Professional Documents
Culture Documents
Lean Manufacturing
March 24, 1999
Jeffrey K. Liker
Principal and Senior Lean Consultant
Optiprise, Inc.
Karl Burr
Vice President
i2 Technologies
Problem
Manufacturing companies throughout the world in many industries are
adopting lean manufacturing methods, a fundamental shift from traditional
mass production. The original model for lean manufacturing is the Toyota
Production System. Toyota runs their system with remarkably little information
technology and relies heavily on simple, visual, manual signals to manage
scheduling and material flow such as kanban cards and strategic buffer
stocks i.e. market places. Yet, in American companies, while lean systems
are being implemented on the shopfloor, in parallel information technology
departments are implementing new information technologies for enterprise
integration and shopfloor control. Thus, the question arises: In what ways can
appropriately applied information technology significantly enhance the
performance of lean systems? In other words, how can we bring together
these parallel activities so they work in concert to drive value?
In this paper we focus in particular on Advanced Planning and Scheduling
(APS) for shopfloor production as an enabler of lean manufacturing.
Advanced Planning and Scheduling systems begin with a representation or
mathematical model that mirrors the actual supply and manufacturing system.
The model can be used to optimize selected parameters of the system as a
whole and can be updated almost instantly as conditions change and new
data is generated. The forerunners to modern APS, like MRP and Finite
Forward Scheduling packages, were used to schedule push systems and
generate schedules down to the level of individual machines and staffing
requirements. This has changed. Modern Advanced Planning and Scheduling
Systems are more sophisticated, simulating pull systems by projecting
consumption by customers. By contrast, lean manufacturing emphasizes that
operations should be directly tied together through pull systems driven by
customer demand.
Our focus on shopfloor advanced planning systems tackles head on the role
of these computerized planning systems in lean manufacturing, as the
shopfloor is the place where lean manufacturing most strongly advocates pull
systems. An analysis of the total supply chain is beyond the scope of this
particular white paper.
Taiichi Ohno realized that material sitting in inventory was wasteno valueadded activity was being performed on it. In fact, Henry Ford had written
about that very idea. And Ohno attacked waste with a vengeance
eliminating warehouses and creating flow of materials whenever possible. For
example, having a set of machines on one side of the plant that built to
inventory which was then moved to the other side of the plant and fed into
another set of machines led to extra handling and material waiting to be
processed. So Ohno would move the machines together into a product layout
(for example, a manufacturing cell) and material would flow. In the process
he discovered that quality improvedsince defects coming from the
preceding process were immediately detected by the succeeding process
before a lot of defects had time to be made. He also found that it was better
to stop and fix defects than to let them build up to be repaired laterthe basis
for the now famous Andon system. Inspection and repair were waste.
Up to this point there is no contradiction with APS as it is applied in traditional
systems. APS seek to optimize total systems and it is up to the analyst and
client to specify the scheduling strategies and optimization goals. If the
objective is to get material to flow through the system very quickly, subject to
real world constraints, APS can create a plan which if followed will maximize
throughput and minimize inventory. Thus the goals of lean manufacturing and
APS can be congruent.
However, lean is more than a set of goals. It is a system of tools, beliefs, and
methodologies. One of Taiichi Ohnos breakthroughs was the recognition that
scheduling individual machines led to piles of inventory. In principle if
operation A and operation B are sequential processes, and each makes just
what is scheduled exactly when it is scheduled, and the schedule is a good
one, material can flow through the plant with little buildup of inventory. But
Ohno observed from experience that focusing on planning individual
operations with the goal of product coming together where it is needed and
when it is needed, leads to waste. In fact, individual operations are decoupled
from their immediate customers and build what they expect the next operation
will need, which is generally different from what the next operation really
needs. This is because schedules always change and schedules are never
perfectly followed in the very dynamic environments of manufacturing plants.
Thus, inventory builds up between operations and we are back where we
startedpush production.
