throughput time reduction framework jms version

Journal of Manufacturing SystemsVol. 22/No. 4

2003

283


2003

AbstractManufacturing throughput time reduction can be a daunt-

ing task due to the many factors that influence it and theircomplex interactions. However, there are basic principlesthat, if applied correctly, can be used to reduce manufactur-ing throughput time. This paper presents a conceptual frame-work that illustrates these principles. The framework illustratesthe factors that influence manufacturing throughput time, theactions that can be taken to alter each factor, and their inter-actions. The framework is detailed enough to provide guid-ance to the industry practitioner on how to reducemanufacturing throughput time, while being general enoughto apply to most manufacturing situations.

Keywords: Throughput Time, Throughput Time Reduction,Lead Time Reduction, Quick Response Manufacturing

1. IntroductionManufacturing throughput time is defined as the

length of time between the release of an order tothe factory floor and its receipt into finished goodsinventory or its shipment to the customer. Reduc-tions in manufacturing throughput time can gen-erate numerous benefits, including lowerwork-in-process and finished goods inventory lev-els, improved quality, lower costs, and less fore-casting error (because forecasts are for shorter timehorizons). More importantly, reductions in manu-facturing throughput time increase flexibility andreduce the time required to respond to customerorders. This can be vital to the survival and prof-itability of numerous firms, especially those ex-periencing increased market pressures for shorterdelivery lead times of customized product.

Many firms are struggling in their attempts to re-duce manufacturing throughput time, and the factorchanges that can reduce manufacturing throughputtime are not always understood (Suri et al. 1996).While manufacturing throughput time reduction canindeed be a daunting task due to the many factorsthat influence it and their complex interactions, thereare basic principles that, when applied correctly, canbe used to reduce manufacturing throughput time.

To apply the principles correctly, the basic factorsthat determine manufacturing throughput time mustbe clearly understood. This paper first uses a simplehypothetical manufacturing system to illustrate thebasic factors that determine manufacturing through-put time and explain why each factor occurs. Thistutorial could be used to train workers in these basicconcepts. The paper then presents a conceptualframework that illustrates the factors that influencemanufacturing throughput time, the actions that canbe taken to alter each factor, and their interactions.Because customers are concerned about the responsetime to their order and because the minimum ordersize can be for a single part/product, the focusthroughput this paper will be on the manufacturingthroughput time per part (MTTP).

Information obtained from case studies of leadtime reduction efforts at four different plants (seeJohnson and Wemmerlöv 1998 and Johnson 1999for published versions of two of the case studies),previous research on throughput time reduction fac-tors (see Table 1), and queuing theory principleswere used to construct the framework. The frame-work is detailed enough to provide guidance to theindustry practitioner on how to reduce MTTP whilebeing general enough to apply to most manufactur-ing situations.

The tutorial on factors contributing to MTTP ispresented next. The MTTP reduction framework isthen presented, and the factor changes that will re-duce each component of MTTP are discussed. Thepaper concludes with some general guidelines onfocusing efforts to reduce MTTP.

2. Understanding the FactorsDetermining MTTP

2.1 Processing Time

Consider a simple manufacturing system consist-ing of two workstations (a workstation is either a

A Framework for ReducingManufacturing Throughput Time

Danny J. Johnson, College of Business, Iowa State University, Ames, Iowa, USA. E-mail: [email protected]


284

machine or a workbench where a worker performsthe job, Cox and Blackstone 1998) that manufac-ture parts X and Y. Both parts must first go throughworkstation 1 (WS-1) and then through workstation2 (WS-2). The processing time per part is 10 min-utes at each workstation, move time between sta-tions is instantaneous, parts arrive one at a time tothe workstation, and no variability in arrival ratesor processing time exists. Under these conditions,it would be possible to sequence the arrivals to

the workstation so the next part doesn’t arrive untilthe current part is finished. As Figure 1a illustrates,if X and Y are processed consecutively, the MTTPfor each part type is the sum of the processingtimes at each station for a total of 20 minutes. Giventhe current state of technology used to producethe parts, 20 minutes is the minimum MTTP pos-sible, and it is a perfect system. Any increase inthe processing time per part would increase theMTTP by the same amount.

Table 1Previous Research on Throughput Time Reduction Factors

Factor ReferencesSetup time Burgess, Morgan, and Vollmann (1993)

Flynn (1987)Garza and Smunt (1991)Hopp and Spearman (2001)Leu, Russell, and Huang (1993)Morris and Tersine (1990)Suresh (1991, 1992, 1993)Suresh and Meredith (1994)Wemmerlöv (1992)Yang and Jacobs (1992)

Processing time per part Suresh and Meredith (1994)Move time Hopp and Spearman (2001)

Leu, Russell, and Huang (1993)Morris and Tersine (1990)Shafer and Meredith (1993)Suresh and Meredith (1994)Shafer and Charnes (1995)

Production batch size Ang and Willey (1984)Hopp and Spearman (2001)Karmarkar (1987)Karmarkar, Kekre, and Kekre (1985)Shafer and Charnes (1993)Suresh (1991, 1992, 1993)Suresh and Meredith (1994)Yang and Jacobs (1992)

Transfer batch size Ang and Willey (1984)Hopp and Spearman (2001)Jacobs and Bragg (1988)

Arrival variability Hopp and Spearman (2001)Jensen, Malhotra, and Philipoom (1996)Morris and Tersine (1990)Suresh and Meredith (1994)

Process variability Athersmith and Crookall (1974)Garza and Smunt (1991)Hopp and Spearman (2001)Leu, Russell, and Huang (1993)Moily and Stinson (1987)Suresh and Meredith (1994)

Resource utilization and/or resource availability Ang and Willey (1984)Athersmith and Crookall (1974)Burgess, Morgan, and Vollmann (1993)Hopp and Spearman (2001)Morris and Tersine (1989)


2003

285

2.2 Production and Transfer Batch Sizes

Production batch sizes (that is, the number of partsof the same type processed before the workstation isset up to process a different part) and transfer batchsizes (the number of parts moved at the same time tothe next workstation) of one unit are often unrealis-tic due to machine setup times and material handlingconstraints, respectively. Further realism can thus beincorporated into the example by first increasing boththe production and transfer batch size for each partto 10 units. Under these conditions, each part spends100 minutes at each station for a total MTTP of 200minutes (see Figure 1b). Each part incurs only 20minutes of actual processing time. The remaining180 minutes is either time a part spends waiting forits turn to be processed at a workstation, or time thepart spends waiting for the remaining parts in thebatch to be processed so the batch can be moved.These wait times are sometimes referred to as wait-in-batch and wait-to-batch times, respectively (Hoppand Spearman 2001), or collectively as the wait-for-lot time (MPX 1996). The wait-for-lot time incurredby each part in this case is linearly related to the sizeof the production and transfer batches used. Thiscauses MTTP to also increase in a linear fashion asproduction and transfer batch sizes increase.

