Você está na página 1de 17

int. j. prod. res., 2000, vol. 38, no.

2, 429± 445

Quantifying bene® ts of conversion to lean manufacturing with discrete


event simulation: a case study

RICHARD B. DETTY { and JON C. YINGLING{ *

Simulation is used to assist in the decision to implement lean manufacturing


principles at an existing assembly operation. Models are developed for the
existing assembly system, as well as for a new system (of similar capacity), that
employ these principles. In addition to the manufacturing processes, the
associated warehousing, inventory management, transportation, and production
control/scheduling systems are included in the model to enable the quanti® cation
of lean manufacturing’s impact on the total system. Simulation experiments
measure each system’s resource requirements and performance, thereby quantify-
ing bene® ts to be derived from applying the shop-¯ oor principles of lean
manufacturing.

1. Introduction
1.1. Overview of lean manufacturing
Lean manufacturing is a comprehensive philosophy for structuring, operating,
controlling, managing and continuously improving industrial production systems.
Key tenets of this philosophy include the following.
. Process stabilityÐ establish processes that combine men, machine, and
materials to produce 100% quality products when they are needed to satisfy
customer demand. This involves attaining demanding standards in equipment
reliability, raw material and purchased parts quality, employee knowledge and
skills, and production quality control.
. Standardized workÐ explicitly de® ne current best practice (minimal manpower
and e€ ort, highest quality, highest safety) in performing each job and commu-
nicate this at the workstation. Shop ¯ oor workers make and own this de® ni-
tion and constantly seek to improve it. It is a benchmark for improvement but
never used as an individual performance-rating tool.
. Level productionÐ attain capacity balance and synchronization of all produc-
tion operations over time in a manner that precisely and ¯ exibly matches
customer demand for the system’s products. In the ideal, this means producing
every product every shif t in quantities equal to demand (after smoothing out
high frequency random components). Moreover, production cycle time should
be less than the customer’ s acceptable waiting time for order receipt to enable
demand-based scheduling. Level production is achieved through means such
as rapid machine set-ups/changeover and ¯ exible, multi-machine manning

Revision received May 1999.


{ Manufacturing Systems Engineering, University of Kentucky, 230D M.M.R.B.,
Lexington, Kentucky 40506-0107, USA.
* To whom correspondence should be addressed. e-mail: jyinglin@engr.uky.edu

International Journal of Production Research ISSN 0020± 7543 print/ISSN 1366± 588X online # 2000 Taylor & Francis Ltd
http://www.tandf.co.uk/journals/tf/00207 543.html
430 R. B. Detty and J. C. Y ingling

strategies. Small lot or, preferably, mixed-model sequenced production sched-


uling is employed.
. Just-in-time (JIT)Ð attain the level production ideal with processes that
employ minimal (ideally zero) inventory through a strategy where each opera-
tion supplies parts or products to successor operations at the precise time they
are demanded. JIT is best attained through a continuous ¯ ow process structure
that employs multi-function employees performing only value-added opera-
tions. Pull production control strategies are widely employed (e.g. kanban
systems) to match production with demand and limit in-process inventories.
. Quality-at-the-sourceÐ build rather than inspect quality into the product. This
is achieved through systems that identify and resolve quality problems at their
source. Inspection systems that provide immediate feedback (e.g. self- and
successor-inspection strategies ), monitoring and control of factors that cause
quality problems, and error-proo® ng (poka-yoke ) mechanisms are widely
employed techniques.
. Visual controlÐ clearly display the operational status of the production system
so that work can proceed e ciently and problems are apparent to all.
Techniques such as 5S, inventory displays, and andon are utilized.
. Production stop policyÐ stop the production process when quality or produc-
tion problems occur, and visually display the location of the problem. With
low inventories this, in turn, readily forces wide shutdowns of the system. This
is done intentionally to reveal problems, with a mandate subsequently to
implement permanent, not quick-® x, solutions.
. Continuous ImprovementÐ empowered employees, working in teams, cont-
ribute both their intellectual and physical talents and seek continuously to
improve safety, quality, productivity, and the work environment through the
application of formal problem-solving strategies. Since continuous improve-
ment often results in fewer required workers, worker con® dence in their job
security is essential to bring forth ideas.
For additional information, readers are referred to Womack and Jones (1996),
Womack et al. (1990), Monden (1993), Shingo (1989).

