
and stress
achieving
location
(using transform T)
9/21/2004
Pre-loaded
bearing sets
=Noise
Daniel E Whitney
Line fit
Clearance
AKA redundant
Cant
happen
but needed
for analysis
Duplicated
arrangements
38
Mathematical Family of Joints
It contains 7 kinds of features and 28 possible joints
arbitrary surface
parallelepiped
body of revolution
cylinder
plane
sphere
They all are one-sided, as are all 28 combinations
Elements can be used to build features from scratch
Next slide shows cylinder, plane, and sphere

9/21/2004
Daniel E Whitney
39
Basic Surface Contacts and Motions They Allow
Allowed motion
of the black part
is shown, when
the red part is
stationary
Source:
Table 4-3 in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
9/21/2004
Daniel E Whitney
40
Engineering Family of Assembly
Features
It contains 17 shapes and counting
peg and blind/through hole

peg on plane and blind/through hole
peg on plane and slot
key in keyway
etc.
You can add any that you want
They are both one-sided and two-sided
They are built using basic surfaces

9/21/2004
Daniel E Whitney
41
Examples of Engineering Features

Pin in Hole
x
d1
d2
Pin in Prismatic Slot

x
Pin in Oversize
Hole
y
z
Pin in Slot
P xc
Two Plates
Px
Nxc
Py
ymax
ymin
y
9/21/2004
Daniel E Whitney
Jeff Adams SM Thesis
42
Assembly Feature Construction Options
Build from elementary surfaces

Reveals two-side over-constraints inside a feature
Permits detection of over-constraints caused by too
many one-side constraints
Use basic mechanical shapes like holes and slots

Suppresses over-constraints inside a feature
Permits detection of over-constraints caused by too
many one- or two-sided constraints
The constraint testing methods described next

must be applied with caution, depending on which
option is used
9/21/2004
Daniel E Whitney
43
Assembly Feature Construction Options
Elementary surfaces
(TTRS) - one-sided
Have location, no size
Intersect these
to make new features.
They will contain overconstraint if 2-sided.
They will have location and size
if 2-sided
Make new features

from scratch. They will not
contain over-constraint
even if 2-sided.
They will have location and size
if 2-sided
Need to add clearance

on size to relieve internal
over-constraint
9/21/2004
Daniel E Whitney
Intersect these
to make compound
features - will
contain overconstraint if 2-sided
or if mistakes are made
44
Kinematicians Approach to Constraint
Kinematicians are interested in things that move

So they are interested in mobility M
Assemblies may or may not move
We are interested in redundancy or negative mobility
The Kutzbach criterion attempts to analyze both but it
can give the wrong answer
M = 6(n g 1) + fi
n = number of links
g = number of joints
f i = total dof of all joints

M = 0.
proper constraint
M <0.
M >0.
over - constraint
under - constraint
9/21/2004
Daniel E Whitney
If the linkage is planar,

the Grbler criterion is used.
It is the same as the Kutzbach
Criterion except that 6 is
replaced by 3
45
Does the Kutzbach Criterion Work?
Source:
Figure 4-4, 4-5, 4-6 in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
M = 3(n g 1) + fi
n = number of links
f i = total dof of all joints
9/21/2004
Daniel E Whitney
M = 0 . proper constraint
M < 0 . over - constraint
M > 0 . under - constraint
46
Our Approach
Mobility and constraint require separate analyses
If an assembly is not over-constrained and it is not

under-constrained, then and only then is it
properly constrained
Mobility and constraint analyses are done using
Screw Theory
The Kutzbach criterion is a nave attempt to do
what Screw Theory can do
9/21/2004
Daniel E Whitney
47
Basics of Screw Theory
Kinematic joints permit motion and resist force

Each degree of freedom of motion is called a twist
Each degree of force resistance is called a wrench
A twist has the form T = [. x . y . z vx vy vz]
A wrench has the form W= [fx fy fz Mx My Mz]
Twists and wrenches are reciprocals of each other
(under certain conditions)
That is, directions that allow motion cannot
support force and vice versa
9/21/2004
Daniel E Whitney
48
Constraint is Accomplished by Surfaces
in Contact
The contact permits

some dof to move
with respect to
each other and
prevents motion of
other dof.
Z
Y
X
The black ones

can move.
The red ones cant.
The red ones can support force.

The black ones cant.
9/21/2004
Daniel E Whitney
Different kinds of
surface pairs
permit and prevent
motion of
different dof
49
Twist Space and Wrench Space Describe
the Behavior of Two Surfaces in Contact
Z
Y
WRENCH SPACE
TWIST SPACE
Wrench space and twist space are reciprocal

9/21/2004
Daniel E Whitney
50
Reciprocal of Screw
Reciprocal of a twist is a wrench
Reciprocal of a wrench is a twist
Twist and wrench are reciprocal if wrench cannot

do work along direction of twist (no friction)
This means: Twist Wrench = 0 ( = dot product)
Equivalently, wrench is in the null space of twist
Rank of twist matrix + rank of its reciprocal

wrench matrix = 6
9/21/2004
Daniel E Whitney
51
Forces and Velocities in Constraint
In general, force and velocity (or any pair of

variables whose product = power) are related by
an impedance (v=iR, for example)
When parts are rigid and friction is zero, all
impedances are zero or infinite.
Any velocity!
No force!
Any force!
No velocity!
9/21/2004
Daniel E Whitney
52
Relation Between Motions (Twists) and

Forces (Wrenches)
Any velocity!
No force!
No friction!
Y
Any force!
No velocity!
No compliance!
}
}
T= recip (W) = [0 0 0 0 1 0]
9/21/2004
W = (recip (T))
= [0 0 0 -1 0
0 0 0 0 -1
0 0 0 0 0
1 0 0 0 0
0 0 1 0 0
Daniel E Whitney
0
0
1
0
0]
53
Twist Representation of Plate and

Slotted Pin Joint Shows What Motions
are Allowed
x
x
1
part ctr
coords
twist = [. x . y . z vx vy vz]
= 0 0 1 2 -1 0 Z rotation
0 0 0 0 1 0 Y translation
The twist has two rows
because the feature
allows two different motions
See next two slides for explanation
9/21/2004
Daniel E Whitney
54
The Translation Part of the Twist

x
{
{
[0 0 0 0 1 0]
.
1
part ctr
coords
z
How the red dot moves:
9/21/2004
Daniel E Whitney
55
The Rotation Part of the Twist

x
{
{
[0 0 1 2 -1 0]
.
1
part ctr
coords
z
How the red dot moves:
9/21/2004
Daniel E Whitney
56
Wrench Representation of Plate and
Slotted Pin Joint Shows What Forces and
Moments Can Be Resisted
wrench = [fx fy fz Mx My Mz]
x
1
part ctr
coords
9/21/2004
=1 0 0 0 0 -2 = X force at pin
0 0 1 0 0 0 = Z force
0 0 0 1 0 0 = X Moment
0 0 0 0 1 0 = Y Moment
The wrench has four rows
because the feature
can resist four different forces
Daniel E Whitney
57
Motion and Constraint Analysis When
Parts are Joined by Several Features
Parts are joined by features
Multiple features may constrain properly or they

may contain over/under constraints
Motion analysis permits detection of underconstraint
Constraint analysis permits detection of overconstraint
Proper constraint = not over- and not underconstrained
9/21/2004
Daniel E Whitney
58
Motion and Constraint Analysis Using

Screw Theory-1
Each feature is represented by a twist that shows
the motions it allows, expressed in part center
coordinates as: [. x . y . z vx vy vz]
If a part joins another part via several features,
then the intersection of all the features twists
shows the net allowed motion
If a net motion is allowed, then all the features
allow that motion
All feature twists and the twist intersection are
expressed in the same coordinate system
9/21/2004
Daniel E Whitney
59
Motion Analysis (Twist Intersection)
Algorithm
For each feature i joining two parts, having twist Ti

find the associated wrench Wi using
Wi = recip(Ti)
Form the union of all the Wi = WU

WU = [W1;W2;...] (using MATLAB notation)
Find resultant net twist from

T = recip(WU)
TR=rref(T) (rref simplifies the result for easy reading)
If any motion exists in TR, then every joint this part

has allows this motion
9/21/2004
Daniel E Whitney
60
.m files for Screw Calculations
function R = recip(T)
% Takes the reciprocal of a screw matrix
p = (null(T))';
[i,j]=size(p);
if i>0
R = flip(p);
R=rref(R);
else
disp('empty matrix')
R=zeros(0);
end
% rref finds row-reduced echelon form
% takes transpose
9/21/2004
function W = flip(WU)
% FLIPs columns of WU
% col 1 becomes 4, 2 becomes 5, and 3 becomes 6
% col 4 becomes 1, 5 becomes 2, and 6 becomes 3
[i,j] = size(WU);
if j == 6
for l=1:i
for k=1:3
W(l,k) = WU(l,k+3);
W(l,k+3) = WU(l,k);
end
end
W;
else
end
Daniel E Whitney
61
Example: Cube on Floor at Wall(s)
T2=[0 0 0 1 0 0]
T1=[0 0 0 0 1 0]
X
T12=?
9/21/2004
Daniel E Whitney
62
Example Twist Intersection
T1=[0 0 0 0 1 0]
W1=recip(T1)
WU=[W1;W2]
T1 =
W1 =
WU =
1
0
0
0
0
T2=[0 0 0 1 0 0]
T2 =
0
T=
0
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
0
0
1
0
0
0
0
0
1
0
0
0
0
0
1
W2=recip(T2)
W2 =
0
0
0
0
0
9/21/2004
T=recip(WU)
empty matrix
1
0
0
0
0
0
1
0
0
0
Daniel E Whitney
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
1
[]
So it cant move
63
Motion and Constraint Analysis Using
Screw Theory-2
Each twist has a reciprocal called a wrench expressed
in part center coordinates as [fx fy fz Mx My Mz]
It represents all the forces and torques that the feature

can transmit to a mating part
The intersection of some wrenches shows if there
is/are direction(s) that they all constrain
To find if a part is over-constrained, it is necessary to
intersect all combinations of wrenches until all overconstrained directions have been found
9/21/2004
Daniel E Whitney
64
Constraint Analysis (Wrench
Intersection) Algorithm
For all features joining two parts, each having Ti

Form their union TU = [T1;T2;...]
find W = recip(TU)
find WR = rref(W) - for ease of interpretation
If a wrench appears in WR then every joint can

exert that wrench, which means over-constraint
But if WR is empty, it does not mean that there is
no over-constraint!
You have to check all pairs, triplets, etc.
9/21/2004
Daniel E Whitney
65
Summary of Twist and Wrench Intersection

(X i ) = recip{. [recip(X i )]}
Intersection of twists:
T1 .
T2 .
T3 .
etc.
W1
W2
W3
. (Wi)=WU .
Intersection of wrenches:
W1 .
T1
. (Ti)=TU .
W2 .
T2
W3 .
T3
etc.
. =
TR
WR
reciprocal . = union
9/21/2004
Daniel E Whitney
66
Example: Assembly Made by Combining
Several Features
The assembly
2
The features used to make it
4
x
x
y
y
f2
2
y
x
f3
f1
Y
y
z
Z
f1=6
9/21/2004
Daniel E Whitney
f2=16
f3=Library Feature 9
67
Motion Analysis Results
T1=[0 0 1 2 -2 0;0 0 0 0 0 1]
T1 =
0 0 1 2 -2 0
0 0 0 0 0 1
T2=[0 0 1 6 -2 0;0 0 0 1 0 0;0 0 0 0 0 1;0 1 0 0 0 2]
T2 =
0 0 1 6 -2 0
0 0 0 1 0 0
0 0 0 0 0 1
0 1 0 0 0 2
T3=[0 0 0 1 0 0;0 0 0 0 1 0;0 0 1 0 0 0]
T3 =
0 0 0 1 0 0
0 0 0 0 1 0
0 0 1 0 0 0
W1=recip(T1)
W1 =
1.0000
0
0 1.0000
0
0
0
0
W2=recip(T2)
W2 =
0 1.0000
0
0
W3=recip(T3)
W3 =
0 0 1 0
0 0 0 1
0 0 0 0
WU=[W1;W2;W3]
WU =
1.0000
0
0
0
0 -2.0000
0 1.0000
0
0
0 2.0000
0
0
0 1.0000
0
0
0
0
0
0 1.0000 -0.0000
0 1.0000 0.0000
0 -0.0000 2.0000
0
0
0 1.0000
0
0
0
0 1.0000
0
0
0
0
0
0 1.0000
0
0
0
0
0
0 1.0000
0
TU=recip(WU)
0
0
0
0 -2.0000
0
0 2.0000
0 1.0000
0
0
0
0 1.0000 -0.0000
TU =
0
0.0000
0 -0.0000 2.0000
0 1.0000
0
0
0
0
1
0
0
0
9/21/2004
0 1.0000 2.0000 -2.0000 0.0000
Part can rotate about Z

Rotation center is at (2,2)
Daniel E Whitney
68
Constraint Analysis Results
Intersect all twists:

TU123=[T1;T2;T3]
TU123 =
0 0 1 2 -2 0
0 0 0 0 0 1
0 0 1 6 -2 0
0 0 0 1 0 0
0 0 0 0 0 1
0 1 0 0 0 2
0 0 0 1 0 0
0 0 0 0 1 0
0 0 1 0 0 0
WU123=recip(TU123)
WU123 =
0 0 0 1 0 0
Mx
2
x
X
y
y
f2
f1
Y
y
Intersect all pairs of twists:
TU12=[T1;T2]
TU12 =
0 0 1 2 -2 0
0 0 0 0 0 1
0 0 1 6 -2 0
0 0 0 1 0 0
0 0 0 0 0 1
0 1 0 0 0 2
WU12=recip(TU12)
y
z
WU12 =
0 1.0000 -0.0000
0 0.0000 2.0000
0
0
0 1.0000
0
0
F &M
Mx
TU13=[T1;T3]
TU13 =
0 0 1 2 -2 0
0 0 0 0 0 1
0 0 0 1 0 0
0 0 0 0 1 0
0 0 1 0 0 0
WU13=recip(TU13)
WU13 =
0
0
0 1.0000
0
0
0
0
0
0 1.0000 0.0000
9/21/2004
Daniel E Whitney
TU23=[T2;T3]
TU23 =
0 0 1 6 -2
0 0 0 1 0
0 0 0 0 0
0 1 0 0 0
0 0 0 1 0
0 0 0 0 1
0 0 1 0 0
WU23=recip(TU23)
WU23 =
0 0 0 1 0
Mx
Mx
My
69
0
0
1
2
0
0
0
0
Constraint Analysis Interpretation
WU12 =
0 1 0
Y
Mx
WU12 =
0 0
My
WU23 =
0 0 0
What it means:
Fy=1
Z
1
9/21/2004
What MATLAB says:

Fy & Mz
X
2
Fy=1
Mz=2 Y
0 0
0 0
Daniel E Whitney
X
Y
70
Second Example
x
X
Analysis results:
f2
f1
Y
y
y
2
f3
z
T2 =
0
0
0
1
0
0
0
0
No motion is possible
Over-constraint exists about X
and Y
1
0
0
0
6
0
0
0
-2 0
1 0
0 1
0 -6
WU123 = 0
WU12 = 0
WU13 =
0 0
0 0
WU23 = 0
0
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
T1& T 3 the same
TU = []
empty matrix
9/21/2004
Daniel E Whitney
71
A Way to Eliminate Some Two-Side
Over-constraints
y
X
Y
1
2
y
y
z
9/21/2004
Daniel E Whitney
Use library features

18 and 19 instead of
6 and 16. These
X-Y locators
do not fight with
each other over
orienting the top
plate and do not
fight with the
bottom plate over
orienting the top plate
72
How to Handle More Than Two

Features: Motion Analysis
T1 .
T2 .
T3 .
etc.
W1
W2
W3
. =
9/21/2004
. (Wi)=WU .
TR
Empty: no motion
possible*
Not empty:
some motion
Possible#
*No direction of motion can

be provided by all these features
recip
#One or more directions of

motion can be provided by
all these features
Daniel E Whitney
73
How to Handle More Than Two

Features: Constraint Analysis
W1 .
W2 .
W3 .
etc.
. =
T1
T2
T3
recip
Empty*
. (Ti)=TU .
WR
Not empty#
*No force or moment can be

resisted by all these features
#One or more forces or
moments can be resisted by
all these features
9/21/2004
Daniel E Whitney
74
General Problem of Mechanism Freedom
Our solution is based on Screw Theory and a

method due to Konkar for intersecting twists
It is based on tracing paths through the mechanism
from a link of interest to a fixed link
Separate motion and constraint analyses are done
Our method works for a class of mechanisms

It can be applied to a succession of subassemblies
in an assembly sequence
A complete solution method exists - see the CD
9/21/2004
Daniel E Whitney
75
Path Method for Motion and Constraint
Analysis - 1: Series and Parallel Reduction
PART OR LINK
JOINT
PART OR LINK
TW1
TW1 * TW2
TW1 + TW2
TW1
TW2
TW2
Joints in series are

reduced using union
9/21/2004
Joints in parallel are

reduced using intersection
Daniel E Whitney
76
Analysis - 2: Paths and Branches

START
NODE
START
NODE
START
NODE
SUBBRANCH
BRANCH
END
NODE
9/21/2004
END
NODE
Daniel E Whitney
END
NODE
77
Analysis - 3: The Process

Identify the link to be analyzed and call it start
Call the fixed link end

Identify all paths and branches from start to end
Form union of twists along each path and branch
Intersect twist unions from branches and paths

using the reduction rules until the mechanism
consists of start and end joined by one
equivalent joint
9/21/2004
Daniel E Whitney
78
Analysis - 4: An Example
L1
1
R2
R3
R2
L1
L3
L2
L3
R2
L4
R3
L2
2
L4
R1
L5
R4
R1
R4
R4
L5
(R2 U R1) U (R2 U R4) U (R3 U R4)
The required unions and intersections can be written as

a Boolean expression and fed right into Matlab
9/21/2004
Daniel E Whitney
79
Conditions for Success of This Method
Paths must be independent at link of interest
All the dof of joints connected to the link of interest

must be independent of each other
Same as saying there cannot be any cross-links

between paths near link of interest
L1
This mechanism is both overand under-constrained. Our method

does not work on it.
The Kutzbach criterion also gives the
wrong answer.
Fixed
9/21/2004
Daniel E Whitney
80
Constraint Analysis with Multiple Features

X
Y
F1 F2 = under-constrained
TR = (T1 ,T2 ) = recip[. (W1 ,W2 )] is not empty
X motion is allowed
WR = recip[. (T1 ,T2 )] is empty
[F1, F2, F3] = properly constrained

WR = recip[. (T1 ,T2 ,T3 )] is empty
TR = recip[. (W1 ,W2 ,W3 )] is empty

4
[F1, F2, F3, F4] = not over-constrained,
WR = (W1 LW4 ) = recip[. (T1 ,T2 ,T3 ,T4 )] is empty

Even tho we know it IS over-constrained
9/21/2004
Daniel E Whitney
81
Constraint Analysis with Multiple Features - 2

X
Y
F1 F2 = under-constrained
1
3
1 2
4
(1 2) 3
TR12 = (T1 ,T2 ) = recip(. (W1 ,W2 ))
X motion is allowed
WR = recip[. (T1 ,T2 )] is empty
F3 ( F1 F2) = not over-constrained
TR123 = (T1 ,T2 ,T3 ) = recip[. (W1 ,W2 ,W3 )] is empty

WR = recip{. [T3 ,TR12 ]} is empty
{[(F1 F2) F3] F4} = over-constrained

TR123 = (T1 ,T2 ,T3 ) = recip[. (W1 ,W2 ,W3 )]
WR = recip{. [T4 ,TR123 ]} is not empty
9/21/2004
Daniel E Whitney
82
What Direction is Over-constrained?

(1 3) 4
Y
2
X
9/21/2004
Daniel E Whitney
83
References for Kinematic Design
Slocum A. H.,Precision Machine Design, New York:

Prentice-Hall, 1991.
Smith, S. T., and D. G. Chetwynd, Foundations of
Ultraprecision Mechanism Design, Philadelphia: Gordon
and Breach, 1992
Whitehead, T. N., The Design and Use of Instruments and
Accurate Mechanism, New York: Dover Press, 1954.
9/21/2004
Daniel E Whitney
84
Publications
Whitney, D. E., Gilbert, O., and Jastrzebski, M., "Representation of Geometric

Variations Using Matrix Transforms for Statistical Tolerance Analysis in
Assemblies, Research in Engineering Design, (1994) 6: pp 191-210
Mantripragada,R., Cunningham, T. W., and Whitney, D E, Assemblyoriented Design: A New Approach to Designing Assemblies, Proceeding, IFIP
WG5.2 Workshop on Geometric Modeling in CAD, May 19 - 23, 1996.
Mantripragada, R. and Whitney, D. E., The Datum Flow Chain, Research in
Engineering Design, v 10, 1998, pp 150-165.
Mantripragada, R. and Whitney, D. E., Modeling and Controlling Variation
Propagation in Mechanical Assemblies using State Transition Models, IEEE
Trans on Robotics and Automation, 1, no 1, Feb, 99, pp 124-140.
Adams, J D and Whitney D E, Application of Screw Theory to Constraint
Analysis of Assemblies of Rigid Parts, 1999 IEEE ISATP
Whitney, D E, Mantripragada, R., Adams J D, and Cunningham, T W, Use of
Screw Theory to Detect Multiple Conflicting Key Characteristics in Complex
Mechanical Products, 1999 ASME DFM Conf
9/21/2004
Daniel E Whitney
85
Variation Buildup in Single Parts
Learn about the history of interchangeable parts
See how parts are given tolerances

Learn what geometric dimensioning and
tolerancing is all about
Look ahead to variation buildup in assemblies
Part variations
9/30/2004
Daniel E Whitney
The Business Case for Tolerances

Performance
Better
marketing
More robust design
Revenue
Cost
Tolerances
And Clearances
Better mfr
Baseline
More robust design
Better mfr
Manufacturing Cost
Part variations
9/30/2004
Daniel E Whitney
Definitions
Tolerance
What is allowed or acceptable, defined by specification limits
Specification limits are set by engineers, designers, and/or manufacturing
people
Variation
What actually happens with real parts and assemblies
Variation can be measured
Clearance
Empty space between surfaces on different parts
Often confused with tolerance
Clearances can have tolerances and can vary
These are typical definitions in the academic and professional literature
Part variations
9/30/2004
Daniel E Whitney
History of Interchangeable Parts
Quest for interchangeability

Begins in 1760s as a customer requirement for muskets
Evolved as a means to systematize manufacturing (1830s)
Culminates in Fords moving assembly line

Permits rapid assembly and mass production

Enables supply chains
Avoids coordination
The zeroth interchange occurs at first assembly
Enabled supply chains via standards for gaging and

tolerancing (1915 to today)
Part variations
9/30/2004
Daniel E Whitney
History - 2
Geometric Dimensioning and Tolerancing
(GD&T) replaced dimensions (1940s +)
Replicated gaging procedures on paper
Solid modeling CAD forced reconsideration of

GD&T on a more mathematical basis
Parts tolerancing seems OK but assemblies are
still something of a mystery
Coordination makes a comeback as demand for
quality exceeds capability
Part variations
9/30/2004
Daniel E Whitney
Taniguchis Diagram
Source:
Part variations
9/30/2004
Daniel E Whitney
Geometric Dimensioning and Tolerancing
Seeks to deal with solid objects rather than lines

on paper
The result is definition of zones where surfaces
should lie
This is good from the point of view of being

realistic about solid objects
It does not shed light on what the tolerances
should be in order to achieve any particular
function
Part variations
9/30/2004
Daniel E Whitney
How Big is a Cube?
Conventional drawing
showing one face
1.000
0.003
This arrow really

sets the distance
between two lines.
This means nothing.
Part variations
9/30/2004
Daniel E Whitney
A more realistic view
of the cube
How many arrows

are needed to say
how far apart the
two surfaces are???
8
All Acceptable Cubes Lie Between Two
Perfect Nested Cubes
Part variations
9/30/2004
Daniel E Whitney
Part variations
9/30/2004
Daniel E Whitney
10
Goals of Geometric Dimensioning and
Tolerancing
Account for 3D Geometry of Parts