While building to customer demand is the ideal, Ohno learned that, when
operating with little inventory, if a downstream operation suddenly withdraws a
lot of one product it could overwhelm the capability of the upstream operation
to replenish what was taken away from the supermarket, unless there was a
very large amount of inventory in the supermarket. To minimize the amount
of inventory held in the supermarket the consuming operation needs to
withdraw at a stable, leveled pace. So TPS argues that a condition for using
pull and keeping inventory down is to build at a stable, levelized pace. This
means that instead of building exactly in the sequence the customer
consumes product, actual customer demand plus forecasted demand should
be spread over some time horizon and manufacturing should build to that
leveled demand. Thus, even in the Toyota Production System there is a
place for a schedule and for forecasted demand.
In sum, central tenets of lean manufacturing include:
Takt time and Continuous FlowAll operations should ideally build at the
pace of customer demand. Continuous flow is the ideal, building one
piece at a time, which tends to minimize waste, with all operations
building to takt time. With a pure one piece flow only the first operation in
the continuous flow needs to be scheduled and all other operations follow
sequentially. Takt time is the pacemaker for the continuous flow.
Pull systems should be used when continuous flow is not feasible. In this
case a small buffer (supermarket) is set up between operations and the
feeder operation replenishes what is taken away by the downstream
operation. Ideally, only the final operation (beginning of final continuous
flow) is scheduled and then all upstream processes build to replenish
what has been consumed by their immediate customer.
Production leveling--While ideally the lean system would build only what
the customer needs exactly as they need it, in reality customer demand is
not level. In a multi-product environment, an uneven demand (e.g., a
sudden surge in demand for one of the products) makes it difficult to
service that demand unless there is a large inventory of all end products.
This surge in demand is particularly disruptive for upstream suppliers (i.e.,
the bull whip effect). Lean manufacturing deals with this through
heijunka, i.e., leveling demand by creating an inventory buffer and
replenishing that buffer using a leveled schedule.
LEAD
TIME
Typically there is one kanban per container and the container size is then the
size of the order quantity. Thus, adding a kanban means adding an additional
container of parts, that is, work-in-process, in the system. We can see from
this equation that:
The greater the takt time the greater the number of kanban. Takt time will
vary across different end products which will require different parts. For
example, if manual mirrors represent 30% of automobiles and 70% are
10
automatic, there are different takt times for manual and automatic mirrors
and there should be more kanban for the automatic mirrors.
The greater the lead time the greater the number of kanban. Note that
the lead time, which is the time required to replenish the marketplace
when a kanban has been sent back, includes production time and
conveyance time. Both of these will vary across components and thus
different number of kanban are needed depending on lead time.
The point here is that including the number of containers of product needed
for two hours of production is a highly simplistic decision rule. And once
made it is not likely to be adjusted often, except for periodically when perhaps
someone says, we are doing well, so lets see if we can get down to 1.5
hours. A better approach is to use data to optimally derive the appropriate
reorder point and number of kanban and update these calculations on a
regular basis. This is something APS is very good at maintaining a plan
instead of recreating and re-assembling the data each time. And APS can
look more broadly than between a feeder operation and the consuming
operation to consider stability in supply through the value chain to identify
appropriate marketplace sizes and kanban quantities.
As an actual example, in an automotive axle plant there were many
processes needed to make a wide variety of rear wheel drive axles for many
different light trucks. Ultimately there are well over 100 combinations of axles
produced in this plant on multiple assembly lines. The plant machines ring
gears and pinion gears which must ultimately mate. The ring gears alone go
through ten major processes prior to assembly. Between each process there
is a large inventory buffer of up to 60 hours worth of inventory. Material has
traditionally been moved on large pallets by forklift between each operation.
Material can spend weeks in the system. The plant is in the early stages of
moving to lean manufacturing and is switching over from the pallets to carts
on wheels that hold smaller quantities of material. The vision is to use these
carts as a kanban system, paint a certain number of squares on the floor, and
only build enough to replenish marketplaces of carts. Each stage in the
production process makes multiple types of gears. Some are dedicated to
one size but there are multiple gear ratios while other machines are used for
all sizes and gear ratios.