1.2. Statement of problem


The decision to implement lean manufacturing, as just described, is a di cult one
because of the substantial di€ erences between traditional and lean manufacturing
systems in employee management, plant layout, material and information ¯ ow
systems, and production scheduling/control methods. These di€ erences make it dif-
® cult for organizations that have historically relied on traditional manufacturing
methods to predict the magnitude of the bene® ts to be achieved by implementing
lean principles in their unique circumstances. As a result, the decision whether or not
to adopt lean manufacturing techniques often must be based on a combination of
faith in the lean manufacturing philosophy, the reported experiences of others who
have previously adopted these principles, and general rules of thumb on anticipated
bene® ts. For many management teams, such faith-based justi® cation is insu cient to
convince them to adopt lean concepts.
What is needed is a tool to assist organizations considering lean manufacturing to
quantify, at the planning and evaluation stage, the bene® ts they can expect from
L ean manufacturing with discrete event simulation 431

applying lean principles to their system. This tool should be adaptable to the speci® c
circumstances of the organization, and should be capable of generating resource
requirements and performance statistics for both the proposed lean system and
the existing operation. This information would enable management to assess the
performance of the lean system in absolute terms and, most importantly, relative
to the well-understood, existing system it is designed to replace.
It is the hypothesis of this paper that discrete event simulation can provide
such a tool. Of course, some of the very important bene® ts from applying lean
manufacturing principles do not lend themselves to quanti® cation by computer
modelling, e.g. those that are the result of employee empowerment, continuous
improvement, organizational learning, and revised management structure, roles,
and information systems. However, simulation can quantify the performance
improvements that can be expected from applying the lean manufacturing shop-
¯ oor principles of continuous ¯ ow, just-in-time inventory management, quality at
the source, and level production scheduling. Simulation has the capability of demon-
strating the bene® ts of lean manufacturing throughout the entire manufacturing
system including:

. warehousing and in-process inventory levels;


. transport and conveyance requirements;
. the e€ ectiveness of production control and scheduling systems;
. uniformity of demand and delivery requirements from parts suppliers, and
. system responsiveness to market demand and customer delivery requirements.

Finally, simulation through animation can provide a visual and dynamic illus-
tration to management of how the new system would work.
The literature reports many previous applications of discrete event simulation to
address aspects of lean manufacturing. These applications can be grouped into two
general categories.

. The use of simulation to establish speci® c parameters of a lean manufacturing


system (i.e. the number of kanban, container size, batch size, mixed-model
sequencing approaches, etc). For example, Lummus (1995) used simulation
to study three production-sequencing strategies (mixed-model, minimum set-
up, and demand pull) as well as set-up time e€ ects for a multi-product JIT
system. Savasar and Al-Jawini (1995) studied the e€ ects of variability in pro-
cessing times and demand on push and pull production systems for various
kanban levels and withdrawal policies. There are numerous other studies in
this category including Abdou and Dutta (1993), Philipoom et al. (1987),
Sarker and Harris (1988).
. The use of simulation to design, test and improve lean systems. For example,
Carlson and Yao (1992) used simulation to pretest various ¯ ow layouts for a
low-volume, mixed-model JIT assembly system. Welgama and Mills (1995)
used a simulation to address design problems faced by a chemical company
changing from a traditional to a JIT system, considering alternative designs for
the JIT system. Galbraith and Standridge (1994) used simulation to test and
validate modi® cations to a traditional system as it was in transition, in stages,
to a JIT system.
432 R. B. Detty and J. C. Y ingling

Note that a prior decision to implement lean manufacturing is implicit in most of


these simulation applications. Simulation has not been employed to assist with the
decision to replace an existing, well-understood manufacturing process with a lean
system. Also, most previous work focuses exclusively on the manufacturing pro-
cesses (or an element thereof ), and does not deal with the entire production system.
The objective of this paper, therefore, is to demonstrate the use of discrete event
simulation as a tool to assist organizations with the decision to implement lean
manufacturing by quantifying the bene® ts that accrue from applying lean principles
in their speci® c situation. To this end, simulation models were developed to replicate
the operation of an existing system, as well as that of a proposed new system that
modi® es the existing design to incorporate lean manufacturing shop-¯ oor principles.
In addition to the manufacturing operations and processes, the models included the
warehousing, inventory control, transportation, and production scheduling systems.
The models were then used in simulation experiments to estimate, for both systems,
the shop-¯ oor resources they required as well as their performance in terms of
various time-based measurements.
The case study manufacturing system considered was an assembly process for
a high volume (500 000 units/year) consumer electronic product. The system
included the warehousing, transportation, inspection and assembly of approximately
80 parts (almost all of which are purchased from independent vendors) into a
® nished product.

