Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

THE EFFECT OF WORK-IN-PROCESS

INVENTORY LEVELS ON THROUGHPUT


AND LEAD TIMES

RICHARD CPIM
E. CRANDALL,
TIMOTHY PHD
H. BURWELL,
College of Business, Appalachian State University, Boone, NC 28608

Historically, manufacturing companies have oper-


stability [31, reduced product complexity [1, 2] and
ated with work-in-process (WIP) inventories.Justifi-
reduced process complexity [2, 5]. Consequently,
cation for WIP has been that it provided buffers the relationship between reduced inventories and
be-
tween operations which helped to prevent the shut- throughput has not been clearly defined.
down of a machine for lack of work and, as a corollary,
Blackstone[21 points out this potential problem
to maximize the utilization of equipment capacity or when reporting on the results of a simulation study in
to otherwise improve other measures of manufacturing which he compared the mean times to complete 500
efficiency. The rationale was that the cost of having units under varying process times. He concludes, "It
idle capacity was greater than the cost of carrying the is apparent that as variabilityof job times increases,
inventory. The inventory buffers also tended to reduce the output of a line with little WIP decreases dramat-
the need to formally balance the loads in the process ically." Later, in discussing the importance of quality
steps. In recent years, particularly with the advent of in JIT applications, he says: "Successful JIT imple-
the Just-in-Time (JIT) philosophy, the trade-off be- mentation requires that variations in job times be re-
tween the cost of idle capacity and the cost of carrying duced in order to avoid slowdowns on the entire line
inventory has come under closer scrutiny. A number . very little has been done explicitlyon the rela-
of writers claim that the WIP increases the lead times tionship between variance, WIP, and capacity utili-
(time in system) and reduces the flexibilityof the zation. This relationship needs to be researched in or-
manufacturer in meeting customer demands. They der to assist the design of JIT lines."
believe the objective should be to minimize the WIP,
thereby also minimizing lead times. [6, 7, 8, 91. PURPOSE OF STUDY
Consequently, JIT programs have the reduction of
WIP inventories as one of the major objectives.The A study was designed to examine the effect of re-
general approach suggested is to arbitrarily reduce the duced WIP on throughput, lead time (average time in
allowable inventory and then identify and correctthe system), and utilization (ratio of time worked to time
problem(s) created by this inventory reduction. Once available). Two primary questions were considered:
the problems have been corrected and the system is • Can the reduction of WIP cause a decrease in
operating satisfactorily at the reduced inventory level, throughput? Under what conditions?
the inventory-reduction-problem correction cycle is • Can increased throughput be achieved even as WIP
repeated. This process is continued until a minimum is decreased? How?
level of inventory is reached. The result of reduced If reduced WIP leads to reduced throughput, the
inventory is expected to be reduced lead Hmesand problem then becomes one of evaluating trade-offs
greater flexibility in responding to customer demands. between the costs of reduced throughput and the ben-
A possible negative consequencefrom reducing in- efits of reduced lead times. Obviously, the ideal situ-
ventory is that of reduced throughput (in this article, ation is the combination of reduced WIP, reduced lead
defined as the number of units processedwithin a times, and increased throughput.
given time period). In some casesof JIT implemen-
tation, throughput has increased, thereby suggesting
DESCRIPTION OF MODEL
that reduced inventories can lead to increased
throughput [2 J. However, in most cases, JIT programs The manufacturing process modeled is that of a flow
also call for other changes, such as increased schedule shop where product is processed through a fixed se-