2.3 Setup and Move Time

Further realism can be entered into the hypotheti-cal system by requiring a setup time of 40 minutesbefore each batch is processed and including a 15minute batch move time between machines. If noother changes are made to the process, MTTP in-creases by 95 minutes, causing the total MTTP toincrease to 295 minutes (see Figure 1c). Any furtherincreases in setup and move time would directly in-crease MTTP by the same amount.

2.4 Variability

Assuming the same production cycle continuouslyrepeats in the examples shown in Figures 1a, 1b,and 1c, no idle time will exist at either station oncethe system fills with work (that is, once WS-2 startsprocessing the first part), causing the steady-stateutilization to be 100%. This can only happen in asystem with no variability. Because such systemsdon’t exist in reality, variability is introduced andexamined in the hypothetical system.

Variability can be a result of either controllable orrandom variation (Hopp and Spearman 2001). Con-trollable variation is a result of decisions made andincludes such things as differences in the process-ing time of different parts due to design differences,differences in wait-for-batch time due to productionand transfer batch size decisions, and so on. In con-trast, random variation is a result of events beyondour immediate control. This includes such thingsas natural variation in process time for the sametype of part due to unplanned machine downtimeor differences in machines, operators, or material;variation in the time between arrivals to each work-station, etc. Regardless of the type, variability gen-erates the possibility that a batch of parts arrivingto the workstation will find the workstation stillbusy processing a previous batch. When this hap-pens, the new batch must join the queue and waitits turn for processing.

For example, suppose in Figure 1c that variabilitycaused the processing time for the batch of X at WS-1 to be 110 minutes instead of 100 minutes, and atWS-2 to be 90 minutes instead of 100 minutes. Inaddition, the batch of Y arrives at WS-1 at 130 min-utes, which is 10 minutes earlier than planned. Theimpact on MTTP is shown in Figure 1d. The earlyarrival of the batch of Y to WS-1 coupled with theextended batch processing time of X at WS-1 causedan initial wait time of 20 minutes for the batch of Yat WS-1. This wait time is called queue time. Theextended batch processing time of X at WS-1 alsodelayed the arrival of the batch of Y to WS-2 by 10minutes (when compared to Figure 1c), which caused10 minutes of idle time between the completion ofX at WS-2 and the start of Y. The net result is anMTTP for X that is the same as in Figure 1c (i.e.,295 minutes), but an MTTP for Y that is 20 minuteslonger than in Figure 1c (i.e., 315 minutes instead of295 minutes).

Increases in variability cause queue size and itsassociated queue time to increase. For example, sup-pose variability caused the batch of Y to arrive at thesame time as the batch of X, but all other conditionsare the same as in Figure 1d. As Figure 1e shows,the MTTP for X remains unchanged at 295 minutes,but the MTTP for Y has now increased by the addi-tional 130 minutes of queue time for a total MTTPof 445 minutes. When variability of all kinds is con-sidered, queuing theory indicates that queue size and


286

Figure 1Impact of Batch Size, Setup Time, Move Time, and Variability on MTTP

PX = Processing time for X P

Y = Processing time for Y

SX = Setup time for X S

Y = Setup time for Y

MX = Move time for X M

Y = Move time for Y

WS-1 = Workstation 1 WS-2 = Workstation 2Q

X = Queue time for X Q

Y = Queue time for Y

I = Idle time Fork = Forklift

No variabilityBatch size = 10 partsMTTP

X = 200 min.

MTTPY = 200 min.

No variabilityBatch size = 10 partsSetup time = 40 min.Move time = 15 min.MTTP

X = 295 min.

MTTPY = 295 min.

Arrival variabilityProcessing variabilityBatch size = 10 partsSetup time = 40 min.Move time = 15 min.MTTP

X = 295 min.

MTTPY = 315 min.

Arrival variability greater than in (d)Processing variabilityBatch size = 10 partsSetup time = 40 min.Move time = 15 min.MTTP

X = 295 min.

MTTPY = 445 min.

No variabilityBatch size = 1 partMTTP

X = 20 min.

MTTPY = 20 min.

Part X arrives

Part Y arrives(a) 0 20

I

0 30

WS-1

WS-2

Batch of X arrivesBatch of Y arrives

(b)0 100

PY

PX

WS-1

WS-2

200

0 100 200 300

PY

PX

I


(c)0 40

PY

PXWS-1

WS-2

140

0 155 195 295

PX

SX

I

Fork II

180 280

SX

SY

MY

MX

PY

SY

335 435


(d)0 40

PY

PX

WS-1

WS-2

130

0 165 205 295

PX

SXI

Fork II

190 290

SX

SY

MY

MX

PYS

Y

345 445

150

QY

I

305


(e)0

PY

PX

WS-1

WS-2

0 165 205 295

PX

SXI

Fork II

190 290

SX

SY M

Y

MX

PYS

Y

345 445

150

QY

I

305

Py

Px


2003

287

its corresponding queue time increases at an increas-ing rate as the standard deviation or coefficient ofvariation of interarrival and/or processing time in-creases (see Figure 2).

In assembly or joining operations, variability canalso cause a part to arrive at a workstation before itsmate(s). When this happens, wait time can occur,even though the workstation is available for setupand processing of the part. Although this waiting timeis often included as part of queue time, it is also some-times referred to as wait-to-match time (Hopp andSpearman 2001).