Define Datum Hierarchy

A, B, C
Corresponds to 3,2,1 of Constraint
Standardizes machining, fixturing, and gaging
Make the A surface wide, stable, 3 points separated
Guarantee that any randomly selected pair of parts will

assemble (i.e., worst case tolerancing)
Has become an international standard
Does not apply to assemblies
Part variations
9/30/2004
Daniel E Whitney
11
GD&T Control Frame for Locating and
Sizing a Feature
2.000
.010
Diameter can vary
in this range
2.000
C
.470 - .500
.010 A B C
Geometric
characteristic
(position)
A
Axis location references
Axis orientation reference
Zone descriptor
(cylinder) Location of center
Zone
.00707
X = .00356
Y = .00356
can vary in a cylinder with

this diameter
Part variations
9/30/2004
Daniel E Whitney
12
Virtual Condition Guarantees Assembly
.240 - .280
.020 M A
.001
A
.280
.280
.060
.020
.240
.020
.240
A
.300
.300
MMC
.02
The virtual condition is a perfect round perpendicular pin .300 diam
Part variations
9/30/2004
Daniel E Whitney
13
Virtual Condition Guarantees Assembly - 2
.365
.325
.020 M B
.365
.325
.305
.305
.02 MMC
The virtual condition is a perfect round perpendicular hole .305 diam

.300
At worst, the hole

occupies a region
no narrower than .305
A
B
Part variations
9/30/2004
At worst, the peg

occupies a region
no wider than .300
.305
Daniel E Whitney
14
GD&T is ~Equivalent to Chain of
Frames Inside a Part
.500 - .520
.010 M
4x .750 .010
.030 M A B M
Center Hole (D)
A C M
TCH
4 Holes
1.875
TCD
C Bolt Circle
TBC
B Outer Diameter
TAB
3.260
3.240
.005 M
B
A Base Surface
.001
A
Part variations
9/30/2004
Daniel E Whitney
15
Summary
The goal of interchangeable parts is over 250 years old
Parts can be toleranced by international standard methods

in ways that
Respect our notions of constraint
Locate features with respect to datum surfaces
Can ~be represented by chains of frames similar to the way
assemblies can
Impose worst-case tolerances
No standard exists for tolerancing assemblies
No clear path exists in standard methods for linking

assembly goals to part tolerancing
Part variations
9/30/2004
Daniel E Whitney
16
Variation in Assemblies
Goals of this class
Understand how to represent variation at the assembly

level
Link models to models of variation at the part level
Link models to nominal chain of frames

Get an idea of statistical process control and statistical
tolerancing vs worst case tolerancing
var_assemblies
10/4/2004
Daniel E Whitney
History of Tolerances in Assemblies
Statistical process control and statistical tolerancing

permit a bet on interchangeability
Coordination makes a comeback as functional build
Quality today is more than interchangeability

The product almost certainly will work
Real quality means
Durability
Reliability
Low noise, vibration, etc
All are associated with low running clearances, fine

balance, etc., all requiring closer tolerances, robust
design, etc.
var_assemblies
10/4/2004
Daniel E Whitney
Recent Gains in Productivity
Wall Street Journal, Sept 28, 00, p 2:
Many analysts may have overlooked one of the main

drivers of the productivity gains of the last five years:
machine tools more than half the increase has occurred
outside the computer, software, and telecommunications
sectors...
Greater precision in machine tools has helped cut the energy
use of air conditioners by 10% between 1990 and 1997
using a new kind of compressor whose manufacture required
precision down to 10 microns
One of the principal reasons for more reliable automobile
transmissions ...
Open architecture, digital programming, ceramic tool bits
are part of the picture
var_assemblies
3
10/4/2004
Daniel E Whitney
KCs and Variation
A KC is delivered when it achieves its desired

value within some specified tolerance
This puts the emphasis on behavior at the
assembly level, not the part level
To deliver the KC we must
Design the chain and the location strategy at each joint

in the chain
Ensure that variation at the part level does not
propagate to the assembly level and upset delivery of
the KC
This means understanding variation in each of the joints
in the chain
var_assemblies
10/4/2004
Daniel E Whitney
Logic Tree of Control of Variation
Main Decision
We cant meet assembly
KCs by controlling part
variations
We can meet assembly
variations
Functional Build
Build to Print
Selective
Assembly
var_assemblies
10/4/2004
Belief in
Coordination
Belief
in systems
Adjustment
Statistical
Process
Control
Daniel E Whitney
Two Kinds of Errors
Change in process average (mean shift)

Process variation around the mean (variance)
These are different kinds of errors that require

different approaches
The screwdriver workstation-RCC story
Whiteheads reason for using kinematic design: to

eliminate variation (erratic errors)
Taguchis systematic approach
Correct mean shift first, then reduce variation
The other sequence makes sense, too
var_assemblies
10/4/2004
Daniel E Whitney
Basic Definitions
Mean
Standard
Deviation
-0.04
-0.03
-0.02
-0.01
0.01
0.02
0.03
0.04
3 =99.73% of all
events if Gaussian
var_assemblies
10/4/2004
Daniel E Whitney
Precision vs Accuracy
Source:
Consistent but
consistently wrong
var_assemblies
10/4/2004
Right on the average
Daniel E Whitney
Best Situation
Source:
var_assemblies
10/4/2004
Daniel E Whitney
Models of How Errors Accumulate in
Assemblies
Worst case tolerancing essentially tracks the max error, treats

it as mean shift, and ignores the variance
Assumes all errors are at their extremes at the same time
This is deterministic, not statistical

Errors accumulate linearly with the number of parts
Statistical tolerancing (AKA root sum square) tracks the
variance and assumes zero mean shift
Assumes errors are distributed randomly between limits
Errors accumulate with SQRT of number of parts but only

if the average part error = the desired nominal dimension!
(i.e., average part (& assy) error = 0.)
var_assemblies
10/4/2004
Daniel E Whitney
10
Stapler Variations
TOP VIEW
Images removed for

copyright reasons.
Source:
Figure 5-14
in [Whitney 2004] Whitney, D. E.
Mechanical Assemblies:
Their Design, Manufacture, and
Role in Product Development.
New York, NY: Oxford University
Press, 2004. ISBN: 0195157826.
FRONT VIEW
var_assemblies
Source:

Worst Case
10/4/2004
Daniel E Whitney
Statistical
11
Logic Tree of Statistical and Worst-Case
Tolerancing
Tolerance method
Worst case
Statistical
Errors grow with
sqrt(N) IF
mean = nominal
Errors grow
with N
Throws away parts
that stat tol keeps
var_assemblies
10/4/2004
Process capability
measures adherence
of process mean
to nominal
Daniel E Whitney
Statistical Process
control keeps
process on its mean
12
How Variations Accumulate
Assumptions
All variations are described by the same distribution
All variations are independent of each other
The SSN jackpot in Milwaukee
Then:
The mean or average of a sum of such variations is the
sum of their individual means
xi =
E xi = NE xi
N
i =1
i= 1
if all E[xi] are the same
Actually this is true regardless of the above assumptions
The variance of the sum of such variations is the sum of

their individual variances
var_assemblies
xi =
N
i= 1
10/4/2004
i =1
xi = N 2 xi
Daniel E Whitney
13
Total Error
Total Accumulated Error =
Accumulated Mean Errors
+ Accumulated Standard Deviation Errors

N
2
(3
i=1
i=1
=
meanshifti +
= N * meanshift + N * 3 * stddev
if all meanshifts and standard deviations
are the same
var_assemblies
10/4/2004
Daniel E Whitney
14
What It Means
Sums of zero-mean errors accumulate at the rate N

because the + and - errors tend to cancel
Sums of non-zero-mean errors accumulate at the
rate N because there are no cancellations
Therefore, non-zero-mean errors accumulate much
more rapidly than zero-mean errors
var_assemblies
10/4/2004
Daniel E Whitney
15
Root Sum Square (RSS) Method
Assume tolerance band = T

Assume T = 3 of error distribution
Then accumulated error from N of Ti =
N
Accumulated error =
i =1
But this is really the entire accumulated error only

if there is no meanshift. If there is meanshift then
the RSS method will badly under-estimate the error.
var_assemblies
10/4/2004
Daniel E Whitney
16
Why the Mean Should Seek the Nominal
Source:
How to survive Las Vegas
var_assemblies
10/4/2004
Daniel E Whitney
17
Statistical and Worst Case Compared
Source:
var_assemblies
10/4/2004
Daniel E Whitney
18
Statistical Process Control
SPC aims to get a process in statistical control
The mean varies in a random way and stays within the mean
control limits
The range varies in a random way and stays within the range
control limits
X-bar and R charts plot the mean and range of samples

taken from production parts
Deviations from the mean, consistent errors, and excess
variation can be found quickly and eliminated before bad
parts are made
If a process is in control, then
there is a chance that it can
deliver parts without mean shift
var_assemblies
10/4/2004
Daniel E Whitney
19
Process Capability
Every process has a mean and a natural tolerance range

which should be tighter than the product needs
Process capability index Cpk tracks whether the process
range is within the 3 tolerance band and the process mean
is on the target nominal dimension
USL X LSL X
C pk = min
;
3
USL = upper spec (tolerance) limit
LSL = lower spec (tolerance) limit
X = mean of the process
= variance of the process
var_assemblies
10/4/2004
Daniel E Whitney
20
Process Capability Cpk
Source:
# of
var_assemblies
1
2
3
4
10/4/2004
% within # of
68.269%
95.450%
99.730%
99.994%
Daniel E Whitney
% outside
31.731%
4.550%
0.270%
0.006%
21
Process Control & Capability
Source:
var_assemblies
10/4/2004
Daniel E Whitney
22
Effect of Mean Shift
0.00167 shift in each part

3=.01732
-0.04
Range
1 sigma
2 sigma
2.5 sigma
3 sigma
-0.03
-0.02
-0.01
0.01
% Outside
31.74
4.54%
1.24%
0.27%
-0.04
0.02
0.03
.005 total shift in 3 parts=0.5

3=.03
2.5
1.24% will exceed 0.03
-0.03
-0.02
-0.01
0.01
0.02
0.03
Cpk=1.00
var_assemblies
10/4/2004
0.04
Daniel E Whitney
0.04
Cpk=.833
23
How Could Non-zero Mean Variations
Occur?
For material removal processes, operators stop as

soon as the part enters the tolerance zone from the
high side
For material-adding processes, operators stop as
soon as the part enters the zone from the low side
Source:
Removal Process
var_assemblies
10/4/2004
Additive Process
Daniel E Whitney
24
Statistical vs Worst Case - Summary
Statistical is cheaper because looser tolerances are

allowed at each part or feature (cost rises as
tolerances are tightened)
Worst case guarantees interchangeable parts, even
at batch size of one
Statistical achieves interchangeability with some
probability < 1, enhanced if there is a big bin of
parts to choose from. The likelihood of the worst
combination actually being picked is pretty low
and is usually easy to recover from.
var_assemblies
10/4/2004
Daniel E Whitney
25
var_assemblies
10/4/2004
LOGIC OF SPC ETC
Daniel E Whitney
Enabling Tree for SPC
Source:
26
Math Models of Variation
There are tolerance standards for designing single

parts (ANSI Y 14.5 xx)
There is no standard for tolerancing assemblies
People just extend the part standard or finesse
The following slides show how to use the 4x4

matrix method for both nominal part-part location
modeling and variation modeling
var_assemblies
10/4/2004
Daniel E Whitney
27
Nominal and Varied KC Relationships
Nominal
Varied
A
KC
C
KC
CC
Each relationship, internal or external,
nominal or varied, is described by a 4x4 transform
Internal relationship
External relationship
Key Characteristic
var_assemblies
10/4/2004
Daniel E Whitney
28
Inside a Car Engine
Source:
var_assemblies
10/4/2004
Daniel E Whitney
29
Chain of Frames for Engine
Source:
var_assemblies
10/4/2004
Daniel E Whitney
30
Valve and Lifter Use Selective Assembly
Statistical Tolerancing
cannot deliver this KC
Source:

var_assemblies
10/4/2004
Daniel E Whitney
Source:

ISBN: 0195157826.
31
How Selective Assembly is Done
Normal Distribution
0.5
0.4
0.3
Shaft
0.2
0.1
0
-4
-3
-2
-1
Bearing
Clearance is
important
Pick and measure a shaft.

If it is a bit big, pick a big bearing to get clearance right.
If it is a bit small, pick a small bearing.
For this to work over a long stretch, there
must be about the same number of big
shafts as big bearings, and the same for small ones.
var_assemblies
10/4/2004
Daniel E Whitney
32
Assembly Process Tolerances
NOMINAL
SHOWING TOLERANCES
TC
TB
TC
TB
TD
TD
T4
TA
T3
T4
TA
T3 + DT3 (PARALLEL)
T2
WORLD
ORIGIN
T1
T0
T2 + DT2 (SIZE)
WORLD
ORIGIN
T1
T0
TC
TB
TD
T4
TA
T3
Not OK!
OK!
T2
T1
T0
var_assemblies
10/4/2004
Daniel E Whitney
33
Nominal Mating of Parts
Source:
var_assemblies
10/4/2004
Daniel E Whitney
34
Varied Part Location Due to Tolerances
AF
TFB'
T AB'
B'
TBB'
The varied location of Part B can be calculated

from the nominal location of Part A. This process
can be chained to Part C, etc., including errors on
Part B. It uses the same math as the nominal model.
var_assemblies
10/4/2004
Daniel E Whitney
35
Equations for Connective Models

Nominal
Source:
Varied
Source:
var_assemblies
10/4/2004
Daniel E Whitney
36
Error Transform (Linearized)

1 . z . y
. z 1 . x
DT =
. y . x 1
0
0
0
T = T * DT
var_assemblies
10/4/2004
Daniel E Whitney
dx
dy
dz
1
dp
d.
T
T
DT
37
Angle Errors Are Important

Please see:
Kedrosky, Paul S. "In the Fray: What the Duffers Can Teach Tiger Woods."
The Wall Street Journal (April 14, 2004): D.16.
var_assemblies
10/4/2004
Daniel E Whitney
38
Transform Order is Important
This is a frame 1 event

2
2
T
D
T=DT11T12
T12
DT11
This is a frame 2 event
1
Tells us what
everything looks
like from frame
1s pov
var_assemblies
10/4/2004
T=T12DT22
Moves the focus to frame 2 and
converts frame 2 events
to frame 1 coords.
39
Daniel E Whitney
Peg Position Error
>> TAD = trans(3,2,4) * roty(dtr (90))

% dtr converts degrees to radians
Y'
Z'
D'
Y
Z
Part A
TAD'
Z'
Y'
D' Y errt
D
X'
TAD'
TAD
errt
D
X'
X
TAD
3
0
0
=
1
0 1
1 0
0 0
0 0
3

2
4
% First method
>> DZ = errt
>> DTAD1 = trans(0,0,DZ)
T AD'1 = DTAD1TAD
% Second method
>> DX = errt
>> DTAD2 = trans(DX,0,0)
>> T AD' 2 = TAD DTAD2

>> T
var_assemblies
10/4/2004
Daniel E Whitney
AD' 2 =
TADTDD
40
Peg Length Error
>> TAD = trans(3,2, 4) * roty(dtr(90))

% dtr converts degrees to radians
errt
Y
D
Y
Z
Part A
TAD
TAD
3
D Y errt
D
TAD X
TAD
0
0
=
1
0
1
0
0
1
0
0
0
3
2
DTpp=trans(0,0,randn*.003)
Z
Y
2
Z
var_assemblies
10/4/2004
Daniel E Whitney
41
Hole Angle Error
F
F
Part B
TEF
X
TEF
X
X'
EE'
1
Y'
Y
Z
Z'
>> TEF = trans (6,0,1)

1 0 0 6
0 1 0 0
=
0 0 1 1
0
0
0
1
>> DTE' E =roty (dtr (erra))

>> TE 'F = DTE 'ETEF
Y
2
1
var_assemblies
10/4/2004
Daniel E Whitney
42
Final Variation at Assembly Level
Where does this point go?
>> TDE = rotz (dtr (180))

1 0 0 0
0 1 0 0
=
0 0 1 0
0 0 0 1
>> TAF = TAD TDE TEF
Part B'
TEF
X
X
E'
E
D' Z'
Part A'
T AD'
X'
0 0 1 4
0 1 0 2
=
1 0 0 10
0 0 0 1
>>T A' F = TAD 'TD ' E 'TE 'F

%TD' E ' = TDE
Y
var_assemblies
10/4/2004
X
Daniel E Whitney
43
MATLAB for Example
TAD=trans(3,2,4)*roty(dtr(90))
TAF=TAD*TDE*TEF
TAD =
TAF =
0
0
-1
0
0
1
0
0
1
0
0
0
3
2
4
1
TEF=trans(6,0,1)
TDE=rotz(dtr(180))
TDE =
-1 0
0 -1
0 0
0 0
TEF =
1
0
0
0
0
1
0
0
var_assemblies
0
0
1
0
0
0
1
0
6
0
1
1
10/4/2004
0
0
1
0
0
0
0
1
0
-1
0
0
1 4
0 2
0 -2
0 1
TAF(1,4)
for i=1:10000
DTad=trans(randn*.003,0,0);
DTdpp=trans(0,0,randn*.003);
DTee=roty(dtr((randn*.003)));
TAF=TAD*DTad*DTpp*TDE*DTee*TEF;
m(i)=TAF(1,4);
end
hist(m,200)
Daniel E Whitney
44
Result for 10000 Samples
200
180
160
140
120
100
80
60
40
20
0
3.985
var_assemblies
3.99
10/4/2004
3.995
4.005
4.01
Nominal Value
Daniel E Whitney
4.015
45
Histogram of 100 Samples
4.5
3.5
2.5
1.5
0.5
0
3.992
var_assemblies
3.994
3.996
10/4/2004
3.998
4.002
4.004
4.006
Nominal Value
Daniel E Whitney
4.008
4.01
46
Constructing Matrix Representations of
GD&T
var_assemblies
10/4/2004
Daniel E Whitney
47
Position and Angle are Coupled by Rule #1
Source:
fopt is derived by simulation to give the best Gaussian

fit to the actual diamond
var_assemblies
10/4/2004
Daniel E Whitney
48
Tolerance Zones as 4x4 Transforms

Two dimensional example:
. 1 = 2 T /L
S
F2
2TS
T 0,1 =
D
DT1 , 2 :
F
0
1
0
0
0
0
1
0
0
0
0
1
0
0
0
D
1
1
0
0
0
T1,2 =
0
1
0
0
0
0
1
0
0
0
0
1
dX = 0
= fo p t x 2 Ts / L y
dY = 0
= fo p t x 2 Ts / L x
d Z = fo p t x T s
= 0
opt
= 0.95
T0,2 = T0,1 * T1,2 * DT1,2

var_assemblies
10/4/2004
Daniel E Whitney
49
Table removed for copyright reasons.

Source:
Table 6-1 in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
var_assemblies
10/4/2004
Daniel E Whitney
50
Two Ways to Use This Model

Direct Monte Carlo simulation (can use Matlab)
randomly select dX, dY, . x, etc (eliminate parts that violate Rule
#1) (use fopt = 1)Assume that Ts = 3
dX = randn * Ts / 3; .
= randn *2 *Ts / (3* LY)
multiply out all the Transforms

repeat many times, collect histograms
Plug into a closed form solution (use fopt to approximate

Rule #1)
requires assuming a Gaussian distribution for each of the errors
delivers ellipsoids of location and angle error at the end of the
transform chain
var_assemblies
10/4/2004
Daniel E Whitney
51
Logic Tree of Control of Variation
Main Decision
We cant meet assembly
variations
We can meet assembly
variations
Functional Build
Build to Print
Selective
Assembly
var_assemblies
10/4/2004
Belief in
Coordination
Belief
in systems
Adjustment
Statistical
Process
Control
Daniel E Whitney
52
Functional Build
Accept the fact that the mean cannot be brought to the

desired value:
Too much trouble, time, or cost
Too much variation in the mean
Some other mean may be shifted the other way
Coordinate the parts involved

Keep an eye on things and make adjustments
This is often done on tools and dies if there is one source

(one die) for each part involved
Then the focus is to control variation around the mean
The SPC metric is then Cp

USL LSL
Cp =
var_assemblies
10/4/2004
Daniel E Whitney
53
Cp Ignores Mean Shift
Source:
var_assemblies
10/4/2004
Daniel E Whitney
54
Logic Tree for Tolerancing
Source:
var_assemblies
10/4/2004
Daniel E Whitney
55
Assembly Sequence Analysis
Goals of this class
Understand one algorithmic approach to finding all

feasible assembly sequences
Make connection between algorithm and assembly
feature models
See how assembly sequences can be designed
Look at some examples

See a video of computer-aided assembly analysis
assy seq anal
10/6/2004
Daniel E Whitney
History
Assembly sequence analysis applied to line
balancing (Prenting and Battaglin, 1964)
Heuristics such as the fastener method (1978)
Bourjault method (1984)

De Fazio/Whitney method (1987)
Gustavson exploded view method SPM (1989)
Baldwin onion skin method (1989)

Sukhan Lee method (force paths, subassemblies,
1989 +)
Wilson method (free directions, 1992+)
assy seq anal
10/6/2004
Daniel E Whitney
Role of Sequence Analysis in Concurrent
Engineering
Line balancing applies sequence analysis after the

product is designed
Our goal is to push assembly sequence analysis to the
beginning of the development process
It can be an important lever in concept design
It interacts with architecture and affects supply chain,

build to order processes, JIT, etc.
To keep up with designers during fluid concept
design, the assembly engineers need a tool that gives
fast turnaround
assy seq anal
10/6/2004
Daniel E Whitney
Analysis Alternatives
Find all feasible sequences

Find all linear feasible sequences
add one part at a time
Find one feasible sequence

Find one linear feasible sequence
The first one is of the most interest to assembly line

designers
assy seq anal
10/6/2004
Daniel E Whitney
Process Phases
Eliminate all truly impossible sequences
parts physically block other parts

sequences dead end before completion
What remain are called feasible = not

impossible
the good, the bad, and the ugly
By various criteria, throw out bad and ugly
Criteria include technical and business issues
This is traditional design:

generate requirements
generate alternatives
use requirements to narrow the alternatives
assy seq anal
10/6/2004
Daniel E Whitney
Classes of Approaches
Most methods assume one hand

Forbids joining 3 things at one step
Graph theory analysis of liaison diagram
Systematic textual analysis of lists of liaisons that

contain blockers
Cut-set methods applied to the liaison diagram
Onion-skin methods that peel off outside parts
Most of these methods utilize disassembly as the

paradigm but it is not necessary
can you remove part X from parts Y,Z,...? is the same
as can you put part X onto parts Y, Z,...? under most
circumstances
6
assy seq anal
10/6/2004
Daniel E Whitney
Non-Assembly Steps Can be Included
Reorientation
Tests, lubrication
Temporary disassembly
These can all be handled one way or another if

you are creative
assy seq anal
10/6/2004
Daniel E Whitney
The Liaison Diagram

A simple graph that denotes parts as nodes and
connections as arcs
Can be augmented with information about the
connection
threaded hole
a
face/face
& peg/hole
assy seq anal
10/6/2004
face/face
b
Daniel E Whitney
a
b
c
8
Rules of Liaison Diagrams

Each part is a node, each arc is a liaison
Each part has no more than one liaison with any
other part
In a loop of n liaisons, if n-2 arcs are closed, then
attempting to close 1 of the 2 remaining will
automatically close the whole loop:
a
threaded hole
a
face/face
& peg/hole
face/face
b
b
assy seq anal
10/6/2004
Daniel E Whitney
Generating Sequences
Source:
assy seq anal
10/6/2004
Daniel E Whitney
10
Selecting Sequences
Source:
Source:
assy seq anal
10/6/2004
Daniel E Whitney
11
assy seq anal
10/6/2004
Daniel E Whitney
Two Alternator Sequences
Source:
12
assy seq anal
10/6/2004
Daniel E Whitney
13
Two More Alternator Sequences
Source:
Rules of Sequence Analysis
Parts are rigid
Liaisons (connections between parts) are also

rigid
Once a liaison is made, it stays made
assy seq anal
10/6/2004
Daniel E Whitney
14
Ask and Address Precedence Questions
Goal of questions is to find out what moves are

forbidden
This is done various ways by different methods:
Computer searches for free paths, using local escape

directions and checking for interference
Person detects these
Typical questions:
can this part be added to those parts
can this set of parts be added to that set of parts
must these parts be present/absent in order to add that/those
parts
assy seq anal
10/6/2004
Daniel E Whitney
15
Subset and Superset Rules Cut the
Number of Required Questions
These are true only if parts and liaisons are rigid
Subset rule:
if you can add part X to parts {Y} then you can add part
X to any subset of {Y}
fewer parts cant contain blockers that arent in the
original set
Superset rule:
if you cant add part X to parts {Y} then you cant add
part X to a superset of {Y}
adding parts cant remove blockers that are in {Y}
counter-example in Sony tape deck with motor
assy seq anal
10/6/2004
Daniel E Whitney
16
Local and Global Freedom
Local freedom means that the combined escape

directions of all liaisons in the query have a common
direction (dot product of escape vectors = 1)
Global freedom means that there is a long range
escape path that completely separates the parts in the
query
Local freedom can be detected by the computer just
by inspecting the escape directions - easy
Global freedom requires solving the piano movers
problem - difficult or impossible
assy seq anal
10/6/2004
Daniel E Whitney
17
Finding Local Freedoms
Use escape direction vectors:

