Push and Pull Production Systems

You say yes.

I say no.
You say stop.
and I say go, go, go!
The Beatles

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

The Key Difference Between Push and Pull

Push Systems: schedule work
releases based on demand.
inherently due-date driven
control release rate, observe
WIP level

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Pull Systems: authorize work

releases based on system status.
inherently rate driven
control WIP level, observe
throughput

http://www.factory-physics.com

Push vs. Pull Mechanics

PUSH

PULL

(Exogenous)
Schedule

(Endogenous)
Stock Void

Production Process

Job

Job

Push systems do not limit

WIP in the system.
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Production Process

Pull systems deliberately

establish a limit on WIP.

http://www.factory-physics.com

What Pull is Not!

Make-to-Order:
MRP with firm orders on MPS is make-to-order.
But it does not limit WIP and is therefore a push system.

Make-to-Stock:
Pull systems do replenish inventory voids.
But jobs can be associated with customer orders.

Forecast Free:
Toyotas classic system made cars to forecasts.
Use of takt times or production smoothing often involves
production without firm orders (and hence forecasts).

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

Push and Pull Examples

Are the following systems essentially push or essentially pull?

Kinkos copy shop:

Soda vending machine:
Pure MRP system:
Doctors office:
Supermarket (goods on shelves):
Tandem line with finite interstation buffers:
Runway at OHare during peak periods:
Order entry server at Amazon.com:

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

Push and Pull Line Schematics

Pure Push (MRP)

Stock
Point

Stock
Point

...

Pure Pull (Kanban)

Stock
Point

Stock
Point

...

Stock
Point

CONWIP

Authorization Signals
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Stock
Point

...

Full Containers

http://www.factory-physics.com

Pulling with Kanban

Outbound
stockpoint

Production
cards

Completed parts with cards

enter outbound stockpoint.

When stock is
removed, place
production card
in hold box.

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Outbound
stockpoint

Production
card authorizes
start of work.

http://www.factory-physics.com

Inventory/Order Interface
Concept:
Make-to-stock and make-to-order can be used in same system.
Dividing point is called the inventory/order interface.
This is sometimes called the push/pull interface, but since WIP could be
limited or unlimited in both segments, this is not a strictly accurate term.

Benefit: eliminate entire portion of cycle time seen by customers by

building to stock.
Implementation:
kanban
late customization (postponement)

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

Example Custom Taco Production Line

I/O Interface
Make-to-Stock

Refrigerator

Cooking

Make-to-Order

Assembly

Packaging

Sales

Customer

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

Example Quick Taco Production Line

Make-to-Order

Refrigerator

Cooking

I/O Interface

Assembly

Packaging

Make-to-Stock

Sales

Warming
Table

Customer

Notes:
I/O interface can differ by time of day (or season).
I/O interface can differ by product.

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

10

Example IBM Panel Plant

Original Line
Treater

Prepreg,
Copper

Lamination

Machining

I/O Interface

Prepreg,
Copper

Drilling

Copper
Plate

Procoat

Sizing,
Test

Copper
Plate

Procoat

process that gives

boards personality

Revised Line
Treater

Circuitize

Lamination

Machining

Core
Blanks

Circuitize

I/O Interface

Drilling

Sizing,
Test

Notes:
Moving I/O interface closer to customer shortens leadtime
seen by customer.
Small number of core blanks presents opportunity to make them to stock.
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

11

Example HP Deskjet Supply Chain

Integrated Circuit
Manufacturing

Printed Circuit
Assembly & Test

Final Assembly
and Test

U.S.
DC

Customer

European
DC

Customer

Far East
DC

Customer

I/O Interface

Notes:
I/O interface located in markets to achieve quick
response to customers
Delayed differentiation of products (power supplies for
different countries) enables pooling of safety stocks

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

12

I/O Interface Conclusions

Basic Tradeoff:
responsiveness vs. inventory (time vs. money)
moving PPI closer to customer increases
responsiveness and (usually) inventory

Optimal Position of I/O Interface:

need for responsiveness
cost of carrying inventory product diversification

Levers:
product design (postponement)
process design (quick response manufacturing)

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

13

Advantages of Pull Systems

Low Unit Cost:

Good Customer Service:

short cycle times
steady, predictable output
stream

low inventory
reduced space
little rework

High External Quality:

high internal quality
pressure for good quality
promotion of good quality
(e.g., defect detection)

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Flexibility:
avoids committing jobs too
early
encourages floating
capacity

http://www.factory-physics.com

14

The Magic of Pull

Pulling Everywhere?
You dont never make nothin and send it no place. Somebody
has to come get it.