2.5 Utilization

Variability has less impact on queue time whenworkstation utilization is low than when worksta-tion utilization is high. When utilization is low andsignificant slack workstation capacity exists, it isfairly easy for a batch to arrive when the worksta-tion is idle and be processed immediately. However,as utilization increases and less slack capacity is avail-able, it becomes more difficult for a batch to arrivewhen the workstation is idle. This increases the prob-ability that the batch must join the queue, resultingin longer queue times and MTTP.

For example, suppose batches of parts arrive toa single workstation on average every 10 hours,each batch contains 10 parts, and the averagebatch processing time is 6 hours. However, due tovariability, the actual interarrival and processingtimes deviate from the average. The actualinterarrival and processing times for four batches

of different parts arriving to this workstation areshown in Figure 3a. In this case, the workstationis idle when each batch arrived, the average utili-zation is 60%, and the average MTTP is 6 hours.In Figures 3b, 3c, and 3d, the average utilizationof the workstation is increased to 70%, 80%, and90%, respectively, by decreasing the average timebetween batch arrivals to 8.6, 7.5 hours, and 6.7hours, respectively, while simultaneously keepingthe absolute deviations from the averageinterarrival time for each batch the same as in Fig-ure 3a. As shown, the increases in utilizationcaused the batch of X to incur queue time of 0.4hours in Figure 3b; batches of X and Y to incurqueue times of 1.5 and 1.0 hours, respectively, inFigure 3c; and batches of X and Y to incur queuetimes of 2.3 and 2.6 hours, respectively, in Figure3d. If each part cannot leave the station until theentire batch has been processed, these queue timescaused the average MTTP to increase at an increas-ing rate with successive increases in utilization.

The magnitude of the impact that utilization andvariability have on MTTP will vary from systemto system. However, queuing theory indicates thegeneral pattern of results shown in Figure 3 holdsfor all systems, namely that queue time and itsassociated MTTP increase at an increasing rate asutilization increases (see Figure 4). Furthermore,queue time and MTTP at a workstation with highvariability will increase faster as utilization in-creases than will queue time and MTTP at a work-station with low variability.

2.6 Factor InteractionsThe preceding discussion indicates that MTTP

is equal to the sum of the processing, setup, move,queue, wait-in-batch, wait-to-batch, and wait-to-match times. Because queue, wait-in-batch, wait-to-batch, and wait-to-match times all involvewaiting, and because actions to reduce one typeof waiting may also reduce other forms of wait-ing, they are collectively referred to as waiting timein the MTTP reduction framework. Reductions inMTTP thus require reductions in one or more ofthese components. While setup time, processingtime per part, and move time are independent ofeach other (i.e., a reduction in move time doesnot affect setup time or processing time per part,and so on), changes in any of these three compo-

Figure 2Queue Time vs. Interarrival and Process Time Coefficient

of Variation. Note: Graph constructed using GI/G/Mqueuing formula in Whitt (1983).


288

Figure 3Impact of Utilization on MTTP

PW

= Batch processing time for W PX = Batch processing time for X

PY = Batch processing time for Y P

Z = Batch processing time for Z

QX = Queue time for X Q


I = Idle time

Average batch interarrival time = 10.0 hrs.Average batch processing time = 6.0 hrs.Average utilization = 60%Average MTTP = 6.0 hrs.

Batch ofW arrives

(a)

I

0 8

PX

PW

Batch ofX arrives

Batch ofY arrives

Batch ofZ arrives

9 16 19 23 30 35 40


Batch ofW arrives

(b)

0 7.6

QX

Batch ofX arrives

Batch ofY arrives

Batch ofZ arrives

8 15 16.2 20.2 25.8 30.8 34.3

I I IPY

PZ

PX

PW I IP

YP

Z I


Batch ofW arrives

(c)

0 6.5

QX

Batch ofX arrives

Batch ofY arrives

Batch ofZ arrives

8 14 15 19 22.5 27.5 30

PX

PW IP

YP

Z I

QY


Batch ofW arrives

(d)

0 5.7

QX

Batch ofX arrives

Batch ofY arrives

Batch ofZ arrives

8 12.4 15 19 20 25 26.7

PX

PW IP

YP

Z I

QY


2003

289

nents can affect waiting time (Hyer andWemmerlöv 2002). Consequently, one way to re-duce waiting time is to manipulate the other threecomponents of MTTP.

For example, if the average processing time perpart is reduced to 5 minutes for each part type ateach workstation in Figure 1e while all other con-ditions remain the same, Y would only wait 100minutes at WS-1 and the MTTP would be 295 min-utes (see Figure 5). Reducing batch processingtime by 100 minutes for each part (i.e., 50 min-utes at station 1 and 50 minutes at station 2) inthis case actually caused a 150 minute reductionin MTTP for Y due to the additional impact onwaiting time at WS-1.

3. Manufacturing ThroughputTime Reduction Framework

3.1 Overview of Framework

Figure 6 presents the MTTP reduction framework.The framework can be described as a flowchart withfive columns. Column 1 lists the objective of theframework as the reduction in MTTP. Column 2 pre-sents the components of MTTP. Setup time is thesum of the times spent setting up all workstationsrequired to process the part through the productionsystem. Processing time is the sum of the times spentprocessing a part at each workstation required in theproduction routing for the part. Move time is the sumof times spent moving a part between each worksta-tion in the production routing for the part. Waitingtime is the sum of the queue, wait-in-batch, wait-to-batch, and wait-to match times at all workstations inthe production routing for the part. Waiting time isusually the largest of the four components, account-ing for as much as 90% of manufacturing lead timein some systems (Houtzeel 1982). Column 3 illus-trates the factors that will reduce each component.Column 4 specifies actions that will alter each factorshown in column 3, and column 5 presents im-portant changes that might be required to enablesome of the actions shown in column 4. The fea-sibility of accomplishing some of the actions andchanges shown in columns 4 and 5 are directlyrelated to the type of production layout used (i.e.,

Figure 5Impact of Processing Time per Part Reduction on MTTP Compared to Figure 1e






Y = Move time for Y




I = Idle time Fork = ForkliftBatch of X arrivesBatch of Y arrives

0

PY

PX

WS-1

WS-2

0 115 155 195

PX

SXI

Fork II

140 190

SX

SY

MY

MX

PY

SY

245 295

100

QY

I

205

MTTPX = 195 min.

MTTP Y

= 295 min.