11
The lean manufacturing support group in the plant is not sure how to size the
marketplace. They are not aware of what calculations are necessary. While
they could get a formula like we presented and calculate the appropriate
quantities they would have to do this for each stage in the process and for
each product meaning dozens of such calculations. Once these calculations
were made they are not apt to recalculate these quantities as conditions
change, something APS can easily do. APS also can help them analyze and
determine where they could get the most leverage in improving the system
using gaming techniques or what-if scenario development , e.g., where to
focus efforts on creating continuous flow or reducing changeover times.
An illustration of the value of APS for planning purposes can be seen by
walking through Figures 1-3. Figure 1 shows an empty kanban board with
hooks for hanging kanban. Each column is set up for kanban for that part.
Shown in parentheses are the maximum number of kanban that have been
allocated and thus the maximum number of containers of parts in the system
at any point in time. A material handler will pick up cards from the consuming
operation which represent containers of parts that have are being used in
production and bring them to the producing operation. The material handler
will fill the board by placing the Kanban on available hooks from the bottom
up. The Green zone indicates no rush to make these parts. Cards in the
yellow zone are higher priority and cards in the red zone mean the
downstream operation may be starved for parts unless you build these right
away. For example, in Figure 1 there are 2 hooks in the red zone for Part A
so someone decided two bins is the minimum reorder quantity.
F ig u r e 1 :E x a m p le K a n b a n S e q u e n c e B o a r d
P a r t A (1 0 )
P a r t B (7 )
P a r t C (2 1 )
Red
. .
Y e llo w
.. . .
. . .
G re e n
.. . .
. . .
...
. .. .
. .. .
. .. ..
. ....
N O T E : T h is e m p ty k a n b a n b o a r d h a s h o o k s f o r h a n g in g k a n b a n . E a c h c o lu m n is s e t u p f o r k a n b a n f o r th a t
p a r t. S h o w n in p a re n th e s e s a re th e m a x im u m n u m b e r o f k a n b a n th a t h a v e b e e n a llo c a te d . T h e b o a rd is
f ille d a s th e d o w n s tr e a m o p e r a tio n c o n su m e s th e p r o d u c t m a d e a n d k a n b a n a re re tu r n e d f o r m o r e
p ro d u c tio n . K a n b a n a re p la c e d o n a v a ila b le h o o k s fr o m th e b o tto m u p . T h e G re e n z o n e in d ic a te s n o r u s h to
m a k e th e s e p a rts . C a r d s in th e y e llo w z o n e a r e h ig h e r p r io r ity a n d c a r d s in th e re d z o n e m e a n th e
d o w n s tre a m o p e r a tio n m a y b e s ta rv e d f o r p a r ts u n le s s y o u b u ild th e se r ig h t a w a y .
12
Part B (7)
Red
. .
Yellow
....
Green
....
Part C (21)
...
. . . .. .. .. ..
. . . .. .. .. .. ..
Part currently running
Part ready for next set-up
Figure 2 shows a board loaded with kanban that have been coming back from
the consuming operation (e.g., assembly). In this case we have color coded
the Part A segment of the board with green indicating the cards for Part A are
almost into the red zone and production should be set up to make those parts
next. By contrast, there are only two cards back for Part C out of a total
possible 21 cards so there is plenty of Part C in the supermarket and that is
the lowest priority for production. It should be clear that this simple manual
system is literally the schedule.
Figure 3 illustrates that when conditions change, e.g., customer demand rates
(volume or mix or both) change or perhaps there are problems with parts or
machines which require more safety stock for some parts, cards should be
added or subtracted from circulation. In this case conditions changed and the
APS systems calculated cards should be reallocated. For example, four more
kanban should be added for Part A to increase the maximum WIP of Part A.
These will be allocated by adding two hooks to the green zone and one each
to the yellow and red zones. APS recommended the change in kanban and
can also recommended how to allocate the kanban (e.g., what should be the
minimum trigger point). The board and cards can be a very powerful, visual
control system which is loaded manuallybut the optimized planning for this
can be done by computer almost instantly.
13
. .