6 PRODUCTION AND INVENTORY MANAGEMENTJOURNAL—FirstQuarter, 1993


quence of operations/ machines/ departments. The also with a mean of 50 time units and a coefficient of
machines are not directly linked such as in a contin- variation of 1.000,is used as an alternative distribution
uous process; therefore, inventory buffers can be ves- to simulate the existence of a few "specials" with ex-
tablished between each machine.
traordinarilylong processing times and many standard
In all alternatives, several assumptions are made so items with relatively short processing times. A constant
that the process conditions are very close to those de- processing time (zero variation) is used as a theoretical
scribed as ideal for a JIT environment. reference point.
I. The processing sequence of first-come, first-served Queue capacity. Four different levels of queue ca-
is used throughout the study. pacity are used: 50 units, 10 units, five units and one
2. Setup times are zero. This assumption precludes unit. Limiting queue capacity causes blocking of ma-
the need for batching, so lot sizes can be equal chines, which imposes the discipline of a "pull" system
to one. as queue capacities approach one. Queue capacity can
3. There is a constant supply of product at the first be considered as performing a function similar to the
machine so that the downstream machines are number of kanbans allowed in a JIT environment. If
never idle for lack of work in the total system. the kanbans are reduced, the average WIP decreases.
4. There is a constant demand for finished goods so The effect of reducing the queue capacity is the same
that the last machine is never blocked. as reducing the number of kanbans.
5. The mean processing time for each operation is the In summary, the model contains many of the con-
same so that, in theory, the result is a perfectly ditions usually specified for a JIT operation, such as
balanced line. constant loading, minimal setup time, lot size of one
The following variables are considered: unit, pull system at low queue capacities,and no major
Number of machines. The number of machines in machine breakdowns. The major variables to be con-
the sequence represents a measure of process com- sidered are product and process variability. As pointed
plexity. In the study, runs are made with four, six, and out by Crawford and Cox [41, "JIT works because its
eight machines to see the effect on throughputand elements are designed to continuouslyidentify and
lead time. eliminate causes of process variation. "
The combination of the above variables resulted in
Processing time variance. The machine processing
the selection of 18 combinations (six levels of pro-
time is a key variable in the study. All machinesare
cessingtime distribution times three lengths of lines)
assumed to have the same mean processing time and
of product/ process variability for each of four levels
the same processing distribution.The processing mes of queue capacity.
are varied to simulate the effects of different levels of In addition, the simulation model provided status
product or process variation. Product variation can be ( % operating), block ( % of time blocked), and queue
the result of a range of sizes, customizing featuresre- (average units of inventory) for each machine.
quired, number of process steps, and quality require- Measures were taken to achieve steady-state status
ments. Process variation can be caused by variances in the simulation runs, and appropriate statistical tests
in machines, tooling, setup adjustments, operators, were used to confirm mean differences between
materials, and yields. In this study, no attempt was groups. The results of the simulation are based on 30
made to distinguish among the causes of variation. runs for each of the combination of input variables
Type of process distribution. Two differenttypes described above. At the beginning of each run, all
of distribution are used to create processing time vari- queues were cleared and each station was idle. Each
ation: uniform and exponential.The uniform distri- run had a length of 20,000 time units.
bution is used to provide a closed-end form of distri-
bution, one that suggests that the product line is fairly
RESULTS
standard and does not contain "specials" that could
require greatly extended periods of Hme. Four different The primary output measures are average WIP,
uniform distributions are used: (45, 55 ), (40, 60), (30, throughput, lead time, and utilization. The results for
70), and (10, 90) to show the range from very little each simulation run are shown in the Appendix. Fig-
variation to wide variation, around the mean of 50 ures 1 through 4 show the results for the six-station
time units per unit produced. The coefficients of vari- line for each processing time distribution type. Even
ation (the square root of the variance divided by the though these graphs are for a six-station line, the same
mean) for these distributions are 0.058, 0.115, 0.231 relationships are observed in the four- and eight-sta-
and 0.462, respectively. The exponential distribution, tion lines.
THE EFFECTOF WIP ON THROUGHPUTAND LEAD TIMES 7
4

270 2

o 10
so 1

FIGURE 1: Average throughputvs. queue capacity FIGURE 3: Average WIP inventoryvs. queue capacity

Throughput same processing distribution, say uniform (40, 60),


a four-station line has greater throughput than an
Increasing, or at least maintaining,the level of eight-station line at any queue capacity.
throughput is considered to be a desired objective. The above suggests that reducing the average WIP can
• Throughput decreases as queue capacity decreases. cause adverse effects on throughput unless product/
This relationship is observed for all distributions and process variability and process complexity are reduced.
is more dramatic for distributions with a higher de-
gree of variability. Lead Time
• Throughput decreasesas product/ processvariation
increases. For example, the throughput for a uniform There is a direct relationship between lead time and
(45, 55) distribution is higher than a uniform (10, work-in-process inventory, in that lead time is the sum
90) distribution at all queue capacities. of the processing time and the queue, or waiting, time.
• Throughput decreases as the process complexity The following equation closely approximates the re-
(number of stations) increases. Hence, given the sults for lead time obtained in the simulation runs.