HighLowLow

High

Que

ue t

ime

Utilization

High variability

Figure 4Queue Time vs. Utilization. Note: Graph constructed

using GI/G/M queuing formula in Whitt (1983).

Lowvariability


290

job shop/functional layout, cellular layout, or prod-uct layout/assembly line). The issue of layoutchoice will be included in the following discus-sion where appropriate.

Based on these definitions, one or more of thesefour components must be reduced in order to re-duce MTTP; by following the flowchart from left toright, actions that will reduce each component canbe identified. This flowchart is intended to provide astructured way to examine the types of actions thatcan be taken to reduce MTTP and the relationshipsbetween these actions. The following sections brieflydiscuss how to reduce each component of MTTP.

3.2 Setup Time Reduction

Column 3 of Figure 6 indicates that setup timereductions can be accomplished by reducing the timeper setup and/or the number of setups. Time per setupcan be reduced by purchasing equipment with shortsetup times, improving setup procedures, dedicat-ing workstations to families of parts with similar setuprequirements so that common fixtures can be usedand developed, and/or by using family schedulingto group batches that have common setup require-ments. Workstation dedication and family schedul-ing can also reduce the number of setups required.Further information on improving setup procedurescan be found in works by Steudel and Desruelle(1992) and Shingo (1985).

3.3 Processing Time per Part Reduction

Column 3 of Figure 6 indicates that reductions inprocessing time per part can be accomplished byreducing the number of operations required, reduc-ing the processing time per operation, and/or reduc-ing scrap and rework. The number of operations perpart may be reduced through the adoption of newtechnology that allows a single operation to do whatwas previously done by several operations, or byredesigning the part so that fewer operations are re-quired. Processing time per operation can be reducedby redesigning the part to require less processing,incorporating faster technology to process the part(if available), or dedicating labor to a family of partswith similar processing requirements. Labor dedica-tion allows the workers processing the parts to be-come more familiar with a smaller family of parts,thus potentially reducing the amount of time spent

reading blueprints, setting machine speeds, perform-ing quality inspections while the parts are on themachine, and so on.

The best way to reduce scrap and rework is toimprove raw material quality to prevent defectivematerial from entering the system, and to improveequipment capabilities, processes, and proceduresto prevent scrap and rework from happening in thefirst place. Implementing poka-yoke (fail-safe) de-vices can be especially beneficial in this respect.Using one-piece flow (or very small transfer batches)can also reduce scrap and rework because defectiveparts can be quickly detected at the next operation.One-piece flow is often impractical in a job shop/functional layout due to the increased material han-dling, production control, scheduling, and/or infor-mation systems requirements such a change wouldentail. In contrast, one-piece flow can often be usedin a cellular or product-oriented layout with littleimpact on the same requirements. As a last resort,increased inspection of the parts to identify defec-tive units and prevent them from being transferredto the next operation can be used to improve scrapand rework.

3.4 Move Time Reduction

Column 3 of Figure 6 indicates that reductions inmove time can be accomplished by reducing eitherthe time required per move or the number of movesrequired. The time required per move can be reducedby increasing the speed of the material handlingequipment (which may not be possible due to safetyimplications), or by reducing the move distance re-quired. If the speed of the material handling systemis increased through the installation of conveyors orother automated handling equipment, it is question-able how realistic this option would be when a jobshop/functional layout is used. While move distancecan sometimes be reduced by reorganizing the equip-ment to optimize the material handling between de-partments in a job shop/functional layout, the amountof reduction is greater if the equipment performingsequential operations on a part is grouped to formmanufacturing cells.

If a job shop or functional layout is currently be-ing used, the number of moves requiring materialhandling equipment can often be reduced by group-ing workstations performing sequential operations


2003

291

Figure 6Manufacturing Throughput Time per Part (MTTP) Reduction Framework

ReduceManufacturing

ThroughputTime Per Part

(MTTP)

Column 1

Objective

Column 2

ComponentChanges ThatWill Reduce

MTTP

Column 3

Factors ThatWill ChangeComponent

Column 4

Actions That Will AlterFactors in Column 3

Column 4

Important Changes PossiblyNeeded to Enable Actions in

Column 4

ReduceSetupTime

ReduceProcessingTime Per

Part

ReduceMove

Time PerPart

ReduceTime/Setup

Reduce# Setup

Reduce Scrapand/or Rework

Reduce Numberof Operations

ReduceTime/Operation

ReduceTime/Move

Reduce Numberof Moves

ReduceProductionBatch Size

Reduce TransferBatch Size

ReduceProcessingVariability

ReduceWaiting

Time PerPart

Reduce SetupTime, ProcessingTime/Part, and/or

Move Time(See Column 2)

Reduce ArrivalVariability

ReduceWorkstationUtilization

IncreasesResource Access

Reduce Numberof Queues

Purchase Equipment WithShort Setup Time

Increase Material HandlingCapacity (if constrained ) or

Group Equipment PerformingSequential Operations

Improve Procedures

Dedicate Equipment

Family Scheduling

One-piece Flow

Improve Procedures

Improve Equipment Capability

Improve Raw Material Quality

Increase Inspection

Improve Technology

Dedicate Labor and Equipment

Increase Move Speed

Reduce Move Distance


Part Redesign

Faster Technology

Improve Technology

Change Production BatchSize Policy

Change Transfer BatchSize Policy

Group Similar Jobs

Standardize Part Design

Dedicate Labor and Equipment

Stabilize Batch Sizes

Improve Preventative Maint.

Reduce Processing Variability(See Column 2)

Control Order Releases

Improve Coordination

Increase Time Available

Reduce Time Required

Cross-Train Workers

Increase Equipment Pooling

Increase # Successive Operations/Worker or Machine

Increase Production Control,Scheduling and/or InformationSystem Capabilities(if needed) or

Reduce Need for TheseSystems

Optimize Current Layout


Increase Workstation Capacity(if constrained) or

Reduce Setup Times

Increase Material HandlingCapacity (if constrained) or


Increase Production Control,Scheduling and/or InformationSystem Capabilities(if needed) or

Reduce Need for TheseSystems


292

into manufacturing cells. In some cases, technologi-cal improvements that allow more sequential opera-tions to be done by a single machine can achievethe same result (for example, a CNC milling ma-chine is used to perform the operations previouslydone by several machines).