.
.. . . . . .
.. . . . . .
.. .
.. ..
.. ..
.. ..
.... ..
Red
Key: Red dots=kanban hooks added to board
Gray dots=kanban hooks taken away
Yellow
Green
. .. ..
. .. . . . . .
.... .. . . .
.. .
.. .
.. ...
.. .
.... ...
Production LevelingIt is not accurate to say Toyota does not schedule its
system but uses pure pull. Nor does Toyota build exactly what the customer
wants when we want it. In its pure form this would mean building exactly to
customer demand without consolidating orders in any way. In practice this
would mean orders would be very uneven with respect to mix and volume.
Orders are to a degree random and random means at times you will get a
sudden rash of red vehicle orders or a rash of vehicles with manual mirrors. If
the assembly plant built directly to these orders they might quickly use of their
red paint supplies and manual mirror supplies in the marketplace. They would
then send an urgent message to the supplier to ship more, which may well
outstrip the small amount of inventory in the marketplace of the supplier. That
supplier will soon run out of unique components for the manual mirror from
their suppliers.
In fact, the ripple effect through the supply chain gets amplified at each stage
backward in the processreferred to as the bull whip effect. With a small
flick of the wrist at the handle end, a bull whip creates a large movement with
great destructive force at the end of the whip. Many studies using
mathematical modeling or simulation have demonstrated how small deviations
from the production schedule in the final assembly plant can lead to very large
amounts of inventory held by upstream suppliers to protect themselves from
the amplified impact on them.
14
It has long been recognized by Toyota that a lean system with little inventory
depends on having a level schedule:
The smoothing of production is the most important condition for production by
Kanban and for minimizing idle time in regard to manpower, equipment, and
work-in-process. Production smoothing is the cornerstone of the Toyota
Production System. (Toyota Production System by Yasuhiro Monden, 1998)
Toyota represents TPS as a house and often draws it with production
smoothing, referred to as heijunka, as the foundation. Why would heijunka be
so critical? Because the entire production process and supply chain have
been finely balanced to takt time which assumes a certain mix of product.
When all the marketplaces and kanban quantities are sized to this takt time, a
wild swing in demand for a particular product cannot be handled by the
kanban. Recall that with kanban each stage of the process, e.g., the mirror
plant, only needs to know what has been taken away from their marketplace
to schedule production. They look and replenish what is taken away. A
sudden surge in demand, like a rash of orders for manual mirrors, will quickly
empty out the marketplace and overwhelm the ability of the production
operation to replenish it. Besides there are not enough kanban cards to place
on all the containers of material needed to fill a very large order.
While there has been a lot of discussion in the bull whip effect literature on
using current and accurate information to mitigate its effects, Toyotas solution
is to level production at each stage of the process and develop stable
manufacturing processes that can build to the leveled schedule. The
assembly plant uses a leveling algorithm to take the demand (actual orders +
forecast) and create a levelized sequence to spread out all variations of
product across the day so the producers of components for that product see a
level stream of orders coming to them. As a general rule they assume the
suppliers should plan for fluctuations of +/- 10% deviations from the levelized
schedule in a given day. Toyota commits to keeping their schedule within
those parameters and goes to extremes to build the mix and volume of
product committed to by the end of the day.
Suppliers also have a responsibility to keep their schedules level so their
suppliers do not see dramatic fluctuations in demand. Most suppliers do not
build directly to a truck, but generally have a finished goods marketplace as
well as a safety stock of product as a buffer to changes in their customers
demand. They then can build at a levelized pace to replenish the marketplace
and the safety stock (if needed).
15
The box is used for scheduling production of multiple end products coming off one line. Material
handlers pick up the cards (kanban or a shipping label) in the verticle slots at the times indicated
across the top. The cards are sorted into the slots to level (spread out) the building of product types
and set the pace of production. The box above ends at 10:30 but would normally cover the work day.
A card is picked up every 15 minutes and delivered to manufacturing.
