Simulation of Lean Construction of HighRise Apartment Buildings

Simulation of Lean Construction of HighRise Apartment Buildings A lean model has been proposed for the construction management of highrise apartment buildings with customized apartment designs. It incorporates a number of changes to traditional management practice, including singlepiece flow with pull scheduling, work restructuring, and multiskilling. A simulated construction process scenario was devised for experimental evaluation of the model. The simulation was first implemented as a live management game, in which participants played the roles of the clients, general contractor, and subcontractors. Eleven runs with different teams indicated that the lean model increased throughput, improved cash flow, and reduced apartment delivery cycle time. However, the limitations of the live simulation led the writers to implement a discrete event computer simulation of the same process. The computer simulation reinforced the findings of the live simulation and emphasized the specific beneficial effect of singlepiece flow under pull scheduling. The lean model may be of immediate interest to construction planners and managers because it enables full customization with minimal waste and no additional resources. The demonstrative clarity of the lean model simulation, both live and computerized, makes it a powerful tool for education and research.

LEAPCON: Simulation of Lean Construction of High-Rise

Apartment Buildings

R Sacks1; A Esquenazi2; and M Goldin3

Abstract: A lean model has been proposed for the construction management of high-rise apartment buildings with customized apartment

designs It incorporates a number of changes to traditional management practice, including single-piece flow with pull scheduling, work restructuring, and multiskilling A simulated construction process scenario was devised for experimental evaluation of the model The simulation was first implemented as a live management game, in which participants played the roles of the clients, general contractor, and subcontractors Eleven runs with different teams indicated that the lean model increased throughput, improved cash flow, and reduced apartment delivery cycle time However, the limitations of the live simulation led the writers to implement a discrete event computer simulation of the same process The computer simulation reinforced the findings of the live simulation and emphasized the specific beneficial effect of single-piece flow under pull scheduling The lean model may be of immediate interest to construction planners and managers because it enables full customization with minimal waste and no additional resources The demonstrative clarity of the lean model simulation, both live and computerized, makes it a powerful tool for education and research

DOI: 10.1061/共ASCE兲0733-9364共2007兲133:7共529兲

CE Database subject headings: Computer aided simulation; Buildings, residential; Buildings, high-rise; Lean construction; Project

management; Scheduling

Introduction

Traditional production systems, designed for particular product

types and quantities, often encounter difficulties when the nature

of the market and clients’ requirements change 共Womack and

Jones 2003兲 In many developed economies, the housing

con-struction market changes as the demand for cheap housing

de-clines and the demand for larger and more sophisticated dwellings

increases Construction companies, using production processes

established to build large quantities of cheap housing, are faced

with the challenge of providing customized apartments for

in-creasingly sophisticated homeowners 共Frutos and Borenstein

2003; Rosenfeld and Paciuk 2000兲 Companies applying

tradi-tional construction planning with centralized control and

exten-sive subcontracting to construction of highly customized

apartments experience difficulties such as long cycle times to

complete individual apartments, loss of control by project

man-agement, and ultimately budget and schedule overruns

In response to this problem, Sacks and Goldin 共2007兲 pro-posed a management model for the construction of high-rise apartment buildings The model applies lean construction prin-ciples to reduce cycle time, improve cash flow, and increase flex-ibility to provide varied interior designs with short lead times As part of the development of the lean management model, simula-tion was used to explore the impacts of the model prior to imple-mentation in practice This paper reports on the experiments conducted using live and computerized simulation experiments to evaluate and refine the model The remainder of this section presents the background to the domain and to simulation in con-struction The next section presents the model, after which the simulation scenario is detailed Both the live simulation and the discrete event computer simulation are described and their results reported The final sections discuss the implications and draw conclusions

Conventional Construction Planning and Management

of Residential Projects

Cost and schedule overruns in construction in general and in housing construction in particular are common 共Egan 1998; Josephson and Hammarlund 1999; Koushki et al 2005; Serpell

et al 1995; Walsh et al 2003兲, and rework to correct errors rep-resents a significant part of the cost of most projects共Love et al 2000兲 The beginnings of an explanation for this can be found in Koskela and Howell’s proposition that the fundamental theories underlying construction project management practices and tools are obsolete: “Project management has not achieved the goals set

to it: it does not perform in a satisfactory way In small, simple and slow projects, the theory-associated problems could be solved informally and without wider penalties However, in the present big, complex, and speedy projects, traditional project

manage-1

Senior Lecturer, Faculty of Civil and Environmental Engineering,

Technion—Israel Institute of Technology, 840 Rabin Bldg., Technion

Campus, Haifa 32000, Israel E-mail: cvsacks@technion.ac.il

2

Graduate Student, Faculty of Civil and Environmental Engineering,

Technion—Israel Institute of Technology E-mail: albertoe@techunix.

technion.ac.il

3

Graduate Student, Faculty of Civil and Environmental Engineering,

Technion—Israel Institute of Technology E-mail: goldm@techunix.