3.5 Waiting Time Reduction

3.5.1 Overview

Column 3 of Figure 6 indicates that reductions inwaiting time can be accomplished by reducing setuptime, processing time per part, move time, produc-tion batch sizes, transfer batch sizes, processing timevariability, arrival variability, resource utilization,and/or the number of queues. It can also be reducedby increasing access to resources. Reductions in setuptime, processing time per part, and move time havealready been discussed. The remaining factorchanges will be discussed in the following sections.

3.5.2 Production Batch Size Reduction

Production batch size reduction is often the easi-est and most cost-effective way to reduce waitingtime and MTTP in most plants. Not only does it re-duce the wait-for-lot time for the part in question,but it also reduces queuing time for parts in otherbatches as well. For instance, consider the examplein Figure 1e. The average processing time per part

for X is 110 / 10 = 11 minutes at WS-1 and 90 / 10 =9 minutes at WS-2. Because only one part in thebatch is processed at a time, 9 parts are always wait-ing, resulting in a wait-for-lot time of 11 * 9 = 99minutes at WS-1 and 9 * 9 = 81 minutes at WS-2.The queue time for Y is 150 minutes at WS-1 andthe wait-for-lot time at both WS-1 and WS-2 is (100/ 10) * 9 = 90 minutes. This produces a MTTP for Xand Y of 295 minutes and 445 minutes, respectively.In contrast, if production and transfer batch sizesare reduced to 5 parts for both X and Y but all otherconditions remained the same, the wait-for-lot timefor X is reduced to 11 *4 = 44 minutes at WS-1 and9 * 4 = 36 minutes at WS-2 (see Figure 7). Queuetime for Y drops to 95 minutes at WS-1 and the wait-for-lot time at both WS-1 and WS-2 drops to (50 / 5)* 4 = 40 minutes. The net result of the batch sizereduction is that MTTP is reduced to 195 minutesfor X and 290 minutes for Y.

To reduce batch sizes, the plant needs to imple-ment a policy to schedule production of smallerbatches. However, if demand stays constant, smallerbatch sizes increase the number of setups required.As the number of setups increases and more of theavailable capacity is used for setups, workstationutilization increases, which causes queues to grow.Eventually, the increased queues negate any benefitto be obtained from batch size reduction and MTTPincreases rapidly (see Figure 8). Reducing setup timewould allow further batch size and MTTP reduction.

Figure 7Impact of Batch Size Reduction on MTTP Compared to Figure 1e






Y = Move time for Y




I = Idle time Fork = ForkliftBatch of X arrivesBatch of Y arrives

0

PY

PX

WS-1

WS-2

0 110 150 195

PX

SXI

Fork II

135 185

SX

SY

MY

MX

PYS

Y

240 290

95

QY

I

200

MTTPX = 195 min.

MTTP Y

= 290 min.


2003

293

When batches are transferred between worksta-tions by forklift, handcart, or another similar con-veyance device, batch size reduction also increasesthe number of trips required. The increased numberof trips raises the utilization of the forklift, whichcauses increased queuing. If utilization increasesenough, the increased queues counteract any ben-efit to be obtained from batch size reduction, andMTTP increases rapidly in the same manner as pre-viously described for the impact of batch size re-duction on setup time.

Batch size reduction also increases the numberof different batches of product on the shop floorat any one time, which may increase the load onthe production control, scheduling, and/or infor-mation systems.

Based on this discussion, if MTTP is to be re-duced through batch size reduction, one or moreof the following changes are often required (seeColumn 5 in Figure 6):

1. Workstation capacity must be increased (if ca-pacity is constrained) or setup times reduced.

2. Material handling capacity must be increased(if capacity is constrained) or the workstationsrequired to process a batch be consolidated sothat material handling equipment is not neededas often.

3. The capabilities of the production control,scheduling, and/or information systems mustbe increased (which may included increases inboth labor and computer capacity) to handlethe increased requirements or the need for thesesystems reduced.

If production is performed using a job shop/func-tional layout, the spatial separation of workstationsand labor resources required to produce the batch ofparts will likely require increases in workstation andmaterial handling capacity and production control,scheduling, and/or information systems capabilitiesas batch sizes are reduced. In contrast, if cells areformed, workstations and labor are dedicated to fami-lies of parts and grouped in close proximity. Thisdedication and grouping reduces setup time and of-ten allows the parts to be transferred between work-stations by hand or by small conveyors, thuseliminating the need for forklifts and other materialhandling equipment. Cells reduce the amount of cen-tralized scheduling required because only the cellmust be scheduled rather than each workstation.Tracking of parts is less because the parts are eitherin one of the cells or the order hasn’t been startedyet. Finally, reduced scheduling and tracking require-ments may reduce the amount of computer informa-tion system capacity needed (if a computerizedinformation system was used) and the amount of timeneeded to enter data, maintain the system, etc. Thus,converting a job shop/functional layout to a cellularlayout would likely allow batch size reduction with-out corresponding increases in machine capacity,material handling, production control, scheduling,and information system capacity/capabilities. In fact,the use of cells may result in less need for these sys-tems, even though batch sizes are reduced.

3.5.3 Transfer Batch Size ReductionIf production batch sizes cannot be reduced, wait-

ing time can still be reduced through the use of trans-fer batches smaller than the production batch size.For example, suppose in Figure 1e that setup timescannot be reduced below 40 minutes, which pre-vents production batch size reductions. However,material handling capacity is such that transferbatches of five parts could be used. The impact ofthis change is illustrated in Figure 9. As shown, thetransfer batch size reduction reduced the wait-to-batch time for the first transfer batch of X and Y atWS-1 to 11 * 4 = 44 and 10 * 4 = 40 minutes, re-spectively. This allowed these transfer batches to bemoved to WS-2 earlier than in Figure 1e, and be-cause WS-2 was idle, it could begin processing thetransfer batches immediately. Thus, the first trans-fer batch of X was being processed at WS-2 at the

Figure 8MTTP vs. Batch Size. Note: Graph constructed using standard

queuing theory formulas.

HighLowLow

High

MT

TP

Batch size

MTTP with original setup time MTTP with reduced setup time


294

same time as the second transfer batch for X wasbeing processed at WS-1. Similar results occurredfor Y. Even if both transfer batches must be com-bined before leaving WS-2, the net result is a re-duction in MTTP for X of 295 – 240 = 55 minutesand for Y of 445 – 395 = 50 minutes when com-pared to Figure 1e.