2.4 1.03
2.3
2.2
2 0.93

0.0

0.83

o.e

a. 73

0.7
0.9
o.ø
0.7 0.65

0.6
0.4
0.3 so
0.55
1 10 s 10 so

a COSTANT + O 70)
X V DÄ<SO)

FIGURE 2: Average time in system vs. queue capacity FIGURE 4: Average utilizationvs. queue capacity

8 PRODUCTION AND INVENTORYMANAGEMENTJOURNAL—FirstQuarter, 1993


Lead time (LT) one) in the entire system decreases as the number
—processing time (PT) + queue time (QT) of stations decrease. For instance, with the uniform
(40, 60) distribution, the average WIP when queue
where capacity is 10 is 1.20, .98, and .88 units, respectively,
PT = is processing time per station times number of for four-, six-, and eight-station lines, respectively.
stations The average total WIP is 3.60 (1.20 times 3) units
QT = is processing time per station times average WIP for a four-station line, 4.90 (.98 times 5) units for
times (n —1) number of stations (first station the six-station line, and 6.16 (.88 times 7) units for
has no WIP). an eight-station line.
Example: (6 stations, uniform (40, 60) distribution,
queue capacity of 10 AverageUtilization
LT = (50 X 6) + (50 X 0.98X 5) Average utilization follows the same pattern as
= 300 + 238 = 538 time units. throughputin that it decreases as product/ process
variation increases, queue capacity decreases, and the
This is identical to the simulation result; however, the number of stations increase.
calculated LT may vary slightly from the simulation
result under different conditions. DISCUSSION OF RESULTS
As a result of the relationship described above, the
following positive results can be expected. For the sake of discussion, let's assume that we have
• Lead time decreases as queue capacity decreases. a six-station operation with exponential (50) process-
This relationship is observed for all distributionsand ing times. Hence, the curves in Figures 1 through 4
is more dramatic for dish-ibutionswith a higher de- labeled Expon(50) show the relationship between the
gree of variability. performance measures and queue capacity. By ob-
• Lead time decreases as product/ process variability serving these curves, the first point that can be estab-
decreases. As an example, average lead time for an lished is that the advocates of conventional manufac-
exponential (50) is higher than a uniform (45, 55) turing thinking and JIT are both correct, to a point.
The advocates of JIT are correct that average lead
at all queue capacities.
time and average WIP inventory will decrease as we
• Lead time decreases as process complexitydecreases.
reduce allowable inventory levels. This follows since
Hence, given the same processing distribution, a
these respective curves drop in value as queue capacity
four-station line has a smaller lead time than an is decreased.However, conventional manufacturing
eight-station line at any queue capacity.
thinking is also correct since both average throughput
Clearly, actions that reduce average work-in-process and average utilization drop as allowable inventory
inventory will reduce lead times. levels are decreased. These reductions in performance
measures are least in those situations with the smallest
Average WIP time variations at each step (i.e., uniform (45, 55))
and greatest in those situations when the process time
Reducing lead times is a desirable objective that de- variation at each step is the greatest (in this case, the
pends on reducing WIP; therefore, reducing WIP is an exponential form).
important action to be taken. Suppose in the six-station, exponential processing
• Average WIP decreasesas queue capacity decreases. time operation that the processing time variability is
This relationship is observed for all distributions with reduced as the allowable inventory levels at each sta-
a higher degree of variability.However, increases tion are also reduced. In terms of our graphs of the
in queue capacity do not necessarilyresult in in- performancemeasures, this is the same as moving to
creased WIP. This occurs when the product/ process points on adjacent curves as we reduce process Hme
variation is so small that the extra queue capacity is variability. In particular, let's assume that we start with
not needed. exponential processing times at a queue capacity of 50
• Average WIP decreasesas product/processvaria- units. We are able to reduce the processing time vari-
tion decreases. ability to that of a uniform ( 10, 90) distribution and
• Average WIP for each work station increases as the drop allowable inventory levels to ten units. For each