technion.ac.il

Note Discussion open until December 1, 2007 Separate discussions

must be submitted for individual papers To extend the closing date by

one month, a written request must be filed with the ASCE Managing

Editor The manuscript for this paper was submitted for review and

pos-sible publication on December 16, 2005; approved on December 28,

2006 This paper is part of the Journal of Construction Engineering and

Management, Vol 133, No 7, July 1, 2007 ©ASCE, ISSN 0733-9364/

2007/7-529–539/$25.00.

ment is simply counterproductive; it creates self-inflicted

prob-lems that seriously undermine performance.”共Koskela and

How-ell 2002兲

The contrast between the conventional approach to

construc-tion management and the lean model, as exemplified for the

domain of high-rise apartment buildings demonstrated in the

simulation presented in this paper, provides a tangible example

The conventional approach described below is the basis of the

control part of the simulation experiments that are the subject of

this paper This approach was observed and documented in the

residential construction divisions of the three largest apartment

contractors in Israel共Sacks and Goldin 2007兲; readers may judge

to what extent the observations herein represent construction

management practice in their own countries

For most building structures, the natural sequence of

construc-tion activities is to progress from the foundaconstruc-tions upward through

the floors and to conclude with erection of the roof Conventional

construction planning of the interior finishing works, in high-rise

apartment buildings as in most other buildings, continues in the

same direction upward through the building In theory, this

method results in the minimum overall project duration because

finishing works can commence on any floor as soon as its

struc-ture is completed Conventional wisdom also dictates that all of

the work of any particular crew on a floor should be completed as

a single work package to minimize the time spent moving

equip-ment from floor to floor Planning based on the critical path

method is the rule, and the master plan devised forms the basis for

project monitoring and control systems Assumptions are made a

priori regarding all task durations and sequence dependencies

More sophisticated planning methods for repetitive building

projects, founded mostly on line-of-balance techniques and

incor-porating the use of expert systems, are available in construction

management literature共Arditi et al 2002; El-Rayes and Moselhi

1998; Peer 1974兲 However, linear scheduling has not seen

wide-spread adoption in construction共Arditi et al 2001兲, partly owing

to the lack of appropriate comprehensive software tools

Com-mercial tools that perform location-based planning, for example,

are only now becoming available共Seppanen and Aalto 2005兲

Client-initiated design changes are common in apartments but

are delivered through the life of the project as apartments are sold

and clients determine their preferences for floor layout 共location

of partitions兲, HVAC, and other systems; sanitary fixtures; and

floor, wall, and ceiling finishes These are dealt with in the

fol-lowing ways:

1 Where apartments have been sold prior to construction,

clients are encouraged to make design decisions with

suffi-cient lead time to meet preset milestones in the overall

con-struction schedule Changes are allowed or denied subject

to construction progress according to the preset schedule

共Frutos and Borenstein 2003兲 In the absence of client

deci-sions, work proceeds according to standard designs in the

knowledge that rework may be required later, a practice

termed “making do”共Koskela 2004兲

2 Client-initiated changes are executed largely as change

or-ders Change orders have been shown to have a strong

nega-tive impact on labor productivity 共Moselhi et al 2005兲 and

wasted materials The later a change order is introduced, the

greater its negative impact on labor productivity 共Hanna

et al 1999; Rosenfeld and Paciuk 2000兲 In many cases, a

change order requires demolition of previously executed

work, for which the client is required to pay

3 Changes are invariably priced to clients at rates significantly

higher than standard construction unit costs

4 Contingency is built into the overall project budget and schedule, usually as a lump sum estimate and a single time buffer at project end No attempt is made to allocate time buffers judiciously to absorb variability and shield produc-tion, as advocated in the literature共Ballard and Howell 1998; Goldratt 1997; Laufer 1996兲

The typical pattern resulting from the conventional approach can

be seen clearly in the Gantt chart of Fig 1, which was reported at the conclusion of a building with 4 apartments on each of 16 floors The period during which finishing works were executed in each apartment is shown by the black bar on each row; the hatched bar above it shows the period during which client design information was defined The batch start for each floor 共four apartments兲 is apparent, as is the large variability in the timings of provision of final design information共end of hatched bar兲

Lean Construction in Residential Building

A number of researchers have investigated the application of lean principles to housing construction, but none have proposed a holistic approach to the management of high-rise apartment construction Gann共1996兲 compared industrialized housing with car manufacture in Japan by highlighting similarities and differ-ences in production and supply chain strategies Ballard 共2001兲 suggested multiskilled teams as a key factor in achieving even-flow production for reducing cycle times and enhancing stability for construction of single family houses The multiskilling con-cept is analogous to the notion of production cells in manufactur-ing and has been used in precast concrete construction 共Ballard

et al 2003兲 The relationship between cycle time and quantity of work in progress 共WIP兲 predicted by Little’s Law 共Hopp and Spearman 1996兲 has been shown to hold at a macro project level for customized housing projects共Bashford et al 2005兲 Naim and Barlow 共2003兲 proposed the application of lean and agile ap-proaches to housing construction in the United Kingdom, but focused on supply chains and did not tackle the fundamental con-struction planning and control practices