Transfer batch size reduction has the same impli-cations for material handling capacity, productioncontrol, scheduling, and information system capa-bilities as those previously discussed for batch sizereduction, but it does not influence the number ofsetups required if all transfer batches of the sameproduction batch are processed consecutively be-fore parts of a different type are processed. Transferbatch size reduction also has less impact on materialhandling capacity, production control, scheduling,and information system capacity if manufacturingcells are used versus a job shop layout.

3.5.4 Processing Time Variability Reduction

Variability in processing time comes from severalsources: variance in setup time for a workstation,variance in the processing time per part, variance inthe size of the batch processed, and variance due to

unplanned downtime and repair of the workstation.Reducing any of these sources of variability will re-duce processing time variability and, consequently,waiting time as well. Grouping similar jobs basedon part family affiliation, dedicating equipment andlabor to these part families, and/or standardizing partdesign will help reduce the variance associated withsetup times and processing time per part. Stabilizingor establishing similar batch sizes for all jobs in thefamily will help reduce variance associated with batchsize differences. Improvements in preventive main-tenance will help reduce variance associated withunplanned downtime and repair of the workstation.

3.5.5 Arrival Variability Reduction

Reductions in arrival variability will also reducewaiting time. Arrival variability is more complex thanprocessing variability and is dependent on the vari-ability of new orders released directly to the work-station, as well as the departure variability from anyupstream workstations that feed the station in ques-tion. When workstation utilization is high, each jobis extremely likely to arrive when the workstation isbusy and, consequently, is likely to have to join thequeue. As a result, the departure variability from the

Figure 9Impact of Transfer Batch Reduction on MTTP Compared to Figure 1e

PXi

= Processing time for X, transfer batch i SX = Setup time for X

PYi

= Processing time for Y, transfer batch i SY = Setup time for Y

MXi

= Move time for X, transfer batch i WS-1 = Workstation 1M

Yi = Move time for Y, transfer batch i WS-2 = Workstation 2

QXi

= Queue time for X, transfer batch i I = Idle timeQ

Y = Queue time for Y, transfer batch i Fork = Forklift


PY1

PX1

WS-1

WS-2

0 110 150 195

PX1

SXI

Fork II

SX

SY

MX1

PY1S

Y

295 345

QY1

& QY2

I

240

MTTPX = 240 min.

MTTP Y

= 395 min.

255 395

PY2

PX2

QX2

QY2

MX2

MY1

MY2

II

110 165 305255

PX2

PY2

0 40 95 190 290240150


2003

295

workstation is primarily dependent on the process-ing variability at the station. In contrast, when work-station utilization is low, the workstation is idle asubstantial portion of the time and each job arrivingto the station is more likely to find the station idle. Inthis case, variability in the time between arrivalstends to directly impact departure variability. In ad-dition, departure variability is reduced as the num-ber of identical copies of the resource at the stationincrease (Hopp and Spearman 2001, p263). This isa direct result of resource pooling. More will be saidabout this impact in section 3.5.7.

Regardless of the utilization level, any changesthat reduce variability in the time between arrivalsor in the actual processing at the workstation willreduce departure variability. Processing variabilityhas already been discussed. Variability in the timebetween the arrivals of new orders can be reducedthrough the use of controlled order release mecha-nisms. Such mechanisms stabilize the productionschedule by releasing new orders to the workstationwhen the queue reaches a set level. For assembliesthat require two or more components to start pro-duction of the job, any changes in production con-trol that improve the coordination of the arrival ofthe components will also reduce arrival variability.

3.5.6 Workstation Utilization Reduction

As discussed in section 2.5, wait time is heavilyinfluenced by workstation utilization. Workstationutilization can be defined as “the total workstationtime required per period divided by the total work-station time available per period.” In this framework,the total workstation time required per period is equalto the sum of the times spent setting up the worksta-tion, processing parts, waiting for labor to becomeavailable, and waiting for the equipment to be re-paired. This is similar to the definition used in queu-ing packages like MPX (MPX 1996). The totalworkstation time available per period is equal to thesum of the times each identical unit of the resourceat the workstation is available to be used. Thus, forexample, if the workstation has two semi-automatedmachines operated by a single worker, each machineis available eight hours per day, and on average atotal of two hours are spent setting up the machines,ten hours are spent processing parts, one hour is spentwaiting for labor, and unplanned downtime equalsone-half hour each day, the average workstation uti-

lization = (2 + 10 + 0.5 + 1) / (8 + 8) = 84.4%. Work-station utilization will decrease if the total time re-quired per period is reduced proportionately morethan the total time available per period is reduced,or if the total time required per period is increasedproportionately less than the total time available perperiod is increased.

The time available per period can be increased byadding equipment if capacity is machined con-strained, adding workers (and possibly extra shifts)if capacity is worker constrained, and reducing ab-senteeism. The capacity or time required can be re-duced by reducing the arrival rate of jobs to theworkstation (which will reduce output), and/or byreducing setup time, processing time per part, equip-ment downtime, scrap and rework, and delays dueto unavailability of workers. Reducing delays dueto unavailability of workers may require adding ad-ditional workers (which also increases capacity), re-assigning worker responsibilities to better balancethe load, or cross-training workers to handle mul-tiple tasks. In the case of cross-training, workers canfloat to the workstation or resource experiencing themost delays. This will reduce the utilization of equip-ment, but it will not necessarily increase the overallaverage worker utilization because it may simplychange when and which worker is idle, rather thanthe total amount of idle time. If this occurs, resourceavailability is increased without increasing utiliza-tion, and wait time goes down.

3.5.7 Increase Resource Access

Figure 6 indicates that waiting time can also bereduced by increasing access to resources. Whileresource access can be increased by purchasingequipment, hiring workers, working overtime, etc.,the intent of this factor is to increase resource accesswithout incurring these additional costs. Using cross-trained workers and increasing equipment poolingcan sometimes accomplish both of these goals. Us-ing cross-trained workers has previously been dis-cussed and will not be mentioned further.

