You are on page 1of 26

The current issue and full text archive of this journal is available on Emerald Insight at:

www.emeraldinsight.com/0969-9988.htm

ECAM
27,1 Multi-agent simulation for
managing design changes in
prefabricated construction projects
270 Juan Du
SHU-UTS SILC Business School, Shanghai University, Shanghai, China and
Received 28 November 2018
Revised 23 May 2019 School of Building Construction, College of Design,
Accepted 23 June 2019 Georgia Institute of Technology, Atlanta, Georgia, USA
Hengqing Jing
Shanghai University, Shanghai, China
Daniel Castro-Lacouture
School of Building Construction, College of Design,
Georgia Institute of Technology, Atlanta, Georgia, USA, and
Vijayan Sugumaran
Department of Decision and Information Sciences,
School of Business Administration,
Oakland University, Rochester, Michigan, USA

Abstract
Purpose – The purpose of this paper is to develop a multi-agent-based model for quantitatively measuring
how the design change management strategies improve project performance.
Design/methodology/approach – Based on questionnaires and interviews, this paper investigates the
coordination mechanism of risks due to design changes in prefabricated construction (PC) projects. Combined
with all the variables related with design change risks, a multi-agent-based simulation model is proposed to
evaluate the design change management effect.
Findings – The coordination mechanism between design change factors, design change events, risk consequence
and management strategy in PC projects is described and then the simulation-based design change management
mechanism in PC projects is used to assess the effect of management strategies under dynamic scenarios.
Originality/value – PC projects have rapidly increased in recent years due to the advantages of fast
construction, high quality and labor savings. Different from traditional on-site construction, the impact and
risk from design changes are likely to be greater due to the prefabricated project being multi-stage, highly
interactive and complex. The simulations presented in this paper make it possible to test different design
change management strategies in order to study their effectiveness and support managerial decision making.
Keywords Risk management, Simulation, Decision support systems
Paper type Research paper

1. Introduction
Prefabricated construction (PC) has rapidly developed in recent years due to its
advantages for increasing construction efficiency, improving project quality and decreasing
environmental impact (Aye et al., 2012). The development of PC is allowing the construction
industry to move from a labor intensive environment to one where component
industrialization and standardization facilitate more mechanized and efficient construction
(Martinez et al., 2008). Compared to conventional on-site casting projects, PC includes design
Engineering, Construction and
and deep design stages, off-site prefabricated component production, logistics and inventory
Architectural Management
Vol. 27 No. 1, 2020
pp. 270-295 This work is supported in part by the National Natural Science Foundation of China under Grant
© Emerald Publishing Limited 71701121, and the Chinese Ministry of Education of Humanities and Social Science Project under Grant
0969-9988
DOI 10.1108/ECAM-11-2018-0524 17YJC630021.
and on-site assembly and construction. In order for PC to achieve a high efficiency and Multi-agent
effective implementation, intense connections and accurate communications are required simulation for
across the whole project stages. Such requirements make PC management more dynamic, managing
interactive and complex (Luo et al., 2015).
Among all the bottleneck problems in PC project management, design change is one of the design changes
typical risk factors (Li et al., 2016) and it refers to the process of modifying the original design
data during the project execution (Wesz et al., 2018). Even in successful industrial construction 271
projects, the costs associated with design changes typically cause a 5–8 percent budget
increase (Isaac and Navon, 2013). Different from traditional construction projects, PC projects
are more likely to trigger design changes because of their multi-stage and highly interactive
characteristics. Some of these characteristics consist of component splitting and embedded
parts design, impacted by production factors (e.g. production capacity, production line, mold
type) and on-site situations (e.g. assembly process, pipeline layout, technical level). Insufficient
considerations of the predisposing factors at subsequent stages will cause design change to
happen ( Jaillon and Poon, 2014).
Design change causes deviations from project objectives such as project cost, date of
completion, performance requirements, and could even increase conflicts among stakeholders.
This situation can be aggravated by engineering claims and contract disputes arising from
the evolution of the design change. Thus, there is a great need to examine the relationship
between design change causes, change events and change consequences, and then establish a
mechanism of coordination and strategy selection to manage the design change.
In order to manage the risk of design changes in PC projects, it is necessary to select
appropriate methods and tools according to the characteristics of design changes. First, the risk
of design change involves many project participants. There is a strong risk correlation among
the participants. It is necessary to establish the complex relationship among participants.
Second, design change management strategies need to consider the global impact of the design
change events, and build a correlation between design change factors and consequences.
Finally, all participants in the prefabricated project interact dynamically. Their behaviors are
interrelated and influenced, and their goals can be inconsistent or even opposite.
Some scholars have proposed related measures to improve design change, but not involving
a quantitative approach. Meanwhile, some studies considered design change as a link to risk
management, without specific analysis based on the characteristics of the design change.
Based on the characteristics of design change management in PC projects and measuring the
value of management strategies, this research proposes a multi-agent simulation model, which
has become a very useful approach to solve complex system problems (Karakas et al., 2012).
As a discrete and dynamic complex system, PC projects are composed of many stakeholders
with different objectives, whose knowledge and experience are very different. Meanwhile, the
discrete distribution of information at different stages increases the difficulty of complex
decision making. To sum up, the flexibility of MAS provides a new way for studying this
complex system (Dharmalingam and Eswaran, 2018; Tan et al., 2019; Yan et al., 2019). Many
scholars have used MAS to solve complex problems in engineering fields: Anumba et al. (2002,
2003) simulated design coordination in engineering with MAS. Obonyo et al. (2002) and
Hadikusumo et al. (2005) investigated the procurement of construction materials based on
MAS. However, these works mainly focused on solving optimization problems in specific areas
and did not consider the interaction factors among the processes within the whole project.
This paper aims to develop a multi-agent-based model for quantitatively measuring how
design change management strategies are able to improve project performance. First, a
coordination mechanism between design change factors, design change events, risk consequence
and management strategy in PC projects is described and then a simulation-based design change
management mechanism in PC projects is proposed. Second, a multi-agent model is put
forward to simulate the whole project process, including the interaction between stakeholders,
ECAM while considering the impact of design changes on each stage of the project. Finally, based on the
27,1 PC projects status, the corresponding design change management strategy can be adopted.
Through MAS, the results can test the implementation effect of different strategies and assist in
the formulation of management decisions.
The structure of this paper is as follows: In Section 2, previous studies are reviewed and
research gaps are analyzed. In Section 3, the coordination between design change cause
272 factors, consequences, and PC projects are described in detail. Section 4 presents the design
change management framework based on multi-agent modeling, and Section 5 gives a
detailed description of the multi-agent model. To show how the model can be applied, a case
study of a PC project in Shanghai is discussed in Section 6, and the conclusions and future
work are described in Section 7.

2. Related works
2.1 Design change in prefabricated construction projects
Design changes address all the modifications to existing design data during the whole
project stages and alter previous decisions made by the project team (Isaac and Navon,
2013). They also affect the economic, environmental and social goals of PC projects, which
makes the design change risks of a PC project more prominent than those of a traditional
one (Li et al., 2017). Compared with short-term goals of traditional construction, PC targets
also include many long-term goals, such as long-term performance of the project, its
environmental impact, and so on (Li et al., 2017). Achieving these goals in the entire project
lifecycle means that the design change risks are also pertinent to the whole lifecycle. In
addition, overall planning and new technology are the basis of scientific management in PC
(Pan and Sidwell, 2011). The management regarding PC is not mature, and experience is
limited, which increase the risk of design change (Luo et al., 2015).
In summary, design change is likely to occur in all stages of the PC project, and the closer
it occurs to the later stages, the greater the impact (Isaac and Navon, 2013). Meanwhile, the
factors that lead to the design change are complex, including force majeure and human
factors. These factors involve different stakeholders and form a complex system. Many
scholars regard design change as a subclass of risk event, which is studied using a relevant
method for risk management (Taillandier et al., 2015).

2.2 Risk management in prefabricated construction projects


Based on the literature related to risk management in PC projects, the methodology is
usually classified into two types: risk factor recognition and analysis by empirical analysis,
and risk evaluation and control by mathematical modeling and optimization.
Some scholars studied the risk hindering PC adoption based on empirical analysis. Mao
et al. (2013) analyzed 83 valid questionnaires and used ranking analysis to identify critical
risk factors of PC. Similarly, based on the investigation of 65 public housing pilot projects,
Lam et al. (2007) applied analytic hierarchy process to identify the major barriers in PC
development. On the basis of identifying risk factors, Luo et al. (2015) conducted a
questionnaire survey to quantify the significance of these risks, and Cronbach’s coefficient
was employed to measure the internal consistency among the risk factors. Arashpour et al.
(2016, 2017) proposed a holistic risk analysis approach that assessed the integrating impact
of interaction uncertainties in on-site and off-site projects. However, these studies did not
propose strategies for risk management. Based on structured questionnaire surveys and
face-to-face interviews, Hwang et al. (2018) and Zhang et al. (2018) investigated the
significant constraints demotivating PC adoption and proposed a group of feasible
mitigation strategies to tackle these constraints.
Optimization methods related to mathematical models and systems analysis provide an
opportunity to evaluate risks and propose management strategies. Li et al. (2017) adopted
system dynamics to analyze the investment risk management of PC in China and classified Multi-agent
the risk factors into five categories: economic risks, internal risks, technical risks, policy and simulation for
legal risks and market risks. Faludi et al. (2012) used the lifecycle assessment modeling to managing
evaluate the sustainability and optimal performance of PC projects. Through the identification
of investment risk, Wang et al. (2012) established the risk evaluation index system to obtain a design changes
PC investment risk evaluation system.
Based on the literature above, the quantitative assessment of the effect of risk 273
management has been insufficient, and the interaction between stakeholders in PC projects
has not been given enough attention. Simulation methods need to be explored to investigate
the improvement of risk strategy under complex dynamic scenarios.