Look at escape direction vectors for each feature
Look for common vector for them all
Conventional screw theory will not work

Its too hard to distinguish one-sided motion limits
Dr. Whitney, what do you do about the facets?
assy seq anal
10/6/2004
Daniel E Whitney
18
Generate Precedence Relations
Example of cookie jar and cookies

Can you put the cookies in the jar if the lid is on?
No.
Therefore: cookies to jar > lid to jar
assy seq anal
10/6/2004
Daniel E Whitney
19
Diagram Feasible Sequences
Network of sequences represents
States of assembly = feasible subassemblies showing

which liaisons have been completed
Transitions between states
A path through the network is a feasible sequence
Cookies to jar, then lid to jar

Cookies to lid upside down, then jar to lid, then flip
We decide later which is better

assy seq anal
10/6/2004
Daniel E Whitney
20
Simple Assembly Sequence Example

Assembly
Liaison Diagram
B
C
1
B
4
C
A
3
2
D
D
Local escape directions

shown by arrows
assy seq anal
10/6/2004
Daniel E Whitney
21
Bourjaults Process as Textual Analysis
Analysis question:
R(a; b, c, d)=Can you make liaison a when

b, c and d are already made?
Ask this question for every liaison a
combined with every other liaison b, c, d
Bourjaults original process uses graphical
analysis based on circuit theory
assy seq anal
10/6/2004
Daniel E Whitney
22
Assembly
Liaison Diagram
B
C
Bourjault Method
A
3
D
R(1;2,3,4) Cant answer because 2,3,4 forces 1
Eliminate 2, 3, or 4
Eliminate 2: R(1;3,4) = No (need to know why)
Eliminate 3: R(1;4) = Yes (so 4 is not why)
Eliminate 4: R(1;3) = Yes (so 3 is not why)
So 1>= 3,4 (i.e., 3,4 together is why)
Eliminate 3: R(1;2,4) = No
Eliminate 2: R(1;4) already answered Y
Eliminate 4: R(1;2) = Y
So 1 >= 2,4
So 1>= 2,3
assy seq anal
10/6/2004
Daniel E Whitney
D
R(2;1,3,4) Cant answer
Eliminate 3: R(2;4) = Yes
So 2>=3,4
Eliminate 3: R(2;1,4) = Yes
Eliminate 1: R(2;3) = aaY
So 2>=1,3
Done by symmetry:
4>=1,2
3>=1,2
4>=1,3
3>=2,4
4>=2,3
23
Results of Bourjault Method
these two together cant be first
2,4
1
3,4
2
2,3
1,4 do not appear on the RHS,

so they are unprecedented,
so they can be first, in either order
3
1,3
4
Then 2 and/or 3 can be next
1,2
3,4>=1,2
assy seq anal
10/6/2004
Daniel E Whitney
1,2>=3,4
1,3>=2,4
1,4>=2,3
2,4>=1,3
24
Simple Assembly Sequence Results

Assembly
Liaison Diagram
B
C
4
C
A
3
2
D
D
PRECEDENCE RULE:
1&4>2&3
DIAGRAM OF
FEASIBLE
SEQUENCES
assy seq anal
10/6/2004
Daniel E Whitney
25
Portion of Cutset Method
B
C
1
4
C
A
3
2
D
D
All questions except the last are answered by inspecting local freedom
R(1,2;3,4)? No: 1,2 >=3,4

R(1,4;2,3)? No: 1,4 >=2,3
R(3,4;1,2)? No: 3,4 >=1,2
R(2,3;1,4)? Yes: 1,4 unprecedented
so they can be first.
assy seq anal
10/6/2004
Daniel E Whitney
26
Other Methods
Randall Wilson checks global freedom using the

weighted blocking graph
This is essentially a search for unidirectional
escape paths along any of the local escape
directions
The escape directions are generated by inspecting
individual surfaces on adjacent parts that touch
each other, essentially rediscovering the mating
features
assy seq anal
10/6/2004
Daniel E Whitney
27
Other Methods, contd
Gustavson and Wolter each generate exploded

views by different methods and then generate
precedence relations from the order along major
explosion directions
A reasonable assumption is that unless there is
some blockage, all moves along one such direction
will be done before starting on another
Gustavson finds a heuristic sequence along major
explosion directions using part c.g.s and asks the
user to fix any errors
assy seq anal
10/6/2004
Daniel E Whitney
29
SPAS and Onion Skin Methods
All local freedoms are checked by the computer
All global freedoms are queried to the user

This has two benefits
the computer does the easy part without attempting the
impossible part or pretending to do it
the user must confront the design and become very
familiar with it
Ref: Daniel Baldwin SM Thesis, MIT, Feb, 1990
Ref: Russ Whipple SM Thesis, MIT, June 1990
assy seq anal
10/6/2004
Daniel E Whitney
31
Stability Checking (Simplified)
e
Bas
Esc
Start with the base part or the part in the fixture
By definition, it is stable
Check each of its liaisons
compare local escape direction to gravity
if part cant slide out then mark it stable
Check the liaisons of each of the newly defined

stable parts the same way
If all parts in the liaison diagram can be marked
stable then the assembly is stable
Screw theory can be used to find mobility
assy seq anal
10/6/2004
Daniel E Whitney
32
Real Assembly Sequence Analysis Example
Source:
ASSEMBLY DATA MODEL

PARTS OF REAR AXLE
PARTS
A = CARRIER ASSY
B = BACKING PLATE
C = SHAFT
D = BRAKE DRUM AND T'NUT
E = WITHDRAWN PINION SHAFT & BOLT
F = INSERTED
"
"
"
G = (PUSH IN SHAFT & ) C-WASHER &
PUSH SHAFT OUT
H = OIL
I = COVER
J = BRAKE CABLE, COILED
K = FINAL PRESS TEST
L = AIR TEST PLUG
M = FIRST PRESS TEST
assy seq anal
10/6/2004
LIAISONS
1 = C TO A
2 = B TO A
3 = J TO B
4 = D TO C
5 = G TO C
6 = E TO A
7 = F TO A
8 = L TO A
9 = I TO A
10 = H TO A
11 = K TO A
12 = M TO A
Daniel E Whitney
PRECEDENCE RELATIONS
2>1
5>4
1&2&6>5
5>7
11 > 8
10 > 9
12 > 10
12 > 11
3>1&4&5
7 > 10
9 > 11
33
Rear Axle
Source:
assy seq anal
10/6/2004
Daniel E Whitney
34
Rear Axle Differential Subassembly
After Liaison 1 Shaft to Carrier
Source:
assy seq anal
10/6/2004
Daniel E Whitney
35
assy seq anal
10/6/2004
Daniel E Whitney
Pinion Shaft Out (Liaison E) and
Insert C Washer (Liaison G)
Source:
36
assy seq anal
10/6/2004
Daniel E Whitney
Pinion Shaft In (Liaison F)
Source:
37
Example assembly
sequence graph
Source:
Each path from top

to bottom is a valid
sequence. Each box
is a valid intermedia
assembly state.
assy seq anal
10/6/2004
Daniel E Whitney
38
Six Speed Truck Transmission
Second
First
Drawn by T. L. De Fazio
assy seq anal
10/6/2004
Daniel E Whitney
39
Juicer
Source:
assy seq anal
10/6/2004
Daniel E Whitney
40
Network Complexity Metric
The liaison diagram is a network

How complex is an assembly?
Network complexity metric k: (#arcs) / (#nodes)
Node = part, arc = connection between 2 parts
If n = # parts, then
Min k = (n-1)/n
Max k = n(n-1)/2n = (n-1)/2
Which product will have more assembly

sequences?
Product with big complexity metric
Product with small complexity metric

assy seq anal
10/6/2004
Daniel E Whitney
41
Chinese Puzzle
assy seq anal
10/6/2004
Daniel E Whitney
42
Some Data on Liaisons per Part
Liaisons per
Part
Liaisons
per
Part
Source:
assy seq anal
10/6/2004
Daniel E Whitney
43
More Data: 34 Products
assy seq anal
10/6/2004
Daniel E Whitney
44
V-8 Engine Liaison Diagram
assy seq anal
10/6/2004
Daniel E Whitney
45
Constraint Limits Liaisons/Part
M = 3(n g 1) + joint freedoms fi

where
n = number of parts
fi = degrees of freedom of joint
= liaisons / part
= average dof per joint

g = n
f i = g = n
and
planar
spatial
1 .5
1 .2
1 .5
M = 3(n n 1) + n
If M = 0
3 3n
.
=
n( 3)
assy seq anal
3
as n gets large
10/6/2004
Daniel E Whitney
46
y = 276.31x - 1906.3
R2 = 0.9139
3500
3000
2000
# Sequences
Linear (# Sequences)
1500
1000
500
0
0
10
15
20
Number of Liaisons
# Parts
Throttlebody
Ballpoint Pen
Juicer
Rear Axle
Transaxle
6 Speed Transm
Chinese Puzzle
assy seq anal
# Liaisons
5
6
8
13
9
11
14
10/6/2004
7
5
9
12
15
18
84
Liaisons/Part # Sequences
1.4
10
0.83333333
12
1.125
71
0.92307692
938
1.66666667
2450
1.63636364
3318
6
1
Daniel E Whitney
Number of Sequences vs
Number of Liaisons
2500
47
Video
Made by Randy Wilson and co-workers at Sandia

National Labs in 1996
Obtains one feasible sequence by using feasible
escape cones derived from local escape directions
Permits user to edit this sequence
assy seq anal
10/6/2004
Daniel E Whitney
48
The Datum Flow Chain
Goals of this class
Pull together features, constraint, key characteristics,

and tolerances
Present a diagramming technique for designing
assemblies top-down
Do examples
Datum Flow Chain
10/14/2004
Daniel E Whitney
What Happens During Assembly?
Most people seem to think that assembly is

fastening
Assembly is really chaining together of
dimensional relationships and constraints
The success of these chains determines the success
of the products quality from an assembly point of
view
A goal of assembly design is to permit these
chains to be defined as the basis for designing
assemblies first, then designing parts
Datum Flow Chain
10/14/2004
Daniel E Whitney
Assembly Design Overview
The theory has three elements
constraint defines how parts are located with

respect to each other
assembly features on parts define where parts
are located with respect to each other
tolerances on feature size and location define
how accurately parts should be located with
respect to each other
The DFC creates a top-down model that supports
all three elements
Datum Flow Chain
10/14/2004
Daniel E Whitney
Range of Application of the DFC
Smallest: inside each part (what people do now)
Mid-range: across a set of parts (done occasionally

or in response to problems)
Across parts and assembly machines and fixtures
Across items obtained from or designed by
suppliers (rarely is the need recognized)
The link between top level quality and all

supporting steps and processes
Datum Flow Chain
10/14/2004
Daniel E Whitney
The DFC is Not Tolerance Analysis
The DFC allows us to design the scheme by which

the tolerances will be achieved
It does this in the context of creating the constraint
structure by which the parts will be located
Once we design the constraint structure and the
DFC, we can analyze the tolerance capability of
this structure
Often we will find that assembly sequences or
tooling schemes will have to be modified
Datum Flow Chain
10/14/2004
Daniel E Whitney
Datum Flow Chain
A DFC is a directed acyclic graph that defines the

relationships between assembled parts as well as
assembly fixtures, tooling, and equipment such as
X, z
robots
C
A
Y, Z,
x, y
(6)
A DFC identifies the part mates that convey

dimensional control and identifies the hierarchy that
determines which parts or fixtures define the locations
of which other parts
DFCs also contain information on the type of mating
feature and the amount of motion constraint applied
by that feature
Datum Flow Chain
10/14/2004
Daniel E Whitney
Mates and Contacts
We have separated joints between parts into two

classes:
Mates convey dimensional location and constraint
from one part to another
when all of a parts mates are complete, it should be
constrained in all 6 degrees of freedom unless free motion is
part of its function
Contacts are redundant and provide strength or partial

constraint
Datum Flow Chain
10/14/2004
Daniel E Whitney
Example: car wheel, hub, and studs
Hub face
The KCs are:

wheel perpendicular to axle axis
wheel concentric with axle axis
Part features that deliver the KCs are:
hub face perpendicular to axle axis

hub rim concentric with axle axis
wheel rim hole concentric with wheel body
wheel face parallel to rim hole plane
Rim
The rim-to-hole joint is the mate, the studs and nuts
are contacts (no geometric KC circumferentially)
Alternate design has 5 or 6 studs, no rim

Datum Flow Chain
10/14/2004
Daniel E Whitney
Liaison Diagrams and DFCs
The liaison diagram shows parts and joints
The DFC emphasizes the mates

Liaison Diagram
Datum Flow Chain
10/14/2004
Datum Flow Chain
Daniel E Whitney
Stapler KCs
Datum Flow Chain
10/14/2004
Daniel E Whitney
10
Each KC is Delivered by a Chain
Chains of Delivery
HANDLE
CARRIER
PIN
PUSHER
RIVET
STAPLES
ANVIL
BASE
Datum Flow Chain
10/14/2004
Daniel E Whitney
11
Mates, Contacts, and KC Delivery
HANDLE
CARRIER
PUSHER
Contact
PIN
Mate
RIVET
STAPLES
ANVIL
BASE
Datum Flow Chain
10/14/2004
Daniel E Whitney
Mates give location.

Contacts reinforce location.
Variation travels from part
to part along the chain
of mates.
12
DFCs and KCs
The DFC delivers the KCs by linking the mates
There must be a chain of mates from one side of

the KC to the other
When there is no direct mate or chain of mates, a
fixture or gage is needed
Datum Flow Chain
10/14/2004
Daniel E Whitney
13
Nominal and Variation
The DFC is the constraint plan for getting the

parts to the right places in space so as to deliver
the KC
Each KC should have its own DFC
Often this is impossible, creating KC conflict
If there is variation in the parts, then it flows to the

KC along the DFC
The DFC is the tolerance chain
Variation is passed by the mates, not the contacts
Variation, like constraint, is passed along in the wrench

space of surface contacts inside features
Datum Flow Chain
10/14/2004
Daniel E Whitney
14
Twist Space and Wrench Space Describe
the Behavior of Two Surfaces in Contact
Z
Y
WRENCH SPACE
TWIST SPACE
Datum Flow Chain
10/14/2004
Daniel E Whitney
15
Rigorous Definition of Mates and
Contacts
JOINTS
CONTACTS
MATES
PERFORM EFFECTOR FUNCTION

STABILIZE OR REINFORCE
THE LOCATORS
WRENCH
SPACE
TWIST
SPACE
ADJUSTMENTS CAN BE MADE

USING DOF IN THIS SPACE
PERFORMS LOCATOR
FUNCTION
NO CONSTRAINT EXERTED
NO VARIATION PROPAGATES
CONSTRAINT IS
EXERTED ALONG
THESE DIRECTIONS
ADJUSTMENTS CAN BE MADE

USING DOF IN THIS SPACE
VARIATION PROPAGATES
ALONG THESE DIRECTIONS
Datum Flow Chain
10/14/2004
Daniel E Whitney
16
Type 1 and Type 2 Assemblies
Type 1 assemblies arrive at the assembly line with

all their assembly features on them
These features are determined by functional needs
Assembly consists of putting them together by

joining the mates. Fixtures are not needed
After the mates locating a part are joined, any
remaining contacts can be joined
Examples: crankshaft into block, wrist pin into
piston and conn rod, head onto block (dowel pins
+ head face = the mate, head bolts = the contact)
Datum Flow Chain
10/14/2004
Daniel E Whitney
17
Type 1 and Type 2 Assemblies - 2
Type 2 assemblies have some incomplete

assembly features on them when they arrive at the
assembly line
these features do not provide enough constraint to
completely determine where the adjacent part will be
These incomplete features are called contacts
they dont carry any KCs
Assembly operations that join contacts require

additional constraint provided by measurements or
fixtures (fixture = passive measurement)
Examples: slip joints in car bodies and aircraft
fuselages
Datum Flow Chain
10/14/2004
Daniel E Whitney
18
A Hybrid Type
Selective assembly is an intermediate form

between Type 1 and Type 2
The parts have complete mate features when they
arrive at the assembly line
However, a fixture or measurement is required
Instead of a slip joint, a selected intermediate part

is used
The resulting joint can be a mate or a contact
select main crankshaft bearing is a mate

select solid valve lifter is a contact
Datum Flow Chain
10/14/2004
Daniel E Whitney
19
Types of Variation Problems
Passive Assembly
1. Type-1: Finished assembly configuration
2. Type-1: In process: Assemblability
Active Assembly
3. Type-2: Adjustable using fixtures

Datum Flow Chain
10/14/2004
4. Hybrid: Adjustable using selective assembly

Daniel E Whitney
20
Front Wheel Drive Transmission
See parts and

operation on
next slide
Datum Flow Chain
10/14/2004
Daniel E Whitney
21
OUTPUT
POWER
INPUT POWER
DIFFERENTIAL
HUB
SELECT THRUST WASHER
TRANSFER
CHAIN
TRANSFER
CHAIN
ROTATING CLUTCHES
UNCERTAIN
STACK
HEIGHT
CASE
OUTPUT
POWER
BAND
CLUTCH
THRUST WASHER
OIL PUMP
Datum Flow Chain
10/14/2004
Daniel E Whitney
22
Transmission Parts
BELL
HOUSING
How Oil Pressure Gets to Pistons
OIL PUMP POST
KC
OIL PASSAGE
PISTON
OIL GROOVES
OIL PASSAGE
CLUTCHES
OIL PUMP
Thrust Washer
Datum Flow Chain
10/14/2004
Daniel E Whitney
23
Select Washer Drives Assembly Sequence
Contact
KC
5
KC
4
1
3
2
Root
Mate
Datum Flow Chain
10/14/2004
Daniel E Whitney
DFC
24
Type 1 and Type 2 Assemblies - 3
Required assembly sequence is
join the mates first

join the contacts second
the Neon problem
In type 1 assemblies, it is usually enough to make

each part well; assembly means snapping the
mates together by mating the features
In type 2 assemblies part accuracy is not enough
because the parts do not mate just to each other
but instead need adjustments or fixtures, which
contribute additional errors
Datum Flow Chain
10/14/2004
Daniel E Whitney
25
From Parts to Assemblies...
The Key Characteristic
The DFC for that
key characteristic
A piece of the DFC inside one part

Each piece of the DFC inside a part
is a DFC, too. Note that it joins features
that are by definition under exact constraint
with respect to each other as long as the parts
are rigid or are not distorted by locked-in stresses.
Datum Flow Chain
10/14/2004
Daniel E Whitney
26
Example - Car Floor Plan - A Type-2
The dimension to be controlled (the Key Characteristic or

KC) is the width L of the floor pan
liaison diagram
contact
Source:

Datum Flow Chain
10/14/2004
Daniel E Whitney
B
mate
KC
A
27
First Candidate Datum Flow Chain
Step 1
contact
Datum Flow Chain
Step 2
L2
mate
L3
Step 1
C
B
Step 2
L1
L2
KC
L3
If there is some uncertainty in the shape or size of part A,

this assembly process may not yield the correct overall size.
Datum Flow Chain
10/14/2004
Daniel E Whitney
28
Second Candidate Datum Flow Chain
mate
Step 1
Datum Flow Chain
L1
L2
Step 2
Step 2
C
A KC
contact
B
Step 1 B
C
L3
L1
L2
This DFC directly controls the delivery of the KC and is likely to be better
but it may be impossible, in which case another assembly sequence may be needed
Datum Flow Chain
10/14/2004
Daniel E Whitney
29
Properties of DFC
No loops are allowed (else over-constraint?)
In each DFC there is one root node with only

outgoing arcs
Every joint where the DFC passes is a mate
Sum of DOFs constrained = 6 - designed freedom
Incoming arcs are labeled with number of DOFs
constrained or names of directions constrained
DFCs also contain information on the type of

mating feature and the amount of constraint
applied by that feature
Datum Flow Chain
10/14/2004
Daniel E Whitney
30
Example Assembly
Controls air flow to engine via

accelerator pedal cable
Reports throttle opening to ECM
Source:
Throttle Body
Datum Flow Chain
10/14/2004
Daniel E Whitney
31
Example Assembly
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
32
Alternate DFCs for Throttlebody
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
33
3D DFC/Tolerance Analysis
We need to find a chain of mates from one side of

a KC to the other
The designer should do this
In 3D it gets interesting because
different directions may have different chains
The same feature may be a mate in one direction and a
contact in another direction
A new symbol is needed:
It is both a mate and a contact, depending on the

direction
Datum Flow Chain
10/14/2004
Daniel E Whitney
34
Rules for Drawing DFCs
Draw the liaison diagram for the assembly
Choose one of the KCs and identify the dof or

dofs that are involved
Identify the base part or origin for the DFC
This should be a part that provides location for other

parts that are involved in the KC
Label the liaisons of the base part with arrows

pointing out
Datum Flow Chain
10/14/2004
Daniel E Whitney
35
Rules 2
For each part at each end of the KC, label the liaisons with
arrows pointing toward the part
Label each of these liaisons with the dof that need to be
constrained in order for the KC to be delivered, as well as
those needed to provide proper constraint
Assume temporarily that the part or parts at the other end
of each of these arrows is properly constrained
Think up an assembly feature joining the KC end part and
each of its immediate neighbors that is capable of
constraining the required dof
Datum Flow Chain
10/14/2004
Daniel E Whitney
36
Rules 3
Remove the assumption that the previous part was properly

constrained, and figure out how to constrain it by repeating
the above process for it
Proceed backwards in this fashion until you reach the base
part, which is always assumed to be properly constrained
If full constraint cannot be provided all along the chain just

by using part-to-part mates, then fixtures (temporary parts)
have to be added
Datum Flow Chain
10/14/2004
Daniel E Whitney
37
Rules 4: Separating the DFC into
Individual KC Directions
Choose a KC and one of its dof

Choose a part at one end of this KC
Identify the mate or mates on this part that are needed to

constrain the chosen dof
The best way to do this is with Screw Theory
Other than that, you have to use high school geometry and
geometric reasoning
Erase liaisons to the chosen part that do not carry these dof
Continue in this fashion until you reach the base part

Repeat for the part at the other end of the KC
Datum Flow Chain
10/14/2004
Daniel E Whitney
38
Building a Stapler DFC
Hammer
Hammer
Handle
Hammer
Handle
Handle
6 Plane-plane
Carri er
Pusher
Pin
Step 2
Carrier
Rivet
Staples
Anvil
6 Plane-plane
Base
Staples
The KC requires 5 dof
Staple
Carrier
Staple
Hammer
Carrier
Handle
6 Plane-plane
Carrier
5 Pin-hole
Step 1
Pin
Step 3
Carrier
5 Pin-hole
Pin
6 Plane-plane
Staples
Datum Flow Chain
10/14/2004
Daniel E Whitney
Pin
39
Door Assembly
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
40
GM and Ford Hinge Mounting
Source:
GM Method
Datum Flow Chain
10/14/2004
Ford Method
Daniel E Whitney
41
GM Door Mounting to Car
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
42
Ford Door Mounting to Car
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
43
DFCs for Car Doors

Liaison Diagram
DFC if this were a Type-1
DI
DI
KC1
Body
Seal
Body
Seal
KC2
DO
DO
Ford Process
GM Process
Source:

Datum Flow Chain
10/14/2004
Daniel E Whitney
Source:

ISBN: 0195157826.
44
KC Delivery Chains for Each KC - GM
Method
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
45
KC Delivery Chains for Each KC - GM
Method
Seal KC
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
46
Door Process Comments
The designer should create these DFCs, with
participation of manufacturing engineers
The chain is the thing to manage, not the parts or

the fixtures separately
The chain will never exist all at once, in one place,
at one time. This makes management hard.
When a fixture like F1 sets door inner to door
outer, the parts remember the positioning and its
errors
The DFC captures organizational interactions
Datum Flow Chain
10/14/2004
Daniel E Whitney
47
GM Door DFC with Organizations
All Directions
Engineering
Door Outer
Manufacturing
F1
Appearance KC
Car
Body
Door
Inner
Seal KC
Seal
F2
Purchasing
Datum Flow Chain
10/14/2004
Hinges & Cones

Daniel E Whitney
48
767 Wing Skin Subassembly
Source:
Figure 8-64(a) in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
Datum Flow Chain
10/14/2004
Daniel E Whitney
49
Assembly Using a Fixture - Type-2
Fixture
Splice Stringer 3
Aft Skin
Splice Str 3
Forward Skin
Aft Skin
Stringers 1 & 2
Fwd Skin
Str 1-2
Plus Chord
Stringers 4-11
Liaison Diagram
Str 4-11
Plus Chord
Datum Flow Chain
Plus
Chord
Forward Skin
Splice stringer
Aft Skin
Fixture
Datum Flow Chain
10/14/2004
Daniel E Whitney
50
Assembly Using Features on the Parts -
Type-1
Splice
Str3
(6)
Aft
Skin
(6)
(6)
Fwd
Skin
FORWARD SKIN
Str1-2
PLU
(5)
(1)
ORD
S CH
SPLICE STRINGER
(6)
Plus
Chord
Str4-11
AFT SKIN
(a)
ASSEMBLY LEVEL DATUMS

PART LEVEL DATUMS
MATING FEATURE (SLOT)
MATING FEATURE (HOLE)
Datum Flow Chain
10/14/2004
Daniel E Whitney
51
Datum Flow Chain for Car Front End

F1
L. Body Side
R. Body side
Underbody
L. Door
R. Door
F
6
L. O. Rail
y,z
F
x, y, z
y
x, x,y, z
L. Apron
L. O. Shot.
z, x y
F
x, y, z
z, y
y, z, y, z
R.I. Rail
x, z, x, y
F
x, x,y, z
x, x
x, x
F
6
Hood
x, z
x, y
Hood fixture
L. I. Shot.
y, z
R. Apron
x, y, z
y
z, x y
R. O. Shot.
x, y, z
R. Fender
L. Bracket
Hood Latch
y, z
Bolster
z, y
R. Hinge
L. Bracket
y, z, y, z
R. I. Shot.
L. Hinge
L. Fender
R. O. Rail
L. I. Rail
Dash
R. Lamp
L. Lamp
x, z, x, y, z
x, z, x,
y, z
Datum Flow Chain
Fascia
10/14/2004
Daniel E Whitney
Drawn by Gennadiy Goldenshteyn, MIT Student
52
DFC for Car Doors
Seal
Bodyside
Door Outer
x, z, y
Glass Set Fixt.
y, x, z
Glass
x, z
x, z, y
x, z, y
Rear Door Inner

F2
x
x, z
y, z
Latch reinf.
Rear Reg. Rail
Lower Hinge
z
x, z
Frt. Reg. Clamp
y, z, x
x, z, y
Frt. Reg. Rail
y, z
F6
x, z, y
Datum Flow Chain
F3
Inner Belt Reinf.
y, x,, y
Upper Hinge
y, x, z x, z, y
x, z, z
F5
6
F7
Rear Reg. Clamp
x, z, y
y, x, z
Frt. Glass Guide
Latch
Outer Belt Reinf.

x,z
Pin on hinge
F4
x, z, y
F8
y, x,
Striker
10/14/2004
Daniel E Whitney
y, z
Frt. Door Inner
y, z,
x
x, y, z
Lower tap. plate

x, y, z
Upper53
tap. plate
Drawn by Gennadiy Goldenshteyn and Jagmeet Singh, MIT Students
Assembly Precedence Constraints
Contact Rule
Permissible subassemblies are connected subgraphs in a
DFC
In other words, subassemblies with only contacts between parts are not
permitted, so all mates must be completed before any contacts are completed
Constraint Rule
Subassemblies with incompletely located parts are not
permitted
Every subassembly must have fully located parts, so all incoming mates must
be done simultaneously (a>=b & b>= a)
If necessary, a fixture (considered to be a part)

may be added to help obey the rule
A practical consequence is that assembly
sequences will build the DFC from the root out
Datum Flow Chain
10/14/2004
Daniel E Whitney
54
Variation propagation analysis using DFC
Tolerance chain for any KC can be derived by

traversing the DFC
Since DFC is a directed acyclic graph, there is a
unique tolerance chain for any KC
For type-1 assemblies, all assy sequences will
have identical tolerance chains and variation
Type-2 assemblies are process-determined, hence

have to evaluate each process and its fixtures
Datum Flow Chain
10/14/2004
Daniel E Whitney
55
DFC and Constraint
Over-constraint can show up if the DFC has
A branch followed by a merge

A loop (should not occur)
A mate consisting of multiple features
You need to check each case individually
Datum Flow Chain
10/14/2004
Daniel E Whitney
56
DFCs, Tolerances, and Constraint
Constraint not OK
at nominal
Constraint OK
at nominal
dimensions
Constraint is robust
to variation:
there is a unique and
permanent DFC:
mates stay mates,
contacts stay contacts
Datum Flow Chain
10/14/2004
Constraint is not robust:

mates and contacts do not
maintain their identity there is no unique
and permanent DFC
Daniel E Whitney
Overconstrained:
there is no DFC
as we define it
(or else you need
a stress analysis
to find it)
57
Assembly Design 1: Nominal Design Phase
Identify the KCs

Decide type 1 vs type 2 (*could be revised)
Identify constraint plan plus fixtures and
adjustments if any*
Define the DFCs, features, mates, and contacts for

each KC*
Check for proper constraint
Find feasible assembly sequences and choose one*
Check for KC conflict

See if different assembly sequence removes KC
conflict
Datum Flow Chain
10/14/2004
Daniel E Whitney
58
Assembly Design 2:
Variational Design Phase
Analyze tolerances to see if DFC is robust

Analyze tolerances to see if KCs are delivered
See if a different assembly sequence gives better

variation
Revise * as necessary
Datum Flow Chain
10/14/2004
Daniel E Whitney
59
Datum Flow Chain
10/14/2004
Daniel E Whitney
Design Procedure for Properly

Constrained Assemblies
Source:
60
Summary of Assembly Theory -
Nominal Design
An assembly is a set of parts that deliver their quality,

as defined by the KCs, as a result of achieving proper
geometric relationships between the parts
Designing an assembly means designing these
relationships in terms of one DFC per KC
The DFC documents the nominal constraint relationships
The DFC passes from part to part via mates
The nominal design is a constraint structure onto

which we paste parts
Assembly features generate and enforce the constraint
relationships at each mate
Datum Flow Chain
10/14/2004
Daniel E Whitney
61
Summary of Assembly Theory -
Variation Design
Tolerances should assure the robustness of the DFC
KC delivery is verified by a tolerance analysis of
each DFC. Variation passes through the mates.
Tolerances on parts flow from tolerances on the

KCs
Type -1 assembly-level variation comes from part
variations
Type - 2 assembly level variation can be altered by
adjustments to the assembly process
Datum Flow Chain
10/14/2004
Daniel E Whitney
62
Summary of Concepts
Source:
Datum Flow Chain
10/14/2004
Daniel E Whitney
63
Assembly in the Large: Basic Issues
Goals of this class
put assembly in the large in the context of product

development
relate it to customer expectations
start to think about architecture
AITL Basic
10/22/2004
FRONT-END PRODUCT DEVELOPMENT PROCESS

CUSTOMER
NEEDS
(DOCUMENT
AS PKCs)
THE BASIC VALUE PACKAGE

FUNCTIONALITY
PRICE
PRODUCT
SPECIFICATIONS
RANK BY
NEED
SATISFACTION
RANK BY
COST AND RISK
ASSESS VALUE
OF SATISFIED NEEDS
AGAINST COST & RISK
AITL FLOW
Basic
PDP PROCESS
PRELIM ARCHITECTURE,
KC FLOWDOWN, AND
REUSE STRATEGY
CONCEPT
GENERATION
PRELIM TECHNOLOGY PLAN

AND SUPPLIER STRATEGY
PRELIM MFR COST EST
ID HIGH RISK AREAS
CONCEPT SELECTION
REFINE SPECIFICATIONS
AITS & AITL occur

here during concept
design
WRITE CONTRACT BOOK:

SELECTED CONCEPT
TARGET SPECS
TARGET COST
TARGET PROCESSES
FOR DESIGN, MFR, IT
TARGET SCHEDULE
AND RESOURCES
ID RISKS AND FALLBACKS
SUPPLIERS,
10/22/2004
Daniel E Whitney
1997-2004
Daniel Whitney 1997 OUTSOURCING,
AND PARTNERS
PKC: PRODUCT KEY

CHARACTERISTIC
KC FLOWDOWN: TRACING
ABILITY TO DELIVER
EACH PKC DOWN TO
SUBASSYS, PARTS,
2
SUPPLIERS
A Little History
Is my product ready for robot assembly?

Well, is your product ready for assembly at all?
What are the requirements for assembly?

Can we explain them to a machine?
Do we understand the product well enough that
our suggestions
make sense
do not compromise performance
We may have to reverse engineer it to find out
AITL Basic
10/22/2004
Product Character
Which of the following products is most like a fire

extinguisher?
(a) sewing machine
(b) hand grenade

(c) lawn sprinkler
What are the issues that go into answering this

question?
AITL Basic
10/22/2004
Two Kinds of Copiers
Industrial strength and capacity

Costs a lot
Is finnicky: design is not robust
Customer can afford full time service person
Home or small business

Must be low cost
Must work
Cant afford service person on site
The manufacturer did OK with the first but failed

with the second
AITL Basic
10/22/2004
Manual Sewing
Source:
Source: "Real Robots Do Need Jigs," Daniel E Whitney,
Harvard Business Review, May-June 1986, pp 110-116
AITL Basic
10/22/2004
Machine Sewing - 1
Source:
AITL Basic
10/22/2004
Machine Sewing - 2
Source:
The needle pokes through the cloth and leaves a loop.
The bobbin is shown passing through the loop.
In fact, a hook catches the loop and slips it under the bobbin.
When this step is finished, an arm above pulls the loop tight.
AITL Basic
10/22/2004
Comparison of Manual and Machine Sewing Methods
Manual
Machine
Number of
"hands"
Two
One
Number of
threads
One
Two
Grasp of
needle
Repeated
grasp/ungrasp
Never
ungrasp
Location of eye
Rear of needle
Tip of needle
Needle movement
Passes through
Flips 180
Point penetrates
Never flips
Joining method
One thread passes

through repeatedly
AITL Basic
10/22/2004
Two threads
interlock but
never pass through
9
Source:
AITL Basic
10/22/2004
Four Ways to Print
10
Comparing 4 Ways to Print
Basic
actuation
method and
power source
# DOF
# of parts
Structure
Typewriter
Manual,
complex
linkages
Carriage:2
Ribbon: 2
Keys: 1
each*50+
keys*many
links/key
Key carrier: 1
Many hundreds
Heavy metal
Shapes printed Fixed character

shapes
Colors
Media
Assembly
AITL Basic
Two
Paper, two or
three sheets
Manual,
lengthy, tedious
10/22/2004
Ballhead
Manual input,
solenoid
actuation,
simple linkages
Platen: 1
Ribbon: 2
Keys: 1
each*50+ keys
electrically
actuated
Ball carrier: 3
Hundreds
Heavy metal
Fixed character
shapes bu t
different balls
have different
fonts
Two
Paper, several
sheets
Manual,
lengthy, easy
Dot Matrix
Electro-magnet
for each dot
maker
Inkjet
Piezo-electric
for each color
of ink
Platen: 1
Ribbon: 2
No keys
Dot carrier: 1
Each dot: 1
Platen: 1
No ribbon
No keys
Jet carrier: 1
25-50
Metal and
plastic
Unlimited
shapes bu t low
resolution
10-20
Almost all
plastic
Unlimited
shapes and high
resolution
Two
Paper, many
sheets
Automatic &
manual
Unlimited
Any, but one
sheet
Manual, quick,
easy
11
Takeaways
There are many ways to implement a function
They differ in technology choice, materials,

degrees of freedom, allocation of dof, number of
parts
Different implementations have different
capabilities for function, customization, upgrade
They also have different assembly requirements
Sometimes assembly requirements can drive

redesign IBM ProPrinter example
AITL Basic
10/22/2004
12
Steps in AITL - 1
Understand the business context
product character, type of market, customer expectations

sales volume anticipated
model variety anticipated
plans for new versions
delayed commitment
supplier logistics and make vs buy
cost limits
labor costs and any regulations
cost calculation and ROI methods
ROI targets
AITL Basic
10/22/2004
16
Steps in AITL - 2
Understand the factory context

labor conditions, training, shift policies
space and facility constraints
Understand the as -is assembly (AITL)

study the existing manual process, if any
inspecting fiber
ignore the existing manual process and focus on

technical and economic requirements
may give rise to a new level of DFA especially if

automatic assembly is under consideration
sewing, Sony VCR line, RAM with fuses
do not ever imply that performance might have to be

compromised!
AITL Basic
10/22/2004
17
Steps in AITL - 3
Identify system requirements

alternate assembly sequences
tentative cycle time
production flow and floor layout
parts presentation
feasible methods and equipment
required sensing and communication

required displays and controls
fixtures and parts carriers
AITL Basic
10/22/2004
18
Steps in AITL - 4
Design a concept assembly system
system architecture
equipment selection and task assignment
cost and economic performance
simulation
average flow and production rate
model changeovers and maintenance (scheduled downtime)
failures, repair time (unscheduled downtime)
queues, blockage, starvation (unscheduled downtime)
AITL Basic
10/22/2004
19
Steps in AITL - 5
Make final recommendations

additional design improvements
line design or sequence options
remaining risk areas

cost estimates
AITL Basic
10/22/2004
20
Structure of System Design Issues
Local
Global
Economics and market

targets
Volume growth
Model varieties
Design volatility
Quality, reliability, safety
Make or buy decisions
Build to order/stock
Cost and productivity goals

How it interfaces to the factory
Labor policies
Failure modes and repair policies
Space needs
Product
Assembly
System
AITL Basic
10/22/2004
Assembly sequences
Types of operations
Geometric constraints
Part size and weight
Shape, stiffness
Tolerances and
clearances
Tests and inspections
System layout
Equipment choice
Task assignment
Part feeding and
logistics
21
Product Architecture
Goals of this class
understand product architecture and its role in product

development
see the effect of different architectures on some AITL
issues
look at examples of different architectures
Product Arch
10/26/2004
Daniel E Whitney
Definition of Product Architecture
Product architecture is the scheme by which the
functional elements of the product are arranged
into physical chunks and by which the chunks
interact
This definition links architecture to system-level
design and the principles of system engineering
Architecture also has profound implications for

how the product is designed, made, sold, used,
repaired, etc
Architecture makes its influence felt during
assembly
Product Arch
10/26/2004
Daniel E Whitney
Front and Rear Wheel Drive Architectures
Same elements, arranged differently
fwd & rwd
Product Arch
10/26/2004
Daniel E Whitney
Role of Architecture in Product Development
MARKET OPPORTUNITY
MODULE
CHOICE
ARCHITECTURAL
CHOICE
MARKET
SEGMENTATION
RE-USE
STRATEGIES
FLEXIBILITY
AND RISK
MITIGATION
PLATFORM
OPPORTUNITIES
TECHNOLOGICAL
OPPORTUNITY
TECHNOLOGY
PLANNING
EXAMPLES: BOEING JETS

DENSO PANEL METERS
TELEMECHANIQUE CONTACTORS
B & D HAND TOOLS
architecture
Product Arch
10/26/2004
Daniel E Whitney
PROJECT
PORTFOLIO
MANAGEMENT
PRODUCT
FAMILY
MANAGEMENT
LONGER RANGE
OUTSOURCING
OPPORTUNITIES
AND RISKS
Architectures Influences
During product
development
How families and platforms are

structured
How functions are realized
How reuse and standardization
are accomplished
How development work is
divided up
Where subassembly and
module boundaries are
Where DFCs go
In the product
In the organization
Along the supply chain
Product Arch
10/26/2004
Daniel E Whitney
Architectures Influences - 2
During production system

design
Product Arch
10/26/2004
Assembly sequences
Reuse of facilities and

knowledge
Planning for flexibility
Sharing of facilities to
match capacity to demand
Daniel E Whitney
Architectures Influences - 3
During manufacturing and Where production happens
assembly
How customer orders are
fulfilled
How unpredictable
demand patterns are met
During use
Product Arch
10/26/2004
How service is delivered
How the product is

updated
How the product is
recycled
7
Daniel E Whitney
Integral and Modular Architectures
Integral = functions shared

by physical elements
Examples...
Some reasons integral is
used...
Modular = each function is

delivered by a separate element
Examples...
Some reasons modular is used...
Is either kind of architecture better than the other?
Product Arch
10/26/2004
Daniel E Whitney
Architecture and the S-Curve
New product:
Stress on innovation
Companies and
and customers
are exploring
Reliability not important
Consistency irrelevant
No dominant
architecture
Dominant design (architecture) emerges:
Customers know
what they want
Companies know
what they have to do
Mature product:
Stress on process innovation
Reliability, consistency important
Reuse and standards dominate
Source: Utterback, von Hippel
In some cases, there is evolution from rapid change strategy

Time
to high variety strategy [Sanderson and Uzumeri] over time
Architecture may evolve from integral to modular [Christensen] or not!
Product Arch
10/26/2004
Daniel E Whitney
Integral/Modular Comparisons
Modul ar
Integ ral
Ch un ks ma y be in teg ra l ins id e bu t are

ind ep ende nt from each oth er
funct iona lly and ph ysically
Ch un ks ma y be in teg ra l ins id e and

int er d ep ende nt amon g each oth er
Standa rd , p re-d es igned int er faces can

be us ed that can re ma in the sam e even
if in te rna l cha ract er istics ch ang e
Int erf aces a re ta ilored to the chu nk s

an d are de p end en t on func tion al
be hav ior
Modu les can be s p ecialized to th eir

ind ividua l cont ribu tion s to over all
funct ion and can be us ed
int er chan ge ab ly
Ch un ks a re tai lore d to th eir app licati on

an d can n ot b e in te rchan ge d w ith out
re qui ri ng cha n ges to other chu n ks
Unp re d icta bility of mo d ul e cho ice

re qui re s ov er- d esign elsew here to
acc ommo d at e poss ible mis m atch es
Ove ra ll d es ign can be opt imiz ed for a

p re d icta ble set of fun ctions an d
impl em entat ions
Standa rd int er faces are ph ysically

sepa rat e from th e mo d u le an d th us
w ast e oth er d esign re sou rces su ch as
spac e or wei gh t; in ter faces are we ak
Int erf aces can be i nt egr al to th e chun k,

sav ing spac e or w eight ; int erf aces a re
s trong
Int erf ace mana ge m ent, if p lann ed

p rop er ly, can p rov id e flexibility d u rin g
p rod uct ion
Int erf ace mana ge m ent occu rs en tirel y

d ur ing de sign an d is froz en; it is not
aim ed at flexibility a fter de sign
Busi n ess p erf ormanc e ma y be favo red
Techn ical p er forman ce ma y be favore d
Product Arch
Adapted from J-P MacDuffie Automotive Build to Order The Modularity - e-Business Link IMVP talk 9/27/00
10/26/2004
Daniel E Whitney
10
Integral/Modular Options*
Each function is
realized by
Cast or molded parts

Transaxle
case
VLSI
Modular
architecture
One part
Typical
simple
assembly
Many parts
Many functions are

realized or shared by
Integral
architecture
Ulrich and
Eppinger
Most assemblies
Car door
Integral
chain or holistic
architecture
Integral or
coupled
architecture
Mixed architectures are the most common, in which some functions are realized
by some options and others by other options
*Tim Cunningham, MIT PhD thesisChains of Function Delivery: A Role for
Product Architecture in Concept Design Feb 1998
Product Arch
10/26/2004
Daniel E Whitney
11
Lehnerd-Meyer Platform Concept
MARKET SEGMENTS
MARKET
TIERS
DERIVATIVE PRODUCTS
AIMED AT SPECIFIC
MARKET TIERS
OR SEGMENTS
PRODUCT
PLATFORMS
CUSTOMER
NEEDS
PRODUCT
TECHNOLOGIES
MANUFACTURING
PROCESSES
ORGANIZATIONAL
CAPABILITIES
COMMON BUILDING BLOCKS
Courtesy of: THE POWER OF PRODUCT PLATFORMS by Marc H. Meyer and Alvin P. Lehnerd, (c) 1997 The Free Press.
Used with permission.
Product Arch
10/26/2004
12
Daniel E Whitney
Example Platform
Source:
Originally one mechanical board and one circuit board using through hole electronics
Later one board for all using surface mount electronics
Outside shell changed to meet different style and customer needs while guts stay the same
Product Arch
10/26/2004
Daniel E Whitney
13
Denso Panel Meters: Architecture and
DFA Driven by JIT in Assembly
Denso makes many kinds of panel meters for Toyota.

Toyota orders different ones in different amounts every day.
ND designed an assembly family of meters and can make
any quantity of any kind at any time by selecting the right
parts. Assembly interfaces were standardized for all parts. The result is
assembly-driven manufacturing.
Source:
Each path yields

a different kind
of meter.
288 different kinds of meters can be made
with no additional cost or delay, and
almost no changeover time.
Product Arch
10/26/2004
Daniel E Whitney
14
Panel Meter Family Assembly Architecture
Source:
ONE INSTANCE OF
THE PANEL METER
Product Arch
10/26/2004
Daniel E Whitney
SEVERAL INSTANCES
OF THE PANEL METER
15
Panel Meter Assembly Architecture and DFC
KC (Function)
NEEDLE
FIXED
SHAFT
M
SH OVI
AF NG
T
NEEDLE
DIAL
FACE
FIXED
SHAFT
DIAL
FACE
BASE
BASE
BASE
BIMETAL
NEEDLE
MOVING
SHAFT
BIMETAL
CASING
BIMETAL
(a) BASIC PART-PART RELATIONSHIPS
CASING
CASING
(b) DIMENSIONAL FLOW

BETWEEN PARTS
Product Arch
10/26/2004
Daniel E Whitney
(c) DIMENSIONAL
FLOW WITHIN
PARTS
16
Panel Meter KCs
Function of each meter

Alignment of parts stack
Placement of moving pin and stationary pin
Function of the family
Negotiation of family variety dimensions
Common architecture along stack of parts
Same DFC for each family member

Each family member mechanizes each function the
same way using similar part designs
All inter-part mates standardized for each version of a
part type or placement on the stack
Product Arch
10/26/2004
Daniel E Whitney
17
Fabrication-Driven Manufacturing
FABRICATION
ASSEMBLY
ORDERS
RELIES ON FABRICATION TO EXPRESS MODEL

MIX AND ACHIEVE FLEXIBILITY:
IN RESPONSE TO ORDERS, COMPLEX
PARTS ARE MADE AND THEN ASSEMBLED
INTO FINAL ITEMS.
THIS IS A LOW BANDWIDTH METHOD BECAUSE
FABRICATION TAKES SO LONG.
Product Arch
10/26/2004
Daniel E Whitney
18
Assembly-Driven Manufacturing (Denso

Panel Meters)
FABRICATION
ASSEMBLY
ORDERS
RELIES ON ASSEMBLY TO EXPRESS MODEL MIX

AND ACHIEVE FLEXIBILITY:
SIMPLE PARTS ARE MADE TO STATISTICAL TRENDS.
IN RESPONSE TO ORDERS, ITEMS ARE ASSEMBLED.
THIS IS A HIGH BANDWIDTH METHOD BECAUSE
ASSEMBLY HAPPENS SO QUICKLY.
Product Arch
10/26/2004
Daniel E Whitney
19
Densos Flexibility Dimensions
Volume change (market controls)

aggregate, distribution within aggregate
Product variation (customer controls)
configuration
size
model
type
Design change (Denso controls)
minor, model, next generation
Product Arch
10/26/2004
Daniel E Whitney
20
Demand Fluctuations
One year
Model C
One
years
needs
Model Cs plant
Model Bs
plant
Model B
Model As plant
Model A
One switchable plant
can make all models and
needs only full years
Capacity - Denso does
it with product design
Product Arch
10/26/2004
Three focused plants must each

have max capacity anticipated
for its product
Daniel E Whitney
21
Lightweighting Cars
Steel approach
high strength steel in small amounts
integral design
welded
Aluminum design
rely on lower weight material
modular design
tinker-toy assembly
Product Arch
10/26/2004
Daniel E Whitney
22
Ultralite Steel and Aluminum Car Bodies
Source:
Rib-stiffened Shell
Photo from Am Iron and Steel Inst
Product Arch
10/26/2004
Daniel E Whitney
Space Frame
Photo from Audi
23
Airbus: Architecture Driven by Politics
Airbus Industrie was a consortium that shared revenue

and profits according to a work-content formula
This formula was based on a decomposition of the
plane
wings to British Aerospace
fuselages to Deutsche Aerospace
tail sections to CASA (Spain)
final assembly and integration to EADS
The A380 caused problems

The wings are too big to transport to Toulouse by air
A special handling system was developed for land and water
transport
Product Arch
10/26/2004
Daniel E Whitney
24
Source:
Product Arch
10/26/2004
Daniel E Whitney
27
Airbus A319, 20, 30, 40
Source:
Product Arch
10/26/2004
Daniel E Whitney
28
Commonality, Carryover, Reuse
A mystery: how could Toyota design cars with so

few engineering hours?
By outsourcing (1970s-80s)
By carrying over parts (1980s-90s)
paradox: heavyweight program manager -> parts proliferation
By overlapping whole car programs (1990s)
Toyotas definitions
OK to reuse parts the customer doesnt see
Reuse of a body shop means reuse of assembly
sequence and assembly locators
Whoops! There goes $75M.