2. Description of the existing system


The manufacturing system analysed in this study, is an assembly process for a
consumer electronic product. The system uses multiple, identical cells to assemble
over 80 individual parts into the ® nished product. With the exception of the main
circuit board, which is manufactured in-house, parts are purchased from outside
suppliers. The assembly steps are very similar for the di€ erent models of this prod-
uct, and each cell is capable of assembling all models. A 15± 20 minute changeover is
required for the cells to switch from one model to another, and most of this time
involves removing and restocking the cell with those parts that are not common to
both models.
There are six identical cells in this system. The cells are either fully sta€ ed or shut
down as product demand dictates. The product has a seasonal demand and, at the
time of the study, four to ® ve cells were operated on ® rst shift and two on second
shift. Each cell contains 12 sequential, adjoining workstations, 1 inspection station, 4
testing workstations, 1 repair station, and 2 packing stations. Following assembly
and initial in-cell testing, a second testing area, designated PPVT (for Product
Performance Veri® cation Testing) conducts additional performance and extended
run tests on a statistical sample of the ® nished product. There are three PPVT
areas, each testing the ® nished product from two cells.
After assembly and PPVT testing, the lot of ® nished product (a `lot’ is generally
one shift of production from the cell, approximately 340 units) is transported to a
® nal inspection and testing operation, designated AOQL (for Average Outgoing
Quality Limits). At AOQL, a statistically determined sample (per Dodge-Romig
Tables) is inspected and tested. The AOQL sample is ® rst examined for appearance
and to con® rm that the required items have been packaged in the product container.
The functional and extended-run tests, previously conducted in the assembly cell, are
then rerun on the AOQL sample. If the lot passes AOQL inspection and testing, it is
L ean manufacturing with discrete event simulation 433

transported to a distribution area for shipment. Lots that fail AOQL procedures are
returned to the assembly cell for rework. After rework, depending on the nature of
the defects, the lot is either shipped to distribution, or returned to AOQL for a
second inspection.
The assembly cells are supplied with parts from two parts storage areas and two
satellite subassembly areas as follows:
. the W arehouse supplies approximately 50 of the smaller parts directly to the
assembly cells. These parts are transported via forklift truck approximately
170 m from the storeroom to the assembly cells and satellite subassembly
operations;
. the Exchange part storage area that is the assembly cell’ s source of over 20 of
the larger parts. The exchange area is located adjacent to the assembly cells,
and a supply person from each cell, using a handcart, transports pallets of
parts from exchange. The parts in exchange are packaged in pallets with 4± 50
containers/pallet, with the containers holding from 18± 250 individual parts.
Pallets of parts from vendor trucks are unloaded to the warehouse for later
transport to the exchange area via forklift trucks, a distance of approximately
210 m;
. the Pre-Assembly operation that supplies two small subassemblies to the
assembly cells. Each of these subassemblies comprises 3± 4 individual parts
assembled with the aid of a small, hand- or air-actuated ® xture. Pre-assembly
is supplied with parts from the warehouse via forklift trucks, and maintains a
substantial parts and ® nished-assembly inventory. The assembly cell supply
person transports the subassemblies from the pre-assembly area to the cells.
The operation has 8 employees, including a supervisor;
. the Kitting operation combines approximately ten small items into a kit of
miscellaneous items that are included in the product container. The kit items
are supplied both from the exchange area and the warehouse. Like pre-
assembly, kitting has a large inventory of parts and completed kits. There
are nine people including a supervisor employed in this operation.
Additional features of production control and information ¯ ow for this system
include the following.
. The cells order parts from the warehouse via an electronic ordering system.
. Each pallet of parts in the exchange area has a kanban card attached to it,
which is removed and placed in a kanban collection box when the pallet is
removed from the exchange area by the cell utility person. (Parts are removed
from the exchange area in pallet quantities. ) The kanbans are collected the
following morning and orders to replace the withdrawn parts are electronically
transmitted that same morning to the supplier for delivery two days later.
. The exchange area has a limited amount of space that is partitioned into pallet-
wide rows with each row labelled for the parts stored in that row. As pallets of
parts are removed to meet the needs of the assembly cells, a forklift driver
(based on visual observation ) brings a replacement pallet from the warehouse.
. Inventories for non-exchange-stored parts are controlled by an MRP system.
The details of this system, due to proprietary concerns, were not available to
this study. Therefore it was assumed (based on observation ) that warehouse-
434 R. B. Detty and J. C. Y ingling

Figure 1. Layout of existing system.

supplied parts were reordered weekly for delivery three business days later.
Also, a two-day warehouse safety stock was assumed.
. Empty part containers from the exchange area are transported back to the
warehouse by forklift and returned to suppliers for reuse.
. Production scheduling results in large batches of individual models, with
model changes limited to once or twice a week for the entire plant.
Figure 1 shows the layout of the entire system including the assembly cells, the
warehouse and exchange areas, the subassembly operations, and the PPVT and
AOQL testing areas.