2.3 Agent-based simulation methodology in project management


Agent-based models can be very helpful when modeling a collection of independent
decision-making entities as agents that sense and respond to a dynamic environment
(Walczak, 2018). Each agent assesses its own situation, analyzes its relationship with
other agents, and makes decisions based on a set of rules. In recent years, multi-agent
models have been widely applied in the field of construction engineering. In the process of
project development, there are often a series of interdisciplinary complex problems that
consider not only the external environmental factors of the system, but also the interaction
among different stakeholders. Compared with the traditional modeling method, the
multi-agent model has many advantages that are more suitable in project management
research. Bonabeau (2002) proposed three core advantages of the multi-agent model: in
contrast to traditional aggregate models, the multi-agent model builds a system from
bottom to up and is more likely to build a nonlinear complex model; the ontological
correspondence between the agents in the model and real-world actors makes it easy to
represent actors and the environment and their relationship; and the multi-agent model
has a high degree of flexibility, which can be defined in any given system environment,
and freely adjust its complexity.
Based on the above advantages, many scholars have used multi-agent models to solve
related problems in project management. The whole lifecycle of a construction project can
be divided into two phases: project planning and implementation. The key issues derived
from the two stages are as follows.
In the project planning stage, project design and bidding are the crucial issues affecting
the project progress and quality:
(1) Project design: Gerber et al. (2017) and Anumba et al. (2002) proposed a multi-agent
system framework for architectural design and collaborative design, respectively.
Moreover, Nguyen et al. (2012) put forward a distributed agent-based optimization
method for structural design.
(2) Project bidding: Awwad et al. (2014) proposed an agent-based model to optimize the
bidding strategy. Considering the significant effect of risk on project bidding, Asgari
et al. (2016) developed an agent-based model to study the risks on bidding behavior
in various market scenarios. Based on similar models, Kim and Reinschmidt (2010)
investigated the impact of risk coping attitudes.
The important issues involved in project implementation mainly include three aspects: planning
and scheduling, resource and task assignment and conflict resolution and negotiation:
(1) Planning and scheduling: the scheduling problem widely existing in construction is
typically considered a nondeterministic polynomial problem (Pinedo, 2012). Phanden
et al. (2011) contend that agent-based system has significant advantages for this
problem: parallel computation, dynamic adaptation and robustness.
ECAM (2) Resource and task assignment: in recent years, the developing complexities of projects
27,1 demand effective approaches to manage resources (El-Diraby and Zhang, 2006). Watkins
et al. (2009) developed an agent-based bottom-up approach that defines the efficiency of
every labor activity, so that task allocation can be planned accurately. Moreover, Koo
et al. (2012) proposed an intelligent knowledge management system based on agents.
(3) Conflict resolution and negotiation: Duan et al. (2012) proposed that agent-based
274 negotiation is a useful tool to help resolve conflicts and balance profits among
participants. Similarly, Ren and Anumba (2004) developed a multi-agent system
representing participants, who negotiate with each other to resolve construction
claims. In addition, Mahmoud et al. (2015) and Dzeng and Lin (2004) proposed an
automated negotiation-based framework via multi-agent system to support
negotiation between construction companies and suppliers.
In summary, the agent-based simulation methodology has provided effective ideas for solving
complex decisions in project management. However, these studies are limited to the project
specific domain, and do not consider the uncertainties and dynamic characteristics of typical
projects in various domains. The main observations from these models can be applied to PC as
follows: the related entities in the whole lifecycle of PC projects can be abstracted as agent
classes, and the relationship between entities can be quantitatively studied. The logic of the
model can be clearly expressed; considering the influence of external environment changes on
model variables, the model can be a better representation of the actual situation, which
guarantees the accuracy of the model; and the model can focus on the dynamic interaction
process among entities, design change and project variables. The dynamic characteristics of
the model will enable the model to adapt to different complex systems.

3. Design change management in PC projects


This section describes the typical design change scenarios in PC projects, analyzes the cause risk
factors of design change and proposes the design change coordination mechanism in PC projects.

3.1 Design change scenarios and cause factors in PC projects


Design change refers to the modification of design data during the whole project
lifecycle, which causes repeated work and process delays, and ultimately increase project
cost and duration (Isaac and Navon, 2013). A PC project is a complex system composed of
multiple stakeholders, and the operation flows between each stakeholder are shown
in Figure 1.
Different from the traditional cast-in-place architectural pattern, the design stage of PC
needs to increase the component deepening design process, where the components are
split into standard sizes to facilitate factory production. According to the drawings
provided by the designer, the fabricators obtain the mold processing drawings and the
rebar reinforcement sample drawings for production. Once completed, the transporter will
deliver the components to the spot for assembly. After reaching the construction site, the
assemblers will prepare for hoisting, which mainly includes on-site inspection, numbering
and detailed size verification. Finally, if the inspection is passed, the components will
be hoisted.
Design change may occur among various stages in the PC project as shown in Figure 1.
Based on interviews with project managers and related research on PC, the scenarios where
design change may occur were summarized as follows:
(1) Design change due to component deepening design: different from traditional on-site
casting projects, the PC project needs to add deep design of components, which is the
main factor triggering design change. The following aspects are important for
Design
Off-site Production Multi-agent
Design Change Bid for orders
Deepen the Make
simulation for
Structural
design
Architectural
design
Conceptual
design
Production delay
design drawing production plan
managing
Production error
Technology Mold drawing
(used to produce molds for mold factory)
design changes
restriction
Rebar pattern drawing
Water supply Heating and Component (used for reinforcing bar binding for workers)
Electric
and drainage ventilation deepening design

Component Component Component 275


yard maintenance production

Bidding
Design Change Design Change Design Change
Design error Component
Increase quantity
Design omission mismatch
Decrease quantity Owner
Unreasonable Uncontrollable
Change function Bidding
Design collision factor

Bidding
Figure 1.
On-site assembly Logistics Design change
among various stages
Component Hoisting
hoisting preparation Transportation Inspection in PC projects

understanding how the deep design of components are a main factor that triggers
design changes:
• Unreasonable design: this means that the deep design scheme, including
components splitting, design of embedded parts and pipelines and reserved hole,
do not conform to the actual situation. When splitting components, the designer
fails to take into account the mold type, production process and equipment of the
component factory. This will lead the fabricator to put forward design changes
because the design requirements cannot be met due to technical restrictions in
the component factory. In addition, the designer also fails to consider the factors
of on-site assembly, such as assembly process, equipment and technology. It
could cause design changes and lead to the entire project delay.
• Design omissions: design omission and incompleteness is a serious problem existing
in the current design market, which lays a potential risk for the subsequent stages.
The production and assembly stage requires constant design changes to improve
the original design, which increases the project duration and cost.
• Insufficient design depth: design depth refers to the degree of design refinement.
Insufficient design depth and unclear representation of drawings lead to
constant design changes, forming a vicious circle.
(2) Design change caused in production and assembly: for the fabricator, with the
increase of PC projects, the number of orders in component factories has considerably
increased. Backlogs of orders lead to production delays and cannot guarantee
component supply for the assembler, who may change the precast part to cast-in-place
through design change to prevent project delays. Many unpredictable situations can
happen in the assembly site, e.g., foundation deepening, water gushing, accidents, due
to inaccurate survey data, thereby leading to design changes.
(3) Lack of process management and coordination: institutional process management is
typically associated with the maturity of the industry. On the other hand, institutional
loopholes, tedious processes and other factors will increase the probability of
design changes:
• Poor process management: the turbulent management process and mechanism
loopholes often bury the risk of design change. For example, the imperfect
ECAM drawing audit mechanism of some design institutes leads to the decentralization
27,1 of responsibilities of all parties and weakens the auditing effect. This makes it
difficult to correct problems that can be avoided, thereby triggering design
changes at the later stages.
• Lack of coordination: the tasks in various stages are interconnected and improper
coordination will lead to design changes. Generally speaking, after the completion of
276 the previous task, the person in charge should provide relevant information and
parameters to the next task timely and accurately. However, in practical work, there
are many obstacles in communication between the parties, resulting in difficult data
interaction and cause frequent design changes. Meanwhile, the negative impact of
multi-channel management increases the possibility of design change.
(4) Design change triggered by external factors: the market and economic environment
are constantly changing, and the client will often change the original demand to
adapt to this change:
• Design change for project scale: objective environmental factors, such as
macroeconomic policies and economic cycles, can be conducive to adjusting
project decisions and then lead to design changes. In the downturn period of the
industry, the tightening of bank credit can directly affect the client’s investment
decisions, leading to the change of design schemes. If the client reduces the
project quantity, and the design work has been carried out, it will cause a large
number of design changes.
• Design change for project usage: affected by the fluctuations of market demand, in
order to save investments or further improve the economic benefits of the project,
the client can change the use function of the project, leading to design changes.
When the use function of the building changes, relevant design requirements and
specifications are also different, which will directly lead to design changes.