Simulation in Construction

Computer simulation has long been recognized as an efficient method to improve planning for construction projects 共Halpin 1977; Martinez and Ioannou 1999兲 The primary motivation for the use of simulation in construction management is that it pro-vides a cheap, fast, and effective method to evaluate multiple alternative courses of action without having to suffer the conse-quences of failure that follow the unsuccessful courses共Back and Bell 1995兲 It is particularly useful for evaluating the distinct impact of each one of a set of process changes 共Farrar et al

2004兲; however, given the likelihood of interactions and interde-pendence between changes, this is best done by running the simu-lation with all the changes and then eliminating each one in turn

to evaluate its marginal contribution 共Warszawski and Sacks

2003兲 In the context of lean construction, simulation has been used to model the impact of pull-driven scheduling for process plant construction共Tommelein 1998兲, to model the impact of pro-cess changes for semiconductor plant delivery共Gil et al 2004兲, and in other projects

Conversely, the use of live role-play management games is rare in construction management Role playing is a recognized training tool in lean manufacturing 共Verma 2003兲 and has been used in various ways in activities of the Lean Construction Insti-tute共LCI 2004兲 The “Parade of Trades” game illustrates the

pact of variability on trade performance共Tommelein et al 1999兲;

it can be played live but has also been implemented in a computer

simulation The simple “penny fab” simulation presented by

Hopp and Spearman共1996兲 provides an effective explanation of

various aspects of flow in production systems

Lean Management Model

The conceptual leap from the conventional approach to a lean

approach in this context requires consideration of the project from

the client’s perspective For the clients, the primary product of

value is their individual apartment, while the public areas of the

building are of secondary value This contrasts with the general

contractor’s view of the project, as completion of the building as

a whole From the client’s point of view, the production measure

of “cycle time” is defined as the period from their first design

change meeting with company representatives to the time when

the apartment is ready for occupation The production measure

of quantity of work in progress共WIP兲, applied to individual

ments as the basic unit of production, counts the number of

apart-ments being worked on at any given point in time In these terms,

conventional practice has been shown to exhibit long cycle times

and high rates of WIP Work on individual apartments is not

con-tinuous, with up to 75% of the cycle time not utilized共Sacks and

Goldin 2007兲 Batch production is common, with activities

com-monly assigned to all the apartments on a floor as the basic work

unit Waste is present in all of the forms defined in lean

produc-tion texts共Womack and Jones 2003兲

The Achilles’ heel of the conventional approach is the implicit

assumption that all of the prerequisite conditions for successful

execution of each work package at each floor will be available at

the time the work is scheduled to begin on that floor However,

in reality, a key resource—design information—is often lacking;

in the context of customized housing, in markets where supply exceeds demand, schemes to enforce clients to make design de-cisions at times convenient to general contractors are difficult to enforce In many cases, essential materials that are subject to client choice are also unavailable Management and subcontrac-tors respond to the absence of timely information in a number of ways They may

1 Delay or slow work progress until the missing information and material is obtained;

2 Postpone parts of the work package, assuming that they will return to them at a later date; or

3 Abandon the construction plan and proceed to other work packages with complete information

The lean management model proposed by Sacks and Goldin 共2007兲 includes four basic measures designed to improve stability

of information and product flow and to reduce WIP and cycle times:

1 Reduce the batch size from full floors to single apartments The basic “unit” of planning for high-rise construction is traditionally a building floor, even in sophisticated planning systems共Arditi et al 2002; Shaked and Warszawski 1995兲

2 Replace the fixed activity network schedule for execution of finishing works with a dynamic method of pulling apart-ments through the finishing process according to the maturity

of the clients’ design changes The finishing works in indi-vidual apartments are performed according to the sequence

in which design information is provided

3 Restructure work to decrease or remove the dependency of early activities on client design change information Separat-ing electrical and plumbSeparat-ing systems from structural or facade work is one example: relocating electrical conduits and water supply pipes to false ceilings and dedicated chases instead of

embedding them in concrete slabs, walls, and columns means

that the structural concrete work becomes independent of

client design change information constraints

4 Reorganize work packages to reduce their number by

con-centrating more work in the hands of fewer multiskilled

subcontractors

Live Simulation

The goal for the simulation was to explore the potential impact of

applying the lean management model to the execution of finishing

works in high-rise apartments The following questions were

posed: Can the lean model enable customization of apartments

according to clients’ needs? Can the time from start to finish of

each apartment共cycle time兲 and the corresponding total number

of apartments worked on at any time共WIP兲 be reduced? Can the

stability of workflow and the throughput rate be improved without

adding resources? What is the impact on the contractors’ cash

flow?

