reinertsen+lkce+2012+wip+constraints (1)
DESCRIPTION
OkTRANSCRIPT
The Science ofWIP ConstraintsLean Kanban Central Europe
Vienna, AustriaOctober 23, 2012
Donald G. ReinertsenReinertsen & Associates
600 Via Monte D’OroRedondo Beach, CA 90277 U.S.A.
(310)-373-5332Internet: [email protected]
www.ReinertsenAssociates.com
No part of this presentation may be reproducedwithout the written permission of the author.
2Copyright 2012, Reinertsen & Associates
Objectives• What is the science behind WIP constraints?• How do WIP constraints affect economics?• What is the difference between a WIP
constraint and PULL?• What is the difference between WIP control and
WIP constraints?
3Copyright 2012, Reinertsen & Associates
SomeBackground
4Copyright 2012, Reinertsen & Associates
The TPS Emergency Room
• We desire to rigorouslyimitate the practices ofToyota.
• We will set a strictupper limit on WIP.
• When we reach ourlimit, no new patientscan enter until anotherdeparts.
5Copyright 2012, Reinertsen & Associates
Two Approaches• We can treat WIP constraints with two approaches:
• Science-based• Faith-based
• A science-based approach:• Forces you to understand underlying
mechanisms of action.• Permits you to engineer specific solutions for
different and changing contexts.• A faith-based approach:
• Requires much less thinking.• Works well in stable contexts like manufacturing.
6Copyright 2012, Reinertsen & Associates
Thus, since the Toyota Production System hasbeen created from actual practices in thefactories of Toyota, it has a strong feature ofemphasizing practical effects, and actualpractice and implementation over theoreticalanalysis. – Taiichi Ohno
From Foreword to 1983 FirstEdition of Toyota ProductionSystem by Yasuhiro Monden
7Copyright 2012, Reinertsen & Associates
The Science
8Copyright 2012, Reinertsen & Associates
Queueing Theory
• The key to a science-based approach isqueueing theory.
• Queues form when processes withvariability are loaded to high levels ofutilization.
• Queues are not intrinsically evil.• They create quantifiable economic costs.• By understanding the behavior of queues
we can discover how to control them.
Traffic at rush hourillustrates the classiccharacteristics of aqueueing system.
Pho
to C
opyr
ight
200
0 C
omst
ock,
Inc.
10Copyright 2012, Reinertsen & Associates
The Effect of Capacity Utilization
Queue Size vs. Capacity Utilization
0
5
10
15
20
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%Capacity Utilization
Que
ue S
ize Deterministic
Stochastic
Notes: Assumes M/M/1/ Queue, = Capacity Utilization
1
2
qL
11Copyright 2012, Reinertsen & Associates
Total Cost
Cost of Delay
Cost of Excess Capacity
The Economics of Queues
Excess Product Development Resource
Dollars
To Maximize Profits,Minimize Total Cost
How many product developers can draw the red curve?
12Copyright 2012, Reinertsen & Associates
Arrivals
Departures
Time in Queue
Quantityin Queue
QueueTime
CumulativeQuantity
Cumulative Flow Diagram
13Copyright 2012, Reinertsen & Associates
Arrivals
Queue
Time
CumulativeQuantity
Little’s Formula
You can calculate size of thequeue either by integratinghorizontal or vertical slices.
Departures
)( tTimeat Queuein Customers ofNumber
)( Customer nth for Queuein Wait time
)()( 1
tL
nW
dttLnW
q
q
N
n
b
aqq
Rate Departure Average
)(Queuein Customers ofNumber Average
Queuein Time Wait Average
tL
W
LW
q
q
14Copyright 2012, Reinertsen & Associates
How WIP Constraints Work• Queueing systems randomly drift into
high queue states.• These high queue states can be relatively
persistent and, when present, they candelay many jobs for a long time.
• If we prevent a system from entering thesehigh queue states, we will producesignificant cycle time benefits at relativelylow cost.