3.2 Design change coordination mechanism in PC projects


Design changes occur throughout the lifecycle of PC projects. Effective design change
management is an important way to ensure project quality and schedule and control cost in
PC projects. Based on the cause factors in different scenarios, design change management
strategies could be proposed and evaluated, just as showed in the upper part of Figure 2. In
order to quantify the value of the design change management strategy, the first step is to
clarify the relationships among design change cause factors, change events, change
consequences and project variables. When the project variable value exceeds a certain
threshold, the corresponding change cause factor is triggered. The combined effect of
multiple change factors leads to the occurrence of design changes and then results in change
consequences that alter the original value of the project variables.
As shown in Figure 2, if the deadline for the production of the prefabricated component is
tight, there might be a component production error such as the wrong position of the embedded
component, causing a design change. With the consent of the design department, the components
will be reproduced or modified. Thus, the time required to complete the task will be increased and
the overall project will be delayed. In turn, the delay will prompt the component factory to speed
up production and form a vicious circle. Then, if the factory standardizes the production process,
the probability of design change will be reduced to ensure that the project schedule is met.

4. Simulation-based design change management mechanism in PC projects


In order to select effective design change management strategies to improve project performance,
a design change management mechanism framework for PC projects is proposed, consisting of
components: coordination component, transition component and agent component.
Coordination Multi-agent
simulation for
Adjust Improve
Management strategies
Variables of project
and design change
managing
design changes
Propose

Design change Induce


Design change
Generate Change Impact Project cost 277
cause factors consequences and time

Project variables

Example

Standardized production process Adjust Task resource Improve


Ensure the operation time of key links Probability of change
...

Component Design change Impact


Increase task time Project time
production error in assembly

Production delay
Figure 2.
Coordination
Short task duration mechanism of design
change management

In the coordination component, and based on the design change coordination mechanism
described in Section 3, the interactions between the design change cause factors, the change
consequences, the design change strategy and the impact to the project are illustrated
in Figure 3.
In the transition component, the selected variables are extracted so as to quantitatively
measure the impact of management strategy on the project. Each type of design change event
(Ty) is triggered by the combined effect of multiple change factors. The risk level of design
change factors (IP) mainly depends on the importance of the factor on design change event
(IMa). For the consequences of change, two criteria, namely, impact on cost and time (a, b) and
probability ( p), are used to measure the consequence of design change (Li et al., 2018).
In the agent component, the multi-agent system is used to simulate the whole lifecycle
operation process of the PC project. Based on the typical roles in the field of PC, the model in
this paper considers five types of agents: project agent, stakeholder agent, task agent,
contract agent and design change agent. The five agents contain different elements or
variables, and the simulation results can be used to select the appropriate strategy, assisting
with the management decision making.
The project agent is a global agent that is characterized by a set of tasks, the project
objectives and constraints (i.e. time and cost constraints). The stakeholder agent
corresponds to the various people working on the project, namely, clients, designers,
fabricators, transporters and assemblers. As the task leader, stakeholders allocate resources
to the task, and the task agent describes the completion status of all the tasks, including task
priority, time and cost. The responsibility of all tasks is directly assigned by the contract
agent, which is a core agent in the entire project as it contains all the documents in the entire
construction lifecycle. The design change agent contains the change factor, design change
ECAM Coordination Component

27,1 Design change


cause factors
Design change of
PC projects
Design change
consequences
Project Performance
(cost and duration)

Improve
Standardization
Stakeholders Induce Design Generate Impact
Stakeholder
Technology Off-site production Delay project schedule
Task
Macro policy Logistics Increase project cost
Contract
Uncontrollable On-site assembly
environment

278 Propose
The impact of change on project cost (ai) Stakeholder (IMc, IMp)
Importance of the factor for the event (IMa) Threshold value the event is impacted (Th)
The impact of change on project time (bi) Contract (Rm, Ra, Ta)
Impact of the factor for the event (IP) Type of change event (Ty)
The probability change occurs (P) Task (D, P, Pr, S, Ru, Ts)

Strategies Selected
(A, B, C) strategies
Design change variables Project variables

Transition Component

Agent Component Multi-agent simulation model


Alter

Figure 3. Basis Affect

Design change Input Contract agent Stakeholder agent Task agent Project agent Design change Output
agent
management Time
mechanism framework Reward
Resource
Cost preference
Distribute
Priority
Form
Duration
Factors
Type
Time preference Progress Resource
in PC projects Duration
Resource
Consequence

events and the risk consequence. These five types of agents work as a basic unit and
cooperate with each other to complete the project.
Considering the variables and parameters of design change and the project, the
corresponding variables are abstracted from qualitative strategies. Then, these variables
are input into the multi-agent model, and the changes of project time and cost under
different strategies are obtained by simulation. Finally, the simulation results are used to
select and optimize the appropriate strategy, assisting with the management decision
making. The multi-agent model is described in detail in Section 5.

5. Multi-agent simulation model


5.1 Variables and ranges of the multi-agent model
The agent component of Figure 3 shows the five types of project agents and its corresponding
elements. Table I summarizes the agent characteristics.

5.2 Modeling
The following items are considered when modeling the process: the project plan considers the
relationship between tasks in the project, as these tasks are performed in a certain order. The
contract specifies the person in charge of each task, who is responsible for managing the
specific implementation process of the task. The progress of the task depends on the resources
obtained and the preferences of stakeholders. Design changes randomly occur at any stage of
the PC projects, affecting the cost and time of the project with different degrees of impact.
Each step of the simulation corresponds to a working day and the flowchart of Figure 4
details how the process works during each working day. Each task goes through eight
steps, and the input and output variables of each step are shown in the diagram. The
detailed calculation process of each step is showed in Sections 5.2.1–5.2.8.
These steps are briefly explained below:
(1) Each task updates its current state, including finished, executable and non-executable.
(2) The project agent checks whether the project is completed, which means whether all
tasks included in the project have been completed.
Agent Code Description Nature
Multi-agent
simulation for
Project PTn Project tasks Integer (Task Id) managing
PRu Resource used for the project Float
PRa Resource allowed for the project Float design changes
PTs Time spent for the project Integer
PTa Time allowed for the project Integer
Stakeholder IMc Importance of the cost criterion for the stakeholder Float [0, 1] 279
IMp Importance of the productivity criterion for the stakeholder Float [0, 1]
IDc Indication of cost given by the stakeholder Float [0, 1]
IDp Indication of productivity given by the stakeholder Float [0, 1]
Task D Normal number of days necessary to carry out the task No. of days
P Task priority Real (0, 1]
Pr Progress of the task Real [0, 1]
RDmin Minimal quantity of resources per day to carry out the task Float
RDmax Maximal quantity of resources per day to carry out the task Float
S State of the task progress {Non-executable,
executable, finished}
Ru Resource used for the task Float
Ts Time spent for the task Integer
Contract Rm Money allowed for the manager Float
Ra Resource allowed for the task Float
Ta Time allowed for the task Integer
Design IMa Importance of the factor for the event Float [0, 1]
change
IP Impact of the factor for the event Float [0, 1]
Si Indication of the positive or negative influence of the event {−1, 1}
Va The value of the origin variable Float
Th Threshold value for the variable from which the event is Float
impacted
Ty Type of change event Int [0, 3] Table I.
Pp Initial probability that this event occurs Float [0, 1] Agent characteristics

(3) The ongoing task calculates its priority. The higher the priority, the more important
the task is for the entire project.
(4) The task stakeholders issue instructions that represent the stakeholders’ emphasis
on cost and time. This depends on the current status of the task and preferences of
stakeholders.
(5) The person responsible for the task allocates resources to the tasks based on the
instructions received.
(6) Each task moves forward based on the resources obtained and the actual conditions
of the project.
(7) When the task is over, the person in charge of the task is paid according to the
completion of the task (i.e. contrast between the actual situation and the contract).
(8) The status of the project is updated, and the resources and time spent by the project
after a working day are calculated.
5.2.1 Update task status. Every task needs to update its state before running. There are
three kinds of task statuses: finished, executable and non-executable. When the task
progress is 1, the task is completed; if the task schedule is not equal to 1 and the task’s
predecessor has been completed, then the task is in progress; otherwise the task has
ECAM 1
Task agent
2
27,1 Task agent
Update task status S Check end condition End
Pr
PT, S