Erection of the structure and closure of the facade are excluded

from consideration, as they are assumed to have been completed

The finishing works are to be subjected to a dynamic information

flow, in which changes to individual apartment designs are

deliv-ered during construction The public areas of the building are

excluded from the simulation

Game Setup and Execution

The Technion Lean Apartment Construction Simulation Game,

LEAPCON共LEAPCON 2005兲, simulates the execution of

inte-rior finishing activities required for construction of an eight-story

building with four apartments on each floor It assumes that the

structure is complete and excludes consideration of public areas

Four participants are assigned the roles of project manager, client

change manager, quality controller, and tower crane operator共all

work for a general contractor兲; four more are independent

spe-cialty subcontractors The participants’ task is to carry out the

interior finishing works for all 32 apartments in as short a time as

possible The execution of the finishing work is simulated by an

assembly of small apartment models using LEGO bricks in four

distinct steps, each step performed by one of the subcontractors

共see Fig 2兲 Four such apartment models are assembled in situ on

each of eight floors of the building The floors are represented by

eight pages placed in order on a long table; each page is marked

with the floor number, shows the elevator and stairwell core at its

center, and has four blank rectangles, each marked with the

rel-evant apartment number

Two additional players represent the apartment clients One

selects design variations at regular time intervals throughout the

game and delivers them to the general contractors client change

manager; the other checks completed apartments and pays the project manager $1,500 for each that is found to be free of defects and to match its final design

In the first round of play, simulating the conventional manage-ment model, the project manager is provided with a suggested construction plan that calls for the specialty contractors to progress up the building floor by floor, one after the other, in logical sequence according to the technological dependences of their work type The batch size is four, and only one contractor may work on a floor at a time However, the expected durations for each trade were not known to the project manager or other players, and they did not attempt to develop detailed construction plans

After the first minute of play has elapsed, the first client representative begins selecting design variations at random for apartments in random sequence Fig 3 shows a typical design variation 共E, one of seven variations兲 to the standard default apartment design共A兲 A design variation and the apartment num-ber to which it applies are delivered to the general contractor’s client change manager every 15 s through the eighth minute of play Design variations received before work on an apartment is commenced are easily executed, although they do destabilize flow because they require more time Apartments for which design variations are received after work has begun, but before the apart-ment has been delivered to the client, must be changed However, only the appropriate subcontractor can make the changes called for in the variation change order The project manager must de-cide whether to withdraw specialty subcontractors from the floors they are working on and send them to make local changes or to delay the changes The driving factor is that the company is paid only for completed floors of four apartments, while they must invest working capital 共$1,000兲 for any incomplete apartment Subcontractors, in contrast, are paid $1,000 for completing their work package on a floor, regardless of the state of completion of individual apartments Their incentive is therefore to complete as many floor work packages as possible in the time given The second client representative records the time at which each complete floor is delivered Play is stopped after 11 min, and the team’s performance is assessed in terms of apartments deliv-ered, quantity of WIP, cash flow, defective apartments, and time required to deliver the first nonstandard apartment

In the second round, simulating the lean management model, the following changes are made, while all other conditions remain

as before:

• Pull flow control replaces push scheduling: The work sequence

Fig 2 A standard apartment Assembly steps represent flooring,

partitions, HVAC ducts, and acoustic ceilings

Fig 3 Typical apartment design variation

is changed to follow the random sequence in which design

variations are selected Work is begun on an apartment only

after its design variation has been selected;

• The batch size is reduced from four to one: Work may be

performed on any apartment on a floor simultaneously with

work on other apartments on the same floor, and payment is

made per apartment delivered; and

• The four specialty subcontractors become four multiskilled

apartment finishing teams; each one can perform all of the

trades needed to complete an apartment

No work is done during the first minute of play because no

de-signs have been delivered As each design is delivered, the project

manager assigns a subcontractor to perform the work on that

apartment The subcontractors are no longer dependent on one

another; as soon as one completes an apartment, he or she is

assigned another Play continues for the same total duration of

11 min

Results

The results measured through 11 executions of the game, each

with different groups, are provided in Tables 1 and 2 The

param-eters of interest are the number of apartments completed, the

number of defective apartments, the time to delivery of the first apartment to the client, the quantity of WIP, the cash flow state at the end of the simulation, the throughput, and the cycle time The first four are easily measured; the latter three are calculated ac-cording to the formulas shown in Table 3

Participants included project managers, construction engi-neers, and site supervisors from various companies共five groups兲; specialty contractors共one group兲; graduate construction manage-ment students共two groups兲; and undergraduate civil engineering students 共three groups兲 Each group numbered 10 participants The construction industry groups, and particularly the project managers and subcontractors among them, reported that the first round faithfully simulated their day-to-day experience in projects

of this kind

Between the first and second rounds of the game, representing the change from the conventional management model to the pro-posed lean management model, average throughput increased from 1.3 units/ min to 2.0 units/ min, cash flow changed from negative to positive 共from −$9,100 to +$6,100兲, and average cycle time was reduced from 5 min 26 s to 2 min 18 s WIP was drastically reduced, from an average of 14.1 units to 2.0 units 关The pull mechanism and reduction of the batch size to one make

Table 1 Simulation Game Results: Round 1: Conventional Model

Group

number

Apartments completed

Defective apartments

Time to first apartment

Throughput 共units/min兲 Cashflow

Cycle time 共min:s兲

Table 2 Simulation Game Results: Round 2: Lean Model

Group

Apartments completed

共NOK 兲 apartmentsDefective

Time to first apartment

Throughput 共units/min兲 Cashflow

Cycle time 共min:s兲

this a CONWIP production system共Hopp and Spearman 1996兲, in

which WIP is explicitly controlled The maximum theoretical

WIP for the second round is four, equal to the number of

contrac-tors, plus any that may be waiting for the quality controller.兴

However, the live experiment has seven limitations:

1 Even though roles were changed and multiskilling was

intro-duced, it is likely that there is a learning curve effect for the

group as a whole, which could be assumed to exaggerate the

observed improvement

2 Variations of the management model could not be tested

be-cause it proved impractical for participants to execute more

than two rounds of the live experiment

3 Production rates and resource capacity utilization could not

be evaluated or adjusted because each subcontractor was

rep-resented by a single player

4 The impact of each change could not be measured in

isola-tion from the impact of the other changes

5 The simulation was not continued through to completion of

the building in either round because of the length of time that

would have been required

6 It was not possible to record the trends of the performance

indicators as the simulations progressed but only at their

conclusions

7 The sample size共11兲 is small, and the degree of variation is

large While the trend is clear, drawing definitive conclusions

as to the scale of improvement would be questionable

For these reasons, a detailed discrete events computer simulation

of the management game was programed and experimented with;

the simulation setup, the experiments, and the results are the

sub-ject of the next section

Discrete Event Simulation

Goals

The overall goals of the computer simulation were threefold:

1 To validate the results of the live experiment by overcoming its technical limitations: removing the learning curve effect, executing a much larger sample, and continuing the simu-lated project to its natural conclusion 共i.e., when all apart-ments are completed兲;

2 To establish and use an improved base construction plan by aligning production rates more closely; and

3 To extend the scope of the results: testing for the marginal contribution of each lean intervention, monitoring the indi-cators through time, and testing variations on the manage-ment model

The series of simulations described in Table 4 was defined to meet these goals The lean interventions considered are reduced batch size, pull flow, and multiskilled work teams The simula-tions include a control simulation, which models the conventional round 共Simulation A兲, the lean round 共Simulation F兲, and four intermediate experimental scenarios共Simulations B to E兲 Table 5 shows the basic durations measured for each activity in each apartment 共the measurement procedure is described below兲 Variation was applied to the durations by assigning normal distri-butions with the measured standard deviations

Three additional simulations共Aimp, Cimp, and Eimp兲 were run to establish an “improved” baseline against which to explore the benefits of the lean interventions The improved baseline plan represents a hypothetical situation in which advanced location-based construction planning methods for high-rise construction would have been used As shown in the line-of-balance charts in Fig 4, the two main differences from the conventional plan

mod-Table 3 Simulation Game Results: Formulas for Calculated Parameters

Model

Throughput

Conventional

共Ntotal − 4 兲

共Tlast− Tfirst 兲 NOK⫻ $1,500−共Ntotal + WIP 兲⫻ $1,000

Cycle _ time = Tfirst or WIP

Throughput

Tfirst ⫽lower bound for cycle time, valid for the first batch Cycle time will grow as the game progresses, peaking at WIP/Throughput when WIP peaks.

Lean

共Ntotal − 1 兲

共Tlast− Tfirst 兲 NOK⫻ $1,500−共Ntotal + WIP 兲⫻ $1,000

WIP 共=4兲 Throughput= 4

共Tlast− Tfirst 兲

共Ntotal − 1 兲 The theoretical WIP 共⫽4兲 should be used rather than the circumstantial WIP measured at the game’s end.

Note: NOK⫽number of apartments completed and approved; Ntotal= NOK + defective apartments; WIP⫽work in progress; Tfirst ⫽time of delivery of first

apartment; and Tlast ⫽time of delivery of last apartment.

Table 4 Schedule of Simulation Experiments

Simulation

code

Flow type

Batch size

Work teams

Production rates

A imp Push 4 Specialty contractors Adjusted

Cimp Push 1 Specialty contractors Adjusted

E imp Pull 1 Specialty contractors Adjusted

Table 5 Task Durations and Standard Deviations for Single Apartment,

Conventional, and Improved Plan

Conventional 共A兲

Task Duration

Std.

dev.

Increase/reduction

of resources Duration

Std dev.

eled in Simulation A are that production rates are more closely aligned by supplementing or reducing work team resources and

by using appropriate time buffers between work team start times Fig 4共c兲 shows the improved plan for single-piece flow, corre-sponding to Simulation Cimp Table 5 also shows the adjusted productivity rates for the improved plan

Simulations B and D test the marginal impact of multiskilling,

C tests for batch size change, and E and F evaluate pull flow In addition, each simulation was repeated in a configuration without changes to assess the influence of design changes on the process and the efficacy of each intervention in coping with the effects of changes during the process The simulations without changes are denoted A⬘, B⬘, etc Each simulation was run to its natural con-clusion, when all apartments were completed Results were re-corded not only at the end of completion of all 32 apartments, but also after the 11-min duration of the live game The results at

11 min are denoted A11, B11, etc., and A11⬘, B11⬘, etc

Implementation

The simulation was implemented using STROBOSCOPE 共Mar-tinez and Ioannou 1999兲 Each simulation scenario 共A to F兲 re-quired a unique model of its construction process Interestingly, the most complex scenario to model was the basic Scenario A, which represents traditional construction management This model comprised 31 activities, 26 links, and 115 queues A small portion of the graphic representation of this simulation model is shown in Fig 5 The full details of the simulation mechanism are beyond the scope of this paper Simulations Aimp, Cimp, and Eimp were modeled by changing activity durations and standard devia-tions in Simuladevia-tions A, C, and E, respectively

