Automation in Construction xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

Agent-based modeling and simulation of earthmoving operations

Ahmad Jabri⁎, Tarek Zayed
Dept. of Building, Civil and Environmental Engineering, Concordia University, Montreal, Quebec, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: Discrete-Event Simulation (DES) models are constructed from sequential duration-based activities. DES is used in
Agent-based modeling modeling several construction operations including earthmoving. Current earthmoving models cannot accom-
Agent-based simulation modate equipment units with different specifications performing the same task. In addition, activity durations
Multi-agent systems are calculated based on primitive methods such as interpolating existing durations of similar activities in pre-
Earthmoving planning
vious projects. Finally, model elements behave in a predetermined manner, ignoring special operational real-life
Earthmoving productivity
scenarios that occur due to resource constraints. These limitations often lead to inaccuracies in calculating
productivity and equipment utilization. This paper applies Agent-Based Modeling and Simulation (ABMS) as an
effective bottom-up tool to modeling earthmoving operations. An Agent-Based (AB) earthmoving model con-
sisting of smart, adaptive agents is developed. Each agent is assigned a state chart and a set of static and dynamic
properties (attributes and variables) to direct its interactions with the environment and with other agents. This
framework proves how modeling earthmoving from the agent's prospective and basing agents' interactions on
their properties allow for modeling equipment units with different specifications performing the same task (e.g.
trucks of different capacities), as well as for an accurate representation of activity durations, resource handling
and resource constraint scenarios. A Java-based application named Agent-Based Simulator for Earthmoving
Operations (ABSEMO) is developed as an implementation of the proposed model. ABSEMO will be helpful to
contractors in planning earthmoving operations according to the AB approach. A real-life case study of a riv-
erbed excavation in a dam construction project is simulated using ABSEMO, and the results are compared with
those obtained from existing simulation models to verify ABSEMO's logic. A percentage difference of 0.42% from
the existing results is obtained, indicating that the model's flow of resources is indeed accurate.

1. Introduction making aspect of construction management, recent efforts suggested

One of the most common applications of simulation in the con-
struction industry is the simulation of earthmoving operations. Since
these operations are typically lengthy in duration and fall on the critical
path of construction projects, accurate planning is crucial in ensuring
project success. And unlike other stages of the construction project,
where manpower is the most relied-on resource, earthmoving opera-
tions are considered equipment-intensive, utilizing large and expensive
fleets of trucks, loaders, bulldozers, etc. Thus, improving the efficiency
of earthmoving operations is a primary objective from the contractor's
point of view [20]. The cyclic nature of earthmoving operations and the
type of work tasks they involve make the simulation process a valid
planning tool for forecasting productivity and costs of these operations
[31].
The nature of Discrete-Event Simulation (DES) provides, in most
cases, a sufficient solution to modeling most construction operations,
especially on the technical level. Furthermore, to address the decision-
the utilization of System Dynamics (SD) to account for the complex
strategic level when simulating construction operations [3]. The main
objective of this paper is the introduction of Agent-Based Modeling and
Simulation (ABMS) into earthmoving operations for the purpose of
enhancing current practices and overcoming limitations of current re-
search work. These limitations include primitive methods for calcu-
lating activity durations, the inability to accommodate equipment units
with different specifications performing the same task and ignoring
unique scenarios of resource constraints. This major objective can be
broken down into the following sub-objectives:
• Develop a detailed agent-based (AB) model of earthmoving opera-
tions, which captures the properties and interactions of model ele-
ments.
• Design a stand-alone ABMS software system for earthmoving op-
erations and verify the model using a real-world case study.

⁎
Corresponding author.
E-mail addresses: ahmadjabri@outlook.com (A. Jabri), zayed@encs.concordia.ca (T. Zayed).

http://dx.doi.org/10.1016/j.autcon.2017.06.017
Received 11 April 2016; Received in revised form 1 May 2017; Accepted 10 June 2017
0926-5805/ © 2017 Elsevier B.V. All rights reserved.

