Professional Documents
Culture Documents
Taichi Ohno describes the pull system as being one in which "the
downstream operation comes to take away the materials," as opposed to
having upstream operation bringing them unsolicited. This conjures up the
image of the output buffer of the upstream operation being like a retail
store where the downstream operation comes to help itself. While is a
powerful image, it should not be taken to mean that an operator from the
downstream line leaves it to go fetch parts. Between two lines, the
issuance of a pull signal causes a materials handler to move the parts in
a process that is more akin to mail order than to walking to the nearest
convenience store.
Holding parts in the output buffer of the operation they just completed
hardly seems like the way to get them out the door faster. On the
contrary, moving them as soon as they are done to the next operation
appears to be the way to expedite their completion. 'You can't get
anything to move" is the most common complaint of production managers
about the materials organization.
But if the next operation is not ready for the parts, the only effects Of
moving them are as follows:
Management does not want forklifts in the Production area, and therefore
production operators need to bring the full pallets to a location where
forklifts are allowed in addition, the output area of the upstream
operation used to stage empty containers and pallets.
Pallets jacks cannot store and retrieve pallets in the upper levels of a
pallet rack
The upstream operation has longer setup times than the downstream
operation, and therefore longer production runs. The production run is,
say, six bins upstream and bin downstream. The pallet rack is needed to
hold at least one upstream production run of every item.
The materials handler has the time to do it and the space is available.
The space next to the upstream operation is no longer used to store empty
addition to the one currently being filled. The output buffer shown in
figure 10-2 has parallel flow racks, each of which can hold one
production run Of one item, and the end of the rack is sufficiently low
for a pallet to be picked up with a pallet jack. Setup time reduction
(SMED) efforts at.
the upstream operation have made if possible to cut the of the size of
the production runs in half to three bins.
The content of the output buffer is an order of magnitude lower than the
content Of the staging area and the pallet rack Of figure 10-1, and, as a
result, the throughput time of the material is reduced from days to
hours. It is possible because the pull signals synchronize production
between the upstream and downstream operations.
In Figure 10-2, we can see why the output of the upstream operation must
be held next to it until pulled: it provides visibility for the operator
When the output is held in a separate staging area, as in Figure 10-1,
the operator Of the upstream operation loses sight Of the downstream
operations comsumption rate. If, for anv reason, the downstream operation
falls behind, the upstream operation is likely to overproduce until the
staging area and the pallet rack overflow.
Because the amount Of WIP between the two operations is limited to the
output buffer of the upstream operation, it does not require a multilevel
pallet rack, and therefore it needs neither a forklift nor a driver. Note
also that, by contrast with Figure 10-1, in Figure 10-2, the pallet jack
is pushed by a materials handler instead of the production operator,
which allows the production operator to focus on production, as he or she
should.
The downstream operation can communicate its needs upstream in many ways.
An approach that predates lean manufacturing is the reorder point system,
in which a replenishment request is issued when the inventory on hand
crosses a lower limit. In the reorder point system, the quantity
requested is intended to bring the inventory level back up to a maximum
quantity.
Where these limitations are not a problem, variants of the reorder point
system are used in lean manufacturing, but the more common process of
issuing pull signals over one shift is as shown in Figure 10-4. Each pull
signal is for a fixed quantity, and is issued whenever the cumulative
consumption crosses the corresponding levels.
The flow of signals provides the upstream operation with visibility into
the status of the downstream operation. As we shall see in the coming
chapters, the generation and timely delivery of these signals is the most
difficult challenge in making a pull system work.
For mixed-flow lines, Figure 10-5 shows the simple example of two
products, with pull signals that are issued continuously during each run.
If the production runs are longer than the replenishment lead times, then
the naive forecast at the end of the run for product P1 is wrong, and
parts for P1 are delivered during the beginning of the run for product
P2. This effect is minimal when the replenishment lead time is very short
compared to the production run, and worst when they are equal. On the
other hand, the shorter production runs are compared to the delivery lead
times, the better the forecast and the smoother the delivery pattern.
As shown in Figure 10-5, the pull system determines how pull signals are
issued and their fulfillment lead time. There are formulas for
calculating the number of signals to put into circulation. The idea
behind these formulas is to calculate the number of signals that would be
needed if production and logistics worked perfectly, and apply a safety
factor to compensate for their failure to do 50.
The formulas are based on a very general principle called Little's law,
illustrated in Figure 10-6. Littles law is obvious when the pace of work
is constant. Blanks arrive at the plant at a constant rate, stay in for
one lead time, and ship out. The difference between the cumulative
numbers in and out is the number of blanks that have come in but not yet
gotten out. In Other words, it is the total inventory inside the plant.