The durations for each activity in all of the models and their standard deviations were established rigorously A team of five students each repeated every step required to assemble each of the eight variants of the basic apartment model ten times each in random sequence The steps performed included not only the basic assembly steps but also every possible work task that might

be required to convert a standard apartment model at any stage of its assembly to any of the eight variations on the basic design

Fig 4 Line of balance charts for conventional construction plan共a兲

and for the improved plans共b and c兲

Fig 5 A portion of Simulation Model A, conventional management model

The logic of each simulation scenario model was checked in

relation to the marginal effects expected in comparison with the

other models Models A and F provided baselines for these

checks, as their results could be compared with the live game

results Finally, each simulation model was run 1,000 times, and

the parameters of interest were recorded at 10 s intervals

The computer simulation models were checked against the live

simulations by comparing the results achieved up to 11-min

du-ration for Experiments A and F with the results of the

conven-tional and lean live simulations, respectively; the results are

shown in Table 6 Small differences were observed between

com-parable results for the number of apartments completed and the

cycle times; these can be attributed in part to the very small

sample size of the live simulations共11兲 compared with the large

computer simulation sample size共1,000兲 The larger variations in

the WIP and cash flow stem from the fact that, in the live games,

players consciously avoided starting work on new apartments

to-ward the end of the simulation time because they knew that they

would adversely affect their cash flow at 11 min The relative

change from the conventional to the lean model from A11to F11

was considered sufficiently close to the relative difference

be-tween the live rounds, both in mean and in standard deviations, to

validate the computer simulations for qualitative relative

com-parison between the various computerized configurations Direct

comparisons should not be drawn between either of the live

rounds and any of the computerized simulations

Results and Analysis

The first comparison drawn is between Experiments A and F,

which simulate the lean and conventional rounds run to

comple-tion of all apartments The lean model has significantly lower

cycle times and WIP levels than the conventional model, as can

be seen from the data in Table 7 Overall project time for the

conventional model is 16% longer, despite its failure to customize

all of the apartments for which clients required changes Perhaps

the most important result for construction contracting companies

is the significant improvement in the cash flow achieved by the lean model that is apparent in Fig 6

The data in Table 7 and Fig 6 also illustrate the effect of the need for customization on the project outcome for each manage-ment model The conventional model results in higher cycle times, higher WIP, and a much less favorable cash flow when customization is needed In particular, apartment delivery times become unstable in the conventional model, as evidenced by the large increase in standard deviation of cycle time共from 0 min 50

s to 2 min 8 s兲 The lean model is resilient to the need for cus-tomization, although it does experience an increase in average cycle time

The results for the improved construction schedule 共Aimp兲 共Table 7兲 show the efficacy of careful scheduling according to best practice in shortening project duration as opposed to the conventional approach Employment of resources is also im-proved, it rises from 68% 共A兲 to 78% 共Aimp兲 overall However, WIP is higher and the capacity for customization is reduced The second aspect investigated is the independent and mar-ginal impact of each intervention The independent impact of re-duced batch size, from four to one, is obtained by comparing Experiments A and C, in which the reduced batch size is the only intervention applied The marginal impact is the difference be-tween Experiments B and D 共with multiskilling only, pull flow with a batch size of four was not simulated兲 These results are provided in Table 7 and Fig 7

Simply reducing batch size significantly reduces cycle times 共by 78%兲, improves cash flow 共the change from curve A to curve

C in Fig 7兲, and enhances predictability 共reduced standard devia-tions兲 The effect in the presence of multiskilling is similar 共68% reduction in cycle time兲 However, in the absence of pull flow,

reducing batch size harms the ability to provide customized

apart-ments because many of those apartapart-ments completed early in con-struction will not yet have client change information available

Table 6 Simulation Results up to 11 min: Means and Standard Deviations共in Parentheses兲

Table 7 Simulation Results to Completion: Means and Standard Deviations共in Parentheses兲

Simulation

Batch size

Duration to completion 共min:s兲 Cycle time共min:s兲 customizationDegree of

Maximum WIP

Net labor input 共min:s兲

Gross labor input 共min:s兲

For sake of simplicity in the live simulation, and thus also in the

computer simulation, if a change order was received for an

apart-ment after that apartapart-ment had been completed and accepted by the

owner, it was not returned to for rework; had this been accounted

for, the duration to completion would have been significantly

longer

In the specialized teams共A, C, and E scenarios兲, wait time is

common for teams with faster production rates than their

prede-cessors; wait time also arises as a result of the variability in

production rates The scale of this problem is reduced if the

con-struction schedule is improved by aligning production rates more

closely and applying appropriate buffers, as modeled in Scenarios

Aimp, Cimp, and Eimp Multiskilling increases the utility of the

work teams because each work team can proceed independently

of the other work teams As long as work is available, each

mul-tiskilled work team can experience continuous work without

wait-ing A comparison of Experiments A and B 共Table 7兲 indicates

that the impact of application of multiskilling alone is as follows:

for A, without multiskilling, the teams work 38 min 41 s, which is

just 68% of their gross 56 m 42 s spent on site; while for B, with

multiskilling, the teams work 43 min 48 s, which is 80% of the

total 54 min 39 s spent on site Comparison between Aimpand B

shows that the benefits of multiskilling alone are perceptibly

lower when compared with specialty teams working under an

improved construction plan

While efficiency measured locally is improved, the gross time

on site is only marginally improved because a larger number of

changes become necessary as a result of quick initial progress

The effect, from the general contractor’s point of view, is that cycle time is reduced, cash flow is improved, and project duration

is reduced, but these are all relatively small improvements when compared to the effect of reducing batch size In addition, the degree of customization that can be achieved is reduced and maximum WIP may in fact increase Thus, multiskilling alone is not necessarily beneficial

The marginal impact of multiskilling, when pull flow and unit batch size are applied, is given by comparing Experiments E and

F The data in Table 7 show that in the presence of pull flow and small batch size, multiskilling reduces project duration 共from

17 min 5 s to 15 min 34 s兲 but increases cycle time and WIP The reduction in overall duration is less impressive when compared with the result for Eimp共15 min 52 s兲 The fundamental contribu-tion of multiskilling is in improving the efficiency of the produc-tion operaproduc-tion and reducing wait times that result from variability, i.e., by increasing the ratio of net to gross time worked Without multiskilling, the average ratio of productive work time to total duration on site 共from start of work to finish for each team兲 is 65% 共37 min 34 s out of 62 min 51 s for all four trades兲; with multiskilling, this ratio improves to 87% 共40 min 8 s out of

46 min 17 s兲 Here too, the result for Eimp共71%, or 37 min 53 s out of 53 min 35 s兲 shows that much of the benefit can be ob-tained by adjusting production rates between specialty teams without the deeper change to completely generic multiskilled teams The reduction of total time on site is significant, but it is achieved at the cost of longer actual work durations, which appear

to occur because the multiskilled teams are less efficient than the specialty teams in performing each distinct task

Last, the marginal contribution of pull flow scheduling can be assessed by comparing Experiments D and F For a variety of reasons, the use of specialized subcontractors has become preva-lent in construction 共Edwards 2003; Hinze and Tracey 1994; Hsieh 1998兲 As a result, multiskilling is more difficult to imple-ment than pull scheduling or single-piece flow Comparing Ex-periments C to E therefore also provides a useful measure of the marginal impact of pull flow In both cases, the most important effect is to enable customization of 100% of the apartments, as opposed to 75% for the push scheduling situations Similarly, comparison between Cimpand Eimpunderlines the major benefit of pull flow control in achieving customization even when compared with the best possible result of project execution according to a location-based schedule with closely aligned production rates Were the simulations to be extended to include rework to make late changes to already completed apartments, then the pull schedule would exhibit significant benefits also in terms of overall cost and duration Pull scheduling also reduces the amount of WIP and improves cash flow, but these effects are minor when compared with the contribution of single-piece flow 共reduced batch size兲

Discussion and Conclusions

Of course, the LEAPCON management simulation 共in both live and computerized forms兲 differs in many ways from real high-rise apartment construction projects Among the features of real projects that differ from the simulations are the following:

• The information on design changes for each individual apart-ment will commonly be delivered by owners over time and not

in one cohesive package;

• Existing commercial and legal arrangements make

multiskill-Fig 6 Cash flows for conventional and lean models, with and

without customization

Fig 7 Cash flows for batch sizes of four and one, with and without

multiskilling

ing in apartment construction applicable to a very limited

degree; and

• Real projects include public areas 共lobbies, grounds,

stair-wells, etc.兲, which are more stable than apartments in terms of

their design information When used wisely, these areas can

serve as buffers to absorb excess capacity and decrease the

amount of downtime for specialty trade contractors

While the LEAPCON management simulation does not reflect

any individual real customized apartment building project

per-fectly, it does provide a reliable means to qualitatively compare

different management strategies It isolates and highlights the

im-pacts of three specific changes, drawn from lean production

prin-ciples, to traditional construction management practice for this

type of building: pull flow, reduced batch size, and multiskilling

Their impacts are magnified in the simulation because actual

con-struction activities outside of the scope of the finishing works of

the apartment units are ignored

The live simulation indicates the overall success of the lean

construction system in meeting clients’ needs better than the

tra-ditional management approach, while consuming fewer resources

However, its practical limitations cause the live game to be more

useful as an instructional tool for construction management

edu-cators than as a platform for experimentation The discrete event

computer simulations, in contrast, enable detailed analysis of the

independent, marginal, and combined effects of the lean

interven-tions Performance indicators can be monitored continuously, the

simulations can be run to their natural end, they can be repeated

sufficiently to provide statistical convergence, and local

modifi-cations can be tested

Although the three interventions interact, the specific

contribu-tions of each can be clearly distinguished from the experimental

results:

• Pull flow enables meeting the needs of all clients’ requests for

changes without rework;