3
Task agent
Contract agent
Access task priorities P

280 Pr, D, Ru,


Ta

RDmin, RDmax, Ts
Stakeholder agent
Stake holder agent
IMc, IMp 4
IMc, IMp
Give instructions
IDc, IDp

Change agent 5
Allocate resources
Si, Va, Th, IMa

6 Task agent Task agent


Stakeholder agent Contract agent
Make progress Ru Pr, P, Ts
IDp, IMi Ra, Rm, Ta Ru
RAT, RDmin,
Ru, Ts
RDmax, Pr

Task agent 7
Pr, Ts Pay the manager
Pu, Ts
Figure 4. Project agent
Process workflow and PTs, PTa
8 Project agent
variables interaction PTs Update project status PTs, PRu

not begun. The contents in the parentheses represent the agent corresponding to the
variables:
8
< If Pr ¼ 1
> )S ¼ Finished
If Pr a1 and Ta A fPTg; PrðTaÞ ¼ 1 ) S ¼ Executable ; (1)
>
: If S aFinished and S a Executable ) S ¼ Non  Executable

where Pr is the task progress (task); S represents task status (task); and PT represents the pre
tasks (task).
5.2.2 Check end conditions. The project agent checks the status of all the tasks. If all
project tasks are finished, the project is completed and the simulation ends. At the same
time, if the time spent on the project is more than twice the target duration, the project is
considered as failed and the simulation ends:
(
If8Ta S ðTaÞ ¼ Finished ) Simulation ends
; (2)
If PTs 42PTa ) Simulation endsðfailure endingÞ

where PTs and PTa are, respectively, the time spent/allowed for the project (project).
5.2.3 Evaluate task priority. Each task’s priority is calculated according to the critical
path method. The tasks on the critical path have the highest priority. The tasks on
non-critical path can be allowed to delay without affecting the total project progress, and
their priority depends on the interval between the latest start time and the earliest start time. Multi-agent
The greater the time interval, the lower the task priority: simulation for
Md managing
P ¼ 1 ; (3) design changes
maxð1; Mdm Þ
where P stands for task priority (task); Md represents the time interval on the critical path;
and Mdm is the maximum value of Md. 281
5.2.4 Give instruction. The task stakeholders need to give instructions on the two
dimensions of the task: cost and productivity. The instruction is expressed by a numerical
value between 0 and 1. The greater the value, the more attention is paid to this factor. In
particular, the greater the productivity index, the more the stakeholder wants to increase
production; the greater the cost index, the more the stakeholder wants to limit the cost. The
sum of the cost and productivity index is 1, which depends on the current state of the task
and the preference of the stakeholders for cost and time:
CSi þ IMi
8i A fp; cgIDi ¼ P ; (4)
i A fp;cg CSi þIMi

where IDi (stakeholder) represents stakeholder’s index for task cost (IDc) and time (IDp)
(IDc+IDp ¼ 1). CSi is the current state of cost and time of the task, and from experts’
judgment, whose computation is shown in Equations (5) and (6). IMi represents specific
interest the stakeholder has regarding these two dimensions of project cost and time:
 
2Ru
CSc ¼ min 1; ; (5)
ðRDmin þRDmax Þ  Pr  D
where CSc represents the current state of the cost factor of the task, and it depends on the
comparison between the actual cost when allowing consumption of resources. The more
resources are actually consumed, the more control the cost will be for stakeholders, so as to
avoid exceeding the budget. RDmin and RDmax, respectively, represent the minimum and
maximum amount of resources per day allowed by the task (task). Pr is the progress of the
task (task), which is expressed as a percentage. D is the standard time for the task (task), and
Ru represents the actual consumption of resources at present (task):
 
Ts
CSp ¼ P  min 1; ; (6)
Ta  Pr
where CSp is the current state of the productivity factor of the task, and it depends on the
ratio between the expected progress and the actual progress of the task. The smaller the
actual progress, the more the stakeholders will increase the productivity so as to complete
the task within the limited time. P stands for task priority (task). Ts and Ta, respectively,
represent the time spent and time allowed for the task (task):
maxð0; Si  ðVaThÞÞ
IP ¼ IMa  ; (7)
1Th
where IP represents the influence of change factors on change events (change). The greater
the value, the more likely that the factor will lead to design change. The IP value depends on
the gap between the original variable value and the threshold (Table I). There are many
factors that cause change events. IMa indicates the weight of the change factor (change). The
greater the weight, the more important the factor is to the event. Si represents the sign
ECAM component (change) of a variable, ensuring that IP is positive. Va is the value of the original
27,1 variable, and Th stands for the variable threshold (change). The probability of change
events is affected by the factors. Referring to the calculation of risk event probability
(Taillandier et al., 2015), the design change probability is shown in the following equations:
!
  X
P dj ¼ k ¼ Pp  1 þ IPi ; k A f0; 1g; (8)
282 i A ffactor linked to the eventg

where Pp is the probability of the change event in the last round of simulation (change).
5.2.5 Allocate resource for tasks. Each task has a corresponding person in charge
responsible for the implementation and supervision of the task. After receiving instructions
from stakeholders, the person begins to allocate resources to the task. The amount of
resources that the task ultimately obtains depends on the quantity stipulated in the contract
and the coefficient of adjustment. The adjustment coefficient includes three aspects: the
correlation coefficient of stakeholders’ cost constraints, the productivity adjustment and the
penalty coefficient of task delay. Furthermore, the allocation of resources is also affected by
the design change, which will result in a large amount of waste of resources:
Ra  
RAT ¼  1 þKc þKp þKdþaj dj ; (9)
Ta
where RAT is the number of resources that the task has acquired per day. Ra and Ta denote
the resource amount and time permitted by the contract (contract). Kc, Kp and Kd are,
respectively, the correlation coefficient of stakeholders’ cost restriction, productivity
adjustment and the penalty coefficient of time delay. aj ( j∈Task Id) represents the impact of
design change on task resource, and δj is {0, 1}, indicating whether the change occurred:
Kc ¼ 0:5IDc ; (10)

Kp ¼ IDp 0:5; (11)


  
Ts
Kd ¼ min 0; P  Pr ; (12)
Ta

Ru ¼ Ru þRAT; (13)
where IDc and IDp are the index on task cost and productivity of stakeholders (stakeholder).
If they receive equal attention, then IDc ¼ IDp ¼ 0.5(IDC + IDP ¼ 1). Ru represents the actual
resource consumption of the task (task). The probability of design change is calculated by
the same Equations (8) and (9).
5.2.6 Make progress on the task. Task progress depends on three factors: the amount of
resources obtained, the characteristics of the task and the impact of design change. Based on
the previous studies in this field, the impact of the amount of resources on progress is
expressed by the quantity modification (MR), see Equation (15). The impact of the task
characteristics on the progress is expressed by the speed modification (MP), see Equation (16):
1 þPTs Ubj Udj þ MRþMP
PrD ¼ ; (14)
D
where PrD is the task’s progress on a specific day; PTs the time spent for the project (project);
D represents the standard time for the task (task); MR the quantity modification and MP
represents the speed modification; bj( j∈Task Id) represents the influence of the design change Multi-agent
on task time, and the δj is taken as {0, 1}, to indicate whether the event occurs. simulation for
The greater the value of MR, the greater the amount of resources the task obtains, managing
leading to a higher progress. When the value of RAT is exactly in the middle of the RDmin
and RDmax, then MRm is equals 1.5. RDmin and RDmax, respectively, represent the minimum design changes
and maximum amount of resources per day allowed by the task (task):
8 283
> RATo RDmin ; MRm ¼ 0
<
RATRD
MR ¼ MRm 1:5 ) RDmin oRAT oRDmax ; MRm ¼ 1 þ RDmax RDmin ; (15)
>
:
min

RAT4 RDmax ; MRm ¼ 2

MP ¼ 2IDp 1; (16)

Pr ¼ minð1; maxð0; Pr þPrDÞÞ; (17)

Ts ¼ Ts þ1; (18)
where MP is the speed modification which depends on the instructions of stakeholders on
productivity. If the index is more than 0.5, the stakeholder attaches importance to the
productivity factor and does not allow the task to be delayed, which will bring a faster speed
and promote the task progress. IDp is the stakeholders’ instructions on task productivity
(stakeholder).
5.2.7 Pay the manager. When a task is finished, this step is triggered, and the person in
charge of the task can get the corresponding remuneration. The amount of remuneration
(RM) is adjusted based on the stipulations of the contract. The adjustment is based on the
completion of the task, that is, the less time the task is spent and the lower the cost, the
higher the remuneration can be obtained by the person in charge:
    
TaTs RaRu
RM ¼ Rm  1 þ min 0; þ min 0; ; (19)
Ta Ra

Ru ¼ Ru þRM ; (20)
where Rm represents the remuneration that is available to the manager for the task specified
in the contract; Ts and Ta, respectively, represent the time spent and time allowed for the
task (task). Ra and Ru are the resource consumed and resource allowed for the task (task).
5.2.8 Update the status of the project. After each round of simulation, the time and
resource status of the project should be updated:
PRu ¼ PRu þRu; (21)

PTs ¼ PTs þTs; (22)


where PRu is the resource consumed by the project (project); and RM the remuneration of the
task leader.