3. Proposed modi® cations of the existing system


While the present processes for warehousing, transporting and assembling parts,
and for testing the ® nished product, provide an opportunity to apply several lean
manufacturing factory ¯ oor principles, it should ® rst be noted that the existing
system already employs several important lean manufacturing concepts:
. multiple assembly cells to provide ¯ exible capacity to meet demand ¯ uctuation;
. a pull-system from suppliers for parts supplied from the exchange area;
. one piece ¯ ow in assembly cells with limited WIP inventory between work-
stations;
. standardized work charts at each workstation;
. 5S and visual control for parts storage in the exchange area and in the assembly
cells;
. a Poka-Y oke scanning operation to ensure all items are in the product
container.
L ean manufacturing with discrete event simulation 435

The additional lean manufacturing concepts proposed for implementation in this


system are discussed below.

3.1. Improve ¯ ow by integrating pre-assembly and kitting into the assembly cells
The ¯ ow of the total assembly process is disrupted by the satellite pre-assembly
and kitting operations. As the system is presently laid out, pre-assembly and kitting
are isolated islands of activity that create waste in the form of unnecessary handling,
inventories, transportation and system delay, in addition to under-utilizing employee
capabilities.
A review of the work performed at the individual workstations in the assembly
cell indicated that the allotted `takt’ time (80 seconds) for these activities exceeded
the assembly task requirements by 10± 20 s, depending on the workstation. The pre-
assembly and kitting operations can readily be consolidated into the assembly cell
operations by adding another person and workstation to each cell.
The bene® ts anticipated from this modi® cation include reduced inventories,
¯ oor space, product cycle time, transportation requirements, and freeing nine
people for other projects. A minor negative is the cost of duplicating ® xtures for
each assembly cell.

3.2. Built-in quality or quality-at-the-source


The traditional quality control process for the present system relies on a series of
redundant inspection and testing procedures in the assembly cells, PPVT, and
AOQL areas. This system attempts to inspect quality into the ® nal product. It is
proposed that it be replaced with processes that are consistent with the lean manu-
facturing principle of building quality into the product at the source. We recommend
that the system be modi® ed to:

. upgrade the skills of team members through training on problem solving, and
kaizen methods;
. enhance procedures at each workstation to inspect the work from the previous
station and to self-inspect its own work;
. increase use of visual management methods to increase/maintain awareness of
potential problem areas, quality goals and results;
. establish poka-yoke and visual systems at critical assembly steps;
. install an andon system that cell operators can trigger when problems are
encountered;
. improve quality-at-the-source practices at supplier operations.

The AOQL process should be eliminated. Initially it may be necessary to increase


the sampling frequency for the testing procedures in the cell and PPVT area and to
assign the existing personnel in AOQL (nine people, on two shifts) to the cell/PPVT
sta ng. As the quality-at-the-source procedures demonstrate e€ ectiveness, these
people should be available for other assignments. The ¯ ow improvements from
eliminating AOQL will also reduce cycle time, ¯ oor space, transportation, and
® nished goods inventory.
436 R. B. Detty and J. C. Y ingling

3.3. Reduced in-cell inventory and model changeover time


As mentioned previously, pallets of parts are supplied to each assembly cell
from the warehouse, exchange, kitting and pre-assembly areas. Depending on
the part number, pallet quantities range from 72 parts/pallet (18 parts/container,
4 containers/pallet), to as many as 4320 parts/pallet (240 parts/container, 18 con-
tainers/pallet). This pallet-quantity restocking policy enhances the output of the
transportation process, but does not consider the impact on total system e ciency
in terms of increased inventories, ¯ oor space and extra handling. Recommended lean
changes for this area include:
. use in-cell ¯ ow racks to supply full parts containers and remove empties from
each workstation;
. apply standards for the number of containers, by part number, at each
workstation;
. employ a standardized parts restocking route for the cell supply person (using
a handcart ) to transport part containers (not pallets) to the cells;
. resupply the cells with container quantities of parts rather than pallet
quantities.
Reducing cell parts inventories and utilizing ¯ ow racks facilitates shorter model
changeovers. At present, switching a cell from one product model to another takes
from 10± 15 minutes to remove the parts for the completed model run, and to restock
parts for the next model to be assembled. Reducing the number of containers and
providing ¯ ow racks at each workstation allows parts for two or more product
models to be available and easily accessible, thereby, facilitating single-minute
model changeovers, smaller lot sizes, and segmented mixed-model production
scheduling.