• Reducing batch size reduces the level of work in progress and

so has a very positive effect on the project’s cash flow; and

• Multiskilling removes the dependence between work teams

and therefore reduces wait times between them and improves

their productivity

If applied alone, multiskilling can have a negative impact on cash

flow and on the degree of apartment customization that can be

achieved Increased local efficiency of the work teams means that

high WIP is accumulated before sufficient information is

avail-able; this leads to significant amounts of rework Multiskilling

represents the opposite trend to increased fragmentation of tasks

through subcontracting The effect of reducing the number of

in-dependent work teams by combining work packages is to increase

productivity and eliminate wait time, which is a desirable effect

even if full-scale multiskilling cannot be achieved in practice

Reducing batch size is the most significant intervention in

terms of improving cash flow However, reducing batch size for

customized construction projects is not recommended unless

ap-plied with pull flow In fact, in the experiment, reduced batch size

applied without pull flow reduced the number of apartments in

which client changes could be accommodated Pull flow enables

reduction of batch size without suffering this deficiency

Unlike the traditional approach to apartment building

con-struction management, the lean approach is resilient to the need

for customization of apartments The production system design is

such that it can cope with instability in the sequence and timing of

the flow of design information for each apartment without

expe-riencing great increases in delivery times, without suffering a

de-pleted cash flow and without the need for extensive rework The

abilities of the system to cater to apartment owner design changes and to provide a much improved cash flow, while consuming fewer resources, are significant reasons for property developers and general contractors to seriously consider what aspects of the lean model can be applied in their operations

The initial testing of the concepts presented has been carried out by a large construction contractor on a large luxury apartment complex and will be reported in future publications, but much more extensive long-term research is needed before definitive conclusions can be drawn about the direct applicability and effect

of each of the three interventions in construction projects Early indications are that the obstacles to implementation are existing commercial and trade arrangements that prevent reduction in the number of trades In contrast, pull flow and reducing batch sizes appear to be fairly straightforward to implement, especially if facilitated by appropriate management information and commu-nication tools

References

Arditi, D., Sikangwan, P., and Tokdemir, O B 共2002兲 “Scheduling

sys-tem for high-rise building construction.” Constr Manage Econom.,

20 共4兲, 353–364.

Arditi, D., Tokdemir, O B., and Suh, K 共2001兲 “Scheduling system for

repetitive unit construction using line-of-balance technology.” Eng.,

Constr., Archit Manage., 8共2兲, 90–103.

Back, W E., and Bell, L C 共1995兲 “Monte Carlo simulation as a tool for

process reengineering.” J Management in Engineering, 11共5兲, 46–53 Ballard, G. 共2001兲 “Cycle time reduction in home building.” 9th Int.

Group for Lean Construction, D Chua and G Ballard, eds., National

Univ of Singapore, Singapore, 1–9.

Ballard, G., Harper, N., and Zabelle, T 共2003兲 “Learning to see work flow: An application of lean concepts to precast concrete fabrication.”

Eng., Constr., Archit Manage., 10共1兲, 6–14.

Ballard, G., and Howell, G 共1998兲 “Shielding production: Essential step

in production control.” J Constr Eng Manage., 124共1兲, 11–17 Bashford, H H., Walsh, K., and Sawhney, A 共2005兲 “Production sys-tem loading: Cycle time relationship in residential construction.”

J Constr Eng Manage., 131共1兲, 15–22.

Edwards, D J 共2003兲 “Accident trends involving construction plant: An

exploratory analysis.” J Constr Res., 4共2兲, 161–173.

Egan, J. 共1998兲 Rethinking construction U.K Department of Trade

and Industry, 具http://www.constructingexcellence.org.uk/pdf/ rethinking%20construction/rethinking_construction_report.pdf典 El-Rayes, K., and Moselhi, O 共1998兲 “Resource-driven scheduling of

repetitive activities.” Constr Manage Econom., 16, 433–446.

Farrar, J M., AbouRizk, S M., and Mao, X 共2004兲 “Generic

implemen-tation of lean concepts in simulation models.” Lean Construction J.,

1 共1兲, 1–23.

Frutos, J D., and Borenstein, D 共2003兲 “Object-oriented model for customer-building company interaction in mass customization

envi-ronment.” J Constr Eng Manage., 129共3兲, 302–313.

Gann, D M 共1996兲 “Construction as a manufacturing process? Similari-ties and differences between industrialized housing and car production

in Japan.” Constr Manage Econom., 14共5兲, 437–450.

Gil, N., Tommelein, I D., and Ballard, G 共2004兲 “Theoretical compari-son of alternative delivery systems for projects in unpredictable

envi-ronments.” Constr Manage Econom., 22共5兲, 495–508.

Goldratt, E M. 共1997兲 Critical chain, North River Press, Great

Bar-rington, Mass.

Halpin, D W 共1977兲 “CYCLONE: Method for modeling job site

pro-cesses.” J Constr Div., 103共3兲, 489–499.

Hanna, A S., Russell, J S., Gotzion, T W., and Nordheim, E V 共1999兲.

“Impact of change orders on labor efficiency for mechanical

construc-tion.” J Constr Eng Manage., 125共3兲, 176–184.

Hinze, J., and Tracey, A 共1994兲 “The contractor–subcontractor