Please cite this article as: Jabri, A., Automation in Construction (2017), http://dx.doi.org/10.1016/j.autcon.2017.06.017
A. Jabri, T. Zayed
2. Background
ABMS is based on the idea of simulating the interactions of smart
and autonomous objects, in order to identify, explain, generate and
design emergent behaviors (Chan et al. 2010). Agents are self-contained
entities that have the ability to control their own actions based on their
perception of other agents and their surrounding environments [11].
Unlike DES and SD, which are considered as top-down modeling tech-
niques, ABMS is a bottom-up modeling approach, in which model ele-
ments are built before the process is examined as a whole. Plus, ABMS
has no specific convention on time progression during the model run; it
can be discrete, continuous, or a hybrid of both (Chan et al. 2010).
ABMS has the potential to have extensive effects on the way researchers
use laboratories to support their research, and businesses use computers
to support decision-making [21]. In ABMS, there is no need to make
excessive assumptions or to direct the model in a way which is accep-
table by the capabilities of the simulation technique.
2.1. ABMS applications in civil engineering and construction management
The majority of ABMS applications in civil engineering and con-
struction management are focused on the following areas: 1) supply
chain management [19,29]; 2) construction claims management [9,26];
3) infrastructure management [23,27]. Supply chain management in-
volves highly complex chains of interacting entities. Sharing informa-
tion about stocks, costs, quantities and schedules is vital to assuring
successful supply chain operations. Likewise, construction claims
management involves interaction among project participants such as
contractors and consultants. It involves discussions, sharing of in-
formation and organizing work tasks. Infrastructure management using
ABMS is a promising topic, in which components of an infrastructure
system are treated as interacting agents. Governments, infrastructure
management agencies, infrastructural assets and users are all modeled
as intelligent agents with attributes and goals. This can help anticipate
performance, plan maintenance and manage budgets. There are a few
other ABMS applications in other construction management areas in-
cluding construction equipment management [30,35], bidding strate-
gies [5], procurement [8], construction site safety [12,18,24] and
construction workers' behavior [1,28].
2.2. The need of ABMS in earthmoving planning
There is a need to incorporate AB technologies with significant
construction operations such as earthmoving. This must be completed
in a smart and flexible paradigm that accepts various types of data,
maintains a presentable view of model operation and analysis, produces
accurate results, and most importantly, overcomes the limitations of
currently used techniques. This research aims at rebuilding the meth-
odology of creating earthmoving simulation models, based on the AB
approach. While the application of simulation in earthmoving remains
a well-researched area ([13,16,22,31], Smith et al. 2000), the vast
majority of work is based solely on a DES or on a DES-SD hybrid system
[3]. The following points discuss the major limitations of earthmoving
modeling and simulation and indicate how ABMS would address these
limitations:
1. Basic methods for calculating activity durations: Earthmoving ac-
tivity durations used in simulation models are obtained from his-
torical data calculated mainly by field observations. These durations
are basically abstract numbers obtained for specific types of equip-
ment units working together. For example, based on data from a
previous project, a contractor using DES to plan an earthmoving
operation can design the truck loading activity to last 3 min. But if
the capacities or conditions of the equipment units to be used in the
project have changed, this number is scaled up or down based on the
planner's experience. An AB model would better address this issue,
2
Automation in Construction xxx (xxxx) xxx–xxx
since it would administer interaction scenarios from the agent's
(equipment unit) prospective. Each unit would have a set of prop-
erties that directs its interactions with other agents, making it easier
to account for different combination of equipment units working
together. Instead of assigning a certain duration for the interaction
between units X and Y and assuming both are passive objects in the
operation, both units would be assigned a set of properties. The
output (duration, quantity moved, etc.) of the interaction between
units X and Y relies on the properties of both agents in a way that if a
new agent is introduced to interact with unit X or Y, no added as-
sumptions to the model is necessary to force a new output; the new
interaction will automatically produce appropriate output. This
issue is further addressed in the AB model development section,
when discussing the properties, state charts and interactions of
agents in the proposed model.
2. Inability of accommodating equipment units with different specifi-
cations performing the same task: Since it is not possible to assign
more than one duration for a certain activity in a DES model,
planning an earthmoving operation with trucks of different capa-
cities becomes a major challenge. Assigning larger loading durations
for larger trucks is not possible in DES models, unless each truck
type has its own path and queues in the model. However, this cannot
be done in most cases, especially if one server is working with
multiple units and therefore needs to process them in different
durations. A simplifying assumption that is often used in such cases
is to treat all units as if they are of the same type by assigning their
service duration a weighted average of the service duration of all
units. Since an AB model is based on individual agents performing
certain tasks, every agent can have different properties, and ac-
cordingly different outputs when interacting with other agents or
with the environment. For example, three trucks with three different
capacities can all be modeled by an AB model, and every truck can
have a different loading duration. These trucks would also queue up
together, be served by a single or multiple loaders, etc.
3. Ignoring unique scenarios of resource constraints: Model elements in
DES behave in a predetermined manner, ignoring special opera-
tional real-life scenarios that occur due to resource constraints. In
DES earthmoving models, the quantity of earth to be excavated is
usually added to the model in the form of truckloads. So, a 100 m3 of
earth is equal to 20 truckloads when trucks of 5 m3 capacities are
used. But if 7 m3 trucks are used instead, after 14 trucks get loaded,
2 m3 is left out. In practice, this 2 m3 is assumed to be a truckload.
An AB model would address this issue through its representation of
agents' properties. Equipment units in an AB model possess both
static and dynamic properties, referred to as attributes and variables
respectively. Attributes are used in regular interaction scenarios,
while variables are flexible and can be used in special interaction
scenarios. Whenever two agents are engaged in an atypical proce-
dure, they can accommodate the change using their variables. In the
previous example, both the loader and the truck agents can be as-
signed a capacity attribute and a carried earth variable. During the
process of loading 14 trucks with 5 m3, both the loader and the
trucks utilize their capacity attribute. However, the 15th load re-
quires the loader to carry 2 m3 only, setting its carried earth variable
to 2 m3. Consequently, the loader fills the truck with that quantity,
increasing its carried earth variable by 2 m3. Similar scenarios are
investigated when building agents' state charts in the AB model
development section.
Marzouk and Moselhi [17] developed an object-oriented simulation
model for contractors to plan earthmoving operations with un-
certainties. Their system utilized DES but with equipment units being
modeled and treated as individual objects. Although their work was the
first to account for equipment units as individual entities with proper-
ties (e.g. ID and capacity), there model was still based on DES, and no
solution to address the aforementioned limitations could be introduced.
A. Jabri, T. Zayed
There are a few research works that applied ABMS to earthmoving
operations. However, the main goal of these research works was the not
utilizing the strengths of ABMS to improve the accuracy of planning
earthmoving options. Vahdatikhaki et al. [32] utilized game theory to
improve the safety of a multi-agent system of earthmoving equipment
units. By tracking the real-time location of agents and using sensors, a
Location-based Guidance System (LGS) was developed to automate the
earthmoving process in a high-precision manner. Zankoul et al. [34]
created an AB model of an earthmoving operation using AnyLogic, and
compared the results of its simulation with that of a DES model. The
model consists of three agents which are road segments, excavators and
trucks. The authors concluded that an AB model would better represent
earthmoving operations due to several reasons including its ease of use
and its informative states of agents (empty vs. full, hauling vs. re-
turning, etc.). A major limitation of this study is the fact that the AB
model was duration-based, making it a simple DES model but with
agents and state charts. The most important aspect of an AB model was
ignored, which is having smart and autonomous objects that possess
characteristics and goals, and interact with each other and with their
environment accordingly. Having agents in the model progress from
one state to another based on fixed durations is a strongly constricts the
potential of ABMS.
3. Research methodology
A generic flow chart that summarizes the research methodology and
key tasks is depicted in Fig. 1. Major components discussed in this paper
are highlighted in grey. Primarily, the literature was reviewed on the
basis of: 1) different simulation techniques being used in scientific re-
search; 2) applications of simulation in construction management in
general and earthmoving in particular; and 3) ABMS principles and
applications. The authors' main contribution in this research work is the
development of a comprehensive AB earthmoving model developed for
earthmoving operations which include the excavation of earth from a
certain location and the transportation of the excavated earth to an-
other location for dumping. The model contains four agents; bulldozers,
loaders, haulers and spotters. The environment setup, the types and
properties of agents, the interaction logic, the earth handling procedure
and other process logistics are carefully examined to develop the pro-
posed model. Finally, the proposed system is implemented in a com-
puter program referred to as ABSEMO (Agent-Based Simulator for
Earthmoving Operations) using AnyLogic 7. The program is a stand-
alone Java application which is comprised of the environment setup,
equipment and labor properties as well as operation logistics obtained
from the proposed AB earthmoving model. ABSEMO was tested by
examining a real-life case study of earthmoving in a dam construction
project. Productivity results from ABSEMO were compared to those
obtained from DES models for verification purposes.
4. AB model development for earthmoving operations
The proposed AB methodology and the developed model aim at
enhancing current practices of modeling earthmoving operations by
viewing them from an individual AB prospective. This allows the cap-
turing of realistic behaviors, through crafting agents' attributes, roles
and interactions. Earthmoving operations utilize different equipment
based on the construction project's scope of work and available re-
sources. The most common earthmoving operation represents the ex-
cavation of earth from a certain location and its transportation to an-
other location where it gets dumped. However, the mechanism of the
excavation, transportation and dumping activities differ from one pro-
ject to another, leading to different combinations of equipment units
and interactions.
3
Automation in Construction xxx (xxxx) xxx–xxx
4.1. Model scope
The AB model developed in this research is of an earthmoving op-
eration that includes bulldozers, loaders, haulers (trucks) and dump
spotters. Accordingly, the operation would be as follows: 1) bulldozers
excavate earth from a certain area and push it in stockpiles; 2) loaders
fill trucks with earth from the stockpiles; 3) trucks transport the carried
earth to a certain dumping location; 4) with the help of dump spotters,
trucks dump their loads in stock piles; and 5) trucks return to the
loading area to carry and transport more earth. The model also ac-
commodates earthmoving operations which do not include bulldozers.
This would be the case if the operation was simply the transportation of
earth from a stock pile into a dumping area, or if the loaders, or any
type of excavators, are performing both the excavation and truck
loading activities. The aforementioned equipment units and dump
spotters are considered as agents. However, earth as well as transpor-
tation roads are considered as passive objects that are embedded in the
model's environment. Earth and the transportation road components
are represented by the model in a passive manner in which they are
controlled by agents. Unless model components possess characteristics
and goals that directly influence the operation, and unless they actively
communicate and interact with other components and agents, it is a
common practice to represent them as embedded objects in the model
rather than agents. Earth in the developed model is an object that is
handled and transported by equipment units; it does not have any
atypical roles or interactions. Similarly, the transportation road is an
object that has no influence on the operation. This is why it was decided
for earth and transportation roads to be non-agent objects in the model.
Two types of properties are defined for each agent in the proposed AB
model: 1) attributes, which refers to fixed properties of agents that do
not change at any time and are unaffected by interactions with the
environment or other agents; and 2) variables, which refers to prop-
erties of agents that can change due to interactions with the environ-
ment or other agents. Table 1 lists agents along with their attributes and
variables in the proposed AB model.
4.2. Agent development
In AB models, diagrams referred to as state charts are often used to
describe the different stages (states) that agents pass through while
performing their roles. State charts are used to construct agents and
govern their roles as well as their communication mechanism. The
overall behavior of the AB model is generated from the interaction of its
agents' state charts. After creating state charts, they are implemented in
computer programs for simulation purposes to obtain the emergent
behavior of the system. For the proposed earthmoving model, four state
charts are created; one for each agent, as depicted by Fig. 2. A color
legend for the agents' state charts is going to be used throughout the
model construction and implementation. A red color refers to an idle
state of the agent. It signifies that the agent is waiting for a certain
condition to be fulfilled or a certain type of interaction to be carried out
in order for it to move on to the next state. Furthermore, a green color
represents a working state of the agent. It indicates that the agent is
currently performing its main role. Finally, a yellow color is used to
describe a transitional state of the agent. It indicates that the agent is
working, but on a minor task that is mostly auxiliary to the main task.
Prior to getting into the details of each agent's state chart, it is im-
portant to define three variables which are crucial to understanding the
agents' breakdown: Earth Ready for Excavation (ERE) which refers to the
quantity of earth available to be excavated by bulldozers; Earth Ready
for Loading (ERL) which refers to the quantity of excavated earth ready
to be loaded in trucks (or the quantity of earth available to be excavated
by loaders or other types of excavators, if the operation does not include
bulldozers); and Dumped Earth (DE) which refers to the quantity of earth
that has already been transported and dumped at the dumping location.
A. Jabri, T. Zayed Automation in Construction xxx (xxxx) xxx–xxx