Since the slope of either of the cumulative lines is the throughput,
then.
The point of Little's Law is that the formula remains valid about
averages
supplier to customer, lead times vary, and finding the right lead drne
num-
makes lead times predictable. Inside the plant, a train of carts may
deliver an
item every 20 minutes, while a truck may bring parts from 5 suppliers
runs, as shown in Figure 10-7. Clocks on the shop floor display the cur-
rent pitch number instead of the regular time, replenishment lead mes
arg
instead of the familiar hours and minutes, and the organization ofbteaks
is
Boa challenge. For example, if you want pitches to represent only worked
then you could stop the pitch clock during breaks, but then it
the entire plant to take breaks at the same timc or else different
Using pitches, law, and a safety factor, wc can calculate the number
start out with enough signals to support tie usually excessive quantity
of
gnals one by one until reaching a lin.t below which shortages begin to
the pitch demand is allowed to vary randomly by adding the I O-pitch mov-
ing average of a white notse to the average for every pitch. The demand
is
lead time is fixed, so that the end-on-hand Of every pitch is the quanuty
that vas present at the begtnrung, decremented by what Vas shipped, and
and "warms up" through the first few pitches until it reaches a steady
state, which is what we are interested in. It shows the end-on-hand quan-
demand of the past two weeks could be used if it is available. Other pat-
A fixed demand.
One caveat about this approach is that simulation is addictive and can
become a distraction.
mechanisms
"Pull," write James Womack 1 and Dan Jones. "t....l means that no one
stxeam asks for it." This defini60n adds the concepts Of service and
customer
ing final assembly with the goal of smoothing the work flow of
The next process does not have the option to get the parts from
another supplier.
Control. The point of it was to say the next process, or production line,
aphor has become popular to the point of being taken literally, and a
such freeaom.
In fact the flow between stores, cells, feeder lines, and final assembly
inside the plant, where the term "customer" only applies metaphori-
cailv, is rhe primary application area of the pull system. Only when pull
signals are used with outside suppliers does the term apply literally,
and
it imposes constraints on the types of signals used that are not present
In general, the money a customer pays for goods is a form of pull signal,
albeit an excessively generic one: it tells the seller that something has
moved,
but it doesn't say what. Before its sales system was entirely
computerized in
the mid 1980s, the campus bookstore at Stanford University kept an iden-
removed rhe cards from the books and placed them in a colle&n box'
issze sizaais which tricger other tasks. The move to a pull sys-
;tFess Ene- There these lines cross, there must be sidings allowing
on the left
Where applicable, pull systems are effective, because local decisions are
tent with global business needs. Tne concept is simple; making it work OY
storage areas and production lines, and between supplier and customer
plants.
The concept of a pull system is not relevant between two operations
inside
moves through like a train on rails all the way to the end. Pull signals,
on
the other hand, can be used to decide which parts to release to the first
station.
Pull signals are usually not used alone, but in conjunction with other
sys-
tems that forecast activity and generate explicit sequences and schedules
fot pacemaker lines, most commonly final assembly, from which pull
signals are issued that drive the rest of the plant. Explicit sequencing
and
I. line stops for setups and following the raw sequence Of pull Signals
would require so many setups that the line would not have the capacity
to keep up.
mix that are better smoothed out before passing the flow of signals on
signals and the protocols for using them must do the job of imple-
can make them. Setting them up may require hald thinking on the part Of
crossroad with 4-way stops on U.S. .roads, as shown in Figure 10-12. The
rule is that the car that has been waiting the longest from any direction
gets to cross next. "fris is as complex a protocol as human beings in
large
numbers can use, and the only reason it works is that the stakes are
high,
as it is, there are many countries that find it too dangerous to use.
Imagine what would happen if it were made more complex, for example as
follows:
If there is mole than one Mercedes waiting, then one that has been
3. If there are no Mercedes then the first car in the direction with most
cars waiting goes first.
It is not difficult to imagine how any attempt to use this protocol would
fnl, simply because the people involved would get confused or balk at its
rules that are easy to follow for operators, materials handlers, or truck
drivers, and applied consistently. The signals must be easy to see and
understand, and they must ajso have the sane meaning wherever used
icy, Erst comes Ee use of devices that are already on the shop floor and
are
comes the use of physical tokens that circulate along the loop, bearing
var.
ious amounts of data. Finally, comes the use 'of computer, communica-
Chapter Il, which covers manual pull signals, and Chapter 12, which cov_