Hall 1983
No! Its the WIP Cap:
Kanban WIP cannot exceed number of cards
WIP explosions are impossible

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

WIP

15

Pull Benefits Achieved by WIP Cap

Reduces Costs:

Improves Customer Service:

prevents WIP explosions

reduces average WIP
reduces engineering
changes

Improves Quality:

reduces cycle time variability

pressure to reduce sources of
process variability
promotes shorter lead times and
better on-time performance

pressure for higher quality Maintains Flexibility:

improved defect detection
avoids early release (like air
improved communication
traffic control)
less direct congestion
less reliance on forecasts
promotes floating capacity
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

16

CONWIP
Assumptions:
1. Single routing
2. WIP measured in units

...

Mechanics: allow next job to enter line each time a job leaves

(i.e., maintain a WIP level of m jobs in the line at all times).

Modeling:
MRP looks like an open queueing network
CONWIP looks like a closed queueing network
Kanban looks like a closed queueing network with blocking

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

17

CONWIP Controller
Work Backlog

Indicator Lights

PN

Quant

LAN

R G

PC

PC

...
Workstations
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

18

CONWIP vs. Pure Push

Push/Pull Laws: A CONWIP system has the following
advantages over an equivalent pure push system:
1) Observability: WIP is observable; capacity is not.
2) Efficiency: A CONWIP system requires less WIP on average to
attain a given level of throughput.
3) Robustness: A profit function of the form
Profit = pTh - hWIP
is more sensitive to errors in TH than WIP.

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

19

CONWIP Efficiency Example

Equipment Data:
5 machines in tandem, all with capacity of one part/hr (u=THte=TH)
exponential (moderate variability) process times

w
w
rb
CONWIP System: TH ( w)
w W0 1
w4
u
TH
5
Pure Push System: w(TH ) 5
1 u
1 TH

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

PWC formula

5 M/M/1 queues

20

CONWIP Efficiency Example (cont.)

How much WIP is required for push to match TH attained
by CONWIP system with WIP=w?

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

21

CONWIP Robustness Example

Profit Function: Profit pTH hw

w
hw
CONWIP: Profit(w) p
w4

need to find optimal

WIP level

5 TH

Push: Profit(TH) pTH h

1 TH

need to find optimal

TH level (i.e., release
rate)

Key Question: what happens when we dont choose optimum

values (as we never will)?

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

22

CONWIP vs. Pure Push Comparisons

70

Optimum

CONWIP

60

Efficiency

50

Robustness

Profit

40
30

Push

20
10
0
0.00%
-10

20.00%

40.00%

60.00%

80.00%

100.00%

120.00%

140.00%

-20

Control as Percent of Optimal

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

23

Modeling CONWIP with Mean-Value Analysis

Notation:
u j ( w) utilization of station j in CONWIP line with WIP level w
CT j ( w) cycle time at station j in CONWIP line with WIP level w
CT ( w)

n
j 1

CT j ( w) cycle time of CONWIP line with WIP level w

TH ( w) throughput of CONWIP line with WIP level w

WIPj ( w) average WIP level at station j in CONWIP line with WIP level w

Basic Approach: Compute performance measures for increasing

w assuming job arriving to line sees other jobs distributed
according to average behavior with w-1 jobs.

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

24

Mean-Value Analysis Formulas

Starting with WIPj(0)=0 and TH(0)=0, compute for w=1,2,
t e2 ( j ) 2
CT j ( w)
[ce ( j ) 1]TH ( w 1) [WIPj ( w 1) 1]t e ( j )
2
n

CT ( w) CT j ( w)
j 1

w
CT ( w)
WIPj ( w) TH ( w)CT j ( w)
TH ( w)

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

25

Computing Inputs for MVA

MEASURE:
Natural Process Time (hr)

STATION:
t0

1
0.090

2
0.090

3
0.098

4
0.090

5
0.090

Natural Process CV
Number of Machines

c0
m

0.000
1

0.000
1

0.000
1

0.000
1

0.000
1

MTTF (hr)

mf

200

200

200

200

200

MTTR (hr)
Availability

mr
A

2
0.990

1
0.995

2
0.990

2
0.990

2
0.990

Effective Process Time (failures only)

t e'

0.091

0.090

0.099

0.091

0.091

Eff Process CV (failures only)

ce '

0.436

0.110

0.400

0.436

0.436

Jobs Between Setups

Ns

100.000

100.000

100.000

100.000

100.000

Setup Time (hr)

ts

0.000

0.000

0.000

0.000

0.000

Setup Time CV

2
cs

1.000

1.000

1.000

1.000

0.500

Eff Process Time (failures+setups)

te

0.091

0.090

0.099

0.091

0.091

Eff Station Rate

re

11.001

11.056

10.103

11.001

11.001

0.004

0.001

0.004

0.004

0.004

0.436

0.110

0.400

0.436

0.436

Eff Process Time Var (failures+setups)