3.4. Reduce system ¯ oor space, inventory and transportation requirements by


increasing supplier delivery frequency and by small lot production scheduling
The existing system has excessive ¯ oor space, inventory and transportation
requirements due to:
. large lot production runs;
. long three-day cycles for the exchange pull system;
. the absence of an external pull system for warehouse parts;
. excessive ¯ oor space requirements which substantially increases transport
distances.
Recommended improvements in this area include:
. increase the frequency of deliveries from the exchange area parts suppliers to
once each shift. Parts consumed each shift should be reordered at the end of
the shift for delivery the next day (two shifts later);
. for parts supplied through the warehouse storeroom, each day’ s (two shifts)
consumption should be reordered at the end of the day for delivery 2 days
(four shifts) later;
. reduce production lots to 100 units and alternate lots between product models
in proportion to each model’s share of overall demand;
L ean manufacturing with discrete event simulation 437

Figure 2. Layout of proposed system.

. reduce inventory and ¯ oor space for the warehouse and exchange areas to
re¯ ect the revised supplier delivery cycles, reduced lot sizes, and mixed-
model scheduling;
. relocate the warehouse adjacent to the assembly cells;
. establish a separate storage area for safety stock. Safety stock should equal one
day’ s supply for the lean system, and two days for the existing system.
These modi® cations reduce the exchange and warehouse ¯ oor space requirements
by approximately 50% and allows the warehouse and exchange areas to be located
close to the assembly cells, thereby reducing transport distances between the ware-
house, the exchange area, and the assembly cells to under 80 m. Incoming supplier
trucks can now be unloaded directly into the exchange area, eliminating the double
handling and much of the warehouse area. The proximity of both the warehouse and
the exchange areas to the assembly cells enables the standardized part restocking
route mentioned previously. Forklifts are needed only for moving parts from the
supplier trucks to the consolidated exchange/warehouse area (and, in rare instances,
to move safety stock). Figure 2 illustrates the layout of the proposed lean system.

4. M ode lling assumptions and simulation experiments


Simulation models were developed for both the existing and proposed lean
systems using System Modeling Corporation’ s Arena software package. The
models were programmed in the Siman V language.
Due to run-time constraints, the system models were limited to three cells and
two product model types. (Advances in PC/Workstation capabilities since the
study was completed would no longer require this limitation. ) The assembly and
subassembly cells were modelled simply as a location to where parts are transported
and combined, with an 80-second takt time, into a ® nished product. The minor
variability of process times in the cells and individual workstations was judged not
relevant to the objectives of this study, which focuses on the plant-wide production
system.
438 R. B. Detty and J. C. Y ingling

Transportation was modelled using individual transport units (supply person or


forklift trucks) operating over average travel distances and velocities between the
cells and inventory stations.
The pull system between system operations was modelled by requiring a resource
in the succeeding process to be seized before releasing the corresponding resource in
the current process, with the quantity of these resources e€ ectively representing the
number of kanbans.
Assembly shifts were set at 450 minutes (8 hours less two 15-minute breaks), with
no set-up or transient period between shifts.
Pre-assembly and kitting operating times and inventories were scaled down to be
consistent with the three assembly cells and two product model assumptions.
Extensive work was done to verify and validate model function. This work
included careful trace studies, detailed animation to verify proper system behaviour,
and a review of the model behaviour (via animation) and numerical output results
with system experts at the plant under study.
AOQL reject rates and rework requirements for the existing system were based
on historical experience. For the lean system, reject and rework requirements were
reduced by 75% to re¯ ect the impact of quality-at-the-source principles.

4.1. Statistics accumulated during model runs


The statistics discussed below were gathered during model runs and were
designed to quantify the performance of both systems with regard to time-based
performance measures. With the exception of the counters, and another noted
case, the statistics discussed below represent averages.

(1) System Flow Time. Average time parts spend in the system from arrival in
warehouse to departure to distribution.
(2) Order L ead Times. Time between order receipt and the arrival of correspond-
ing ® nished goods at distribution for shipment.
(3) Average Inventory L evels. A time-weighted average of the quantities of parts
in the various inventories in the system.
(4) Relative Inventory Floor Space Requirements. The inventory storage space
requirement immediately after restocking. This measure is based on the
physical dimensions of the pallets, but does not re¯ ect space required for
access aisles or shelving.
(5) Time Between Departures from ExchangeÐ time between pallet withdrawals
from exchange. This statistic provides insight into the frequency and vari-
ation in the frequency of withdraws for each part, and demonstrates the
impact of production scheduling on supplier demand. The variance of this
statistic is more important than the mean, which is the same for both the
existing and the lean system.
(6) Utilization Rates. Percentage of time a resource such as the assembly cells,
forklift trucks or the cell supply people are busy.
(7) Counters for Absolute Quantities of parts (pallets, containers, parts), sub-
assemblies, ® nal products, orders, shipments, units rejected in AOQL and
reworked.
L ean manufacturing with discrete event simulation 439

(8) Counters to Measure the Cumulative Transport Distances required during


simulation runs.