number of work stations decrease. However, the successive drop in inventory level we are able to reduce
average total WIP (number of work stations minus processing time variability to that of a uniform distri-
9
THE EFFECTOF WIP ON THROUGHPUTAND LEADTIMES
1
bution with a mean 50 and smaller variance. In Figures accept reduced throughput to provide greater mar-
1 through 4, we illustrate the resulting changes in the keting flexibility.On the other hand, there may be
performance measures with a dashed line. The impli- circumstances where maintaining high levels of
cation of these figures is that we can actually achieve throughput and utilization are paramount. In this cir-
increased throughput and utilization in addition to cumstance, sufficient (not excessive) WIP inventories
smaller lead times and WIP inventories while reducing would compensate for the variability associated with
allowable inventory levels if we address the issue of the production process.
process variability. This is consistent with some of the
actual results observed when JIT concepts were em-
SUMMARY
ployed [21.
It follows that the approach suggested by JIT pro- The results of this study should help explain why
grams to first reduce inventory as a means of identi- both conventional and JIT thinking about WIP inven-
fying the problems in a production process may not tory have some points of merit. If there is large product
be successful (except where WIP is excessive) unless or process variability, some WIP is required to prevent
product / process variability is also reduced as a result. a reduction in throughput. If product/ process varia-
Two general approaches can be used to reduce the tion can be reduced, it should be possible to benefit
variability. from reduced WIP inventory and reduced lead times
while maintaining current levels of throughput and
1. Reduce the length of the line (reduce process resource utilization. The more this variability can be
complexity). In the study, this meant reducing the
reduced, the more inventory can be reduced, leading
number of staåons. In practice, this is accomplished
to even greater benefits.
by reducing the labor or machine content of prod-
ucts through methods improvements or through
product redesign. REFERENCES
2. Reduce the process time variation. In this study, 1. Bandyopadhyay, Jayanta K. "Product Design to FacilitateJIT
this meant moving from one processingtime dis- Production." Production and Inventory Management Journal 31,
tribution to one with a smaller variance. In practice, no. 4 (1990): 71-76.
2. Blackstone,John H., Jr. CapacityManagement.Cincinnati, OH:
this can be achieved by reducing the number of South-Westem Publishing Co., 1990: 221—224.
special orders or standardizing the products to be 3. Chapman, Stephen N. "Schedule Stabilityand the Implemen-
marketed. Other ways include the reduction of tation of Just-in-Time." Productionand Inventory Management
machine variations, cross training of operators, Journal 31, no. 3 (1990): 66-70.
4. Crawford, Karlene M. and James F. Cox. "Addressing Manu-
more consistent loading of the shop from one day
facturing Problems Through the Implementation of Just-in-
to the next, and the use of group technologyto Time." Production and Inventory ManagementJournal 32, no. 1
standardize the products being processed. ( 1991): 33-36.
5. Gaither, Norman. Productionand OperationsManagement.4th
In those situations where this variabilitycan be re- ed. Chicago, IL: The Dryden Press, 1990.
duced, companies can expect to receive the benefits 6. May, Neville P. "Measurement and Reduction of Lead Times.'
of reduced WIP inventories and lead times while APICS 1990 Conference Proceedings: 306-307.
maintaining or increasing current levels of throughput 7. Toomey, John W. "Establishing Inventory Control Options for
and resource utilization. Just-in-Time Applications." ProdUctionand Inventory Manage-
mentJoumal 30, no. 4 (1989): 13—15.
If it is not possible to achieve a suffcient reduction 8. Umble, M. Michael and M. L. Srikanth. SynchronousManufac-
in product/ process variability, then the trade-off be- luring: Principles for World Class Manufacturing. Cincinnati, OH:
tween the benefits of reduced lead times and the costs South-Westem Publishing Co., 1990.
of reduced throughput must be evaluated. There may 9. VanDeMark, Robert L. "Process Lead-Time and Cost-Reduction
be occasions when a company would be willing to Simulation." APICS 1990 Conference Proceedings: 475—478.