Product Arch
10/26/2004
Daniel E Whitney
33
Source:
Ford Motor Co.

Product Arch
10/26/2004
Daniel E Whitney
34
Elements of Carryover and Reuse
Consistency of interfaces
between adjacent parts

between parts and tools
between parts and fixtures
assembly sequence
Consistency of performance (out of scope for this course)
strength
materials compatibility
dynamics
coherence as defined by Fujimoto and Clark
Product Arch
10/26/2004
Daniel E Whitney
35
Product Structures
Ulrichs types (called architectures)
integral (parts are combined)

slot (parts share structured common base)
bus (parts share unstructured common base)

sectional (parts are linked)
Redford and Chals
integral (parts are combined)

differential (parts are separated)
stack/serial (Tower of Hanoi)
array/parallel (circuit board)
Product Arch
10/26/2004
Daniel E Whitney
36
Business-Driven Structures
variety
assy seq
Mushroom product (Mather)
inside/outside or assembly sequence chosen to permit

delayed committment or to accommodate long lead time
items
only a few parts are in the variable set (head of mushroom)
Plain vanilla box strategy (Tayur)

similar to mushroom but often focused on inventory
management and smoothness of line operations
Denso panel meters

no mushroom or vanilla box segment can be found
no base plus variations can be found
Product Arch
10/26/2004
Daniel E Whitney
39
How DELL Does It
Product is highly modular

Suppliers own inventory until assembly starts
Order takers know what is in stock

Order takers know what modules are mutually
compatible
Customers are lured to what is in stock that meets
their needs using short delivery time
There are runners and expediters scouring the
earth for parts
Inventory turns were 6 only 4 years ago vs 20 now
Product
Why
cant
Detroit be like DELL?
Arch
10/26/2004
40
Daniel E Whitney
Volvos 21 Day Car
Cars are delivered to customers 21 days after order
All high value items are built less than a few
hours transit time from final assembly line
Engines, transmissions, seats, instrument panels
Many mid-value or highly variable items are built

in the final assembly plant in nearby shops
Steering columns
There are big stocks of low value parts

The biggest problem is unpredictable paint quality
Only 20 colors and 3 body styles avail

Cars are painted on spec and picked to meet orders
from 10/26/2004
a one shift size buffer
Product Arch
41
Daniel E Whitney
Volvos 21 Day Car - 2
Customer can choose:

Seats
Interior colors
Roof rails
A/C, cruise control
Electr windows & mirrors
Random access buffer of ~ 300 painted bodies

= 8 hrs worth - 3 body types,
20 colors = 5 duplicates of any one item
Car assigned
To customer
4 hours before
Final assy starts
Weld
(4-6 hrs)
Seat
plant
Paint
(4-6 hrs)
Final assembly (8 hrs)

Specific order
build starts
4 hrs
165K/yr
2 shifts=
~1.5 min/car
3 hrs
5 hrs
Source: Mr Etienne De Jaeger, Volvo (Ghent)
Product Arch
10/26/2004
Daniel E Whitney
42
Telemechanique Motor Contactors:
Architecture Driven by Customers
CONTACTORS START AND STOP MOTORS FROM WATTS TO KWATTS
THE FAMILY IS A SET OF SIMILAR-LOOKING ITEMS WITH SIZE WATTS

BASIC IDEA: MODULES ARE CUSTOMER'S VIEW
SUBASSEMBLIES ARE MANUFACTURER'S VIEW
THE CUSTOMER PUTS THE PRODUCT TOGETHER FROM MODULES
THEREFORE: DESIGN THE "BUYING EXPERIENCE" FIRST
DESIGN THE CATALOG, CROSS-INDEX, LINK TO USES
DETERMINE THE CORRECT MODULES

STANDARDIZED MOUNTINGS, TERMINALS, BOXES
THEN DEFINE SUBASSEMBLIES

THEN PLAN THE RANGE OF IMPLEMENTATIONS
ACROSS POWER LEVELS
Product Arch
10/26/2004
Daniel E Whitney
43
Architecture of Power Tools
Source:
CLAMSHELL ARCHITECTURE
Brand X late 70s

(still used)
Product Arch
10/26/2004
Daniel E Whitney
45
Alternate Architecture for Power Tools
Source:
AXIAL/STACK ARCHITECTURE WITH COMMON MOTOR MODULE
Black & Decker ~ 1981

(still used)
See Lehnerd, Alvin P, Revitalizing the Manufacture and Design of Mature Global Products
In Bruce Guile and Harvey Brooks, eds, Technology ad Global Industries, Washington,
National Academy Press, 1987.
Product Arch
10/26/2004
Daniel E Whitney
46
Drill Architectures
Bosch
Milwaukee
Craftsman
Product Arch
10/26/2004
Craftsman
Daniel E Whitney
51
Architecture Tradeoffs for Power Tools
Clamshell design easy to assemble manually

Long history of cost reduction
Cant assemble automatically or robotically
Stack architecture easy to assemble automatically
Allows huge economies of scale in motors

Represents a paradigm shift
How did Black and Decker accomplish it?
a shocking answer
Product Arch
10/26/2004
Daniel E Whitney
52
AMP Powerline Splice
Two use case scenarios

how to buy
how to make
Product architecture aligned to both
Looked great on paper and prototype
Product Arch
10/26/2004
Daniel E Whitney
53
Existing Power Line Splice Scenario
PREMADE
PHONE RINGS: SIZE OF CABLE
SELECT TUBE AND INSERTS
CALL FED EX
IN THE FIELD: TRANSPORT SPLICE, GENERATOR, HYDRAULIC POWER

SUPPLY, AND PRESS TO SITE
INSERT CABLE AND INSERTS, SQUEEZE
Source:
Product Arch
10/26/2004
Daniel E Whitney
54
New Power Line Splice Scenario
ASSEMBLED SPLICE
HOW IT WORKS
THE BUSINESS CASE

PREASSEMBLED
PRICE: $100
SELECT JAWS
SWAGE TAPER
CALL FED EX
Product Arch
10/26/2004
Daniel E Whitney
55
What Happened...
PREASSEMBLED
SELECT JAWS
SWAGE TAPER
Product Arch
10/26/2004
Daniel E Whitney
56
Failure and Remedies
Batch size one plus learning curve
The first one has to work

If it doesnt, you cant keep trying until it works
because it kills the business case
Attempted fix:
Redesign so that swage is done before commitment
Insert selected jaws after swaging
Company rejected this: if the charge ignites, it might
shoot the jaws out like bullets
Product Arch
10/26/2004
Daniel E Whitney
57
Summary
The influence of product architecture is pervasive
It affects most aspects of product design and

manufacturing
Product family design
Product structure including outsourcing and main

subassemblies
Fulfillment, including customization and rapid response
The outcomes of architecture decisions are

implemented during assembly
Product Arch
10/26/2004
Daniel E Whitney
58
Discrete Event Simulation
Goals of this class
Simulation
Understand discrete event simulation
See how it applies to assembly systems

Understand its strengths and weaknesses
See some statistics about real systems
11/20/2002
Stoppage? What stoppage?
It is very hard to achieve high uptime
People do not notice the stops that add up the

most: the very short ones
Story (on a later slide) about loop machine with
three tenders
Denso rules for line workers repairing their own
machine
you have 30 minutes to fix it yourself
if you dont think you can fix it or if 30 min are up, call
the repair crew
Simulation
11/20/2002
Histogram of Stoppage Durations
Source:
Simulation
11/20/2002
Where Do All The Seconds Go?
W. M. Chow, Assembly Line Design, Marcel Dekker, 1990
24 hours per day

Scheduled operating time (utilization)
Unscheduled (bad) downtime
part jams
queue block/starve
machine broken
Scheduled (good) downtime

tool change
variety change
preventive maintenance
employee breaks
Source:

and Role in Product Development .
Process efficiency
task time variation, line balance, logistics, scheduling
Process quality
% bad parts or cycles
Simulation
11/20/2002
Block Line Uptime Varies Greatly from
Plant to Plant
80%
scheduled
stops(tools/variants
/TPM)
60%
unscheduled stops
(tool, machine,
parts)
40%
21
16
0%
Note: Data not available

for some lines
11
20%
uptime
percent of scheduled time
100%
individual lines
Simulation
11/20/2002
Uptime Is a Property of the Engine Plant
Uptime Values Provided by Engine Plants
Note that on Average, in one Plant, there is less than a 15% Difference in Uptime between
any of the 3 Machining Lines
100%
90%
Uptime BLOCK
Uptime CRANK
Uptime HEAD
80%
70%
60%
50%
40%
30%
20%
10%
0%
Engine Plants
Simulation
11/20/2002
How to Get 85% Uptime
Loop-style assembly machine for washing

machine feet
Two people with wands to beat stuck part feeder
tracks
Machine with bus-style pull cord
When a track jams, the people whack at it with

their wands and yank the cord to restart the
machine
Result is 85% uptime
Simulation
11/20/2002
Basics of Simulation
Allows for inclusion of a variety of random events
machine stoppages
queue blockages and starvation
absence of workers, pallets, parts, fixtures
resource contention and levels of service
queue capacity
Companies are surprisingly unaware of the effects of

random events and do not take good data on downtime
We made the line lean. It had no buffers. It didnt
work.
Simulation
11/20/2002
Simulation
Uses a system definition to run a time-based

simulation
Often includes random variables
Can be continuous time or discrete event
Simulation
11/20/2002
Continuous and Discrete
Continuous means equal size time steps
Discrete event means that time advances until the

next event can occur
time steps during which nothing happens are skipped
duration of activities determines how much the clock

advances
Simulation
11/20/2002
10
Components of a D. E. Simulation
Simulations contain
activities where things happen to entities during some

time (which may be governed by a probability
distribution)
queues where entities wait an undetermined time
entities that wait in queues or get acted on in activities
entities can have attributes like kind, weight, due

date, priority
Simulation
11/20/2002
11
Things that Can Be Simulated
Factories
entities are products, people, transporters, tools
activities are machines for fab or assembly
queues are conveyors, warehouses
Highways
entities are cars, trucks, cops
activities are go, stop, rage
queues are highways, on-ramps, off-ramps, rest stops
Simulation
11/20/2002
12
Goals of System Simulation
Check the design of the system

Material flows - are there bottlenecks?
Queue locations and sizes - do they get blocked or starved?
Resources - are they sufficient, do they starve important
operations?
Failure modes - what are they and what causes them?
Check if it has the required capacity

See what different types of downtime do to performance
Improve the design
Simulation
11/20/2002
13
Choices for Probability Distributions
Things that have distributions:
Station operating times

Breakdown intervals
Station repair times
Transport system performance
Alternate distributions
Simulation
Normal
Weibull
Uniform
Fatigue Life
Rayleigh
11/20/2002
14
Station Time Data
Obtained by Prof Tim Baines, Cranfield U, UK

Taken at engine dressing line at UK car company
Electronic data gathered from operators release of

pallet to next station
Averaged over many operators at many stations on
this line to protect anonymity
Main conclusion is that times vary much more
widely than is accounted for in typical simulations
Simulation
11/20/2002
15
Activity Time: a typical day
Simulation
11/20/2002
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
Graph removed due to copyright restrictions. (Workstation time [s] vs Time of Day.)
16
Activity Times & Production Standard
Times (PST)
Graph removed due to copyright restrictions. (Frequency vs Activity Time.)
Simulation
11/20/2002
17
Common Probability Distributions
Weibull
Normal
Fatigue Life
Uniform
http://www.itl.nist.gov/div898/handbook/eda/section3/eda366.htm
Simulation
11/20/2002
18
Kinds of Design Choices
Number and location of resources

People - operators and repair crews

Machines
Pallets and their carrying capacity
Tools
Queue locations, sizes, disciplines

Priorities of activities that compete for limited resources
Floor layout distances, separation of people and

machines
Kinds of transport conveyors, AGVs
Assembly sequences
Simulation
11/20/2002
19
Simple Simulation Diagram
Wait-test
Takes 3 minutes
3 % fail
Takes 10
+/- 1 minute
Arrive
& wait
Test
Done
Machine
Takes 15
+/- 10 minutes
Repair
Wait-repair
Note: If Wait-repair and Wait-test both are full, the system will become deadlocked as soon as
the next unit needs repair because the test station will be unable to unload and take the next
unit.
Simulation
11/20/2002
21
What You Might Learn
Products fail too often
Repair takes too long
A queue is so short that it fills up and prevents

products from moving along
Machines break too often, or there are not enough
repair personnel, or it takes them too long for them
arrive, etc
Then you have to do something about it
Simulation
11/20/2002
22
How a Sim Works
Scan the agenda of activities
Find the first ones that can execute (or continue

executing)
all the entities they need are available
Advance the clock to the end of the activity with

the shortest operating time
Send its entities to their destination queues
the destinations must be open, not full
Go back to the top

Simulation
11/20/2002
23
Basic Modeling Considerations
Identify all the elements and activities

Decide which activities need which entities
Separate the activities with queues

Decide the queue discipline
first come first served, latest, heaviest, etc
Decide where entities should go when the activity

is finished
Often, when you have done all this, you have
learned so much that you do not need to run the
simulation
Simulation
11/20/2002
24
Bottlenecks
Every system has a bottleneck
Its the station with the minimum capacity, the one

that paces the systems operations (often the one
with the longest operation time)
A cycle lost on the bottleneck station is a cycle
lost forever
see The Goal by Goldratt
Queues at non-bottleneck stations are illusions
Simulation
11/20/2002
25
Strengths and Weaknesses
Modeling makes you think

You can run systematic experiments, even DOE
Modeling is not a solution, just a model of a proposed

design
GIGO applies
The probability distributions you choose may not be

the correct ones
Simulation
11/20/2002
26
Conclusions
Simulation is valuable and should be part of every

manufacturing system design process
Simulation makes you think about a stochastic
world, which is reality
Too often, simulation is done only when a
problem is discovered with a system that is
already installed and hard to change
In this and other ways, simulation is like variation
analysis
Simulation
11/20/2002
27
Economic Analysis of Assembly
Systems
Goals of this class
understand the basics of economic analysis

unit cost of assembly by different resources
return on investment
particular properties of assembly systems
Econ Analysis
11/18/2004
Daniel E Whitney
Cost and Price Considerations
Development cost
Unit cost
Unit mfr cost:

materials
labor
depreciation
waste, scrap, rework
Production
ramp-up
Marketing
Ongoing support
Supply curve
Prodn volume/yr
Cost
Cost
Loss
Profit
Prodn volume
Sales volume
3
2
Price
Price
Demand curve
Sales volume/yr
Econ Analysis
11/18/2004
Daniel E Whitney
Price Has Nothing to Do with Cost
Price is about value

Value is often perceived and can be influenced
Direct value involves functions and ilities

Perceived value involves or is influenced by
Marketing
Perceived quality
Number of choices available even if most will not be
taken
Value > Price > Cost, otherwise no sales or no

profit
Econ Analysis
11/18/2004
Daniel E Whitney
Cost Analysis is a Murky Area
Engineers need to know the basics of cost analysis

for three reasons
so they can make sound technological choices
so they can judge the suitability of a suppliers bid
so they can argue effectively with accountants
Dont ask us how we do investment justification.

We just fill out a form and after a while an answer
comes back Yes or No.
MAPI means makes a project impossible
MAPI = Manufacturing and Allied Processes Institute
Econ Analysis
11/18/2004
Daniel E Whitney
Kinds of Cost Categories
Fixed cost = what you pay to set up (usually investment in

facilities)
Variable cost = what you pay that depends on how many you
make per unit time

Labor, both direct and indirect (maintenance, supervisors)

Materials cost: what you buy that you add value to
Expendables: energy, lubricants, tool bits, etc
Scrap, rework
Institutional cost = all other costs of doing business
In many cases, labor is a fixed cost, due to contracts or the

inability to lay people off for short periods when business
fluctuates
Econ Analysis
11/18/2004
Daniel E Whitney
Cost Distribution in Engine Plants
Source:
Econ Analysis
11/18/2004
Daniel E Whitney
Sources of Cost in the Supply Chain
Source:
Source: Daimler Chrysler via Munro and Associates

Econ Analysis
11/18/2004
Daniel E Whitney
A Small Problem
Fixed costs are usually expended all at once, usually

before production starts
Variable costs are incurred as production runs
How should these two kinds of costs be combined to

provide a true picture of the cost per unit?
The usual method is to allocate the fixed costs to the
units by choosing a time period during which the
investment is recovered
unit cost = variable cost
+ Some_Fct (fixed cost, # of units made in some time period)
Econ Analysis
11/18/2004
Daniel E Whitney
Cash Flows Over Time
Source:
Econ Analysis
11/18/2004
Daniel E Whitney
Payback Period Method
A payback period P is selected (arbitrarily?)
The fixed cost is allocated equally to each unit

made during P:
+ fixed cost / (P Q)
where Q = quantity made per year
P = a number of years
Econ Analysis
11/18/2004
Daniel E Whitney
10
Internal Rate of Return Method
The payback period is replaced by an investment

horizon H and an interest rate r
This is equivalent to a mortgage for H years at
interest rate r
The annual payment A and the annual cost factor
fAC for an initial investment Io are (for zero
salvage value):
A = I0
Econ Analysis
r 1 + rH
1 + rH 1
11/18/2004
r 1 + rH
fAC = A =
I0
1 + rH 1
Daniel E Whitney
11
Unit Cost Based on IRoR
+ fAC *fixed cost /Q

where Q = quantity made per year
fAC = fraction of fixed cost paid per year,
based on:
r = IRoR (ranges from 15% to 35%)
H = investment horizon (ranges from 2 to 5
years or more)
Econ Analysis
11/18/2004
Daniel E Whitney
12
Annualized Cost Factor vs r
Source:
Econ Analysis
11/18/2004
Daniel E Whitney
13
Simplified Unit Cost for Manual Assembly
COSTUNIT MANUAL =
# People =
A$
# People
Q
TNQ
2000 * 3600
[largest integer]
Q = annual production volume
T = assembly time per part, sec
N = number of parts per unit
A$ = annual cost of a person
A$ = L H * 2000
L H = labor cos t, $ / hr
2000 = hours per shift year
3600 = sec / hr
(assumes no investment required)

Econ Analysis
11/18/2004
Daniel E Whitney
14
Simplified Unit Cost for Fixed Automation
f AC N S $
C UNITFIXED =
Q
where Q = annual production volume, units / year
f AC=fraction of machine cost paid for per year
S$ = cost of one station in the machine
(assumes one station per part)
(also assumes no people required)
Econ Analysis
11/18/2004
Daniel E Whitney
15
f ACI L$
+
Q
Q
where I = total investment in machines and tools

L$ = annual cost of workers associated with the system
I = # MACHINES * $ / MACHINE + # TOOLS * $ / TOOL
TNQ
2000 * 3600
# MACHINES =
# TOOLS = N
L$ = w L H # MACHINES * 2000
where w = number of workers / station
Combining the above yields:
C UNIT FLEX =
fAC
L$
# MACHINES * $ /MACHINE + # TOOLS * $ / TOOL +
Q
Q
f $/ MACH IN E T N
f AC $/ TOOL N w T N L H
+
C UNITFLEX - A C
+
3600
Q
2000 * 3600
Econ Analysis
11/18/2004
Daniel E Whitney
Simplified Unit Cost for Flexible
Automation
C UNITFLEX =
16
Conclusions from Unit Cost Models*
Cost is linearly proportional to number of parts N

one reason for fixation on part count reduction
Cost of flexible automation grows with the pricetime product: $/machine * T

shows that cost and time can be traded
Other costs grow as part, station, and tool count

grow
floor space
support staff
line downtime (see System Design chapter)
*P. M. Lynch, Economic-Technological Modeling and Design Criteria for

Programmable Assembly Machines, MIT ME Dept PhD Thesis, June 1976
Econ Analysis
11/18/2004
Daniel E Whitney
17
Unit Cost Example
Unit Assembly Cost by Three Methods

2
f AC=0.38
1.8
T=5s
1.6
L H=$15/hr
1.4
S$=50000
$/tool = $10000
Unit Cost
1.2
N = 10 parts/unit
1
w = 0.25 workers/sta
0.8
MANUAL $/UNIT
0.6
FIXED $/UNIT
FLEX $/UNIT
0.4
0.2
0
0
100000
200000
300000
400000
500000
600000
Annual Volume
Econ Analysis
11/18/2004
Daniel E Whitney
18
Unit Cost Example - 2
Source:
Econ Analysis
11/18/2004
Daniel E Whitney
19
More Detailed Cost Model
Source:
Econ Analysis
11/18/2004
Daniel E Whitney
20
Caveats About Examples
If T = 2 s, then Q = 3.6 million, or else the line

runs only part of one shift
If # people > # of parts or operations, then extra
people are needed for one shift operation
If Q > 7.2 Million / T, then a 2nd or 3rd shift is
needed
Econ Analysis
11/18/2004
Daniel E Whitney
21
Discounted Cash Flow Analysis
AKA net present value calculation
More detailed and sophisticated than unit cost

comparisons
Seeks to determine if an investment is good
Based on comparing return on investment

a base case is compared to an alternate
the alternate requires upfront investment
it creates a saving stream over time, which is
discounted to present value
do the savings justify the investment?
Econ Analysis
11/18/2004
Daniel E Whitney
22
Discounting Future Cash Flows
Money is a two-dimensional quantity ($,t)
$
Its value now
1
(1 + DF) i
e- t
= DF = discount factor
Its value if delayed

until t
t
Econ Analysis
11/18/2004
Daniel E Whitney
Time
23
Two Cash Flow Formulas
Takeaway:
Takeaway:The
Theearly
earlycash
cashflows
flowscontribute
contributethe
themost.
most.
10
Sum of future
cash flows
without
discounting
9
8
7
present value of
sum of future
cash flows
6
5
CF/ ( 1+DF)^ i
( CF/ ( 1+DF) ^ i)
e^ -DF t
( 1-e^ -DF t ) / DF
3
2
DF = 0.1
present value of
future cash flows
0
1
Econ Analysis
7
11/18/2004
11
13
15
Daniel E Whitney
17
19
21
24
Comparison Analysis
Base case
Alternate case
fixed costs
labor costs
material costs
fixed costs
labor costs
material costs
Comparison:
What discount rate makes the discounted sum of future
savings in labor and material costs greater than or equal
to the difference in fixed cost between base and alternate?
H
Net savingsi / 1 + DF .
Investment alt Investment base = i
=1
i
Alternatively: set discount rate = cost of borrowing