5.3 Process example


This section illustrates the operation of the model through a simple example. The project
consists of five tasks, and task 2 is in the “executable” state. The variables used in the
ECAM example are derived from Table I. The following calculation refers to the 22 equations
27,1 proposed above and the whole calculation process from steps 1–8 is shown in Table II:
(1) Update the task status.
At present, the progress of task 1 is 100 percent, and the progress of task 2 is
30 percent.
(2) Check end conditions.
284 In the project, tasks T2–T5 did not reach the “finished” status, so the simulation
continues.
(3) Evaluate task priority.
As shown in Figure 5, the logical relationships between tasks are represented by
Activity-On-Node, and then the priority of each task is calculated according to the
critical path method.
(4) Give instruction.
Task 2 is in execution, and the progress (Pr) is 0.3. The time spent on T2 (Ts) is
2 days, the allowed time for T2 (Ta) is 5 days, equal to the standard time (D). The
minimum resource (RDmin) and the maximum resource (RDmax) of the task are $3k
and $4k per day, respectively, and the resource used in the task so far is $7.95k. The
preference value of cost and productivity for stakeholders is set to IMc ¼ 0.5,
IMp ¼ 0.6. For demand change, consider the occurrence of one change event as
follows: the owner’s estimate of the cost is low. This will reduce the importance of
cost factors, set the influence value a1 ¼ −0.1 and δ1 is 1. First, calculating the
mid-variables CSc ¼ 1 and CSp ¼ 1, and finally, get the instruction parameters of the
stakeholders to the cost and productivity: IDc ¼ 0.47, IDp ¼ 0.53.

No. Input variable Mid-variable Output variable

1 Pr: (T1 ¼ 100%) – S: (T1: finished


T2 ¼ 30% T2: executable, T3–
T3 ¼ T4 ¼ T5 ¼ 0% T5: non-executable)
2 S: (T1: finished, T2: executable, T3 ¼ T4 ¼ T5: non- – Simulation continues
executable)
3 PT: Figure 4 – P: (T1–T5 ¼ 1,
S: (T1: finished, T2: executable, T3 ¼ T4 ¼ T5: non- T3 ¼ 0)
executable)
4 Task: Pr ¼ 0.3, P ¼ 1, RDmin ¼ $3k/day, RDmax ¼ $4k/day, CSc ¼ 1, CSp ¼ 1 IDc ¼ 0.47
Ru ¼ $7.95k, D ¼ 5 days, Ts ¼ 2 days; contract: Ta ¼ 5 days; IDp ¼ 0.53
stakeholder: IMc ¼ 0.5, IMp ¼ 0.6; a1 ¼ −0.1
5 Task: Pr ¼ 0.3, P ¼ 1, Ts ¼ 2 days, Ru ¼ $7.95k Kc ¼ 0.03 Ru ¼ $11.99k
contract: Ta ¼ 5 days; stakeholder: IDc ¼ 0.47 Kp ¼ 0.03
IDp ¼ 0.53; a2 ¼ 0.05 Kd ¼ 0.1
RAT ¼ $4.04k/day
6 Task: Pr ¼ 0.3, RDmin ¼ $3k/day, RDmax ¼ $4k/day, MR ¼ 0.5 Pr ¼ 0.61
Ts ¼ 2 days, D ¼ 5 days; project: PTs ¼ 7 days, PTa ¼ 30 MP ¼ 0.06 Ts ¼ 3 days
days; PrD ¼ 0.31
stakeholder: IDp ¼ 0.53; RAT ¼ $4.04k/day; a2 ¼ 0.05
7 Task: Ts ¼ 2 days, Ru ¼ $7.95k; contract: Ta ¼ 5 days, RM ¼ $8.4k Ru ¼ $30.4k
Ra ¼ $20k, Rm ¼ $12k
Table II. 8 Stakeholder: Ru ¼ $30.4k; project: PRu ¼ $19k; – PTs ¼ 11 days
Calculation process task: Ts ¼ 2 days PTu ¼ $49.4k
(5) Allocate resource for tasks. Multi-agent
The resources available for the task are affected by three factors. First, the contract simulation for
stipulates the amount of resources that can be consumed (Ra), which is the basis of managing
resource allocation. Second, the instruction from stakeholders (IDc, IDp) and the penalty
mechanism will affect resource allocation. Finally, the change event can lead to waste of design changes
resources. In this process, this paper considers the following change event: the fabricator
is delayed, and the assembler will change the prefabricated part to cast in place to ensure 285
progress, thereby causing waste of resources. The impact of the event on the task is set
to b2 ¼ 0.05 and δ2 is 1. The amount of resources consumed is $7.95k. The mid-variable
values of the process are calculated as: Kc ¼ 0.03, Kp ¼ 0.03, Kd ¼ −0.1, RAT ¼ $4.04k.
Based on the mid-variable, the output variable value is Ru ¼ $11.99k.
(6) Make progress on the task.
The progress of the task depends on the amount of resources and the
characteristics of the task. The resource quantity modification (MR) is determined
according to the comparison between the resource allocation quantity (RAT) and the
minimum and maximum consumption value for the resource. The speed modification
(MP) depends on the instruction value on productivity of the stakeholders (IDp).
Meanwhile, an environmental change takes place: due to inaccurate survey data, the
assembler finds water gushing at the foundation excavation, so the superstructure
needs to be redesigned, resulting in a design change. Since a design change event
occurs at later stage of the project (construction phase), it has a significant impact on
the progress, setting the impact value of a3 ¼ 0.2 and δ3 ¼ 0 (the event has not
occurred). It is assumed that the project is currently taking a total of seven days.
Intermediate variables can be obtained through calculation: MR ¼ 0.5, MP ¼ 0.06,
PrD ¼ 0.31. The output variable is Pr ¼ 0.61, Ts ¼ 3 days.
(7) Pay the manager.
When the task is finished (Pr ¼ 1), process 7–8 starts. The time and resource of
T2 was set as Ts ¼ 6 days and Ru ¼ $22k. The remuneration of the person in charge
stipulated in the contract is $12k, and on this basis, adjustments can be made
according to the delay of the task and the consumption of resources. After obtaining
the mid-variable RM ¼ $8.4k, the output variable Ru ¼ $30.4k is obtained.
(8) Update the status of the project.
After each task, the resources and time consumed by the project need to be updated. The
resource and time used by the preceding task are PRu ¼ $19k and PTs ¼ 5 days,
respectively. It is calculated that the current resource consumption of the project is
PRu ¼ $49.4k, and the time consumption is PTs ¼ 11 days.

Md (T1=T2 = T4 =T5 =0)


ES Task EF Md (T3 = 20–18 = 2)
Mdm =2
P1= P2 =P4 = P5 =1
LS Duration LF
P3 = 0
13 T3 18

13 5 20
26 End 26
0 T1 8 8 T2 13 20 T5 20
Figure 5.
0 8 8 8 5 13 20 6 26 26 0 26 Example of
13 T4 8 calculation of priority
on critical path
13 7 8
ECAM 6. Case application
27,1 To verify the feasibility and applicability of the multi-agent model of design change and to
provide direction for management strategies, a residential PC project in Shanghai was
analyzed. The process of applying the proposed management mechanism and model
included the flowing steps: collect project basic data used in the model based on interviews
with project managers and technicians involved in this project; conduct a survey of involved
286 experts to obtain the design change events in this project and get the probability of the
event and weight on the project based on expert evaluation method; carry out simulation
and quantitatively analyze the effect of different management strategies and assist in
making management decisions.

6.1 Project description


The PC project is the Chenxiang Road Station construction in Shanghai District, located in
Nanxiang Town, Jiading District. The information for the case project is shown in Table III.
The project was divided into 16 sub-tasks according to project characteristics. In order to
describe each task quantitatively, we interviewed project managers and technicians
involved in this project. Initial parameters of each task were obtained, including the
standard time to complete the task (D) and the minimum and maximum resource
consumption (RDmin, RDmax), as shown in Table IV. In particular, the resources mentioned in
this paper include consumable resources (e.g. materials) and reusable resources (e.g.
equipment, human), which are calculated in the form of funds to simplify the operation
process. Minimal resources correspond to the minimum amount of resource necessary to
make the task progress. Maximal resources correspond to the maximum amount of useful
resources to work on a task. The values in Table IV are derived by combining historical data
from similar projects and years of field experience of the project managers.