Eff Process CV (failures+setups)

ce

rb

10.103

T0

0.462

W0
TH
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

PWC
4.669
Best
Worst
Case
Case
CT TH
CT
CT
CT
TH
TH

http://www.factory-physics.com

26

Output of MVA
TH
w
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Actual
2.164
3.807
5.037
5.957
6.651
7.185
7.604
7.938
8.210
8.435
8.624
8.784
8.922
9.041
9.145

CT
CT
CT
CT TH
TH
TH
CT1(w) CT2(w) CT3(w) CT4(w) CT5(w)
Actual
Best
Worst
Best
Worst
PWC
PWC
Case
Case
Case
Case
0.462
0.091
0.090
0.099
0.091
0.091
0.525
0.104
0.100
0.114
0.104
0.104
0.596
0.118
0.111
0.131
0.118
0.118
0.672
0.133
0.123
0.149
0.133
0.133
0.752
0.149
0.135
0.170
0.149
0.149
0.835
0.166
0.147
0.191
0.166
0.166
0.921
0.182
0.160
0.214
0.182
0.182
1.008
0.199
0.173
0.237
0.199
0.199
1.096
0.216
0.186
0.262
0.216
0.216
1.186
0.233
0.198
0.288
0.233
0.233
1.276
0.250
0.211
0.315
0.250
0.250
1.366
0.267
0.224
0.342
0.267
0.267
1.457
0.283
0.236
0.371
0.283
0.283
1.549
0.300
0.249
0.400
0.300
0.300
1.640
0.316
0.261
0.430
0.316
0.316

http://www.factory-physics.com

27

Using MVA to Evaluate Line Performance

12.000

10.000

TH

8.000
Actual
Best Case
Worst Case
PWC

6.000

4.000

2.000

0.000
0

10

15

20

25

30

35

WIP

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

28

Implementing Pull
Pull is Rigid: replenishing stocks quickly (just in time) requires
level mix, volume, sequence.

JIT Practices:
Support Rigidity:
production smoothing/mix stabilization

Mitigate Rigidity in Production System

capacity buffers
setup reduction
flexible labor
facility layout
product design (postponement, etc.)

this is the
genius
of pull!

Mitigate Rigidity in Organization

TQM
vendor management, etc.
Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

29

Capacity Buffers
Motivation: facilitate rapid replenishments with minimal WIP
Benefits:
Protection against quota shortfalls
Regular flow allows matching against customer demands
Can be more economical in long run than WIP buffers in push
systems

Techniques:
Planned underutilization (e.g., use u = 75% in aggregate planning)
Two shifting: 4 8 4 8
Schedule dummy jobs to allow quick response to hot jobs

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

30

Setup Reduction
Motivation: Small lot sequences not feasible with large setups.
Internal vs. External Setups:
External performed while machine is still running
Internal performed while machine is down

Approach:
1. Separate the internal setup from the external setup.
2. Convert as much internal setup as possible to external setup.
3. Eliminate the adjustment process.
4. Abolish the setup itself (e.g., uniform product design, combined
production, parallel machines).

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

31

Flexible Labor
Cross-Trained Workers:
float where needed
appreciate line-wide perspective
provide more heads per problem area

Shared Tasks:
can be done by adjacent stations
reduces variability in tasks, and hence line stoppages/quality
work can float to
problems

workers, or workers
can float to work

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

32

Cellular Layout
Advantages:
Better flow control
Improved material handling (smaller transfer batches)
Ease of communication (e.g., for floating labor)

Challenges:
May require duplicate equipment
Product to cell assignment

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

Inbound Stock

http://www.factory-physics.com

Outbound Stock

33

Focused Factories
Pareto Analysis:
Small percentage of skus represent large percentage of volume
Large percentage of skus represent little volume but much complexity

Drill

Saw

Mill

Drill

Paint

Grind

Mill

Drill

Paint

Weld

Grind

Lathe

Drill

Grind

Paint

Lathe

Warehouse

Saw

Warehouse

Mill

Stores

for low runners

many setups
poorer performance, but only
on smaller portion of business
may need to use push

Lathe

Assembly

Job Shop Environment:

Saw

Assembly

for families of high runners

few setups
can use pull effectively

Stores

Dedicated Lines:

Mill
Drill

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

34

Push/Pull Takeaways
Magic of Pull: the WIP cap
MTS/MTO Hybrids: locating the I/O interface
Logistical Benefits of Pull:
observability
efficiency
robustness (this is the key one)

Overcoming Rigidity of Pull:

capacity buffers
setup reduction
flexible labor
facility layout, etc.

Wallace J. Hopp, Mark L. Spearman, 1996, 2000

http://www.factory-physics.com

35