10 PRODUCTION AND INVENTORYMANAGEMENTJOURNAL—First Quarter, 1993


APPENDIX: DATA SUMMARIES BY NUMBER OF STATIONS

Queue Capacity Queue Capacity

5 10 50 5 10 50
Average throughput: Average time in system:
4 Stations 4 Stations
Constant 396 396 396 396 Constant 200 200 200 200
unif (45, 55) 394 394 394 394 Unif (45, 55) 263 279 279 279
unif (40, 60) 389 391 391 391 Unif (40, 60) 274 380 380 380
Unif (30, 70) 373 384 385 386 Unif (30, 70) 286 503 555 557
Unif (10, 90) 371 376 387 Unif (10, 90) 309 572 790 913
Expon (50) 251 317 337 Expon (50) 379 674 1,010 1,645
6 Stations 6 Stations
Constant 394 394 394 394 Constant 300 300 300 300
Unif (45, 55) 391 391 391 391 Unif (45, 55) 394 421 421 421
Unif (40, 60) 385 387 388 388 Unif (40, 60) 422 538 538
Unif (30, 70) 367 380 380 Unif (30, 70) 445 727 795 780
Unif (10, 90) 324 361 366 367 Unif (10, 90) 493 874 1,097 1,279
Expon (50) 232 301 324 Expon (50) 631 1,084 1,656 2,285
8 Stations 8 Stations
Constant 392 392 392 392 Constant 400 400 400 400
Unif (45, 55) 388 388 388 388 Unif (45, 55) 523 552 548 548
Unif (40, 60) 382 384 383 383 Unif (40, 60) 569 709
705 709
Unif (30, 70) 374 374 375 Unif (30, 70) 581 936 953 1,022
Unif (10, 90) 319 354 359 359 Unif (10, 90) 659 1,149 1,399 1,628
Expon (50) 221 291 313 320 Expon (50) 863 1,562 2,095 2,992
Average work-in-process: Average utilization:
4 Stations 4 Stations
Constant Constant .996 .996 .996 .996
Unif (45, 55) 0.41 0.53 0.53 0.53 Unif (45, 55) .990 .992 .992 .992
Unif (40, 60) 0.47 1.20 1.20 1.20 Unif (40, 60) .978 .989 .989 .989
Unif (30, 70) 0.51 1.98 2.36 2.37 Unif (30, 70) .939 .975 .980 .981
Unif (10, 90) 0.51 2.31 3.80 4.72 Unif (10, 90) .836 .939 .959 .970
Expon (50) 0.51 2.44 4.70 9.44 Expon (50) .629 .804 .868 .928
6 Stations 6 Stations
Constant Constant .994 .994 .994
Unif (45, 55) 0.38 0.53 0.53 0.53 Unif (45, 55) .987 .988 .988 .988
Unif (40, 60) 0.47 0.93 0.98 0.98 Unif (40, 60) .973 .983 .984 .984
Unif (30, 70) 0.50 1.67 1.96 1.91 Unif (30, 70) .926 .970 .972 .972
Unif (10, 90) 0.52 2.11 2.68 3.86 Unif (10, 90) .819 .944
.923 .953
Expon (50) 0.52 2.35 4.16 7.75 Expon (50) .587 .774 .845 .910
8 Stations 8 Stations
Constant Constant .991 .991 .991 .991
Unif (45, 55) 0.35 0.43 0.42 0.42 Unif (45, 55) .983 .984 .984 .984
Unif (40, 60) 0.46 0.88 0.88 0.88 Unif (40, 60) .968 .978 .977 .977
Unif (30, 70) 0.50 1.49 1.57 1.77 Unif (30, 70) .918 .962 .962 .965
Unif (10, 90) 0.52 1.96 2.72 3.46 Unif (10, 90) .809 .914 .934 .941
Expon (50) 0.51 2.42 4.06 7.17 Expon (50) .564 .757 .826 .886

THE EFFECTOF WIP ON THROUGHPUTAND LEAD TIMES


11

You might also like