Type of company Shanghai Xing Xin Real Estate Development Company


Table III. Structure type high-rise PC structure residence Location Shanghai
Basic information of Storeys 22 with one underground floor Land area 16,583 m2
the case project Investment RMB 50m Greening ratio 60%

Stages Tasks D (day) RDmin (¥1,000) RDmax (¥1,000)

Establishment 1. Feasibility studies 18 4.8 6


2. Preliminary technical research 21 3.8 5
3. Prepare bid package 17 3.5 5
Design 4. Conceptual design 5 13 14
5. Architectural design 10 11 12
6. Structural design 7 11.2 15
7. Component deepening design 15 11.4 11.5
Production 8. Mold processing drawing 45 5.6 6
9. Rebar sample drawing 45 0.2 0.245
10. Component production 140 18 24.3
11. Component maintenance 28 14.5 20
Transportation 12. Component transportation 40 4.8 7
Assembly 13. Foundation 28 4.5 7.8
Table IV. 14. Hoisting inspection 10 0.5 1.5
List of tasks and 15. Component hoisting 150 1.2 1.87
their attributes Final inspection 16. Completion acceptance 15 0.8 2
To ensure that all tasks can be achieved, the contract is signed for each task which indicates Multi-agent
the allowed resources and time consumed (Ra, Ta) by each task, and the remuneration (Rm) simulation for
obtained by the task leader. The contract includes all documents establishing a link between managing
stakeholders and the project requirements according to each task. This specific information
is shown in Table V. design changes
In addition, the 16 sub-tasks have parallel and successive relationships in the project, and
each task has a specific task leader, responsible for supervising the smooth execution of the 287
task, as shown in Figure 6.
Stakeholders involved in this project can be classified into two categories: cost-oriented
and time-oriented, due to their different value preferences and project experience. Cost-
oriented stakeholders will give higher preference values to cost (IMc) than time (IMp), and
focus on limiting project costs. A survey was conducted with experts in the PC industry to
elicite their preference values for the cost and time dimensions. The initial values of the
stakeholders’ preferences for cost and time gathered from the survey are used to compute
the average, which are shown in Table VI.

6.2 Design changes and management strategies in the project


Questionnaires were used to collect design change information from experts who were
participants in the PC project. Survey results are shown in Tables VII and VIII.

ID Tasks Rm (¥1,000) Ra (¥1,000) Ta (day)

C1 1. Feasibility studies 10 11 18
C2 2. Preliminary technical research 20 5 21
C3 3. Prepare bid package 10 8 18
C4 4. Conceptual design 15 15 5
C5 5. Architectural design 20 14 10
C6 6. Structural design 20 15 7
C7 7. Component deepening design 40 12 15
C8 8. Mold processing drawing 50 10 45
C9 9. Rebar sample drawing 5 0.3 45
C10 10. Component production 200 25 140
C11 11. Component maintenance 100 20 28
C12 12. Component transportation 50 8 40
C13 13. Foundation 80 7.5 30
C14 14. Hoisting inspection 10 1.5 10
C15 15. Component hoisting 50 1.8 150 Table V.
C16 16. Completion acceptance 10 3 15 Contract information

Project Designer
manager Assembler
13

1 2 4 5 6 7 8 10 11 12 14 15 16

3 9

Figure 6.
Client Producer Transporter Supervisor Tasks relations
ECAM The design change events in the project are classified into four categories based on the
27,1 classification in Section 3: design change due to deepening design; design change caused in
production and assembly; process management and coordination; and design change triggered
by external factors. In order to clarify the impact of design change events on the project, the
expert evaluation method is adopted here. We invited 15 field experts (i.e. project managers,
project assistant manager, field engineers, water and electricity engineers and the project
288 supervisor), to rate the probability of design changes and the influence weights of events on the
project (1–10). Thus, the total score of each category is calculated by the product of probability

Stakeholders IMc (importance of the cost criterion) IMp (importance of the time criterion)

Client 0.7 0.8


Project manager 0.8 0.9
Designer 0.7 0.8
Table VI. Fabricator 0.6 1
List of stakeholders Transporter 0.5 0.7
and their preference Assembler 1 0.6

No. Design change events Probability Weight

E1 Embedded wire pipe on the floor slab cannot be connected on the construction site 0.8 6.5
E2 Steel bar and reserved hole collision in column production 0.75 5.0
E3 Outer wall panel missing pre embedded wire box 0.55 4.0
E4 Laminate panels were cut 1/3 0.4 8.0
E5 Assembler canceled production of laminated floor slabs that have been completed 0.36 8.5
E6 Wall panel could not be demolded in production 0.35 5.5
E7 Designer changed the size of external wall panel opening from 315 mm to 345 mm 0.35 3.5
E8 Designer changed the size of the grouting sleeve from 36 mm to 25 mm 0.35 4.0
Table VII. E9 Prefabricated balcony is returned by the assembler 0.2 8.5
Design change events E10 Client proposed partial adjustment to the kitchen 0.09 9.0

No. Causes Classification E

1 Designer provides unreasonable drawings due to lack of field experience (1) E1


2 Drawings of PC components are inconsistent with construction drawings due to (3)
information opacity
3 Various design disciplines are lacking required “linking” knowledge (1) E2
4 No collision test after design completion (3)
5 Designers lack professional training (1) E3
6 Designers give unreasonable drawings, not considering the actual site (1) E4
7 Component size is not taken into account when assembler arranges reinforcement (2)
8 Insufficient communication between the two parties (3)
9 Due to construction site limitations, only cast-in-place floor can be used (2) E5
10 The assembler failed to notify the manufacturer in time to cancel the components (3)
11 Keyway design makes it impossible to demold (1) E6
12 Designer is short of time and hastily submitted diagrams (1) E7
13 36 mm sleeve matching materials are hard to purchase (1) E8
14 Designer is not familiar with market conditions (4)
15 Excessive production orders cause balcony production delays (2) E9
Table VIII. 16 Information between designer and assembler is out of sync (3)
Design change causes 17 There are some changes in market demand (4) E10
and weight. The scores of the first to the fourth categories are 11.5, 3.4, 7.9 and 0.8, respectively. Multi-agent
Therefore, based on the relevant previous study and experts interviews of this project, the simulation for
corresponding strategies are proposed mainly for the first and third design change types. managing
For type 1 (design change due to deepening design), strategy 1 is proposed: allocating more
resources to key task 4–8 at the design stage. The cause factors of design change at the design design changes
stage mainly include designer’s error and omissions, lack of project experience and so on. In
fact, the root cause of these factors is the lack of time for the task. Some design institutes have 289
many heavy engineering tasks, leaving the designers with a shortage of thinking and design
time. In view of this situation, key links along the whole lifecycle of PC project should be
controlled to increase the time and resources for key tasks, to ensure its normal operation, and
to reduce the probability of design changes. For example, component detailed design is an
intermediate link between the designer and the fabricator. The drawings should not only
represent the architectural design scheme, but should also be feasible. It is necessary to leave
enough time for the component detailed design, otherwise the design change will likely occur
in the project, which is already aimed at progressing according to schedule.
For type 3 (lack of process management and coordination), strategy 2 is proposed:
establishing a department that specializes in design change. In terms of the cause factors of
the design change, it mainly refers to the normalization of the process. The standard
management system and process are the operation criteria of the project and effectively
enhance performance. Therefore, it is necessary to have unified management of design
changes. Furthermore, design changes are managed centrally and uniformly by a business
unit, and all design changes are handled, issued, archived and numbered by the same unit. All
design changes are calculated, entered and counted, so that the dynamic state of design
changes can be grasped in time, which will provide reference for design decisions of
subsequent unfinished projects. Meanwhile, the department should establish an effective
evaluation mechanism for design changes, which is uniformly issued after internal evaluation,
thus thoroughly changing the situation that design changes occur frequently.