Fig. 1. Research methodology.

Literature Review

Simulation in Simulation of
Construction Earthmoving ABMS
Management Operations

Methodology

Equipment Process
Types Logistics

Equipment & AB Model of

Soil Handling
Labor Earthmoving
Procedures
Properties Operations

Interaction Environment
Logic Setup

Implementation

Main Class Agent Classes

Stand-Alone
System
(ABSEMO)
Reporting of Graphical
Results User Interface

Case System
Study Verification

4.2.1. The bulldozer agent
The attributes and variables of the bulldozer agent are shown in
Table 1. Push Quantity refers to the quantity of excavated earth by the
bulldozer at the end of each pass. Push Duration is the time it takes the
bulldozer to excavate that quantity of earth (to make one pass). Time to
Adjust Position refers to the time it takes the bulldozer to turn and rotate
to the opposite orientation. Return Duration is the time it takes the
bulldozer to return back to the location where it can begin another
excavation pass. In addition, Actual Push Quantity and Actual Push
Duration are designed for special cases when the quantity of earth
available for excavation is less than the Push Quantity of the bulldozer.
In such cases, the Actual Push Quantity would be whatever quantity is
available and the Actual Push Duration would be an interpolated fraction
of the regular Push Duration based on the ratio of the Actual Push
Quantity to the regular Push Quantity. Regularly, Actual Push Quantity
and Actual Push Duration are equal to Push Quantity and Push Duration.
The state chart of the bulldozer agent is depicted in Fig. 2(a). The
state chart commences by the bulldozer being at the starting point,
ready to begin excavating. If there is no Earth Ready for Excavation, the
bulldozer remains idle. However, if there is Earth Ready for Excavation,
the bulldozer starts moving and excavating that earth, either to the
regular Push Quantity or to whatever quantity is available (Actual Push
Quantity). After the bulldozer reaches the end of its excavation pass, it
adjusts its position and rotates to be able to move in the opposite