4.2. Simulation experiments conducted with models


Two series of experiments were conducted for both systems, each representing a
di€ erent production scheduling assumption.
. In the ® rst series both systems produced alternating batches of product model
types 1 and 2. The batch size for each cell of the existing system was 900 units
(with 300 unit lots) and, for the lean system, both the batch and lot sizes were
100 units. This series was designed to quantify the bene® ts a€ orded by the lean
system’s ability to operate under a small lot, mixed-model production schedule
as opposed to the large batch scheduling employed in the existing system. The
batch size assumed for the existing system in this comparison is actually smal-
ler than the typical batches manufactured; thus, a number of projected per-
formance di€ erences are conservative.
. In the second series, production from both systems was `pulled’ by actual
customer demand. For this experiment, the times between customer orders
were assumed to be a stationary, compound Poisson process with the mean
demand equal to 85% of the manufacturing system’s capacity. The primary
objective of this series was to measure di€ erences between the two systems in
`order lead time’ (the time between order receipt and shipment of ® nished
goods). Customer orders for both systems were generated in quantities vary-
ing, with equal probability, from 100± 500 units, in 100 unit increments. In the
case of the existing system, orders were accumulated into 3000 unit batches for
each model (approximately 1.5 days of production for the total system) and
then sent to the assembly cells as production orders. Observations of the actual
case-study assembly system indicated this was a conservative estimate of the
batch sizes that are generated by the MRP system. For the lean system, orders
were sent to the ¯ oor in 100 unit batches.
Note that both series of runs may be classi® ed as non-terminating (Law
and Kelton 1991) and were concerned primarily with steady-state system
performance measures. Because the system at time zero was assumed to be empty
of parts, a start-up, or transient, time period was required for the entire system to
® ll itself with parts and, thereafter, reach steady state. For Series 1, the transient
period for each system (the time period in which the distribution of the performance
measures is non-stationary, due to the initial [time zero] state of the system) was
determined based on an analysis employing several 100 000 minute runs. The
extended duration runs monitored successive observations of multiple performance
variables, including cell, exchange, and ® nished goods inventories, time between
departures from exchange, and the average time in system for the parts in the
® nished goods pallets entering the distribution area. The time in system for the
® nished goods parts was determined to be the appropriate statistic for establishing
the transient period for both systems since it took the longest time to reach
steady state.
Figure 3 summarizes the results. The plots shown on this chart represent the
average time across the replications that the parts, making up the assembled ® nished
goods entering distribution, had been in the system as a function of simulated time.
440 R. B. Detty and J. C. Y ingling

Figure 3. Transient period analysis for existing and lean systems.

Figure 4. Order lag time analysis.

Referencing ® gure 3, the transient period was established at 18 000 minutes of run
time for the existing system, and 11 000 minutes for the lean system.
Series 1 results were then established based on the average of data from ® ve
replications. In each replication, the statistics were cleared and reinitialized at the
end of the transient time period, and collected for 40 000 minutes thereafter.
In Series 2, the transient analysis of order lead-time for the existing system, with
its large batch-size pull scheduling posed di culties. As shown in ® gure 4, steady
L ean manufacturing with discrete event simulation 441

state is not reached in the time it takes to receive over 600 orders, approximately
100 000 minutes, or 5 months. (In a very long run, at a lower order arrival rate of
approximately 50% of capacity, the existing system reached steady state after
approximately 260 000 minutes, over 13 months of operation. ) Presumably the
system with a demand rate of 85% of capacity would eventually reach steady
state, but excessive run time duration made such a determination impractical. The
high variance of the performance measures and excessive duration of the transient
interval for the large-batch pull system, moreover, makes the meaning of such a
steady state measure dubious. For the lean system, in Series 2, the transient
period for order lead-time (also shown in ® gure 4) was so small as to be considered
negligible. The performance of the two systems in Series 2 was established by aver-
aging data from ® ve 100 000 minute replications without correction for a transient
period.