6.3 Multi-agent simulation for managing design change


The multi-agent model was implemented in Matlab. The Simulink in Matlab provides an
integrated environment for modeling, simulation and comprehensive analysis of dynamic
systems. Matlab has complete visualization functions for displaying computation results
and programming.
The interaction among design change agent, stakeholder agent, project agent, contract agent
and task agent is the basis of the simulation, which has been illustrated in Figure 4. The project
data (Tables IV–VIII) are input into the multi-agent model as the initial parameters. The eight
steps in the model are implemented in Matlab and Figure 7 shows the variable interaction
process in Step 6. After exporting the simulation results to Excel, the detailed statistical analysis
can be carried out on global performance of the project cost and time. This analysis will enable
to improve the management of the project in order to better account for its stochastic features.
The selected management strategies are simulated. Strategy 0 refers to the normal operation
of the model in accordance with the default rules without intervention. Strategy 1 is to control
the tasks of design stage which are quite critical in this PC project. Based on the actual
experience of project managers and technicians, the corresponding quantitative indicators from
the strategy were extracted. Therefore, the standard completion time of tasks 4–8 (D) were
increased by 10 percent, and the resources consumed (Ru) were increased by 5 percent. Then,
the probability of design change was reduced (P) by 15 percent. Strategy 2 is to set up a special
coordination department based on professional management to monitor the design changes.
This department will add an extra 1 percent to the total cost and reduced the upper limit of
resources (RDmax). At the same time, the probability of design change remains unchanged, but
the impact of design change on the project cost and time (a, b) is reduced by 5 percent.
ECAM
27,1

290

Figure 7.
Simulation execution
in Matlab

The three strategies were executed 2,000 times each in the simulation software, and the
statistical results are shown in Figure 8. The actual time of the project was 498 days (485 days
was originally planned) and the cost was the same as the plan. Strategy 0 is the closest to the
actual situation of the project. In general, the actual time spent on the project is slightly greater
than the simulation results of Strategy 0. The selected strategies have their own advantages
and disadvantages. Strategy 1 spent more time on critical links, resulting in an increase in
total project time. However, because there is enough time to control the critical link, the
verification of the upper and lower parts was more important. This greatly reduced the
incidence of the design change, thereby avoiding the unnecessary waste of resources and
reducing the total project cost. Strategy 2 focuses on specialized management. When design
change occurs, they can be quickly handled, easing the impact of design change on project
progress. However, this strategy is an afterthought and cannot prevent design change.
Based on these strategies, a strict responsibility system for the PC project process should
be established among stakeholders. For the design teams, they should provide careful
examination during and after the design and improve the design process to reduce the
probability of the design change. Meanwhile, as the highest-level managers of the project,
the project manager and project secretary should have strengthened the communication of
the participants based on project real-time conditions. Also, they should give assistance and
coordinate the interaction between the design teams and other stages in this project.

S0 S1 S2 S0 S1 S2
30% 27%
27% 24%
24% 21%
21%
18%
18%
15%
15%
12%
12%
9%
9%
6%
Figure 8. 6%
3%
Simulation results of 3%
0% 0%
project time and cost 474 477 480 483 486 489 482 495 498 501 504 507 510 513 516 519 522 525 528 4,930 4,940 4,950 4,960 4,970 4,980 4,990 5,000 5,010 5,020 5,030 5,040 5,050

Project Time (Day) Project Cost (¥ 10 K)


6.4 User interface development Multi-agent
In order to simplify the application of the model and the Matlab-based simulation platform, simulation for
a user-friendly interface was created in a practical application. It has been integrated into managing
the “Shanghai Urban Rail Transit Prefabricated Components Management System,” which
is used in most of the Urban Rail Transit PC construction projects in Shanghai. This design changes
platform displays the operation status of the whole PC project lifecycle to all participants.
Through quality traceability and information gathering during design, production, 291
transportation and assembly stages, as well as real-time correlation with the BIM model, the
platform provides support for various decision-making activities in project execution.
As shown in Figure 9, the “Change Management” module consists of two parts: data
input and results display. The user can import initial values for the model from the “Data
Center” linked to the platform, and the results are displayed after the simulation is
completed. The Data Center contains real-time data from production and construction sites.
Users can adjust the parameters of each task and adopt different design change
management strategies. The simulation results under these strategies are presented in the
form of charts and graphs. The simulation data can also be exported and various data
analysis can be carried out according to the needs of the decision maker. At present, the
platform is being used in several PC projects in Shanghai.

7. Discussion and conclusions


The proposed model provides a systematic quantitative analysis method for design change
management in PC projects. This paper proposed a decision-making evaluation model based
on MAS, considering the interactive, global and dynamic characteristics of design change risk
in PC projects. As a supplement to empirical methods in risk management research, this model
provides a flexible decision-making evaluation method based on quantitative analysis and
specific scenario problems. This multi-agent simulation model is oriented to the whole process
of PC project. Therefore, it can be used not only for design change risk, but also for risk
decision-making evaluation of other PC projects, which has operable practical value.
Our study clarified design change events and factors in PC projects and put forward
design change management mechanisms based on the proposed multi-agent model. First, we
identified design change scenarios and cause factors in PC projects by using the scientific
literature and previous construction experience. Next, based on the results, a design change
management mechanism framework for PC projects was proposed to improve project
performance. Finally, a multi-agent model was proposed to measure the effect of design

Figure 9.
Change management
interface for the
Shanghai urban rail
transit prefabricated
components
management system
ECAM change management. This model integrated the dynamic characteristics of PC projects and
27,1 design change. The dynamic process of stakeholder interaction was also integrated into the
model. This is crucial for the accuracy of the model, albeit often ignored by traditional design
change management methods. In addition, the model was applied to a practical PC residential
project for simulation. These results render novel and practical ideas for design change
management, particularly to test the effect of different management strategies.
292 There is still much room for improvement on the following aspects. First, before running
the model, we need to determine the design change agent, that is, specify the design change
event type, the probability of occurrence and the consequence. The PC project displayed in
the case study is just an example from China’s residential building sector. However,
different projects have different attributes and stakeholder characteristics, which lead to a
great difference in the design change agent. Therefore, before running the model, it is
necessary to conduct systematic design change analysis and establish a corresponding
agent that conforms to the actual situation, so as to obtain more accurate operation results.
Second, as an initial analysis prototype, the relationship analysis of stakeholders in the
model is simplified. For example, in this model, the person in charge of each task is set as the
stakeholder of one party. In fact, it is possible for a task to be undertaken simultaneously by
multiple participants who form a management chain. If the relationship between stakeholders
is defined in more detail and clarity, the accuracy of the model will be improved.
Finally, a design change case library can be established to store various events in the
form of a change catalog. Therefore, the output function of the model can be expanded. After
each simulation, the design change events of this simulation can be output and stored in the
case library. After many experiments, the case library can be enriched and improved
continuously, providing references for the analysis of design change for later use.
Meanwhile, the combination of the case base and the model also helps improve the efficiency
and accuracy of the model.
There are a number of potential future research directions. Prefabricated buildings are
playing an increasingly important role in rapidly developing countries like China. Compared
with traditional cast-in-place buildings, PC projects are a more complex system. The model
description of the system needs to be more accurate, and entities in the system can establish
more detailed correspondence with agents in the model. On the other hand, the classification
of design change in this paper is based on expert opinion and is subjective. Future research
can form a case library of design changes based on experience and improve the ability to
integrate them with the model.