4
A. Jabri, T. Zayed
Table 1
Agents' attributes and variables in the proposed AB earthmoving model.
Agent Attributes Variables
Bulldozer/s Push Quantity (PQ) Actual Push Quantity
Push Duration (PD) (APQ)
Time to Adjust Position (TAP) Actual Push Duration
Return Duration (RD) (APD)
Loader/s Bucket Capacity (BC) Carried Earth (CE)
Time to Load Full Bucket (TLFB) Time to Load Bucket (TLB)
Time to Adjust Position while Full Unloading Quantity (UQ)
(TAPF) Time to Unload Bucket
Time to Unload Full Bucket (TUFB) (TUB)
Time to Adjust Position while Empty
(TAPE)
Hauler/s Capacity (C) Carried Earth (H-CE)
Time to Get in Load Position (TGLP) Available Space (AS)
Hauling Duration (HD)
Time to Get in Dump Position
(TGDP)
Dumping Duration (DD)
Returning Duration (RD)
Spotter/s Time to Adjust Position (S-TAP) None
orientation. Subsequently, the bulldozer starts returning and heading to
the starting point of the excavation pass, where it adjusts its position
and rotates again to be in the correct form, ready to perform another
excavation run. In view of that, the bulldozer agent in the proposed AB
model has no interactions with other agents. However, considering the
bulldozer participant as an agent is necessary since it is an equipment
unit that is actively participating in the system. In addition, it is better
to keep it as a flexible entity in case its mechanism changes when up-
grading the model in the future. Table 2 provides a detailed elaboration
on the bulldozer's state chart transitions. The actions of transition 2
were explained earlier when studying the bulldozer agent's state chart.
The actions of transition 3 include reducing the quantity of Earth Ready
for Excavation by the Actual Push Quantity. Aside from this state, all
other states and transitions are straightforward.
4.2.2. The loader agent
Table 1 illustrates the attributes and variables of the loader agent.
Bucket Capacity refers to the quantity of earth that the loader can carry
in its bucket. Time to Load Full Bucket is the time it takes the loader to
fill its bucket with that quantity of earth (Bucket Capacity). Time to
Adjust Position While Full refers to the time it takes the loader, when its
bucket is full, to get in an appropriate position for loading haulers
(trucks). Time to Unload Full Bucket is the time it takes the loader to
dump its full bucket load (Bucket Capacity) of earth in the truck. Time to
Adjust Position While Empty refers to the time it takes the loader, when
its bucket is empty, to get in a position for loading its bucket after
dumping its load in the truck. Moreover, Carried Earth represents the
actual quantity of earth in the loader's bucket. This variable is added to
accommodate different bucket quantity scenarios, determined by the
availability of earth. Time to Load Bucket refers to the actual time it
takes to load the bucket based on the Carried Earth. Thus, it is an in-
terpolated fraction of the regular Time to Load Full Bucket based on the
ratio of the Carried Earth to the Bucket Capacity. Unloading Quantity
represents the quantity of earth that the loader chooses to dump in the
truck it is serving, which depends on a variable of the hauler agent
called Available Space. To elaborate, if the Available Space of the truck is
less than the loader's Carried Earth, the loader dumps only a quantity of
earth equal to the truck's Available Space. Accordingly, the Time to
Unload Bucket is an interpolated fraction of the regular Time to Unload
Full Bucket based on the ratio of the Unloading Quantity to the Bucket
Capacity. Since the model is designed in a way that allows for the loader
agent to perform the excavation activity when no bulldozers are in-
cluded, all the loader's attributes and variables in Table 1 are made to
be generic and can accommodate both scenarios. Time to Load Bucket
5
Automation in Construction xxx (xxxx) xxx–xxx
can refer to either the time it takes the loader to fill its bucket from an
excavated earth in stock piles or the time it takes the loader, or other
types of excavators, to perform the excavation activity and fill their
buckets with earth. If needed, other attributes and variables change
accordingly as well. This is an example of flexibility in AB models.
Fig. 2(b) represents the state chart of the loader agent. Accordingly,
the initial state is the loader being ready to begin loading its bucket by
carrying earth from an excavated stockpile, or performing the excava-
tion in the absence of bulldozers. If there is no Earth Ready for Loading,
the loader remains idle. If, however, there is Earth Ready for Loading,
the loader starts filling its bucket either to its maximum capacity
(Bucket Capacity) or to the available quantity of earth if it is less than
that capacity. After the loader's bucket is filled, the loader adjusts its
position to be ready for loading the arriving truck. At that point, the
loader checks the truck's queue. If there are no trucks waiting to be
loaded, the loader remains idle. However, if trucks are available, the
loader signals the first truck in the queue to move to the loading area.
Then, the loader determines the Available Space in the truck. If that
Available Space is larger than or equal to the loader's Carried Earth, the
loader dumps its entire bucket load in the truck. Otherwise, the loader
dumps a quantity equal to the truck's Available Space. Regularly, when
trucks arrive for loading, their Available Space will be equal to their
Capacity. After unloading the bucket's load in the truck, the loader
checks the Available Space in the truck again and determines whether
the truck has reached its Capacity or not. In the event that the truck is
full or if there is no Earth Ready for Loading at that point, the loader
signals the truck that the interaction between the two equipment units
is over and that the truck should start hauling to the dumping location.
The loader then has to adjust its position to get in the bucket loading
setup again and waits for extra quantities of earth to become available.
On the other hand, if the loader determines that the truck has not yet
reached its full Capacity and that there is still Earth Available for Loading,
the loader adjusts its position, fills its bucket and loads the truck again.
This cycle is repeated until the truck reaches its Capacity or until the
Earth Ready for Loading is fully consumed.
The loader in the proposed AB model is a good example of a dy-
namic agent that actively adapts with the changes of model conditions
and other agents' properties. During different cycles of the loader's
operation, the same variable can have different values that depend on
the cycle's conditions and interactions. This demonstrates the strength
of the ABMS technique and its promising capabilities for future en-
hancements of developed models. Table 3 presents a detailed break-
down of the loader's state chart transitions. Consequently, these few
points need to be noted:
• There are two types of position adjustments made by the loader
agent; position adjustment A is performed when the bucket is filled
with a quantity of earth, while position adjustment B is performed
when the bucket is empty. This distinction was made to make pos-
sible the option of using a longer duration for the position adjust-
ment when the bucket contains earth.
• For the 6th transition, since the truck has just arrived for loading, it
may appear that the only valid option is the one in the second row
(Truck's Available Space > Carried Earth). However, designing the
state chart to accommodate the other alternative (Truck's Available
Space ≤ Carried Earth) does not affect the logic of the first cycle and
is necessary in accommodating succeeding cycles without having to
increase the size of the state chart.
• For the second alternative of the 6th and 10th transitions (Truck's
Available Space > Carried Earth), a quantity of (Carried Earth -
Unloading Quantity) is added back to the Earth Ready for Loading to
indicate that the extra quantity of carried earth, which could not be
added to the truck because it was full, is still in the operation and
will be used in upcoming loading activities of other trucks.
• The condition (Earth Ready for Loading = 0) was added to the con-
ditions of the 11th transition to guide the interaction between the
A. Jabri, T. Zayed Automation in Construction xxx (xxxx) xxx–xxx

Start
1 Ready To Fill
Filling Bucket Soil Available
Start
Bucket

2 10
Ready to
Excavate
5 Adjusting Adjusting
1
Position (A) Position (B)
Soil Available
3
9
Truck Full || Soil Unavailable
Waiting for 4
Adjusting Trucks Available
Excavating Trucks
Position
Unloading
Bucket
2 4 Adjusting
8
Position (A)
5
Truck Not Full & Soil Available
7
Adjusting
3 Returning
Position Adjusting
Filling Bucket 6
Position (B)

Start

Getting In
1 Waiting For
Loading Loader Available
Start
Loader
Position
Adjusting 1 Ready To Spot
2 8 Position Truck Available

Being Loaded Returning 2 4

3 7
Adjusting
Spotting 3
Position
Hauling Dumping

4
6

Getting in Idle Transitioning Working

Waiting for 5
Spotter Available
Dumping
Spotter
Position

Fig. 2. Agents' state charts (a) bulldozer's state chart; (b) loader's state chart; (c) hauler's state chart; (d) spotter's state chart.