5. Discussion of results
S eries 1 Results: The bene® ts of the lean system (relative to the existing system)
quanti® ed by this experiment include the following.
(1) Average time parts spend in system reduced 55%.
(2) Model changeover times reduced in the assembly cells from 11 to 3 minutes.
(3) Reduction in the time-weighted, average inventories throughout the system:
. 70% lower warehouse inventories;
. 63% lower exchange inventories;
. 75% reduction in assembly cell inventories;
. 100% reduction in pre-assembly and kitting inventories (parts and WIP);
and
. 10% reduction in ® nished goods inventory.
(4) Reduction in ¯ oor space requirements:
. 37% in Warehouse area due to reduced maximum inventory levels;
. 51% in Exchange area from lower maximum inventory requirements;
2
. 560 m reduction from consolidating AOQL, pre-assembly, kitting into
assembly cell;
Floor space requirements in the assembly cells are also reduced by lower cell
inventories. These savings have not been quanti® ed due to the uncertain
impact on ¯ oor space requirements of the parts handling ¯ ow racks proposed
for the lean system.
(5) Substantially lower variability in demand from parts suppliers. While, as
expected, the average time between withdrawals from the exchange area is
essentially the same for both systems, the coe cient of variation (the ratio of
the standard deviation to the mean of the observations) is from 70%± 90%
lower for the lean system. Figure 5 illustrates, for a typical part, the di€ er-
ences in the two systems in this measure. A level demand from the ® nal
assembly process levels demand throughout the supply chain and allows
the entire supply chain, not just the ® nal assembly process, to be lean.
(6) Substantially reduced forklift truck utilization and transport distance
requirements. As a result, seven forklifts were not needed in the lean system.
442 R. B. Detty and J. C. Y ingling

Figure 5. Comparison of demand pull frequency for a representative part.

(7) Nine employees from pre-assembly and kitting and seven forklift drivers
can be reassigned to other activities. In addition, as quality at the source
improvements are demonstrated, a substantial number of the nine employees
currently assigned to AOQL can be reassigned.

These performance improvements represent the consolidated bene® ts of all of


the changes incorporated in the lean system relative to the existing system. While the
analysis did not isolate the bene® ts of each modi® cation, certain bene® ts can be
associated with speci® c changes.

. The reduction in ¯ ow time in warehouse and exchange is primarily attributable


to the increasing frequency of supplier deliveries, and secondarily to small lot,
mixed model production scheduling.
. The bene® ts of consolidating pre-assembly and kitting into the assembly cells
are the elimination of their respective part ¯ ow time and inventories, and the
freeing of nine employees for other work. Also, over 120 000 m of transport
distance, over the simulation period, were eliminated along with approximately
370 square metres of ¯ oor space, thereby contributing to the reduction in
required forklift trucks and operators.
. The reduction in warehouse and exchange inventories, combined with the
elimination of the pre-assembly, kitting and AOQL operations, substantially
lowered the operation’ s ¯ oor space requirements, thus allowing the warehouse
to be located near the assembly cells and exchange area. This contributed to
the reduction in forklift trucks and operators.
. The changes made in the cell supply system are directly responsible for the
75% reduction in cell inventories and contribute to the transport savings. The
L ean manufacturing with discrete event simulation 443

faster model changeovers enable a small lot, mixed-model production


sequence, which, in turn, reduced the exchange inventories.
S eries 2 Results: The reductions in system ¯ ow times, inventories, ¯ oor space, trans-
port requirements, and time between departures for Series 2 were, as expected,
essentially identical to Series 1. The noteworthy result from Series 2 was the lean
system’s 86% reduction in average order lead time.
For obvious reasons of con® dentiality, this study was not provided access to the
® nancial information that would be needed to convert the savings predicted in shop
¯ oor resources for the Lean Manufacturing system into monetary terms. However,
inasmuch as the study quanti® ed these savings in physical quantities, many of these
savings can be readily converted to ® nancial value. For example:
. inventory savings from Lean Manufacturing is quanti® ed in units, by part
number. These quantities can be converted to an inventory investment savings
based on the cost of the parts. Annual holding cost savings can be established
by applying an annual inventory holding cost percentage (i.e. cost of money) to
the investment savings;
. the reduction in ¯ oor space requirements can be converted to $ based on the
cost per square foot to rent/own and occupy facility space;
. the savings in equipment (7 forklift trucks) and personnel (18 people) can
readily be expressed in ® nancial terms, both in reduced investment and annual
operating costs.
Unfortunately however, bene® ts from Lean Manufacturing such as reductions in
order lead time, and levelled demand on suppliers, are not readily convertible into
monetary terms. Regardless, management will readily agree that these bene® ts have
clear value.

6. Conclusions
Discrete event computer simulation has been shown to be a tool that can help
quantify the bene® ts of lean manufacturing. The use of simulation in the manner
prescribed in this paper can provide creditable estimates of the savings in shop-¯ oor
resources and the improvements in time-based performance statistics that are
achievable with lean manufacturing. In the case study examined, simulation demon-
strated the impact of lean principles in terms of improvements (reductions) in:
. inventory, ¯ oor space, transportation, manpower and equipment require-
ments;
. time-based performance measures, such as model changeover time, order lead
time, and system ¯ ow time; and
. reduced variability in supplier demand.