To understand how equipment pooling can in-crease resource access and reduce waiting time, con-sider the case where parts A and B both require amilling operation (as well as other operations notrequiring a mill) and this operation can be done oneither Mill 1 or Mill 2. However, the two mills arecurrently located in different areas of the plant and


296

Mill 1 is dedicated to the production of A and Mill 2is dedicated to the production of B. While this mayhave its advantages, it does create the possibility thatMill 1 is starved for work due to variability in de-mand, variability in processing times at previous sta-tions, workstation downtime at a previous station,etc., while Mill 2 has a queue of B waiting for pro-cessing. Thus, B incurs waiting time even though amill with the required capabilities is currently sittingidle in another area of the plant. This would not hap-pen if the two mills were pooled (i.e., resource pool-ing is increased) by locating them in close proximityand feeding them with a common queue of work.Whenever a mill in the pool becomes idle, it wouldbegin processing the next job in the queue. This canreduce waiting time and MTTP for A and B, pro-vided the increase in equipment pooling doesn’t in-crease setup times, processing times, move times,variability, etc., to the point where the impact of theseincreases overcomes any potential waiting time re-duction resulting from the pooling increase. Due tothe complex interaction of such factor changes, queu-ing theory or simulation models are often requiredto determine if MTTP would be reduced through in-creases in equipment pooling.

3.5.8 Reduce Number of Queues

The final way to reduce waiting time is to reducethe number of queues by increasing the number ofsuccessive operations that the same worker or ma-chine performs. For example, suppose a metal partrequires several different milling, drilling, and tap-ping operations and these operations are currentlydone on three different machines made specificallyfor that purpose. At each machine, the part may haveto join a queue to wait its turn for processing. Incontrast, if all these operations can be done on aCNC milling machine, the queues between opera-tions are eliminated. The elimination of wait timewill reduce MTTP, provided any increase in setupand processing time resulting from the use of the CNCmilling machine rather than the specialized equip-ment is less than the amount of time the part nor-mally spends waiting. Similarly, cross-trainingworkers to perform multiple assembly tasks that werepreviously done by separate workers will reduceMTTP, provided any increase in task time resultingfrom the loss of specialization is less than the wait-ing time eliminated.

4. ConclusionManufacturing throughput time reduction can of-

ten be a daunting and confusing task due to the largenumber of factors that can be changed and the inter-actions between them. This paper provides a brieftutorial that illustrates the basic factors that deter-mine MTTP and explains why each factor impactoccurs. This tutorial can be used to educate workerson these basic concepts. The paper also presents aconceptual framework that illustrates the actions thatcan be taken to reduce each factor, and the relation-ships between them. This framework provides aneasy-to-use tool that managers can use to determinea course of action to reduce MTTP in their own plants.

Because all of the actions listed in column 4 ofthe framework (Figure 6) can be used to reduce MTTP,manufacturing plants striving to reduce throughputtime must decide where to focus their efforts. Theanswer to this question will vary between plants, butthe following guidelines can be offered:

1. Production and transfer batch size reductionsoffer the largest potential for MTTP in mostplants. If the plant has a job shop/functionallayout in place, significant reductions in batchsize may require conversion to manufacturingcells (for reasons listed in column 5 of the frame-work). Converting to cells may also reducemove time per part, processing time per part,processing variability, and arrival variability,further reducing MTTP beyond that achievedthrough production and transfer batch size re-ductions alone.

2. High workstation utilization is a major contribu-tor to long MTTP, especially in cases wherevariability is high. If variability cannot be re-duced, workstation utilization must be reducedto lower throughput times. In general, work-station utilization levels in the 75–80% rangemay be required on critical resources to keepMTTP low (Suri 1998).

3. Many causes of long MTTP are a result ofpolicies and procedures implemented in thepast that are used to control production batchsizes, transfer batch sizes, workstation utili-zation, resource access, and so on. With thecurrent emphasis on short manufacturingthroughput times, the relationships illustratedin the throughput time reduction framework


2003

297

and the principles discussed in this papershould be used to evaluate whether these de-cisions should be changed.

A version of the framework presented in this pa-per has been used by engineers and managers at asheet metal products fabrication plant to guide theirMTTP reduction efforts. The plant’s engineeringmanager stated that many of the relationships con-tained in the flowchart are concepts they have beentrying to teach their employees. They found that theflowchart organizes these concepts and relationshipsin a way that is easy for the employees to under-stand. The framework has also been used to teachmanufacturing throughput time reduction conceptsto university students enrolled in production/opera-tions management classes. These experiences indi-cate the framework is a useful tool for understandingthe actions that can be taken to reduce manufactur-ing throughput time per part and the relationshipsbetween them and for guiding throughput time re-duction efforts in real manufacturing plants.

ReferencesAng, C.L. and Willey, P.C.T. (1984). “A comparative study of the

performance of pure and hybrid group technology manufactur-ing systems using computer simulation techniques.” Int’l Journalof Production Research (v22, n2), pp193-233.

Athersmith, D. and Crookall, J.R. (1974). “Some organizational as-pects of cellular manufacture based on computer simulation.” Proc.of 15th Machine Tool Design Research Conf., Sept. 18-20, Bir-mingham, UK.

Burgess, A.G.; Morgan, I.; and Vollmann, T.E. (1993). “Cellular manu-facturing: its impact on the total factory.” Int’l Journal of Produc-tion Research (v31, n9), pp2059-2077.

Cox, J.F. III and Blackstone, J.H., Jr. (1998). APICS Dictionary, 9thed. Falls Church, VA: APICS.

Flynn, B.B. (1987). “Repetitive lots: the use of a sequence-dependentset-up time scheduling procedure in group technology and tradi-tional shops.” Journal of Operations Mgmt. (v7, n1-2), pp203-215.

Garza, O. and Smunt, T.L. (1991). “Countering the negative impactof intercell flow in cellular manufacturing.” Journal of Opera-tions Mgmt. (v10, n1), pp92-118.

Houtzeel, A. (1982). “Computer integrated manufacturing.” Proc. of1982 Academic-Practitioners Liaison Operations Mgmt. Workshop,American Production and Inventory Control Society, pp23-37.

Hopp, W.J. and Spearman, M.L. (2001). Factory Physics, 2nd ed.Boston: Irwin McGraw-Hill.