15Copyright 2012, Reinertsen & Associates
• We flip a coin 1000 times, add 1 for each head,subtract 1 for each tail, and keep track of ourcumulative total.
• What is your best estimate of:• How many times the cumulative total will return to
the zero line during the 1000 flips?
Random Processes
Time
H T T H H HTCumulativeTotal
16Copyright 2012, Reinertsen & Associates
Cumulative
-10
0
10
20
30
40
50
0 250 500 750 1000
Note: +1 for each head, -1 for each tailBased on example from “Introduction to Probability Theory and Its Applications”, by William Feller. John Wiley: 1968
One Thousand Coin Tosses
1st Half Crossings = 382nd Half Crossings = 0
Average Time Between Crossings = 25.6
Maximum Time Between Crossings = 732
17Copyright 2012, Reinertsen & Associates
Early
Late
Cumulative Totals Diffuse
Value of Random Variable
Prob
abili
ty
1. Zero is always most probable value.2. But, it becomes less probable with time.3. For large N, a binomial distribution approaches anormal distribution.
Notes:
18Copyright 2012, Reinertsen & Associates
Probability of High Queue States
Number of Items in Queue
Prob
abili
ty
n1State Probability =
for M/M/1/ Queue
19Copyright 2012, Reinertsen & Associates
Impact of High Queue States
Number of Items in Queue (Queue State)
Prob
abili
tyState Probability
Impact onCycle Time
for M/M/1/ QueueA State’s Cycle Time Impact = pn(n)/
Cyc
le T
ime
Impa
ct
20Copyright 2012, Reinertsen & Associates
So, Why Not Chop Off theTail?
Number of Items in Queue
Prob
abili
ty
State Probability
Impact onCycle Time
21Copyright 2012, Reinertsen & Associates
The Economics
22Copyright 2012, Reinertsen & Associates
How WIP Constraints Work• WIP constraints affect three important
economically important factors:• They decrease average cycle time. (+)• They generate blocking costs. (-)• They create underutilization costs. (-)
• We need an economic framework toassess their impact.
23Copyright 2012, Reinertsen & Associates
The M/M/1/k Queue
WIPCAP 1 2 5 10 20 InfiniteAverage Cycle Time 1.0 1.5 2.8 4.6 7.2 10.0 Time in Queue 0 0.5 1.8 3.6 6.2 9.0
Utlilization Percent 47% 63% 79% 85% 89% 90%Empty Percent 53% 37% 21% 15% 11% 10%Blocking Percent 47% 30% 13% 5% 1% 0%
Note: Assumes 90 percent utilization.
WIP constraints tradeoff reductions in cycle timeagainst blocking and underutilization costs.
24Copyright 2012, Reinertsen & Associates
Effect of WIPCAPS
72%
52%
28%
21%
5%
1%
13%
5%
1%
0% 20% 40% 60% 80%
0.5x
1x
2x
WIPCAP
Percent Change
Delay Reduction Underutilization Blocking
Consequences of WIP Constraints
Note: WIPCAP relative to average WIP for M/M/1/ queue loaded to 90 percent utilization.
25Copyright 2012, Reinertsen & Associates
The Big Idea!
We can get significant cycle timereduction, for a relatively low price,by reducing the amount of time ourprocess spends in high queue states.
But, some practical details remain...
26Copyright 2012, Reinertsen & Associates
SomePracticalities
27Copyright 2012, Reinertsen & Associates
What Should I Constrain?
• The span of the WIP constraint can be:• Local - e.g. Kanban• Regional - e.g. QRM• Global - e.g. TOC
• Larger spans require less total inventory,but they can create local WIP starvation.
• Larger spans can cause feedback delayswhich lead to long loop closure times.
28Copyright 2012, Reinertsen & Associates
Setting the Constraint
• WIP constraints are cheap, effective,incremental, and reversible.
• Some useful heuristics:• Start at 2x average unconstrained WIP.• Drop the limit by 20-30 percent.
• You can always reverse direction.• We can use a combination of WIP
constraints and time-slicing to differentiateservice levels for different workstreams.