Table 2
loader and hauler agents in a way that when there is no earth
Bulldozer's state chart transitions. available for the loader to use in filling the truck, the loading ac-
tivity is completed and the truck should be able to leave the loading
Transition Condition/s Action/s area.
1 SRE = 0 No action • It can be observed in the table that the loader is controlling its in-
2A SRE > 0 & SRE > PQ APQ = PQ; APD = PD; Start teraction with the hauler. Hence, the loader is using some of the
excavating hauler's properties in its state chart's conditions and actions.
2B SRE > 0 & SRE ≤ PQ APQ = SRE; APD = PD ×
APQ
Although this interaction depends on the properties of both equip-
PQ
Start excavating ment units, the hauler cannot tell how much earth it is being filled
3 Excavation pass completed SRE reduced by APQ; SRL increased with or when it reaches its full capacity. So, the loader has the
by APQ; Start adjusting position overall authority over the loading activity, and the hauler acts only
4 Position adjustment Start returning
based on commands from the loader. This will be further explained
completed
5 Return completed Start adjusting position when discussing the hauler's state chart (Fig. 2(c)) and transitions
6 Position adjustment No action (Table 4).
completed

6
A. Jabri, T. Zayed Automation in Construction xxx (xxxx) xxx–xxx

Table 3 refers to the duration of the hauler's trip from the dumping area to the
Loader's state chart transitions. truck queue at the loading area. Accordingly, it is important to note that
these two attributes depend on the properties of both the hauler itself
Transition Condition/s Action/s
and the hauling and returning roads. However, in the proposed AB
1 SRL = 0 No action model, the trip durations are assumed to be attributes of the hauler
2A SRL > 0 CE = BC agent only. Time to Get in Dump Position is the time it takes the hauler to
& TLB = TLFB
leave the queue and arrive at the dumping area. Dumping Duration refers
SRL > BC Start filling bucket
2B SRL > 0 CE = SRL to the time it takes the truck to dump its carried load in a stockpile. In
& TLB = TLFB ×
CE addition, an important observation on the hauler's attributes is that
SRL ≤ BC BC
unlike hauling, the truck's loading duration is not an attribute. This is
Start filling bucket
3 Bucket filling completed SRL reduced by CE because the hauler agent does not govern its loading activity, which is
Start adjusting position (A) under the control of the loader agent. Moreover, Carried Earth re-
4 Position adjustment (A) completed Start waiting for truck presents the actual quantity of earth carried by the hauler. This variable
5 Truck unavailable No action is added as an updating mechanism of the quantity of earth present in
6A Truck available UQ = CE
& TUB = TUFB
the hauler at different stages of its cycle. Available Space is a variable
Truck's AS > CE Start dumping UQ in truck dependent on Carried Earth. Always, the hauler's Available Space is equal
6B Truck available UQ = Truck's AS to its Capacity less Carried Earth. The hauler's Available Space is a very
& TUB = TUFB ×
UQ
important variable for the loader agent due to the influence it has on
Truck's AS ≤ CE BC
Start dumping UQ in truck the transition's conditions and actions of its state chart, as proven by
SRL increased by (CE - UQ) Table 4.
7 UQ dumped in truck Truck's AS reduced by UQ The state chart of the hauler agent is illustrated by Fig. 2(c). The
& Truck's AS > 0 Truck's HCE increased by UQ
initial state represents the hauler waiting in the queue for loaders to
& SRL > 0 Start adjusting position (B)
8A Position adjustment (B) completed CE = BC become idle. Once a loader becomes idle, it signals the hauler to move
& SRL > BC TLB = TLFB to the loading area. The hauler then moves and gets in the loading
Start filling bucket position. At this point, the loading activity begins and the loading cycles
8B Position adjustment (B) completed CE = SRL
described in the loader's state chart start to be executed. As indicated
& SRL ≤ BC TLB = TLFB ×
CE
BC when describing the loader's state chart transitions in Table 3, when the
Start filling bucket hauler reaches its Capacity or when there is no Earth Ready for Loading,
9 Bucket filling completed SRL reduced by CE
the interaction between the loader and hauler agents is finalized.
Start adjusting position (A)
10A Position adjustment (A) completed UQ = CE Subsequently, the hauler starts hauling to the dumping location, where
& Truck's AS > CE TUB = TUFB it stops first at a queue of haulers waiting for a spotter to become idle.
Start dumping UQ in truck Once a spotter becomes idle, it signals the hauler to move to the
10B Position adjustment (A) completed UQ = Truck's AS dumping area. The hauler then moves and gets in the dumping position.
& Truck's AS ≤ CE TUB = TUFB ×
CE
BC At that point, the hauler starts dumping its load with the help of the
Start dumping UQ in truck spotter, and once its load is fully dumped, the spotter signals the truck
SRL increased by (CE - UQ)
to leave, ending the interaction between the two agents. The hauler
11 UQ dumped in truck SRL reduced by CE
& Truck's AS reduced by UQ then heads back to the loading area, where it stops first at a queue
Truck's AS = 0 or SRL = 0 Truck's HCE increased by UQ waiting for loaders to become idle and starts repeating the cycle again.
Start adjusting position (B) Table 4 demonstrates the transitions and states of the hauler agent's
12 Position adjustment (B) completed No action state chart. The state chart is simple, owning to the fact that the hauler
is not in control of its loading and dumping activities. As mentioned
earlier, the loader agent performs all the actions in the loading activity.
Similarly, the hauler needs the permission of the spotter before coming
Table 4 to or leaving the dumping area. However, unlike the loading activity,
Hauler's state chart transitions. the dumping activity relies solely on properties of the hauler agent. The
spotter only acts as a regulator of the dumping activity.
Transition Condition Action

1 Loader unavailable No action 4.2.4. The spotter agent

2 Loader available Start getting in loading position The spotter agent is the most basic agent in the proposed AB model.
3 Permission from loader No action Its only attribute, Time to Adjust Position, is depicted in Table 1. This
4 Permission from loader Start hauling attribute refers to the time it takes for the spotter agent to adjust its
5 Arriving at dumping queue No action
6 Spotter unavailable No action
position before and after a dumping activity to accommodate arriving
7 Spotter available Start getting in dumping position and departing haulers. The state chart of the spotter agent is demon-
8 Permission from spotter Start dumping load strated in Fig. 2(d). The initial state represents the spotter waiting for a
9 Permission from spotter HCE = 0 truck to dump its load. The spotter checks the truck's queue, and if there
DS increased by HCE
are no trucks waiting to dump, it remains idle. If, however, trucks are
Start returning
10 Arriving at loading queue No action available, the spotter agent signals the first truck in the queue to move
to the dumping area, while adjusting its position to engage with the
arriving truck. When the truck dumps its load, the spotter agent gives it
4.2.3. The hauler agent the permission to leave the dumping area and return back to be loaded
The attributes and variables of the hauler agent are demonstrated in again. After signaling the truck to depart, the spotter agent adjusts its
Table 1. Capacity refers to the maximum quantity of earth that the position again and waits for the next truck to arrive. Table 5 represents
hauler can carry. Time to Get in Load Position is the time it takes the the transitions' conditions and actions of the spotter agent's state chart.
hauler to leave the queue and arrive at the loading area. Hauling The spotter in the proposed AB model has limited interactions and re-
Duration refers to the duration of the hauler's trip from the loading area sponsibilities, which can be clearly verified by its simple transitions.
to the truck queue at the dumping site. Similarly, Returning Duration Fig. 3 summarizes the interaction between different participants, both