Hyer, N. and Wemmerlöv, U. (2002). Reorganizing the Factory:Competing Through Cellular Manufacturing. Portland, OR: Pro-ductivity Press.

Jacobs, F.R. and Bragg, D.J. (1988). “Repetitive lots: flow-time re-ductions through sequencing and dynamic batch sizing.” Deci-sion Sciences (v19, n2), pp281-294.

Jensen, J.B.; Malhotra, M.K.; and Philipoom, P.R. (1996). “Machinededication and process flexibility in a group technology environ-ment.” Journal of Operations Mgmt. (v14, n1), pp19-39.

Johnson, D. (1999). “Conversion to cellular manufacturing at SheetMetal Products.” Chapter 23 in Handbook of Cellular Manufac-turing Systems, S.A. Irani, ed. New York: John Wiley & Sons,pp707-748.

Johnson, D.J. and Wemmerlöv, U. (1998). “Cellular manufacturingfeasibility at Ingersoll Cutting Tool Co.” Chapter E2 in GroupTechnology/Cellular Manufacturing: A State of the Art Synthesisof Research and Practice, N.C. Suresh and J.M. Kay, eds. NewYork: Kluwer Academic Publishers, pp239-254.

Karmarkar, U.S. (1987). “Lot sizes, lead times, and in-process inven-tories.” Mgmt. Science (v33, n3), pp409-418.

Karmarkar, U.S.; Kekre, S.; and Kekre, S. (1985). “Lotsizing inmulti-item multi-machine jobs shops.” IIE Trans. (v17, n3),pp290-298.

Leu, J.Y.; Russell, R.S.; and Huang, P.Y. (1993). “On the applicabilityof cellular manufacturing.” Proc. of the Decision Sciences Insti-tute. Washington, DC.: Decision Sciences Institute, pp1469-1471.

Moily, J.P. and Stinson, J.P. (1987). “Effects of processing time vari-ability and routing flexibility on group technology implementa-tion in flow shops.” Working paper, Syracuse Univ.

Morris, J.S. and Tersine, R.J. (1989). “A comparison of cell loadingpractices in group technology.” Journal of Mfg. and OperationsMgmt. (v2), pp299-313.

Morris, J.S. and Tersine, R.J. (1990). “A simulation analysis of factorsinfluencing the attractiveness of group technology cellular lay-outs.” Mgmt. Science (v36, n12), pp1567-1578.

MPX (1996). Rapid Modeling Software User’s Manual, 2nd ed.Burlington, MA: Network Dynamics, Inc., 128 Wheeler Road,Burlington, MA 01803.

Shafer, S.M. and Charnes, J.M. (1993). “Cellular versus functionallayouts under a variety of shop operating conditions.” DecisionSciences (v24, n3), pp665-681.

Shafer, S.M. and Charnes, J.M. (1995). “A simulation analyses offactors influencing loading practices in cellular manufacturing.”Int’l Journal of Production Research (v33, n1), pp279-290.

Shafer, S.M. and Meredith, J.R. (1993). “An empirically-based simu-lation study of functional versus cellular layouts with operationsoverlapping.” Int’l Journal of Operations and Production Mgmt.(v13, n2), pp47-62.

Shingo, S. (1985). A Revolution in Manufacturing: The SMED Sys-tem. Cambridge, MA: Productivity Press.

Steudel, H.J. and Desruelle, P. (1992). Manufacturing in the Nineties:How to Become a Mean, Lean, World-Class Competitor. New York:Van Nostrand Reinhold.

Suresh, N.C. (1991). “Partitioning work centers for group technol-ogy: insights from an analytical model.” Decision Sciences (v22,n4), pp772-791.

Suresh, N.C. (1992). “Partitioning work centers for group technol-ogy: analytical extension and shop-level simulation investiga-tion.” Decision Sciences (v23, n2), pp267-290.

Suresh, N.C. (1993). “Job shops and cellular manufacturing: a dual-constrained comparison.” Proc. of the Decision Sciences Institute.Washington, DC.: Decision Sciences Institute, pp1463-1465.

Suresh, N.C. and Meredith, J.R. (1994). “Coping with the loss ofpooling synergy in cellular manufacturing systems.” Mgmt. Sci-ence (v40, n4), pp466-483.

Suri, R. (1998). Quick Response Manufacturing. Portland, OR: Pro-ductivity Press.

Suri, R.; Wemmerlöv, U.; Rath, F.; Gadh, R.; and Veeramani, R. (1996).“Practical issues in implementing quick response manufacturing:insights from 14 projects with industry.” Proc. of Mfg. and Ser-vice Operations Mgmt. (MSOM) Conf., Dartmouth College.

Wemmerlöv, U. (1992). “Fundamental insights into part family


298

scheduling: the single machine case.” Decision Sciences (v23,n3), pp565-595.

Whitt, W. (1983). “The queuing network analyzer.” Bell System Tech-nology Journal (v62), pp2779-2815.

Yang, K.K. and Jacobs, F.R. (1992). “Comparison of make-to-order job shops with different machine layouts and productioncontrol systems.” Int’l Journal of Production Research (v30,n6), pp1269-1283.

Author’s BiographyDanny J. Johnson is an assistant professor of operations manage-

ment at the College of Business at Iowa State University. He holds aBS in business administration from Moorhead State University and anMBA and a PhD in operations management from the University ofWisconsin–Madison. Prior to obtaining his BS, he worked for eight

years in the service sector. Dr. Johnson’s research interests are in thedesign, implementation, operation, and management of quick re-sponse manufacturing systems and the problems faced by firms asthey attempt to develop and use these systems to improve key perfor-mance measures. He has conducted and assisted with studies on theimplementation of cellular manufacturing systems in industry, andtwo case studies from these research projects have been publishedas chapters in books on cellular manufacturing. He has also pub-lished articles on cellular manufacturing in the International Jour-nal of Production Research and Production and OperationsManagement. He is certified in production and inventory manage-ment by the Educational Society for Resource Management (APICS)and is a member of the Educational Society for Resource Manage-ment, the Decision Sciences Institute, and the Production and Op-erations Management Society.

throughput time reduction framework jms version

Documents

quick response

partitioning

decision sciences

manufacturing

times spent

job shopfunctional

material handling

material handling