References
Anumba, C.J., Ugwu, O.O., Newnham, L. and Thorpe, A. (2002), “Collaborative design of structures
using intelligent agents”, Automation in Construction, Vol. 11 No. 1, pp. 89-103.
Anumba, C.J., Ren, Z., Thorpe, A., Ugwu, O.O. and Newnham, L. (2003), “Negotiation within a multi-
agent system for the collaborative design of light industrial buildings”, Advances in Engineering
Software, Vol. 34 No. 7, pp. 389-401.
Arashpour, M., Abbasi, B., Arashpour, M., Hosseini, M.R. and Yang, R. (2017), “Integrated management
of on-site, coordination and off-site uncertainty: theorizing risk analysis within a hybrid project
setting”, International Journal of Project Management, Vol. 35 No. 4, pp. 647-655.
Arashpour, M., Wakefield, R., Lee, E.W.M., Chan, R. and Hosseini, R. (2016), “Analysis of interacting
uncertainties in on-site and off-site activities: implications for hybrid construction”, International
Journal of Project Management, Vol. 34 No. 7, pp. 1393-1402.
Asgari, S., Awwad, R., Kandil, A. and Odeh, I. (2016), “Impact of considering need for work and risk on
performance of construction contractors: an agent-based approach”, Automation in
Construction, Vol. 65, pp. 9-20.
Awwad, R., Asgari, S. and Kandil, A. (2014), “Developing a virtual laboratory for construction bidding Multi-agent
environment using agent-based modeling”, Journal of Computing in Civil Engineering, Vol. 29 simulation for
No. 6, pp. 1-14.
managing
Aye, L., Ngo, T., Crawford, R.H., Gammampila, R. and Mendis, P. (2012), “Life cycle greenhouse gas
emissions and energy analysis of prefabricated reusable building modules”, Energy and design changes
Buildings, Vol. 47, pp. 159-168.
Bonabeau, E. (2002), “Agent-based modeling: methods and techniques for simulating human systems”, 293
Proceedings of the National Academy of Sciences, Vol. 99 No. 3, pp. 7280-7287.
Dharmalingam, J.M. and Eswaran, M. (2018), “An agent based intelligent dynamic vulnerability
analysis framework for critical SQLIA attacks: intelligent SQLIA vulnerability analyzer agent”,
International Journal of Intelligent Information Technologies, Vol. 14 No. 3, pp. 56-82.
Duan, L., Dogru, M.K., Oezen, U. and Beck, J.C. (2012), “A negotiation framework for linked
combinatorial optimization problems”, Autonomous Agents and Multi-agent Systems, Vol. 25
No. 1, pp. 158-182, available at: https://doi.org/10.1007/s10458-011-9172-7
Dzeng, R.J. and Lin, Y.C. (2004), “Intelligent agents for supporting construction procurement negotiation”,
Expert Systems with Applications, Vol. 27 No. 1, pp. 107-119, available at: https://doi.org
/10.1016/j.eswa.2003.12.006
El-Diraby, T.E. and Zhang, J. (2006), “A semantic framework to support corporate memory management in
building construction”, Automation in Construction, Vol. 15 No. 4, pp. 504-521, available at: https://
doi.org/10.1016/j.autcon.2005.07.005
Faludi, J., Lepech, M.D. and Loisos, G. (2012), “Using life cycle assessment methods to guide
architectural decision-making for sustainable prefabricated modular buildings”, Journal of
Green Building, Vol. 7 No. 3, pp. 151-170, available at: https://doi.org/10.3992/jgb.7.3.151
Gerber, D.J., Pantazis, E. and Wang, A. (2017), “A multi-agent approach for performance based
architecture: design exploring geometry, user, and environmental agencies in façades”,
Automation in Construction, Vol. 76, pp. 45-58.
Hadikusumo, B.H.W., Petchpong, S. and Charoenngam, C. (2005), “Construction material procurement
using internet-based agent system”, Automation in Construction, Vol. 14 No. 6, pp. 736-749.
Hwang, B.-G., Shan, M. and Looi, K.-Y. (2018), “Key constraints and mitigation strategies for
prefabricated prefinished volumetric construction”, Journal of Cleaner Production, Vol. 183,
pp. 183-193.
Isaac, S. and Navon, R. (2013), “A graph-based model for the identification of the impact of design
changes”, Automation in Construction, Vol. 31, pp. 31-40.
Jaillon, L. and Poon, C.S. (2014), “Life cycle design and prefabrication in buildings: a review and case studies
in Hong Kong”, Automation in Construction, Vol. 39, pp. 195-202.
Karakas, K., Dikmen, I. and Birgonul, M.T. (2012), “Multiagent system to simulate risk-allocation and
cost-sharing processes in construction projects”, Journal of Computing in Civil Engineering,
Vol. 27 No. 3, pp. 307-319.
Kim, H.-J. and Reinschmidt, K.F. (2010), “Association of risk attitude with market diversification in the
construction business”, Journal of Management in Engineering, Vol. 27 No. 2, pp. 66-74.
Koo, J., Kim, Y.S. and Kim, B.I. (2012), “Estimating the impact of residents with disabilities on the
evacuation in a high-rise building: a simulation study”, Simulation Modelling Practice and
Theory, Vol. 24, pp. 71-83, available at: https://doi.org/10.1016/j.simpat.2012.02.003
Lam, P., Chan, A.P.C., Wong, F.K.W. and Wong, F.W.H. (2007), “Constructability rankings of construction
systems based on the analytical hierarchy process”, Journal of Architectural Engineering, Vol. 13
No. 1, pp. 36-43, available at: https://doi.org/10.1061/(ASCE)1076-0431(2007)13:1(36)
Li, C.Z., Hong, J., Fan, C., Xu, X. and Shen, G.Q. (2018), “Schedule delay analysis of prefabricated
housing production: a hybrid dynamic approach”, Journal of Cleaner Production, Vol. 195,
pp. 1533-1545.
ECAM Li, C.Z., Hong, J., Xue, F., Shen, G.Q., Xu, X. and Mok, M.K. (2016), “Schedule risks in prefabrication
27,1 housing production in Hong Kong: a social network analysis”, Journal of Cleaner Production,
Vol. 134, pp. 482-494, available at: https://doi.org/10.1016/J.JCLEPRO.2016.02.123
Li, M., Li, G., Huang, Y. and Deng, L.M. (2017), “Research on investment risk management of Chinese
prefabricated construction projects based on a system dynamics model”, Buildings, Vol. 7 No. 4,
pp. 83-98.
294 Luo, L.-Z., Mao, X., Shen, L.-Y. and Li, Z.-D. (2015), “Risk factors affecting practitioners’ attitudes toward the
implementation of an industrialized building system”, Engineering, Construction and Architectural
Management, Vol. 22 No. 6, pp. 622-643, available at: https://doi.org/10.1108/ECAM-04-2014-0048
Mahmoud, M.A., Ahmad, M.S., Yusoff, M.Z.M. and Idrus, A. (2015), “Automated multi-agent
negotiation framework for the construction domain”, in Omatu, S. et al. (Eds), Distributed
Computing and Artificial Intelligence, 12th International Conference. Advances in Intelligent
Systems and Computing, Salamanca, Vol. 373, Springer, Cham, June 3-5, pp. 203-201.
Mao, C., Shen, Q., Pan, W. and Ye, K. (2013), “Major barriers to off-site construction: the developer’s
perspective in China”, Journal of Management in Engineering, Vol. 31 No. 3, pp. 1-8.
Martinez, S., Jardon, A., Navarro, J.M. and Gonzalez, P. (2008), “Building industrialization: robotized
assembly of modular products”, Assembly Automation, Vol. 28 No. 2, pp. 134-142.
Nguyen, V.V., Hartmann, D. and König, M. (2012), “A distributed agent-based approach for simulation-
based optimization”, Advanced Engineering Informatics, Vol. 26 No. 4, pp. 814-832.
Obonyo, E.A., Anumba, C.J. and Thorpe, A. (2002), “Agent-based specification and procurement of
construction products”, Proceedings of the International Conference on Advances in Building
Technology, Hong Kong, December 4-6, Vol. II, pp. 1713-1720.
Pan, W. and Sidwell, R. (2011), “Demystifying the cost barriers to offsite construction in the UK”,
Construction Management and Economics, Vol. 29 No. 11, pp. 1081-1099, available at: https://doi.org/
10.1080/01446193.2011.637938
Phanden, R.K., Jain, A. and Verma, R. (2011), “Integration of process planning and scheduling: a state-
of-the-art review”, International Journal of Computer Integrated Manufacturing, Vol. 24 No. 6,
pp. 517-534, available at: https://doi.org/10.1080/0951192X.2011.562543
Pinedo, M. (2012), Scheduling: Theory, Algorithms, and Systems, Springer, New York, NY, available at:
https://doi.org/10.1080/15458830.1996.11770714
Ren, Z. and Anumba, C.J. (2004), “Multi-agent systems in construction – state of the art and prospects”,
Automation in Construction, Vol. 13 No. 3, pp. 421-434, available at: https://doi.org/10.1016/
j.autcon.2003.12.002
Taillandier, F., Taillandier, P., Tepeli, E., Breysse, D., Mehdizadeh, R. and Khartabil, F. (2015), “A multi-
agent model to manage risks in construction project (SMACC)”, Automation in Construction,
Vol. 58, pp. 1-18.
Tan, J., Jiang, G. and Wang, Z. (2019), “Evolutionary game model of information sharing behavior in
supply chain network with agent-based simulation”, International Journal of Intelligent
Information Technologies, Vol. 15 No. 2, pp. 54-68.
Walczak, S. (2018), “Society of agents: a framework for multi-agent collaborative problem solving”,
International Journal of Intelligent Information Technologies, Vol. 14 No. 4, pp. 1-23.
Wang, B., Wang, X. and Wang, J. (2012), “Construction and empirical analysis of agricultural science
and technology enterprises investment risk evaluation index system”, IERI Procedia, Vol. 2,
pp. 485-491.
Watkins, M., Mukherjee, A., Onder, N. and Mattila, K. (2009), “Using agent-based modeling to study
construction labor productivity as an emergent property of individual and crew interactions”,
Journal of Construction Engineering and Management, Vol. 135 No. 7, pp. 657-667.
Wesz, J., Bolzan, G., Torres Formoso, G. and Tzortzopoulos, P. (2018), “Planning and controlling design in
engineered-to-order prefabricated building systems”, Engineering, Construction and Architectural
Management, Vol. 25 No. 2, pp. 134-152, available at: https://doi.org/10.1108/ECAM-02-2016-0045
Yan, Y., Zhang, J. and Ma, X. (2019), “Modeling of agent-based complex network to detect the trust of Multi-agent
investors in P2P platform”, International Journal of Intelligent Information Technologies, Vol. 15 simulation for
No. 2, pp. 20-31.
Zhang, W., Lee, M.W., Jaillon, L. and Poon, C.S. (2018), “The hindrance to using prefabrication in Hong
managing
Kong’s building industry”, Journal of Cleaner Production, Vol. 204, pp. 70-81, available at: https:// design changes
doi.org/10.1016/J.JCLEPRO.2018.08.190

Further reading 295


Christodoulou, S. (2009), “Construction imitating ants: resource unconstrained scheduling with artificial
ants”, Automation in Construction, Vol. 18 No. 3, pp. 285-293, available at: https://doi.org/10.1016/
j.autcon.2008.09.005

Corresponding author
Juan Du can be contacted at: ritadu@shu.edu.cn

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com

You might also like