7
A. Jabri, T. Zayed
Table 5
Spotter's state chart transitions.
Transition Condition Action
1 Truck unavailable No action
2 Truck available Start adjusting position
3 Position adjustment completed Start spotting
4 Hauler dumping completed Start adjusting position
5 Position adjustment completed No action
agents and non-agents, in the proposed AB earthmoving model.
5. Development of ABSEMO
AnyLogic 7 was utilized in the development of ABSEMO. AnyLogic
is a Java-based modeling tool that includes libraries for DES, SD and
ABMS. Thus, users can use the graphical modeling language of Anylogic
to prepare simulation models, with the option of extending the model
with additional Java codes. The Java nature of AnyLogic allows for
custom model extensions via Java coding as well as the creation of Java
applications, which can be a basis for decision support tools [33]. These
applications can be easily shared and run on any standard browser.
The developed simulation application has some interesting aspects:
1) it is one of the first ever AB simulation tools to be developed for
planning earthmoving operations; 2) it can model different types of
equipment units performing the same activity (trucks with different
capacities and travel durations, loaders with different buckets, etc.); 3)
it accepts stochastic data for the characteristics of equipment units as
well as for activity durations; 4) it requires neither knowledge in pro-
gramming nor simulation from end-users to operate; and 5) it is a stand-
alone system that can be easily shared and run on different platforms.
5.1. Main class
The main class in the developed system represents the simulation
engine that integrates all model components, manages the interaction
between agents and their environment, performs major actions and
generates results and analysis. In view of that, the main class duties can
be summarized in five point: 1) setting up the earthmoving environ-
ment; 2) creating and governing agent populations; 3) managing agent
queues; 4) controlling model run-time conditions; and 5) performing
analysis and producing results.
5.2. Earthmoving environment
The earthmoving environment of the proposed AB model is
Environment
Bulldozer Loader
Hauling &
Soil Returning
Roads
Spotter Hauler
8
Automation in Construction xxx (xxxx) xxx–xxx
portrayed in Fig. 4. It represents the physical space in which the dif-
ferent activities involved in the earthmoving operation take place. The
earthmoving environment is created in the main class, where agents are
generated to perform their different roles. Elements of the earthmoving
environment include the excavation area, hauler load queue, loading
area, haul road, hauler dump queue, dumping area and return road.
These elements are embedded in the main class, and every time the
model is run, the main class sets up the environment according to the
configuration illustrated by Fig. 4, generating agents at their initial
locations. Locations in the environment are specified by XY coordinates.
Pointers are placed at different locations in the environment to set up
the initial location of agents and guide their movement throughout
different activities. For instance, a pointer is placed on the location
where the loading activity takes places. When the number of loaders is
specified and the model is run, the XY coordinates of that pointer is
obtained so that the loader can be placed at that pointer. Other com-
ponents of the environment including a residential area, a commercial
area and a body of water were added for aesthetic purposes.
5.3. Agent classes
Besides the main class, ABSEMO has four classes that represent the
three equipment units and the spotters participating in the earthmoving
operation. Each of these classes contains an agent's state chart, along
with elements required in the operation of that state chart.
Consequently, the same previously defined states, transitions, attributes
and variables are used in the implementation of agent classes in
ABSEMO. Some extra elements are added to the implemented system
for programming requirements. This was done using both the graphical
programming language of AnyLogic (drawing the state charts, con-
necting states with transitions, choosing equipment shapes, etc.) and
Java scripts (specifying the conditions of transitions, formulating the
actions of transitions, associating each agent with a startup location
using coordinates, etc.).
5.4. Agent queues
The functions and queues in the main class govern the hauler agent's
interaction with the loader agent when it is being loaded, and with the
spotter agent when it is dumping its load. These functions and queues
are necessary prior to the loading and dumping activities. Before trucks
get to the loading and dumping areas, they join queues. These queues
are formed by gathering and storing requests from the arriving trucks in
Java Collections (LinkedList). Then, based on the order of arrival, trucks
are dispatched to the loading or dumping area. This is known as the
First-In-First-Out (FIFO) principle, which is usually utilized in such
Fig. 3. Summary of agents' interactions in the proposed AB
earthmoving model.
Agent
Non-Agent
Interaction
A. Jabri, T. Zayed
Fig. 4. Earthmoving environment in ABSEMO.
operations. Future enhancements to ABSEMO can include adding
priorities for trucks based on their type or current state. For example,
trucks with smaller capacities can be permitted to move to the loading
or dumping areas before larger trucks that are ahead in the queue.
5.5. Graphical User Interface (Java application)
The Graphical User Interface (GUI) of the implemented earthmoving
model was developed after finalizing the structure of the main class and
the four agent classes. The main purpose of the GUI is to create a user-
friendly tool that allows for planning earthmoving operations with
flexibility in inputs to accommodate different case studies. Fig. 5(a)
demonstrates the introductory welcome page of ABSEMO, where gen-
eral information about the makers of the application is presented.
Fig. 5(b) displays the material input page of ABSEMO, where in-
formation about the material type, quantity of earth to be excavated
and quantity of excavated earth to be transported (or quantity of earth
to be excavated by loaders) are entered by users. The material type was
just added for appearance purposes related to future enhancements of
the model. On the other hand, the quantity of earth to be excavated
refers to the quantity of earth that is planned to be excavated by bull-
dozers. The quantity of excavated earth to be transported (or quantity
of earth to be excavated by loaders) is added to give users the ability to
model earthmoving systems with one of the two following scenarios: 1)
a quantity of earth is already excavated and ready to be loaded in
trucks; 2) the earthmoving operation does not include the utilization of
bulldozers, and loaders (or excavators) are the equipment units re-
sponsible for excavating and loading earth in trucks. ABSEMO accepts
and employs inputs based on both cases. Furthermore, Fig. 5(c) de-
monstrates the equipment and labor input page of ABSEMO. It is where
9
Automation in Construction xxx (xxxx) xxx–xxx
users specify the numeric value of each of the previously defined at-
tributes and variables (SRE, SRL, DS and those listed in Table 1). To
insert stochastic input for any of these attributes and variables, the user
can type U(minimum value, maximum value) for a uniform distribu-
tion, N(standard deviation, mean) for a normal distribution and
T(minimum value, mode, maximum value) for a triangular distribution.
Users can select up to three types of each equipment unit to participate
in the earthmoving operation. Having more than three types of the
same equipment unit performing the same task is not realistic. How-
ever, the model can be easily upgraded to accommodate more types of
equipment units if needed. It is important to note that having the ability
to model one to three different types of each equipment unit allows for
modeling operations where there are multiple excavation and dumping
locations. This is due to the fact that the developed model is based on
activity durations and travel times. For instance, if it is required to
model an operation where earth is dumped in two different locations,
then two types of trucks with two different Hauling Durations should be
selected to transport and dump the earth.
A color legend is added for users to identify the state of each
equipment unit during model run. Fig. 5(d) displays the model run
control elements in ABSEMO. Users can choose to stop the model when
the whole quantity of earth is excavated, loaded and dumped, when the
dumped or excavated earth reaches a specific quantity, or at a specific
time instance specified by hours, minutes and seconds of simulation
time. Fig. 6 is a snapshot of the model run in the 2-D view. While the
model is running, users can monitor major statics including the quantity
of earth available for excavation, the quantity of earth available for
loading, the quantity of dumped earth, the productivity of work, the
number of trucks in the loading queue and the number of trucks in the
dumping queue. Users can alternate between the 2-D view, the 3-D view
A. Jabri, T. Zayed
Fig. 5. ABSEMO'S (a) Welcome page; (b) Material input; (c) Equipment and labor input; (d) Model run control.
and the metrics view. Users also have the ability to run the model at a
fast speed, pause and resume the simulation, or stop and terminate the
simulation.
5.6. Reporting of results
ABSEMO reports simulation results both during model run and after
the simulation is concluded. Major statistics are constantly updated
while the model is running. Fig. 7(a) represents basic statistics on the
average percentage of simulation time in which agent units belonging
to the same class are either working or idle. Fig. 7(b) demonstrates a
detailed version of the equipment and labor utilization statistics for the
loader agent. Similar charts are also generated for the bulldozer, hauler
and spotter agents. In these detailed graphs, the average percentage of
time spent by the agent group at each state is separately represented
and given different colors. Major statistics and equipment utilization
data are added to excel sheets which were previously created and
linked to ABSEMO.
6. Model implementation to a case study
The construction of Sainte-Marguerite-3 (SM-3) Dam (1994–2002),
which is located on the Sainte-Marguerite River in Sept-Îles City,
700 km northeast of Montréal, Canada is investigated. The work in-
volved two main operations: 1) the excavation of riverbed soil at the
location of the dam; 2) the backfill of three types of soils in three stages.
The quantity of work in these two operations was estimated from the
structural design of the dam, and the hauling and returning durations of
10
Automation in Construction xxx (xxxx) xxx–xxx
trucks were calculated based on routes' profiles and rolling resistances.
As a proof of concept, only the excavation of the riverbed soil will be
modeled and simulated using ABSEMO. Information about this case
study was obtained from Peer [25], Hydro Quebec [15] and Alzraiee
[2].
6.1. Scope of work
Earthmoving operations in the SM-3 dam construction project were
allotted three years by the management. The actual quantity of ex-
cavated natural riverbed soil was 1,038,000 m3. The quantity of ex-
cavated soil was not used in the construction of the dam; instead, it was
hauled away and dumped in another location. Accordingly, the backfill
operation was performed by borrowing 6,300,000 m3 of soil from three
pits.
6.2. Fleet selection and configuration
Table 6 represent the hauler and loader fleet combinations selected
for performing the different earthmoving operations involved in the
SM-3 dam construction. Triangular probability distributions were con-
sidered for the process durations of loading, hauling, returning,
spreading and compaction. On the other hand, uniform distributions
were considered for the dumping process [2]. The last row of Table 6,
which is highlighted in bold, depicts the equipment units' capacities and
activity durations of the riverbed excavation operation.
A. Jabri, T. Zayed Automation in Construction xxx (xxxx) xxx–xxx