For an organization considering implementing lean manufacturing, this


approach has the signi® cant advantage of re¯ ecting site speci® c circumstances, as
well as comparing the performance of the lean system with the present operation.
The shop-¯ oor resource and performance measures generated provide the basic
information management requires to access the overall ® nancial, strategic and
competitive bene® ts of lean manufacturing.
444 R. B. Detty and J. C. Y ingling

The availability of this information both facilitates the decision to adopt lean
principles and strengthens the organization’s commitment during the implementa-
tion phase. Moreover, by comparing actual plant performance during the implemen-
tation of lean procedures to the simulation results, it also provides a benchmark to
judge the organization’s implementation e€ ectiveness and the need for remedial
action.
Con® rming previous work by others, simulation is also shown here to be an aid
in analysing, designing and improving speci® c elements of lean manufacturing
systems, speci® cally:
. as an aid to establishing speci® c parameters of a system. It this study, the
feasibility of using the cell supply person to transport both warehouse and
exchange parts to the cell and to return empty cartons to the warehouse for
reuse was demonstrated;
. as a tool for quantifying the impact of system improvements. For the
case study assembly operation, the bene® ts of consolidating the satellite pre-
assembly and kitting operations into the assembly cells were clearly translated
in terms of reduced system ¯ ow time, inventory, and transportation require-
ments.
It is hoped that this paper will advance the use of simulation to help companies
make the initial commitment to lean manufacturing, as well as to provide the incen-
tive and motivation for them to stay the course during the inevitable di culties they
will encounter during implementation. As mentioned earlier, many of the bene® ts of
lean manufacturing cannot be quanti® ed with simulation modelling. Instilling proper
organizational values, organization learning and employee empowerment systems,
continuous improvement programmes, and setting up a consistent organization
structure as well as management information systems, are essential, mandatory ele-
ments of lean systems. If these management principles are not fully adopted along
with the factory ¯ oor principles, the lean system will not achieve the bene® ts quan-
ti® ed in this study and its performance could be inferior to the system it replaced.
However, assuming both management and factory ¯ oor principles are rigorously
applied in the new lean system, the performance of the lean system will likely be
signi® cantly greater than predicted.
It is also critical to remember that lean manufacturing requires a factory ¯ oor
focus, and computer simulation analysis should not detract from this `shop-¯ oor
® rst’ philosophy, nor should it substitute for or impede the ¯ ow of information from
the factory ¯ oor. Rather, simulation should be a tool that enhances the shop-¯ oor
® rst philosophy by observing and quantifying long term e€ ects that are not readily
apparent on the ¯ oor, are available from traditional factory information systems, or
are too expensive or time consuming to experiment with on the ¯ oor.

References
Abdou, G. and Dutta, S. P., 1993, A systematic simulation approach for the design of JIT
manufacturing systems. Journal of Operations Management, 11 , 225± 238.
Carlson, J. G. and Yao, A. C., 1992, Mixed-model assembly simulation. International
Journal of Production Economics, 26, 161± 167.
Galbraith, L. and Standridge, C. R., 1994, Analysis in manufacturing systems simulation:
a case study. Simulation, 63 , 368± 375.
Law, A. and Kelton, D., 1991, Simulation Modeling and Analysis, 2nd edn (New York:
McGraw-Hill).
L ean manufacturing with discrete event simulation 445

Lummus, R., 1995, A simulation analysis of sequencing alternatives for JIT lines using
kanbans. Journal of Operations Management, 13 , 183± 191.
Monden, Y., 1993, Toyota Production System, an Integrated Approach to Just-in-T ime,
2nd edn (Norcross, GA: Engineering and Management Press).
Philipoom, P. R., Rees, L., Taylor, B. and Huang, P., 1987, An investigation of the factors
in¯ uencing the number of kanbans required in the implementation of JIT technique
with kanbans. International Journal of Production Research, 26 , 457± 472.
Sarker, B. R. and Harris, R. D., 1988, The e€ ect of imbalance in JIT systems: a simulation
study. International Journal of Production Economics, 26, 1± 18.
Savasar, M. and Al-Jawini, A., 1995, Simulation analysis of just in time production
systems. International Journal of Production Economics, 42, 67± 78.
Shingo, S., 1989, A Study of the Toyota Production System from an Industrial Engineering
V iewpoint (Portland, OR: Productivity Press).
Welgama, P. S. and Mills, R. G. J., 1995, Use of simulation in the design of a JIT system.
International Journal of Operations and Production Management, 15 , 245± 260.
Womack, J. P. and Jones, D. T., 1996, L ean Thinking, Banish W aste and Create W ealth in
Y our Corporation (New York: Simon & Schuster).
Womack, J. P., Jones, D. T. and Roos, D., 1990, The Machine That Changed the W orld (New
York: Harper Perennial).

Você também pode gostar