Choose the alternate investment if NPV > 0
Econ Analysis
11/18/2004
Daniel E Whitney
25
Discounted Cash Flow (DCF) and
Economic Value Added (EVA)
EVA is very similar to DCF. The discount rate used in
EVA is the weighted average cost of capital (WACC)
Cost of capital includes interest rate on debt plus expected rate of

return on stock (not easy to compute)
EVA is usually used to value the whole company but is

being used more and more to value individual investments
See http://www.pitt.edu/~roztocki/abc/abc.htm
See Econ DEMO-Stanley Hammer.xls on MIT Server.
Econ Analysis
11/18/2004
Daniel E Whitney
26
Zero or Net Present Value Calculations
Comparing two investments, the savings Sv are

considered income
You pay taxes on the income at tax rate Tx, yielding
your net income Ni
You can claim depreciation Dp on your investment,
decreasing your taxable income and lowering your
taxes
The IRS specifies how much you can claim in
depreciation each year
the net income is: Ni = Sv - Tx(Sv - Dp)
present value analysis spreadsheet on MIT Server

finds the discount rate that gives NPV = 0
Econ
Can
be used
to find Daniel
NPV
for any discount rate
27
11/18/2004
Analysis
E Whitney
Zero Present Value Analysis
ZERO PRESENT WORTH CASH FLOW ANALYSIS

7 YEARS ECONOMIC LIFE
0% SALVAGE VALUE % OF COST AT END OF ECONOMIC LIFE
EXPENSE FORECAST
YEAR
0
1
2
3
4
5
6
7
8
RATIO
100.00%
INCOME FORECAST
TAX RATE
34.00%
DEPRECIABLE
66.67%
SAVINGS
DEPRECIATION
$100
$181
$198
$150
TOTAL INVESTMENT
DEPRECIABLE INVESTMENT
INTERNAL RATE OF RETURN
TAX RATE
CREDIT
14.29%
24.49%
17.49%
12.49%
8.92%
8.92%
8.92%
4.46%
34.00%
34.00%
34.00%
34.00%
34.00%
SUM OF UNUSED YRS
DEPR=
31.22%
USED FOR SALVAGE VALUE
OF REMAINING DEPRECIABLE INVESTMENT
$400
$267
18.41%
GOAL SEEK
ON CELL G38 = 0
TAX CREDIT IN YR 0 0N
UNDEPRECIATED INVESTMENT
RESULT OF
PRO FORMA CASH FLOW

YEAR
INCOME
0
($400)
1
$100
2
$181
3
$198
4
$150
4
$83
SALVAGE VALUE
IN YEAR 4
DEPRECIATION
TAXES
$38
$65
$47
$33
$0
($45)
$21
$39
$51
$40
$0
CREDITS
$0
$0
$0
$0
$0
$0
NET
($355)
$79
$142
$147
$110
$83
DISC NET
($355)
$67
$101
$88
$56
$42
$355
GROSS INCOME
$713
$183
$152
$0
$561
NET INCOME
$313
$183
$106
$0
$206
Econ Analysis
11/18/2004
Daniel E Whitney
SUM OF UNDISC INC

$79
$221
$368
$478
$561
UNDISC PAYBA
FOR EX IN FIG
($0)
28
How to Use this Spreadsheet
Enter savings, tax rate, depreciation rate

Goal seek to get zero NPV
Or
Put in various discount rates and observe NPV
NPV > 0 is desired
Econ Analysis
11/18/2004
Daniel E Whitney
29
Aircraft Development Cost Quandry
Aircraft Product Development Cost Recovery

PD Investment = $3.5B
all taken in year 0
Depreciated over 7 years MACRS
$1.2B tax credit in year 0
First plane sold in year 2
2500
2000
1500
IRoR = 7.11%
Cost Recovery
1000
Undiscounted Cumulative
Income After Taxes and
Depreciation
500
0
-500
9 10 11 12 13 14 15 16 17
Discounted Cumulative Income

After Taxes and Depreciation
Annual Gross Income Before
Taxes and Depreciation
-1000
-1500
-2000
-2500
-3000
Year
Econ Analysis
11/18/2004
Daniel E Whitney
aircraft pd investment analysis.xls
30
A380 Business Case
A380
Cost and Income
PD Development cost $10.5B
14000
660 planes sold, price $200M ea
10000
$Millions
8000
6000
4000
First planes sold
12000
Avg profit/plane = $50M
IROR = 11.5%
Net Income Before

Depreciation and Taxes
Net Income After Taxes
and Depreciation
Sum of Undiscounted Net
Cash Flows
2000
0
-2000 0
10
20
30
Sum of Discounted Net

Cash Flows
-4000
-6000
Years
Econ Analysis
11/18/2004
Daniel E Whitney
31
NPV vs Discount Rate for A380
NPV VS DISC RATE

FOR A380
5000
4000
NPV, $M
3000
2000
NPV
1000
0
-1000 0
10
15
20
25
-2000
DISC RATE, %
Econ Analysis
11/18/2004
Daniel E Whitney
32
Critiques of DCF
Target IROR is arbitrary

The calculations can be gamed
Cost is a slippery quantity

People know their expenditures and assume that they
know their costs, but these are different even if they add
up to the same amount
Overheads are allocated arbitrarily and can distort the
calculations
Activity-based costing is intended to overcome this
Robert Kaplan is an EE!
Econ Analysis
11/18/2004
Daniel E Whitney
33
Summary of Economic-Technical
Analysis
Source:
Econ Analysis
11/18/2004
Daniel E Whitney
34
Is The Make/Buy Decision a Core
Competence?
Daniel E Whitney
MIT
Cambridge MA
in collaboration with Prof Charles Fine, MIT
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Outline
I. Necessity of outsourcing, importance of

supply webs
II. Generic nature of the outsourcing process
III. Formalization of this process

foundation in system engineering
two kinds of dependency
definition of outsourcability
IV. Examples
V. Conclusions
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Vertical Integration Used to be Good
Now, outsourcing is good

Outsourcing creates dependencies
Knowledge has to be managed up and down the

supply chain
Most companies do not understand the
requirements for smart outsourcing and do not
appreciate the risks
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Drivers of Make-Buy
Cost
others...
Are there risks?
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Dependency and Outsourcing
Modern products are so complex that no

company has all the skills needed
Companies are dependent on each other for
many crucial things - and some are even
proud of it!
Make/buy decisions are made for both
product components and for
manufacturing infrastructure: machines,
CAD, MRP
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Four Stories
Manufacturing managers unable to design a toy

assembly line
Switching suppliers for a washer
HWP discovers what laid-off managers were
doing (Ed Anderson and Geoff Parker: From Buyer to Integrator: The Transformation of
the Supply-chain Manager in the Vertically Disintegrating Firm)
Shifting jobs to Mexico generates 1000%

overhead
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Can We Make These Cars?
Customer Demand / Per Shift:
One Shift = 6 min.
Photos removed due to copyright restrictions.
Sedan
6
Make/Buy for PDD
Semi
2
11/30/2004
Daniel E Whitney
Race car
2
Truck
2
7
Theory of the Firm
What We Dont Do
Transaction
Costs
What We Do
Coordination Costs
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Two Views of the Make-Buy Dilemma
You
Youlearn
learnby
bytrying
trying
not
notby
bybuying.
buying.
Our
Ourbusiness
businessisiscars,
cars,
not
notrobots.
robots.
Product/process
integration
First access to
new technology
Keep new methods
from leaking out
Mastery itself
is the prize
The US has a robust

vendor base
There is economy
of scale
Can capitalize on
vendors learning
Focus on the core
business
It is not obvious who is right.

Make/Buy for PDD
11/30/2004
Daniel E Whitney
Outsourcing and Architecture
Product architecture and outsourcing interact
interfaces must be defined and managed
Architecture of supply chain must correspond to

architecture of product (C. H. Fine)
Many criteria influence choices of architecture
which in turn influence outsourcing options and
problems

System integration
Design and manufacturing methods
Power management
Politics
Knowledge management
Make/Buy for PDD
11/30/2004
Daniel E Whitney
10
Integral and Modular Architectures
Integral = functions shared

by physical elements
Examples...
Some reasons integral is
used...
Modular = each function is

delivered by a separate element
Examples...
Some reasons modular is used...
Is either kind of architecture better than the other?
Or are all architectures more or less integral?
Make/Buy for PDD
11/30/2004
Daniel E Whitney
11
Three Kinds of Modules
Module by design (could be a coherent system)
Module by manufacture (could be a connected set of parts)
Module by use (could contain user functions)
(Baldwin and Clark, Design Rules)
Front module contains parts of several systems

Make/Buy for PDD
11/30/2004
Daniel E Whitney
12
Product and Organizational Architectures
Rear wheel drive IC engine car

Engine dept, transmission dept, brake dept, ABS dept, packaging dept
Electric car with one motor and transmission

Motor dept, transmission dept, brake dept, ABS dept, packaging dept
Electric car with motor at each wheel

Motor(s) dept, no brake dept, ABS dept merged with computer dept
Dont even tell them about that, theyll throw you out
Make/Buy for PDD
11/30/2004
Daniel E Whitney
13
Airbus: Architecture Driven by Politics
Airbus Industrie is a consortium that shares revenue
and profits according to a work-content formula
This formula is based on a decomposition of the plane
wings to British Aerospace

fuselages to Deutsche Aerospace (DASA)
tail sections to CASA (Spain) (now owned by DASA)
final assembly and integration to Aerospatiale
The A380 is a challenge

The wings will be too big to transport to Toulouse by air
No land in Hamburg for final interior dressing
Final decision: truck to river, onto barge, to Bordeaux, truck again to Toulouse
Make/Buy for PDD
11/30/2004
Daniel E Whitney
14
An Emerging Trend
In many industries, first tier suppliers are starting

to consolidate (tier 0.5), horizontally integrating
and technically deepening
The source of their growing power is their ability
to create coherent systems and exploit synergies
Brakes and chassis control systems in cars
Industrial gas, liquid, and power systems in chip factories
In some cases, these systems reach up to the customer directly
OEMs focus more on customer needs and think

tier 1 and below are just manufacturers or
suppliers
Make/Buy for PDD
11/30/2004
Daniel E Whitney
18
The Car Is a Collection of High-
Technology Systems
Customer satisfaction is provided by these systems
MANAGEMENT
SAFETY
ENERGY
OPERATIONS
POWER TRAIN
ENGINE
TRANSM
FUEL
EMISSIONS
CHASSIS
ABS
SUSPENSION
STEERING
Make/Buy for PDD
AMBIENCE
ENTERTAINMENT
COMFORT
A car is a self-mobile
cordless computer
network with peripherals
like engine block, seats,
glass, etc.
11/30/2004
Daniel E Whitney
COMMUNICATION
AND CONTROL
INTERNAL
EXTERNAL
NAVIGATION
GPS
COMPASS
WEATHER
ROAD COND
19
These Systems are Highly Interlinked
NOTE: THERE ARE MANY LINKS INSIDE EACH BOX!

MANAGEMENT
SAFETY
ENERGY
OPERATIONS
POWER TRAIN
ENGINE
TRANSM
FUEL
EMISSIONS
AMBIENCE
ENTERTAINMENT
COMFORT
CHASSIS
ABS
SUSPENSION
STEERING
Make/Buy for PDD
11/30/2004
Daniel E Whitney
COMMUNICATION
AND CONTROL
EXTERNAL
INTERNAL
NAVIGATION
GPS
COMPASS
WEATHER
ROAD COND
20
Different Architecture Brings Different
Outsourcing Options and Restrictions
Generic function:
deliver air+fuel to combustion chamber
meter them together
meter them separately

fuel
injectors
air
integrate TB
with intake manifold
one air meter

each cyl
has its own for each cylinder
inj, air, &
nothing like
spark!
it today
separate TB
from intake
manifold
one air meter
for whole manifold
novel molding
technology
todays TB
carburetor
Increasing innovation
Make/Buy for PDD
11/30/2004
Daniel E Whitney
21
Integration Driven by Power
Whitneys theory says high power drives
integration*
Moores Law drives rise in heat generated in
CPUs
Cost of managing heat has reached economic limit
Evidence that it is happening (Intel):

Patents on fans
Investments in software and heat transfer solutions
Close cooperation with PC designers
Major shift in marketing strategy to de-emphasize processor speed
Ref: SDM Thesis by Sam Weinstein, March 2004
*Why Mechanical Design Cannot be Like VLSI Design, Research in Engineering Design,
1996 (8), pp 125-138
Make/Buy for PDD
11/30/2004
Daniel E Whitney
22
Moores Law
Make/Buy for PDD
11/30/2004
Daniel E Whitney
23
Cost of Thermal Management
>$100
< $10
Make/Buy for PDD
11/30/2004
Daniel E Whitney
24
Vertical Industry Structure Driven by
Technology
with Integral Product Architecture
Computer Industry Example, 1975-85

IBM
DEC
BUNCH
All Products
All Products
All Products
Microprocessors
Operating Systems
Peripherals
Applications Software
Network Services
Assembled Hardware
Make/Buy for PDD
(A. Grove, Intel; and Farrell, Hunter & Saloner, Stanford)
11/30/2004
Daniel E Whitney
26
Horizontal Industry Structure Reshaped by
Technology and Standards
with Modular Product Architecture
Computer Industry Example, 1985-95

Microprocessors
Intel
Intel
Operating Systems
Mac
Mac TITIetcetc
Microsoft
Mac Unix
Peripherals
Mac TIetcetcetc
HP IntelCanon Samsung
Applications Software
Microsoft
Lotus Borland
Network Services
Novell
Lotus
Assembled Hardware
Make/Buy for PDD
HP Compaq IBM
EDS etc
etc
etc
Toshiba etc
(A. Grove, Intel; and Farrell, Hunter & Saloner, Stanford)

11/30/2004
Daniel E Whitney
etc
27
The Double Helix:
The Dynamics of Product Architecture and
Industry Structure
NICHE
COMPETITORS
INTEGRAL PRODUCT
VERTICAL INDUSTRY
MODULAR PRODUCT
HORIZONTAL INDUSTRY
TECHNICAL
ADVANCES
HIGHDIMENSIONAL
COMPLEXITY
ORGANIZATIONAL
RIGIDITIES
SUPPLIER
MARKET
POWER
PRESSURE TO
DIS-INTEGRATE
PRESSURE TO
INTEGRATE
PROPRIETARY
SYSTEM
PROFITABILITY
Christensen "The Drivers of Vertical Disintegration" HBS working paper, Oct 1994
Fine & Whitney, "Is the Make/Buy Decision a Core Competence?" MIT Working Paper, March 1995
Make/Buy for PDD
11/30/2004
Daniel E Whitney
28
Recent Examples
Citigroup
Delphi
Cisco, Microsoft
(gathering,
(buy companies with
integrating
new technologies)
organically)
More Integrated
Intel
(thermal management
integrating supply
chain)
Tending to
Dis-integrate
Make/Buy for PDD
11/30/2004
Ford
(CAE)
GM
(spun off Delphi)
Cisco, Sun
(outsource
mfr of PCBs)
Intel
(processors
that generate
More Dis-integrated
lots of heat)
Tending to
Integrate
Daniel E Whitney
Ford
(interiors)
29
Summary
Outsourcing relates to
knowledge and learning

dependency and power
organization
system coherence
architecture
Products, organizations, supply chains, and

industries have architectures
Architecture is volatile
Make/Buy for PDD
11/30/2004
Daniel E Whitney
30
Make-Buy Frameworks
Link make-buy to system engineering

Link make-buy to product architecture
Classify dependencies that make-buy could create
Combine these ideas into a framework for

assessing make-buy decisions
Place some of our examples in the framework
Sum things up
Make/Buy for PDD
11/30/2004
Daniel E Whitney
31
The Product Development Process
DETERMINE
CUSTOMER
NEEDS
CONVERT NEEDS
TO ENGINEERING
SPECS
At each stage:
"BLACK BOX"
OPTION
DETERMINE
CUSTOMER
NEEDS
CONVERT NEEDS
TO ENGINEERING
SPECS
CONVERT ENG
SPECS TO
PROCESS
SPECS
CONVERT ENG
SPECS TO
PROCESS SPECS
CONVERT
PROCESS
SPEC
TO PROCESS
MAKE
ITEM
CONVERT
PROCESS
SPEC
TO PROCESS
VERIFY THAT
ITEM MEET
SPECS
MAKE
ITEM
VERIFY THAT
ITEM MEET
SPECS
Make/Buy for PDD
11/30/2004
"WHITE BOX"
OPTION
Break down into sub-systems
Write specifications
Seek sources and choose
Transmit spec
Obtain item
Verify compliance with spec
Both customer and source
could be internal or external
Either way, the process and

skills are the same
Daniel E Whitney
32
Principles of System Engineering
CLUSTERED
SUBSYSTEM
+
+
DON'T
DECOMPOSE
HERE
CLUSTERED
SUBSYSTEM
+ +
DECOMPOSE
HERE
Make subsystems from things that are tightly related
Keep relations between subsystems simple, few,

and easy to define
Alexander, C. Notes on the Synthesis of Form, Harvard University Press, 1964
Rechtin, E., Systems Architecting, Prentice-Hall, 1991
Make/Buy for PDD
11/30/2004
Daniel E Whitney
33
Theory of the Firms Relation to
Modularity
What We Dont Do
What We Do
Classic module All complexity inside
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Transaction
Costs - managed
by keeping interfaces
simple and clear
via interface docs
called contracts
34
Design = Outsourcing = System Engineering
Design
Outsourcing
System Engineering
Determine
requirement
Determine need
Determine top
requirement
Break into
sub-requirements
Write spec
Break into
subreqmts
Find someone
to fill reqmt
Find someone
to fulfill spec
Assign subreq
Assess fulfillment
Assess fulfillment
Assess fulfillment
Make/Buy for PDD
11/30/2004
Daniel E Whitney
35
Make-Buy Criteria
1. Buy what you cant make (necessity)

2. Buy what someone else makes cheaper (low bid)
3. Buy what someone else makes better (opportunity)
4. Make what matters most to the customer (visibility)
5. Make what matters most to the technical memory

of the company (strategy)
Condense to two choices:
Be dependent for capacity (you could make it)
Be dependent for knowledge (you cant make it)
Make/Buy for PDD
11/30/2004
Daniel E Whitney
36
Examples of Dependency
Disk drives: dependent for knowledge

Half shafts: dependent for capacity
Make/Buy for PDD
11/30/2004
Daniel E Whitney
37
Disk Drive Engineering and Outsourcing
SPINDLE
ARM
PLATTER
BEARING
HEAD
TOLERANCE
CHAIN
METRICS:
Each major element

was obtained from
a different source,
including assembly.
The company is no
longer in this
business.
VOLUME
WEIGHT
$/MEGABYTE
Make/Buy for PDD
11/30/2004
Daniel E Whitney
38
Toyota as an Example
Toyota appears to buy most car parts and make
much of its infrastructure

It could have made the choice many other ways
Toyota appears to have performed triage on the
things it buys and on the suppliers themselves
black box (supplier does it all)

grey box (co-design)
white box (build to print)
cf Taka Fujimoto: Origin and Evolution of Black Box Practice
in the Japanese Auto Industry
How does Toyota retain skills when it outsources?

Make/Buy for PDD
11/30/2004
Daniel E Whitney
39
Car NVH Illustrates KC Distribution Through
a Chain of Elements
A Toyota specification for an outsourced drive shaft

contains a vibration specification
In fact, the drive shaft is one of many elements in
a complex system that delivers the noise,
vibration, and harshness KC
ENGINE
ENGINE
MOUNT
SUSPENSION
TRANSMISSION
BRAKES
HALF
SHAFT
WHEELS
and TIRES
NVH Chain
Make/Buy for PDD
11/30/2004
Daniel E Whitney
41
Half Shaft with Constant Velocity Joints
Make/Buy for PDD
11/30/2004
Daniel E Whitney
42
Is System Engineering Toyotas Main Skill?
Toyota has chosen to outsource many components
But it keeps control of many key manufacturing and
design technologies; for example:
- machines for dies and engine parts
- CAD technologies, especially info management

Are these choices related?
Is Toyotas dependence on Denso a problem?
Make/Buy for PDD
11/30/2004
Daniel E Whitney
43
Coherent Systems
Examples
Any self-contained, testable module that delivers a measurable function,

plus the knowledge to design, make, and test it
A software suite such as Microsoft Office
A patent network
A product plus its customer support system
What they offer to the owner

Control (over performance, cost, changes and improvements, diagnosis,
underlying product or process knowledge)
Customer reliance on the whole and inability of customer to deconstruct
Efficiency of delivery
Barrier to entry by competitors
A form of complementary asset (required to exploit the main assets)
The risks in breaking them

Loss of control, knowledge, synergy, barrier, future exploitation
Make/Buy for PDD
11/30/2004
Daniel E Whitney
44
MORE IN-HOUSE KNOWLEDGE NEEDED

IT IS INTEGRATION KNOWLEDGE
ITEM IS LESS OUTSOURCABLE

BECAUSE IT IS LESS DECOMPOSABLE
Degrees of Outsourcability
GENUINE COMMODITY WITH STANDARD INTERFACES

(ANSI, NEMA, ETC) BOUGHT FROM CATALOG
DECOMPOSABLE ITEM WITH A FEW WELL-DEFINED
INTERFACES
ITEM CONTAINS BUYER'S PROPRIETARY TECHNOLOGY
ITEM MUST BE INTEGRATED WITH ITEMS MADE IN-HOUSE
OR BOUGHT ELSEWHERE
ITEM IS THE "CORE" WITH WHICH MANY OTHERS MUST
BE INTEGRATED
ITEM CONTAINS MANY MUTUALLY INTEGRATED KEY
CHARACTERISTICS WHOSE DEFINITION AND
MANUFACTURE REQUIRE IN-HOUSE DECISIONS,
TRADEOFFS, AND KNOWLEDGE
Make/Buy for PDD
11/30/2004
Daniel E Whitney
45
Degrees of Dependency
CAN IDENTIFY QUALIFIED
BIDDERS
DEPENDENT FOR
KNOWLEDGE
CAN WRITE COMPETENT
SPECIFICATION
IF YOU GET OFF THE BUS,

CAN YOU EVER GET BACK
ON??
CAN VERIFY THAT ITEM MEETS
SPEC
CAN IMPROVE BID

CAN HELP SUPPLIER
TECHNICALLY
DECREASING
KNOWLEDGE
DEPENDENCY
CAN EVALUATE BIDS
CAN HELP SUPPLIER
OPERATIONALLY
CAN IMPROVE ITEM AFTER

RECEIPT
DEPENDENT FOR
CAPACITY
CAN MAKE IN-HOUSE
WHAT'S THE MINIMUM

THAT'S ECONOMICAL
TO RETAIN??
Make/Buy for PDD
11/30/2004
Daniel E Whitney
46
Questions to Ask Before Outsourcing
How well do we understand our own product?
How sensitive is our products performance to details of

the outsourced items performance?
Who is the technological leader? Us? Them?
Could important technological advances occur in the

outsourced item? Who will own them?
Who will maintain the core competence if we outsource?
Are we creating a future holdup situation?