Fig. 6. ABSEMO's 2-D snapshot of model run.

6.3. DES results
Alzraiee [2] created an EZStrobe DES simulation model of the riv-
erbed excavation of the SM-3 dam construction project. The model
consists of two bulldozers, two loaders, seven haulers and two spotters.
The quantity of soil to be excavated is 1,038,000 m3. Also, equipment
properties and activity durations are the ones listed in the last row of
Table 6. The results of the DES model suggested a duration of 808.33 h
with an average productivity of 1284.52 m3/h to complete the opera-
tion. These results were reported by Alzraiee [2] and verified by the
authors.
6.4. ABSEMO results
Activity durations presented in Table 6 are all of DES nature. Ver-
ifying ABSEMO using this data requires some manipulation. This is

Fig. 7. ABSEMO's (a) general; (b) detailed (loaders) equipment and labor utilization statistics.

11
A. Jabri, T. Zayed Automation in Construction xxx (xxxx) xxx–xxx

Table 6
Hauler and loader fleet configurations in the SM-3 dam construction earthmoving operations.

Hauled material Bucket capacity of Hauled soil Loading process-time Hauling process-time Dumping process-time Returning process-time
loader (m3) volume (m3) distribution (min) distribution (min) distribution (min) distribution (min)

Rock 12.3 49 (3.94, 4.15, 4.57) (4.3, 4.53, 4.98) (1.9, 2.2) (3.17, 3.34, 3.67)
Moraine 9.2 28 (3.01, 3.2, 3.32) (19.47, 20.5, 22.55) (1.6, 1.9) (16.71, 17.59, 19.35)
Granular 6.9 20 (2.3, 2.42, 2.5) (30.6, 32.34, 35.57) (1.3, 1.5) (25.85, 26.51, 29.16)
Riverbed soil 4.59 32 (4.26, 4.48, 4.93) (5.32, 5.6, 6.16) (1.6, 1.9) (2.86, 3.01, 3.31)

because the proposed model breaks activities into sequences of smaller
activities, each with a duration. It is important to note that ABSEMO
accepts more complex data input than what is available in this case
study. To fully utilize the potential of ABSEMO, detailed data input
demonstrated in Fig. 5(c) should be used. However, since such data is
unavailable to the authors, the case study is used as a verification of
ABSEMO's flow of resources and productivity calculations. Conse-
quently, some inputs of the riverbed excavation process used in AB-
SEMO are adjusted based on the following assumptions:
1. Setting the Push Duration in ABSEMO equal to the excavation
duration of the case study. Other durations in the bulldozer agent
inputs have no value (0 min).
2. By dividing the hauler's Capacity over the loader's Bucket Capacity,
the number of loader cycles to fill one truck is obtained (49 m3/
4.59 m3 = 10.675). The loading duration is then divided by that
number to get the duration of the loader cycle ((4.260 min,
4.480 min, 4.930 min) / 10.675 = (0.399 min, 0.420 min,
0.462 min)). Finally, as an assumption, 75% of the duration of the
loader cycle (0.299 min, 0.314743 min, 0.346 min) is inputted as
the Time to Load Full Bucket and 25% (0.100 min, 0.105 min,
0.115 min) is inputted as the Time to Unload Full Bucket in ABSEMO.
Other durations in the loader agent inputs have no value (0 min).
3. Setting the Hauling Duration, Dumping Duration and Returning
Duration in ABSEMO as the haul, dump and return durations of the
case study. Other durations in the hauler agent inputs have no value
(0 min).
4. The Time to Adjust Position input of the spotter agent has no value
(0 min).
After inputting the Earth Ready for Excavation, the equipment ca-
pacities and the adjusted activity durations of the riverbed excavation
in ABSEMO, a duration of 804.95 h with an average productivity of
1289.52 m3/h was obtained at the completion of the operation.
6.5. Comparison between DES and ABMS results
input procedure. This is why the ABMS model behaved like a DES
model when the simulation was run. The goal of this comparison was to
verify ABSEMO's logic rather than validate its accuracy compared to
real life operations. To summarize, the full capabilities of ABSEMO
were not utilized in that case study due to the absence of the real life
project and equipment data. Forcing an AB model to function similar to
a DES model may not outline the strength of ABMS, but it can show
whether the ABMS model would function properly or not when re-
ceiving more complex data. Therefore, it is concluded that ABSEMO can
now be used to plan earthmoving operations with more complex input
that utilizes its capabilities.
6.6. Overcoming stated limitations
To demonstrate how the implemented AB approach and ABSEMO
overcome the previously stated limitations, a simple theoretical case
study is presented. The information in this case study is not necessarily
accurate; it is merely a proof of concept. 100,000 m3 of earth is required
for a certain stage of a dam construction project. It was planned that
this quantity is to be excavated from a certain location using bulldozers.
Then, it is to be loaded by frontend loaders into trucks and transported
for dumping in two different areas at the dam location. Two types of
bulldozers (B1 and B2), three types of font end loaders (L1, L2 and L3),
and three types of trucks (T1, T2 and T3) are available to the contractor
to undertake the earthmoving operation. Due to some access constraints
at the dumping site, one of the two dumping locations can only be
accessed by the small trucks (T1 and T2). T3 will dump in the second
dumping location. After studying the site layout and equipment prop-
erties, the contractor estimated the attributes of the equipment units
and spotters to be as demonstrated by Table 7.
To illustrate the superiority of the AB approach and ABSEMO over
conventional DES models in modeling and simulating this earthmoving
operation, it is vital to think about how such an operation can be si-
mulated using DES. The two following points summarize problems that
arise when trying to model and simulate such an example using DES: 1)
cycle times and resources: DES models are interacting duration-based