29Copyright 2012, Reinertsen & Associates
PULL vs. WIPConstraints
30Copyright 2012, Reinertsen & Associates
Classic Pull• Items are removed from inventory whenever
the customer chooses.• The removal of an item triggers the
production of a replacement item.• This protocol will strictly balance inflows and
outflows and therefore intrinsically maintainsconstant WIP.
• Thus, classic pull entangles two ideas:• A process to trigger replenishment orders• Stabilization of WIP levels
31Copyright 2012, Reinertsen & Associates
Implications of Classic Pull
• Flows are tightly coupled.• The instantaneous rate of replenishment is
equal to the instantaneous rate of removal.• The batch size of replenishment is equal to the
batch size of removal.• This stabilizes inventory as a secondary effect.
Hamburger Chute
32Copyright 2012, Reinertsen & Associates
Classic Push• Classic Push uses forecasts to trigger
removal and replenishment of inventory.• In Push, the balance between inflows and
outflows is neither prescribed, nor prohibited.• Likewise, the synchronization of inflow and
outflow is neither prescribed nor prohibited.• However, unbalanced inflows and outflows
are common, which leads to uncontrolledrandom variation in inventory levels.
• Thus, we could choose to constrain WIP in aPush system.
33Copyright 2012, Reinertsen & Associates
Is Pull Intrinsically Superior?
“Pull in simplest terms means thatno one upstream should produce agood or service until the customerdownstream asks for it…”
- Womack and Jones
34Copyright 2012, Reinertsen & Associates
Jean-Pierre’sBoulangerie
• We could not help noticing that your inventoryof baguettes varies widely between 7 AM and10 AM.
• And, because you are using what are, quitefrankly, medieval methods, you must get upinsanely early to start baking bread.
• We recommend you use customer orders,instead of forecasts, to trigger production.
35Copyright 2012, Reinertsen & Associates
WIP Control
36Copyright 2012, Reinertsen & Associates
Where Pull Fits
Pull WIP Constraints
WIP Control
37Copyright 2012, Reinertsen & Associates
Demand-focusedApproaches
Block Entry
Purge WIP
Redefine theEndpoint
T-ShapedResources
Supply-focusedApproaches
WIP Control
Methods of WIP Control
ResourcePulling
Part-timeResources
FlexibleExperts
SkillOverlap
Mix-focusedApproach
ChangeMix
Toyota’s Kanban Method
38Copyright 2012, Reinertsen & Associates
Finding Better Ideas
• In my opinion, the most advancedWIP control techniques appear intelecommunications systems.
• Let’s take one example.
39Copyright 2012, Reinertsen & Associates
Controlling Internet Flows
Sender Receiver
Packet from Sender
Acknowledgement (ACK) from Receiver
Data Network
Packets + ACKs < Window SizeWindow Size Limits Number of Unacknowledged Packets
40Copyright 2012, Reinertsen & Associates
Congestion Avoidance vs. Control
0
100
Probabilityof
DroppingPacket TCP
RED
REM
RED = Random Early Deletion, REM = Random Early Marking
Quantity in Buffer
41Copyright 2012, Reinertsen & Associates
Contrasting Flow Control Strategies
Toyota Production System• On/Off Flow Control• Static WIP Limits• One Service Level
• Local Feedback Loops
• Static Routing
• Constrains Variability
INTERNET• Progressive Throttling• Dynamic WIP Limits• Multiple Service Levels
• Local and GlobalFeedback Loops
• Dynamic Routing
• Tolerates Variability
Which system is a better fit with the needs of product development?
42Copyright 2012, Reinertsen & Associates
Summary• WIP constraints are cost-effective way to
prevent the accumulation of variability.• Their economic trade-offs favor progressive
tightening.• There are many alternatives to completely
stopping flow when a WIP limit is reached.• Look to telecommunication systems, not
factories, for the most advanced approaches.
1991 / 1997 1997 2009
Going Further
Print + KindlePrint + Kindle
Print Only