What are the criteria for a competent supplier?
Can the supplier deconstruct our coherent system and
reconstruct it in his domain?
Make/Buy for PDD
11/30/2004
Daniel E Whitney
47
Matrix of Dependency and Outsourcing
OUTSOURCED
ITEM IS
INTEGRAL
OUTSOURCED
ITEM IS
DECOMPOSABLE
DEPENDENT FOR
KNOWLEDGE
Make/Buy for PDD
A POTENTIAL
OUTSOURCING
TRAP
YOUR PARTNERS
COULD SUPPLANT
YOU. THEY HAVE AS
MUCH OR MORE
KNOWLEDGE AND CAN
OBTAIN THE SAME
ELEMENTS YOU CAN.
DEPENDENT FOR
CAPACITY
BEST OUTSOURCING
OPPORTUNITY
YOU UNDERSTAND IT, YOU CAN
PLUG IT INTO YOUR PROCESS OR
PRODUCT, AND IT PROBABLY CAN
BE OBTAINED FROM SEVERAL
SOURCES. IT PROBABLY DOES
NOT REPRESENT COMPETITIVE
ADVANTAGE IN AND OF ITSELF.
BUYING IT MEANS YOU SAVE
ATTENTION TO PUT INTO AREAS
WHERE YOU HAVE COMPETITIVE
ADVANTAGE, SUCH AS
INTEGRATING OTHER THINGS
WORST OUTSOURCING
SITUATION
CAN LIVE
WITH
OUTSOURCING
YOU DON'T
UNDERSTAND WHAT
YOU ARE BUYING OR HOW
TO INTEGRATE IT.
THE RESULT COULD
BE FAILURE SINCE YOU
WILL SPEND SO MUCH
TIME ON REWORK OR
RETHINKING.
YOU KNOW HOW TO

INTEGRATE THE ITEM SO YOU
MAY RETAIN COMPETITIVE
ADVANTAGE EVEN IF OTHERS
HAVE ACCESS TO THE
SAME ITEM.
11/30/2004
Daniel E Whitney
48
Outsourcing Pro and Con
Pro
Con
Provides competitive
alternatives
Allows contact with different
sources and kinds of
knowledge
Augments in-house capabilities
Augments in-house capacity
Reminds everyone that there is
no monopoly on skill or
knowledge
Make/Buy for PDD
11/30/2004
Daniel E Whitney
Decentralizes things that need central

oversight
Breaks chains of delivery of quality
Disperses responsibility and
accountability
Opens the door to conflicts of interest
Devolves power, especially power
based on coherence in performance,
processes, or knowledge
Fosters the illusion that everything is
plug and play and that cost is the only
differentiator
Fosters the illusion that risk can be
eliminated while reward can be
retained
49
Conclusions
Outsourcing seems necessary, but is it "good"?
System coherence and integration potential is

a possible guide on what to make vs buy
The choice of what to make and what to buy has long term
strategic implications especially where system control
conveys market power
The main long term core competency may in fact be the
ability to make the make/buy decision
No such decision is permanent
Make/Buy for PDD
11/30/2004
Daniel E Whitney
50
Additional Reading
Have a look at my web page:
http://web.mit.edu/ctpid/www/Whitney/papers.html
on which you will find, among other things, a paper
by Charles Fine and myself called
Is the Make-buy Decision a Core Competence?
Make/Buy for PDD
11/30/2004
Daniel E Whitney
51
The HP Pavilion:
Where Does HP Add Value?*
R&D-Intel, Microsoft, Component Mfrs (HP)
Manufacturing-SCI Systems, Intel

Sales-CompUSA, Best Buy, etc
Service-3rd party maintenance providers
HP adds value in information management,
bringing it all together, knowing the sweet spots
in the components, understanding the market.
*Managing Against the Clock: Lessons from the IT Industry, by Prof Haim
Mendelson, Stanford U Business School, presented at the MIT Symposium Creating
and Managing Corporate Technology Supply Chains, May 12, 1998
Make/Buy for PDD
11/30/2004
Daniel E Whitney
52
Front and Rear Wheel Drive Architectures
fwd & rwd
Make/Buy for PDD
11/30/2004
Daniel E Whitney
57
Outsourcing Driven by System
Integration
Car industry suppliers gaining power by

assembling coherent systems
interiors (seats, entertainment, instrument panel)
chassis (ABS, active suspension, steering, engine controls
see Wards articles on Bosch-AlliedSignal and VarityKelsey-Hayes
Delphi will offer complete drive/brake by wire
Boeing will offer similar products

Ford is taking back interiors integration
Make/Buy for PDD
11/30/2004
Daniel E Whitney
58
Case Study of 767 Horizontal Stabilizer
Goals of this class

Carry through the topics of this course on one product
Look in detail at a real aircraft structural assembly
Define and flow down KCs
Compare different assembly methods
conventional one based on fixtures
proposed one based on part-to-part mating features
Draw datum flow chains for them
Study the economics
767 Case Study
11/30/2004
Daniel E Whitney
History of 767 Horizontal Stabilizer
Project
Fast/Flexible Manufacturing Project 1996

Coordinated Aircraft and Auto industry projects
Vought Aircraft partner via LAI

Voughts goal: cut costs, earn more Boeing
business
Voughts hypothesis: convert from fixed to
flexible assembly tooling
Voughts focus of project: 767 H. S. upper wing
subassembly
767 Case Study
11/30/2004
Daniel E Whitney
Our Challenge: How To Do This
Available data
Existing tooling
No history, people, drawings
Evidence of errors in tooling
Our process
Understand goals of existing process

Reverse engineer from the top down
Expand scope of study to complete horizontal stabilizer
Look up the supply chain to Boeing to get the
requirements
Generate new process to achieve agreed goals

767 Case Study
11/30/2004
Daniel E Whitney
Structure of Horizontal Stabilizer
767 Case Study
11/30/2004
Daniel E Whitney
Top Level Key Characteristics
GAP
GAP
FTB
PLUS CHORD
SKIN
FTE
SPAR
END
SPAR
END
PLUS CHORD
AERODYNAMICS
(gap betw skin and FTB & FTE)
(gap between skins)
STRENGTH
(based on joining plus chords
and ends of spars)
HORIZ ASSY 2
767 Case Study
11/30/2004
Daniel E Whitney
Horizontal Stabilizer Subassemblies
(current decomposition)
STRINGERS
UPPER SKIN
ASSEMBLY
SKIN
D
W
F
FWD SKIN
AFT SKIN
FORWARD
TORQUE
BOX
FOR
PLUS CHORD
D SP
R
A
W
AR
RIB
STRINGERS
RE
PA
S
R
FIXED
TRAILING
EDGE
STRINGERS
PLUS CHORD
HORIZ ASSY 2
LOWER SKIN
ASSEMBLY
LOWER SKIN
PLUS CHORD
767 Case Study
11/30/2004
Daniel E Whitney
PKCs for Horizontal Assembly
PKC #2 & #3: Aerodynamics affected

by these skin gaps
SPAR END FITTINGS
FWD TO
PKC #1:
Joint Strength
PLUS
affected by
CHORD
alignment between
plus chord ends
and spar end fittings.
RQUE B
OX
FORWARD SKIN
SPLICE STRINGER
AFT SKIN
FIXED TRAILING EDGE
SPAR END FITTINGS

JOINT STRENGTH PKC ACHIEVED
END FITTING
RD
HO
767 Case Study
11/30/2004
Daniel E Whitney
END FITTING
SPLICE
PLATE
SHIM
SPAR
RD
HO
S C NE
D
PLU ALIG
S
(M I
SC
PL U
SPLICE
PLATE
SPAR
SHIM NEEDED TO ACHIEVE

JOINT STRENGTH PKC
Current Total Process
PLUS CHORD
ALIGNMENT
TO SPAR
ENDS PKC
PLUS CHORD
ALIGNMENT
TO SPAR
ENDS PKC
SKIN GAPS PKC
FINAL ASSEMBLY
SKIN GAP PKC

DELIVERED
BY SKIN
GAP AKC
PLUS CHORD
ALIGNMENT TO
SPAR ENDS PKC
DELIVERED BY
P.C. ANGLE
TO SKINS AKC
FTB-RIB-FTE SUBASSEMBLY
SKIN SUBASSEMBLY
SKIN GAP PKC

DELIVERED BY
DESIGN OF
FIXTURE TO
HOLD FTB
AND FTE
FORWA
RD TOR
FORWA
Q UE B O
RD SPA
X
R
FORWARD SKIN
S
PLU
RIBS
RD
CHO
767 Case Study

Figures for designing assemblies
SPLICE STRINGER
REAR SPAR
FIXED TRAILING EDGE
AFT SKIN
11/30/2004
Daniel E Whitney
10
Top-Level KCs
FTB
Skin Gap
& rib align KCs
PKC #2
F Spar
PKC #1
Ribs
FS
PC
Skin Gap
& rib align KCs
AS
SS
PKC #1
PKC #3
A Spar
PKC #2
FTE
767 Case Study
11/30/2004
Daniel E Whitney
11
Product Decomposition Based on
Independent KCs
FTB
F Spar
Ribs
FS
PC
SS
AS
A Spar
FTE
767 Case Study
11/30/2004
Daniel E Whitney
12
FTB
Ribs
PC
A Spar
FTB
FTE
F Spar
Ribs
SS
FS
AS
PC
SS
A Spar
FTE
FTB
F Spar
PC
Ribs
A Spar
767 Case Study
11/30/2004
FTE
Daniel E Whitney
FS
AS
Decomposition/Subassemblies
Based on Independent KCs
F Spar
13
Sob
767 Case Study
11/30/2004
Daniel E Whitney
14
Assembly Access Problem Eliminates an
Attractive Assembly Sequence
UPPER PLUS
CHORD
UPPER
SKIN
STRINGER
~24"
SPAR
SHIM
PIVOT RIB
SHIM
~
STRINGER
LOWER PLUS
CHORD
767 Case Study
11/30/2004
If plus chords are assembled

to spar ends before skins
are assembled to plus chords,
then it will be almost
impossible to join skins and
stringers to plus chords
LOWER
SKIN
Daniel E Whitney
15
Actual Subassemblies
FTB
F Spar
Ribs
FS
PC
SS
AS
A Spar
FTE
767 Case Study
11/30/2004
Daniel E Whitney
16
FTB
Ribs
PC
SS
Actual Decomposition
F Spar
FS
AS
A Spar
FTE
FTB
F Spar
PC
Ribs
FS
AS
SS
A Spar
FTB
FTE
Ribs
F Spar
767 Case Study
A Spar
11/30/2004
Daniel E Whitney
FTE
17
Type 1 and Type 2 Methods for One
Subassembly - KC Flowdown Seems OK
FTB
FTB
F Spar
F Spar
Ribs
Ribs
A Spar
FTE
767 Case Study
11/30/2004
Daniel E Whitney
F
A Spar
FTE
18
Second Subassembly Has Lost Its KC
Links to Higher Level Assemblies
Any assembly process for
this subassembly must
provide proxies for the
missing KCs, regardless of

whether the subassembly is
made as a Type 1 or Type 2
These KCs will be coupled
Note that no drawing of this
subassembly could be found
767 Case Study
11/30/2004
Daniel E Whitney
PKC #2
PKC #1
PC
PKC #1
FS
PKC #3
SS
AS
PKC #2
19
Possible Assembly Strategies
767 Case Study
11/30/2004
Daniel E Whitney
20
Our Challenge
Current assembly method relies on costly fixtures
Can a process be devised that does not rely on

fixtures other than for support against gravity?
Can such a process achieve the PKCs?
Would it be economical?
What new worker skills would be needed?
Can we figure out what the old process was doing

so we can reproduce its objectives using new
methods?
767 Case Study
11/30/2004
Daniel E Whitney
23
Diagram of Assembly Analysis Process
ORIGINAL PKCs
RD
FORWA
ACHIEVE PKCs
SEPARATELY
(ASSEMBLY IMPOSSIBLE)
SKIN
NGER
STRI
ASSEMBLY POSSIBLEPKCs COUPLED
Skin Gaps
2
1
PLUS
CHORD
NG
STRI
Forward Skin
ER #3
SK
AFT
IN
CURRENT METHOD
USING FIXTURES
FTB
TWO ALTERNATE METHODS

REQUIRING DIFFERENT AKCs
Plus
Chord
Aft Skin
using edge features
using hole and slot features
FTE
Assembly
Decomposition
Product
Architecture
Joint Strength affected by this alignment.
AKCs
1
KC Flowdown
PKCs
6
Assembly Sequence
Generation
Prune into
Families
Identify most
promising family
Propose Assembly
Features
Analyze
Sequences
FORWARD SKIN
10
Equipment
Requirements
SPLICE STRINGER
Process Selection
and Planning
PKCs
PKC #1
PKC #2
AFT SKIN
PKC #3

PART LEVEL DATUMS
AKCs
VSA RESULTS
767 Case Study
11/30/2004
Daniel E Whitney
Assembly
Feature AKCs
AKC #1
Plus Chord
aft hole &
forward slot
Aft Skin
aft hole &
inboard slot
Fwd Skin
inboard slot
& fwd slot
Stringer #3
inboard
holes
Stringer #3
holes
24
Liaison Diagram
Str4-11
Fwd
Skin
Plus
Chord
Splice
Str3
Str1-2
767 Case Study
11/30/2004
Daniel E Whitney
Aft
Skin
25
Current Skin Assembly Process
SPLICE STRINGER
PLUS
CHORD
FWD SKIN
AFT SKIN
FIXTURE
Everything indexes off the fixture
767 Case Study
11/30/2004
Daniel E Whitney
26
Current Skin Assembly Process - 2
Track f or Rout er t o Tr im Forward Skin Edge
Plus
Chord
Forward Skin
Splice stringer
Aft Skin
Fixture
767 Case Study
11/30/2004
Daniel E Whitney
27
Assembly KC #1 & #2
#2
#1
767 Case Study
11/30/2004
Daniel E Whitney
28
Datum Flow Chain for Current Skin Process
Fwd
Skin
Str4-11
Skin Gap
KC
Plus
Chord
Splice
Str3
Plus Chord
Angle KC
EXPLICIT DATUM TRANSFER

CONTACT WITHOUT
DATUM TRANSFER
KC
Str1-2
Aft
Skin
Fixture
767 Case Study
11/30/2004
Daniel E Whitney
29
New Process #1: Fixtureless (Type 1)
FORWARD SKIN
HOLES CHOSEN
PLU
FOR LOCATION
SLOTS CHOSEN FOR
ORD
S CH
LOCATION, THERMAL &
SPLICE STRINGER
SHOT PEEN GROWTH

ACCOMMODATION
ALL SLOTS DRILLED
OUT TO FULL SIZE
AFT SKIN
HOLES AT FINAL
ASSEMBLY
BIGGEST THERMAL
DIFFERENCE
PART LEVEL DATUMS

767 Case Study
11/30/2004
Daniel E Whitney
SMALLEST
FULL SIZE
HOLE
30
PKC Delivery Map for New Process #1
PKCs
PKC #1
PKC #2
PKC #3
AKCs
AKC #1
AKC #2
AKC #2
Assy features
PC aft hole
& fwd hole
Mfr features
Size/shape
of PC and
hole locations
767 Case Study
11/30/2004
Aft skin
aft hole &
inbd hole
Fwd skin
inbd hole
& fwd slot
Size/shape
of skins
Daniel E Whitney
Str 3 inbd
holes
Hole/slot
locations on
skins & str 3
Skins & str 3

holes & slots
Coord.of
slot lengt h
and hole size
31
Datum Flow Chain for New Process #1
Str4-11
Fwd
Skin
(6)
rd
o
h C
C
K
us le A
l
P g
(1)
An
(6)
Plu
An s Ch
gl e or
AK d
(5)
C
Str1-2
(6)
Skin Gap AKC
Plus
Chord
Note: in existing process,

skin-plus chord is a contact.
In new process it is a mate.
Splice
Str3
(6)
Aft
Skin
EXPLICIT DATUM TRANSFER

CONTACT WITHOUT
DATUM TRANSFER
KC
767 Case Study
11/30/2004
Daniel E Whitney
32
New Process #2
SKIN
STRINGER
SHIM
PLUS
CHORD
PLU
SC
FORWARD SKIN
HO
RD
SPLICE STRINGER
AFT SKIN
SPACER TOOLS
LAYUP TABLE FEATURES
767 Case Study
11/30/2004
Daniel E Whitney
33
Assembly Features for Process #2
FORWARD SKIN
U
S
P
L
D
OR
CH
SPLICE
STRINGER
AFT SKIN
PART LEVEL DATUMS
MATING FEATURE (HOLE OR SLOT)
MATING FEATURE (EDGE)
767 Case Study
11/30/2004
Daniel E Whitney
34
Datum Flow Chain for New Process #2
Str4-11
Plus
Chord
P lu
An s Ch (6)
gle or
AK d
(4)
(5)
C
Str1-2
767 Case Study
11/30/2004
(6)
Daniel E Whitney
Skin Gap AKC
(6)
rd
o
h C
C
K
us le A
l
P g
(1) (1)
T (2)
An
Fwd
Skin
(6)
(5)
Splice
Str3
(6)
Aft
Skin
35
KC Deliverability Map - Process #2
PKCs
PKC #1
PKC #2
PKC #3
AKCs
AKC #1
AKC #2
AKC #2
PC aft hole
hole
Assy features
Tool location
at fwd end of
plus chord
Mfr features
767 Case Study
11/30/2004
Aft skin aft

Hole & inboard
surface
CNC
worktable
Daniel E Whitney
Fwd skin
inboard surf
Skin gap
tool size
Part sizes &

hole
locations
36
Rib-Spar as a Type 2 Assembly
Fwd Torque
Box
Strength PKC
Fairness PKC
Rib
Fixed Trailing
Edge
Step 1: Put FTB and

FTE in Fixture
Step 2: Put in ribs
FTB-Rib-FTE assy
767 Case Study
11/30/2004
Daniel E Whitney
37
Rib-Spar Assembly - 2
Skin
Plus Chord
Step 3: Add skins and adjust

skin gaps and plus chord alignment
to FTB and FTE
767 Case Study
11/30/2004
Daniel E Whitney
38
DFC for Rib-Spar as a Type-2
FTB
(6)
SKIN GAP
& RIB ALIGN
PKC
FIXTURE
(1)
RIBS
(5)
(6)
767 Case Study
11/30/2004
FTE
Daniel E Whitney
39
DFC for Wing Assembly as a Type 2
Two KCs in
conflict
FTB
Skin G
ap
PKC
Fwd
Skin
Str4-11
Plus Chord
Alignment PKC
RIBS
Plus
Chord
Plu
An s Ch
gle or
AK d
C
Plus Chord
Alignment PKC
Str1-2
p PK
a
G
Skin
Skin Gap AKC
SKIN GAP PKC
FIXTURE
Splice
Str3
Aft
Skin
FTE
767 Case Study
11/30/2004
Daniel E Whitney
40
Tolerance Analysis of KC Delivery
Using VSA
VSA was used to on each

candidate new process
Results show that process
1 is unable to deliver AKC
& PKC 1 all the time
because the holes in the
splice stringer cant be
placed accurately enough
This also hurts PKC 2 and
3
767 Case Study
11/30/2004
Process 2 is able to deliver
all 3 PKCs 100% of the
time
Daniel E Whitney
41
Matlab Analysis
(TM)
Assumed assemblers could maneuver the wing

skin laterally and angularly
Assumed smaller variation in hole and slot
placement
Assumed that the rest of the wing was error-free
Determined that only a few assemblies would fail
767 Case Study
11/30/2004
Daniel E Whitney
42
3"
G3
S1
FORWARD SKIN
G1
12"
P2
336"
4
6"
SPLICE STRINGER
PLUS
CHORD
3
63"
3
12"
AFT SKIN
H1
3"
S2
3"
P3
12"
P4
Parts and Their Frames
42"
3"
767 Case Study
11/30/2004
Daniel E Whitney
43
FORWARD
FTB
G4
STRENGTH KC
AERODYNAMIC KC
G2
0.045"
G1
P2
G3
S1
FORWARD SKIN
RIBS
PLUS
CHORD
63"
4
a
SPLICE STRINGER
3
AFT SKIN
P3
H1
b
S2
2
0.045"
AFT
1
FTE
P4
Parts Assembled and Frames

Aligned
END
FITTING
351"
767 Case Study
INBOARD
11/30/2004
Daniel E Whitney
OUTBOARD
44
Sample Space for Tolerance Analysis
X2
0.0127"
ADJUSTMENT POSSIBLE
BY TRANSLATION
AND CCW ROTATION
0.010"
.005"
ADJUSTMENT POSSIBLE
BY TRANSLATION
NO ADJUSTMENT
NEEDED
62.97"
63.00"
-.005"
2
63.03"
Y1
ADJUSTMENT POSSIBLE
BY TRANSLATION
-0.010"
ADJUSTMENT POSSIBLE
BY TRANSLATION
-0.0127"
AND CW ROTATION
0.015*5.25/29.25=.0027
767 Case Study
11/30/2004
Daniel E Whitney
45
Matlab Results
(TM)
f ixt ure and FTB-ribs-FTE modeled as 0.008 and plus chord

modeled as 0.01
0.02
0.015
~120
out of
10000
fall
outside
0.01
0.005
-0.005
-0.01
-0.015
-0.02
-0.01
767 Case Study
11/30/2004
0.01
0.02
0.03
0.04
Daniel E Whitney
0.05
0.06
0.07
0.08
46
Pros & Cons of Proposed Processes
Pros
Cons
767 Case Study
Current Process
Delivers all AKCs
and PKCs
repeatably
Inflexible fixtures
Variation absorbed
at stringer-plus
chord interface
11/30/2004
Proposed Process #1
Delivers AKC #2
and PKC #3
repeatably
Completely flexible
method
No dedicated
fixtures
Uses existing
fab
equipment
Least costly
Controls critical
interfaces
Fails to deliver
AKC #1 on a few
assemblies
PKC #1 & #2 not
delivered
on those same
assembies
Daniel E Whitney
Proposed Process #2
Delivers all AKCs
and PKCs
repeatably
Completely flexible
method
Uses existing
fab
equipment
Controls critical
interfaces
Requires higherfunctionality tack

fixture (higher
cost)
Requires a limited
number of small
fixtures
47
Rib-Spar as a Type 1 Assembly
Step 1: Put FTE on

Support
Step 2: Add ribs
Step 3: Add FTE
FTB-Rib-FTE assy
767 Case Study
11/30/2004
Daniel E Whitney
49
DFC for Rib-Spar as a Type-1
FTB
SK
I
N GAP PKC
RIBS
FTE
767 Case Study
11/30/2004
Daniel E Whitney
50
DFC for Wing Assembly as a Type 1
FTB
Skin G
a
p PKC
Str4-11
Plus Chord
Alignment PKC
Skin
d
hor
C
KC
Plu le A
g
An
Plus
Chord
P lu
An s Ch
gle or
AK d
Plus Chord
Alignment PKC
Str1-2
Skin
figs for designing assemblies
767 Case Study
Skin Gap AKC
S
K
IN

GAP PKC
RIBS
Two KCs still

in conflict Fwd
Splice
Str3
Aft
Skin
KC
P
p
Ga
FTE
11/30/2004
Daniel E Whitney
51
Impossible as a Type 2
FTB
d
hor PKC
t
sC
Plu nmen
Alig
ap PK
C
Str4-11
Plus
Chord
Plus
Alig Chord
nme
nt P
KC
RIBS
KCs do not conflict

Fwd
Skin
Skin Subassembly
Skin Gap AKC
SKIN GAP PKC
FIXTURE
Skin G
Str1-2
pP
a
G
Skin
KC
Splice
Str3
Aft
Skin
FTE
767 Case Study
11/30/2004
Daniel E Whitney
52
Impossible as a Type 1
FTB
p PKC
KCs do not conflict
Str4-11
Skin Subassembly
Plus
Chord
Plus
Alig Chord
nme
nt P
KC
Fwd
Skin
Skin Gap AKC
d
hor PKC
t
sC
Plu nmen
Alig
SK
N
I
GAP PKC
RIBS
Skin G
a
Str1-2
PKC
p
a
G
Skin
Splice
Str3
Aft
Skin
FTE
767 Case Study
11/30/2004
Daniel E Whitney
53
Cost Analysis -1
The basis for analysis was the KC-driven

Precision Assembly (PA) process for the 767
horizontal upper skin assy.
PA time and cost were estimated for the 767 skin
The 767 cost/time analysis was scaled for the

remaining 747 & 767 assemblies Vought makes
for Boeing.
PA assumed to be accomplished in three distinct
cells: Tack, CNC Auto-Rivet, Final Assembly
These cells all require new investment

767 Case Study
11/30/2004
Daniel E Whitney
54
Cost Analysis - 2
Baseline times for each step were taken from

Voughts estimates for its process.
Required cell time for MITs processes was
estimated based on Voughts times and a
distribution of realization factors applied to obtain
an assembly time for each cell.
A computer simulation was conducted to
determine the necessary capital equipment.
767 Case Study
11/30/2004
Daniel E Whitney
55
Simulation Scenarios
Three PA processes were developed and

analyzed.
The 3 processes are Vought, MIT 1, and MIT 2
Vought is Voughts proposed PA process
MIT 1 uses holes and slots. It was derived from Vought by applying the KC flowdown
method. MIT 2 uses NC tack cell
Three scenarios were studied:

- All Boeing assemblies, all programs
- Four representative assemblies
- Introduction of a new assembly
- New assembly would require new fixed tool but not new PA equipment
One and Two shift operations
767 Case Study
11/30/2004
Daniel E Whitney
56
Results - 1
PA estimated to reduce process time by

approximately 50%. At current demand this
results in approximately XX hours saved
annually.* Value of flexibility, image, and
freed-up floor space not included.
Annual savings = $X Million (assumes all
assemblies converted to PA at a rate of
$XX/hour.)
VOUGHT TO BE = 54% OF AS IS TIME
MIT 1 = 43% OF AS IS
MIT 2 = 42% OF AS IS
*ACTUAL NUMBERS ARE PROPRIETARY
767 Case Study
11/30/2004
Daniel E Whitney
57
Results - 2
Estimated equipment investment to implement PA

(example for MIT 1)
All Parts 4 Parts
One Shift
$21.4M
$14.1
Two Shifts
$14.1M
$7.3
(assumes cost per cell is Tack $2M, A-R
$4.8M, Final Assembly $0.5M)
767 Case Study
11/30/2004
Daniel E Whitney
58
Results - 3
Current economics did not justify the new process
The new process becomes economical if Vought
gains new business for which it can use the new
cells, thus saving the cost of new hard fixtures
Training and cultural issues remain to be evaluated
Adjusting by hand becomes adjusting via computer
Ad hoc process becomes a preplanned and designed

one requiring more manufacturing knowledge during
design
More communication between fab and assembly shops
needed
767 Case Study
11/30/2004
Daniel E Whitney
59
Assembly System Design Issues
Goals of this class
understand basic decisions in assembly system design

look at some typical lines for small and large products
different types of assembly machinery
example lines from industry
AITL Sys Des
11/5/2004
Daniel E Whitney
Basic Factors in System Design
Capacity planning - required number of units/year
Resource choice - assembly methods

Task assignment
Floor layout
Workstation design
Material handling and work transport
Part feeding and presentation
Quality
Economic analysis
Personnel training and participation
AITL Sys Des
11/5/2004
Daniel E Whitney
Basic Decision Process
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
Available Methods
Seat of the pants

The suppliers method, using his equipment
Trial and error, using simulation to evaluate
Analytical methods using math programming or

heuristics
Combination of technical and economic factors
and inequality constraints make this a hard
problem
AITL Sys Des
11/5/2004
Daniel E Whitney
The Basic Tradeoffs
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
Unit Cost Example
Unit Assembly Cost by Three Methods

f AC =0.38
T=2s
L H =$15/hr
S$=50000
$/tool = $10000
N = 10 parts/unit
w = 0.25 workers/sta
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
Characteristics of Manual Assembly
Technical
dexterous, able to learn and improve, flexible
can overlap operations - move+flip+inspect
may be too innovative, or may be unable to repeat
exactly the operation or the cycle time
Economic
top speed dictates need for more people to get more

output (called variable cost)
AITL Sys Des
11/5/2004
Daniel E Whitney
Cellular Assembly Line
Source:
One station
AITL Sys Des
11/5/2004
Daniel E Whitney
Whole line
8
Characteristics of Fixed Automation
Technical
simple operations with few DoF and simple alternatives
each station is dedicated to one operation
(place/fasten/confirm) built from standard modules
strung together
small parts, relatively high speed
basic architectures include in-line and rotary
Economic
the investment is in fixed increments regardless of
required capacity (fixed cost)
the payoff is in keeping uptime high (many stories)
AITL Sys Des
11/5/2004
Daniel E Whitney
Typical Cam-operated Assembly Machine
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
10
Typical Dial Machine
Same principle used

by Gillette for Mach 3
razors
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
11
Characteristics of Flexible Automation
Technical
multiple motion axes
motion (gross and fine) modulated by sensing and
decisions
multiple tasks with or without tool change
Economic
multiple tasks (within a cycle or next year)
investment scalable to demand (variable cost)

tools and parts presentation costly (fixed cost)
AITL Sys Des
11/5/2004
Daniel E Whitney
14
Sony VCR Assembly System
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
15
Line Architectures
Single serial line (car or airplane final assembly)
Fishbone serial line with subassembly feeder lines

(transmissions, axles)
Loop (common for automated lines)
U-shape cell (often used with people)
Rotary dial (used for very short production cycle

work with a single long task cycle like filling
bottles)
Transport can be synchronous or asynchronous
AITL Sys Des
11/5/2004
Daniel E Whitney
16
Serial and Parallel Line Arrangements
Source:
How do they compare on tool cost, reliability, time, flexibility?

AITL Sys Des
11/5/2004
Daniel E Whitney
17
Serial Line with Multiple Stations
Source:
(A) THREE COPIES OF STATION 3 ARE NEEDED BECAUSE

ITS TASK TAKES SO LONG
AITL Sys Des
11/5/2004
Daniel E Whitney
18
Serial Line with Uneven Task Assignment
Source:
Figure 16-10(b) in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
(B) GROUPING WORK AT STATIONS IMPROVES

BALANCE OF STATION TIMES
AITL Sys Des
11/5/2004
Daniel E Whitney
19
Multiple Paths Are Good and Bad
Source:
THERE ARE 6 POSSIBLE PATHS

AITL Sys Des
11/5/2004
Daniel E Whitney
20
Buffers - Conservative Design
They insulate the line from stopped stations
The only buffers that matter are the ones just

ahead and after the bottleneck station (the one
whose speed paces the line)
But it is often hard to tell which station is the
bottleneck
Since a blocked buffer is as bad as a starved one,
the ideal state of a buffer is half full
Let a = the average number of cycles to fix a
simple breakdown; b = buffer capacity
Then if b/2=a, there will be enough parts in the
buffer to keep everything going while a simple
breakdown
is fixed
AITL Sys Des
11/5/2004
21
Daniel E Whitney
Single Piece Flow
Necessary for big things like airplanes
Not necessary for little things

The alternative is batch transport
This creates work in process inventory, takes up space,
and seems associated with big inefficient factories (see
research by Prof Cochran)
Errors can hide in the batch and the whole thing might
have to be thrown away
Transport is infrequent so transport resources can be
shared
Creates a transport mafia and finger pointing (VW
engine plant story)
AITL Sys Des
11/5/2004
Daniel E Whitney
22
Kanban and Just in Time Systems
The kanbans are like money

Work is done when a ticket (kanban) arrives
Unwanted work is not done, WIP is controlled
Machines are not used just to use them (misplaced cost idea)
The whole factory operates, as much as possible, like one

big conveyor.
You never dont make the same thing every day.
It doesnt work unless the suppliers are doing it too
Kanban + single piece flow means piece rate = takt time
See To pull or not to pull: What is the question? by

Hopp and Spearman, Mfr & Service Ops Mgt, v 6 #2,
Spring 2004, pp 133-148
AITL Sys Des
11/5/2004
Daniel E Whitney
23
Toyota Georgetown KY Plant
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
24
Sometimes it Isnt Possible

~18-20 HOURS
4 HOURS NOTICE OF NEED

FOR A PARTICULAR ENGINE
1.5 hours to
make one block
~1.5 HOURS
ONE
BLOCK
3 weeks of
finished blocks
in about 30
varieties
ENGINE BLOCK MACHINING
ENGINE ASSEMBLY
400 OTHER PARTS:

6 ARE MADE HERE.
THE REST ARE BOUGHT
AND ARRIVE IN BOXES
AITL Sys Des
AN ENGINE
CAR ASSEMBLY
11/5/2004
Daniel E Whitney
You can make a block

quickly enough but
not soon enough
25
Sub-Delayed Commitment in a Fish
Bone Arrangement
SU
SU
BA
BA
SS
SS
E
EM
MB
DECOUPLING
BL
LY
YB
POINT FOR
A
SUBASSEMBLY A
AITL Sys Des
11/5/2004
Daniel E Whitney
DECOUPLING POINT
FOR SOMETHING THAT
IS NOT IN ANY SUBASSEMBLY
26
Basic Nominal Capacity Equations
# operations/unit * # units/year = # ops/yr

# ops/sec = # ops/yr * (1 shift/28800 sec)*(1 day/n shifts)*(1 yr/280 days)
cycle time = 1/(ops/sec) = required sec/op
equipment capability = actual sec/op
actual sec/op < required sec/op -> happiness
required sec/op < actual sec/op -> misery (or multiple resources)
Typical cycle times: 3-5 sec manual small parts
5-10 sec small robot
1-4 sec small fixed automation
10-60 sec large robot or manual large parts
AITL Sys Des
11/5/2004
Daniel E Whitney
27
Basic Cycle Time Equation
Cycle time = 1 assy time + in out time + tool ch. time * #ch. / unit
# units / pallet
# units / tool ch.
/unit
cycle time = net avg time per assembly
in out time = time to move one pallet out and another in
tool ch. time = time to put away one tool and pick up another
# ch. / unit = number of tool changes needed to make one unit
# units / tool ch. = number of units worked on before tool is
changed (cannot be larger than number
units / pallet)
= station uptime fraction: 0 < < 1

AITL Sys Des
11/5/2004
Daniel E Whitney
28
Example Lines from Industry
First Sony Walkman Line (~1981)

Four programmable robots with XYZ motions
Parts on trays, tools on robot frame

Assembly visits two stations, then person puts it
on a second tray upside down
Assembly then visits the other two stations
AITL Sys Des
11/5/2004
Daniel E Whitney
29
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
30
Example Lines from Industry - 2
Denso Alternator Line (~1986)
~ 20 parts installed
loop arrangement
20 home-made robots
able to switch size of alternator
brushes retained by throw-away pin
cycle time perhaps 10 sec, two or three shifts
inspired by Draper movie of alternator assembly shown
in 1980
AITL Sys Des
11/5/2004
Daniel E Whitney
32
Denso Robotic Alternator Assembly Line
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
33
Example Lines from Industry - 3
Boeing 777 Assembly (~ 1993)
whole airplanes from structural subassemblies

lots of outsourcing
basically single serial line
fuselage segments built upside down on floors, then
flipped for installation of crowns
successive joining from front, rear and sides
a lot of systems installed before final body join

cycle time moving toward 3 days, 3 shifts/day
AITL Sys Des
11/5/2004
Daniel E Whitney
38
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
39
Source:
AITL Sys Des
11/5/2004
Daniel E Whitney
40
Assembly Workstation Design Issues
Goals of this class

understand workstation elements
look at part feeding and presentation alternatives
design a process and a single station system for it
AITL Workstation Issues
11/5/2004
Daniel E Whitney
Assembly = Reduction in DoF Uncertainty
Source:
11/5/2004
Daniel E Whitney
What Happens in a Workstation
An incomplete assembly arrives (or several at once)
Parts to be assembled arrive

as single parts
as a subassembly
Parts may have to be separated, oriented, given a

final check
Parts are joined to the assembly
Assembly correctness is checked
Documentation may have to be filled out
The assembly is passed on to the next station
11/5/2004
Daniel E Whitney
Major Issues
Get done within the allowed cycle, which is

usually short
Avoid the three common errors
wrong part
correct part installed wrong, damaged, or causing
damage to the rest of the assembly
bad part used anyway
Error-proofing or poka-yoke
Handle a lot of distractions

11/5/2004
Daniel E Whitney
Cycle Time
Varies from milliseconds for cigarettes to days for

aircraft
Components
SPEED
work in/out
move to get tool
move to get part
move to insertion point
insert
move to get new tool
RAMP
UP TO
FULL
SPEED
% TIME
% DISTANCE
COVERED
RUN AT
FULL SPEED
DISTANCE
RAMP
DOWN
22
40
38
14
68
18
time
CREEP
60% OF THE TIME IS SPENT COVERING ONLY 32% OF THE DISTANCE
Each move includes accel-steady speed-decel-creep
11/5/2004
Daniel E Whitney
Simulation Software
Source:
11/5/2004
Daniel E Whitney
Coordinate Transfer and Part Control
Machine assembly requires transfer of coordinates
from where part is palletized to where it is gripped
from where it is gripped to where it mates

grip points may or may not be functional features
These coordinates are usually on different

locations on a part
From each of these coordinates runs a chain of
coordinate frames back around until they meet at
the point of assembly
11/5/2004
Daniel E Whitney
SOME OF THE MANY SOURCES

OF ERROR THAT MUST BE
CONTROLLED IN ORDER TO
ACHIEVE SUCCESSFUL ASSEMBLY
A. PART CONSTRUCTION
B. PART JIGGING
C. JIG LOCATION
D. ROBOT ACCURACY AND
CALIBRATION
E. TOOL SOCKET
F. PART GRIP
G. OFFLINE MODEL
Assemblability
= A+B+C+D+E+F+G
E
F&A
A
B
C
11/5/2004
Daniel E Whitney
Some Important Decisions
Choice of assembly resource
cost, reach, speed, multi-task capability, load cap,

dexterity, etc: people, robots, dedicated/fixed
Part presentation at the station or elsewhere
accuracy of palletizing or carrier strips almost the same

as that of assembly
economics of palletizing: how/who; pipeline of WIP
Serial vs parallel parts presentation

vibratory bowl or parts strip vs pallet
Tool change vs multi-purpose tools

similar issues apply to manual and robotic
11/5/2004
Daniel E Whitney
11/5/2004
Daniel E Whitney
Small Parts Feeder
Source:
10
Large Parts Feeder
Source:
11/5/2004
Daniel E Whitney
11
Other Important Decisions
Allocation of degrees of freedom
all in one place

shared between two, as in 4 DOF robot and 2 DOF
workholder
Combinations of fab and part arrangement with

assembly
creates parts or subassemblies on the spot
examples: pre-assembly of valve keepers, spring

winding, lubrication, sorting
11/5/2004
Daniel E Whitney
12
Valve Keepers
Source:
11/5/2004
Daniel E Whitney
13
Workstation Layout
Part presentation
Automatic feeders
Chutes loaded from opposite side
Bulk parts vs kits
Station layout to provide
Parts and tools within easy reach

Things laid out in process sequence
Instructions - paper, video for each version
Instructions - what version is this
Documentation - tests performed, parts installed
Line layout to provide

Space for materials at lineside
Space for transporters
11/5/2004
Daniel E Whitney
14
Sony APOS
Offline shakers fill pallets (~ 10 x 12)
Part jams, if any, occur off line and do not stop the
assembly system
Rather complex parts can be presented automatically
Pallets occupy considerable space at the workstation
The robot spends a lot of time slewing over to the

pallet to get a part
So you trade time for space: do you win?
11/5/2004
Daniel E Whitney
15
Sony APOS - Palletizer
Source:
11/5/2004
Daniel E Whitney
16
Sony APOS - Assembly Station
Source:
11/5/2004
Daniel E Whitney
17
Other Architectures
Escort parts and tools (early Sony FX-1)
Flexibility based on station lockout

one simple station per part or version thereof
assembly passes through unneeded stations
lots of floor space
Roving robot (Hitachi, 1980)

carries assembly in its lap
visits stations that feed parts and hold special tools
Roving robot teams (Denso, 2000)

Robots carry tools, assemblies ride conveyor, parts delivered at
stations
Robots can be added or removed from system to adjust capacity
Robots can share work at highly loaded stations
Parts made in or fastened to carrier strips - separates part

prep from part feeding for higher feeding reliability
11/5/2004
Daniel E Whitney
18
Sony Walkman II
Source:
11/5/2004
Daniel E Whitney
19
Sony Phenix 10 System
Source:
11/5/2004
Daniel E Whitney
20
Parts Tray
Source:
11/5/2004
Daniel E Whitney
22
Source:
Figure 17-21(b) in [Whitney 2004] Whitney, D. E. Mechanical Assemblies: Their Design, Manufacture,
11/5/2004
Daniel E Whitney
23
Pallet Arrangement for Large Parts
UNIQUE
PARTS FOR
ONE ASSEMBLY
ASSEMBLY
PALLET
BULK PARTS
FOR SEVERAL
ASSEMBLIES
11/5/2004
Daniel E Whitney
24
Starter Assembly Automation Line
(Slide from Denso)
Source:
11/5/2004
Daniel E Whitney
27
Making Stacks - Method 1
Source:
11/5/2004
Daniel E Whitney
29
Source:
11/5/2004
Daniel E Whitney
30
Source:
11/5/2004
Daniel E Whitney
AMP Ignitor
Source:
11/5/2004
Daniel E Whitney
32
11/5/2004
Daniel E Whitney
Manual Assembly Sequence
Source:
33
11/5/2004
Daniel E Whitney
Alternate Process
Source:
34
11/5/2004
Daniel E Whitney
Alternate Process One-handed
Source:
35
Single Station System
Source:
11/5/2004
Daniel E Whitney
36
Number every page in your report.

1. Completely describe the product so that you become thoroughly familiar with it at the part and function level
tell where you bought the product and how much it cost
carefully take the product apart, making notes or even taking digital photos
draw all the parts freehand or 2D or 3D computer (except fasteners)
number all the parts for easy reference and make a parts list, including fasteners
make an assembly drawing showing all the parts, including fasteners; make cross-section drawings of important subassemblies
note: an assembly drawing shows all the parts in their final assembled positions
make an exploded view showing all the parts, including fasteners; name and number each part
make a bill of materials (BOM) listing the part names and numbers from your exploded view
include the BOM and exploded view in every project report

determine what materials the parts are made of and how they are made
explain what each part does to help the product deliver its functions
identify any "mystery" parts or features
list and discuss all the mates between parts
note important dimensions on the drawings of the parts, including clearances
measure clearances
estimate difficulty of assembly using clearance ratios and your personal observations
if possible, determine differences between different but similar versions of the product
contact the company or talk to retailers about the product and find out or estimate
where it is made, what is the production volume, how many versions are made, etc
2. Perform a datum flow chain analysis of part of your product in order to understand how one important function is realized
Make a liaison diagram of the product
select a significant set of parts that does an identifiable function
describe and explain that function (report #1 will be helpful here)
identify the KC or KCs associated with that function and identify the DFC or DFCs involved
guess or surmise a root for the DFC
draw the DFC and explain all its elements: features, constraint, important KCs
on each KC, write the degrees of freedom that are important for its delivery
explain why your DFC looks as it does, paying particular attention to which parts
locate which other parts
determine if there are any over-constraints or risk of over-constraints
suggest redesigns that improve constraint or KC delivery
3. Completely choreograph each assembly step in any convenient assembly sequence and do a DFA analysis
so that you understand all your product's assembly operations and problems
generate some feasible assembly sequences, choose one, say why
Note:
there is no need to generate all assembly sequences
List all the:
required gross motions including reorientations
required fine motions
features on parts where they can be gripped
features on parts where they can be mated to a fixture
all chamfers and lead-ins or indicate where there are none
the types of fasteners, how many, how many of each kind or size
auxiliary operations such as lubrication, test, and inspection
possible problems and risk areas:
feeding, fixturing, presenting, orienting
gripping
inserting
damage to parts or people
suggest design improvements to deal with these risks

sketch the fixtures and grippers that are needed for assembling this product
perform a DFA analysis (report #2 will be helpful here)
try assembling the product yourself several times and compare your time to that predicted by DFA
calculate assembly efficiency
if feasible, implement some of the design changes on an example part or parts
4. Create a floor layout for this product's assembly
draw an assembly tree for your assembly sequence, identifying subassemblies
Note:
this tree is simply a diagram of main and subassemblies in assembly sequence order
assume a production rate or annual volume
calculate required system capacity and takt time
assume workstations or people with given realistic assembly speeds and assign assembly operations to them
data from report #3 will be useful here
make a Gantt chart for the entire process and indicate on it which stations do whch tasks
draw or sketch the floor layout, showing where each workstation is and what steps it does
this should be an actual floor plan, not just a schematic drawing
show how parts will flow across the floor to feeding points
provide space for any people who work on the line, keeping them clear of machinery, and provide space for material handling
determine how many direct workers, supervisors, inspectors, material handlers and other people will be needed
if software is available, use it to design the concept line
5. Design a workstation
select a group of operations, say for making a small subassembly, and assume they will be done at one workstation
determine the time needed to complete the operation or operations involved based on data from report #3
lay out the station, including in- and out-flows of assemblies and parts
plan required motions of equipment or people
show how necessary inspections or tests will be carried out
allocate required time of each required activity and draw a Gantt chart for a complete cycle
estimate the purchase and installation cost of this station, using actual costs that you may be able to find on the Web or in standard catalogs
Note: McMaster-Carr and other catalogs have benches, chairs, power screwdrivers, conveyors, etc.
estimate the labor cost of doing the work at this station based on where you think the product is made
6a. Perform an economic analysis of your assembly layout
estimate or obtain approximate costs of equipment and labor rates with OH and benefits
estimate engineering and installation costs
establish a payback period or required rate of return
determine unit assembly cost including equipment and labor
determine ROI by comparing two assembly methods
or by assuming manufacturing costs and sales revenues
6.b Perform a discrete event simulation of your line
diagram the line with activities and queues
write program and make several runs
identify any operating problems and improve the design
Systematic Assembly Analysis and Planning Process

Understand context (addressed in more detail later)
management's objectives for the product or product line
production volume
cost
quality
model mix or evolution
schedule for going into production
status of the design: new, reused
character of the product, nature of the market and customers
customer expectations
nature of customer interaction with the product
reuse, upgrade
Assembly in the Small
Understand each assembly step in detail
the basic requirements
size, shape, weight, dimensions of each part
characterization of each mate between parts
special character of particular parts
assembly difficulty
handling constraints
gripping
feeding
Conventional Design for Assembly
part consolidation opportunities
part feeding difficulty
part handling difficulty
Identify high risk areas
part damage
wrong part
misassembly
safety or regulatory issues
tasks so hard only one person can do them
Identify necessary experiments
Recommend local design improvements
Assembly in the Large (aka Design of Assembly)
Understand the business context
product character and type of market
sales volume anticipated
model variety anticipated
plans for new versions
delayed committment
supplier logistics and make vs buy
cost limits
labor costs and any regulations
cost calculation and ROI methods
ROI targets
Understand the factory context
labor conditions, training, shift policies
space and facility constraints
Identify system requirements
tentative cycle time

production flow and floor layout
feasible methods and equipment
required sensing and communication
required displays and controls
parts presentation
alternate assembly sequences
fixtures and parts carriers
Design a concept assembly system
system architecture
equipment selection and task assignment
cost and economic performance
simulation
average flow and production rate
uptime
queues, blockage, starvation
model changeovers
Make final recomendations
additional design improvements
line design or sequence options
remaining risk areas

2875 Fall 2004

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

2875 Fall 2004

Hochgeladen von

Copyright:

Verfügbare Formate

Assembly System Design Techniques

Goals of this class

Introduce system design methods

Asst Sys Des Tech

Assembly System Design Techniques

Assembly system design algorithms exist

They solve the Equipment Selection and Task

The tasks that need to be done

What each resource costs to buy

Time for transport from station to station

Heuristics by R E Gustavson at Draper and N H

Bruce Lamar, 1979 - bus routing algorithm

Carol Holmes, 1987 - multiple products, dynamic

Holmes-Cooprider method reprogrammed by

System Selection Criteria

Minimize annualized cost

Systems can be forced to be all manual, all robot,

Asst Sys Des Tech

Summary of Required Input

Info about assembly resources with cost, operation

Info about assembly tasks with operation time and

Asst Sys Des Tech

Station-station move time (s)

Asst Sys Des Tech

Data Input (Applicable

For each resource:

Asst Sys Des Tech

Assembly Planning Chart

Applicable Technology Chart

Asst Sys Des Tech

Assembly Planning Chart

Asst Sys Des Tech

Basic Nominal Capacity Equations

# operations/unit * # units/year = # ops/yr

How the Holmes-Cooprider Method

The maximum takt or cycle time is calculated

of selected resources and the tasks they will do

Asst Sys Des Tech

Asst Sys Des Tech

Network Models of Assembly Systems

Model of system as flows in a network

Represents equilibrium state

Outbound probabilities add to 1.0

through and average flow on each branch

Asst Sys Des Tech

pij=pr of going from node i to node j

pij = 1 for each

Conservation of flow at node j:

xj=flow into node j from outside

cost = fij cij

Asst Sys Des Tech

Example System: an assembly with two

subassemblies and several test and rework

Asst Sys Des Tech

Network Equivalent of Example

Build/repair Subassembly #1 and Test it

Asst Sys Des Tech

Asst Sys Des Tech

%FF = total flow in system

Total flow without rework = 5

Goals of this class

Key Characteristics (KCs)