16
Production
Signal
Withdrawal
Kanban
Part 8:00 8:15 8:30 8:45 9:00 9:15 9:30 9:45 10:00 10:15 10:30
Part #
KJ-467
To Shipping
Dock
Pull Parts
to
Assembly
Gravity fed
Store
Stop Producing
Must Replenish
When the process is stable and the operators and material handlers have the
discipline to execute it as planned it is a thing of beauty in its simplicity. The
level-loading box becomes the pacemakerthe heart beat of the system. So
why would we want to add information technology to such a simple and visual
system? Again, if the system is stable enough with a small number of end
products and the heijunka box can work then it is a great execution system
and need not be replaced by IT. But there are a number of reasons why APS
behind the scenes can add greatly to heijunka:
17
"
"
18
One of the pioneers in the United States in promoting JIT systems and
1
explaining how they operate is Robert Hall. He lists a number of conditions
that make pure pull systems very difficult to implement effectively:
1. Despite stable demand, the assembly of the final product cannot be
executed in a level enough fashion to provide steady demand for
upstream operations. This can happen when the assembly plant is not
stable or when the product inherently takes different amounts of time with
every unit (e.g., there is testing involved that takes different amounts of
time per unit).
2. Some operations must be started in advance of pull signals. This
happens when operations require special, lengthy, or difficult setups
which cannot be simplified or significantly shortened and must be
scheduled in advance.
3. The product is made in so many options, and the demand for each option
is so small or unstable that it is impractical to carry buffer stocks for all
parts everywhere in the process. This is the case of mass
customization and is found in industries like furniture where each order is
customized and assembly builds only complete orders regardless of the
mix within it.
4. The high defect level causes too many interruptions to permit continuous
flow, and the state of the technology is such that the defect level cannot
be reduced significantly.
5. Products must be produced as integrated batches throughout the process
for reasons of quality control or certification. An example of this is
pharmaceuticals.
When any of these conditions or a combination of them apply, it will probably
be necessary to schedule in addition to pull. There is no reason why it is
necessary to choose one or anothermixed models are quite possible. For
example, a schedule can be developed for purposes of having material and
people available but then some type of visual signal can actually be sent from
downstream operations to start production.
19
20
Lean Execution
Systems
Advanced Planning
System for Lean
Part
Demand Planning
Levelized Sequence
Kanban Quantities
Supermarket Sizes
Visibility through
supply chain
8:00
8:15
8:30
8:45
9:00 9:15
9:30
9:45
Load-leveling systems
Mixed-model production
Continuous flow processing
Pull systems
lean supply chain
management
21
22
Model
Pure Optimal
Scheduling
What is it?
Optimally plan
and schedule all
work centers and
operations. Each
operation builds
to the schedule.
Benefits
Demand-driven system
will develop globally
optimal plans.
Limitations
Assumes all operations
and plants in the supply
chain execute the
schedule. Lack of a
response to unplanned
deviations might cause
inventory buildups.
However, APS can act as
a fast decision aid in case
of such events.
Transition to Lean
Lean phased in to
product lines.
Scheduled
operations that
have not yet
transitioned to pull
systems and
modeling to help
prioritize lean
initiatives.
Leveled schedule
for final assembly
and then pull from
all upstream
operations (with
supermarket
buffers).
Pure Lean
Operations leadership
might not pursue the most
aggressive and
comprehensive path for a
full lean transformation.
Where Appropriate?
Non-level customer demand
(>10% fluctuation from planned
schedule)
Many products that require shared
resources
Optimized sequence of special,
lengthy setups required.
Low product yield (e.g., complex
paint colors)
Highly customized products with
small demand for each option (i.e.,
mass customization).
Some products or phases of the
manufacturing process have the
characteristics appropriate for optimal
scheduling (see above) and others
have the characteristics that support
lean manufacturing.
Any plant that has not yet fully
implemented continuous flow and pull
systems but is heading in that direction.
Role of APS
Develop global optimal schedule
and dynamically size buffers.
Provide advanced warning and
visibility to problems.
Provide direct link to
procurement outside 4 walls of
the plant.
Provide customer and release
due date visibility throughout the
process.