DES results represent a good verification tool for AB models. The Table 7
nature of DES provides an accurate flow of resources, which allows for Agents' attributes in the theoretical earthmoving case study.
verifying the quantitative aspects of AB models. There are several ex-
Agent Type 1 Type 2 Type 3
amples in literature on the verification of ABMS outputs using DES
results [6,10]. The DES and ABMS durations were 808.33 h and Bulldozers PQ = 5 m 3
PQ = 8 m 3
–
804.95 h with productivities of 1284.52 and 1289.52 m3/h respec- PD = 3 min PD = 5 min
TAP = 1.5 min TAP = 1.75 min
tively. The percentage difference between the DES and ABMS durations
RD = 2.5 min RD = 3 min
was 0.42%, indicating that the representation of agents' properties, the Loaders BC = 1.5 m3 BC = 2.5 m3 BC = 3.5 m3
interaction logic and the flow of resources in ABSEMO are correct. TLFB = 1 min TLFB = 1 min TLFB = 1.5 min
ABMS is superior to DES in many aspects including the graphical TAPF = 0.75 min TAPF = 0.75 min TAPF = 1 min
modeling support, quantity monitoring and tracking, having different TUFB = 0.5 min TUFB = 0.5 min TUFB = 0.75 min
TAPE = 0.25 min TAPE = 0.25 min TAPE = 0.75 min
types of agents performing the same tasks (e.g. trucks of different ca- Haulers C = 28 m3 C = 32 m3 C = 45 m3
pacities), productivity tracking and accommodating complex interac- TGLP = 0.5 min TGLP = 0.5 min TGLP = 0.75 min
tions and accommodating scenarios where one or more agents are ab- HD = 42 min HD = 46 min HD = 63 min
sent (e.g. loaders excavating and loading trucks instead of having TGDP = 0.5 min TGDP = 0.5 min TGDP = 0.75 min
DD = 1 min DD = 1 min DD = 1.5 min
bulldozers). That being mentioned, the reason both techniques yielded
RD = 30 min RD = 32 min RD = 45 min
the same results was because the ABMS model was run using DES data. Spotters S-TAP = 0.2 min – –
And in some cases, the data had to be manipulated to fit ABSEMO's data

12
A. Jabri, T. Zayed Automation in Construction xxx (xxxx) xxx–xxx

cycles of entities. The attributes of the entities participating in the op- Table 8
eration do not directly contribute to the model. They are only used to Agent combinations and their corresponding durations and productivities in the theore-
tical earthmoving case study.
calculate cycle times which are the input of the DES model. So, the
interaction between any two or more entities must be governed by a Combination Type Number of Duration (h) Productivity (m3/h)
cycle time. For instance, a duration has to be calculated for the inter-
action between the bulldozer and the loader. This duration is basically B L T
the time required to prepare one loader load. Then, another duration
1 1 1 1 1 2033.58 49.20
has to be calculated for the interaction between the loader and the 2 1 1 1
truck, which in turn is the time required to load one loader load in the 2 1 2 1 1 2033.33 49.2
truck. One of the drawbacks of this technique is the difficulty in cal- 2 1 1 1
culating cycle times. Usually, contractors use previous data of the same 3 1 1 1 1 2033.60 49.2
2 2 1 1
equipment units to estimate cycle times. For instance, the cycle time of
4 1 2 1 1 2031.61 49.2
loader L1 to load one load in truck T1 is known from previous projects to 2 2 1 1
average 3.85 min. Changing the type of one of the two interacting 5 1 1 1 1 1530.27 65.34
equipment units requires a new estimation of cycle times, which has to 2 1 1 2
6 1 2 1 1 1530.39 65.34
be obtained from the field or calculated using equipment properties.
2 1 1 2
Another drawback of this technique is resource handling. After the first 7 1 1 1 1 1529.51 65.34
cycle, in which the bulldozer interacts with the loader, the produced 2 2 1 2
resource is loader loads. In the second cycle, trucks are loaded with 8 1 2 1 1 1529.15 65.4
several loader loads until they are full and ready to haul. However, in 2 2 1 2
9 1 1 2 1 2034.17 49.14
the presented example, assuming only loaders L1 and trucks T1 exist,
2 1 1 1
T1's capacity is not a multiple of L1's bucket capacity (8/3 = 5.33). 10 1 2 2 1 2033.45 49.2
When having such a scenario in a DES model, there are usually different 2 1 1 1
alternatives to go about it. One option is to round 5.33 loads up to 6 11 1 1 2 1 2033.74 49.2
2 2 1 1
loads, which overestimates the loading duration. The larger the quan-
12 1 2 2 1 2032.67 49.2
tity of handled earth the larger the effect of this overestimation be- 2 2 1 1
comes, leading to significant forecasting errors. Another option is to use 13 1 1 2 1 1530.84 65.34
truckloads instead of loader loads, especially if the cycle time to fill the 2 1 1 2
same type truck with the same type of loader is available. However, the 14 1 2 2 1 1530.42 65.34
2 1 1 2
modeler has to go back to the bulldozer-loader interaction and change
15 1 1 2 1 1529.18 65.34
the produced resource to truckloads as well, because only one resource 2 2 1 2
can exist in a DES model; 2) Agents with different properties performing 16 1 2 2 1 1529.41 65.40
the same task: In DES models, only one type of entities can be re- 2 2 1 2
presented at each queue where an activity takes place. Having two
types of bulldozers excavating the same area, three types of loaders
further prove that, combination #13 is inputted in ABSEMO with an
loading trucks, three types of trucks carrying and hauling earth to be
additional T1 and T3 trucks (one B1, one B2, two L1, one L2, one L3, one
dumped and having two dumping locations, like in this example, is not
T1, two T2 and three T3). The duration for this combination is 1300.18 h
possible. When similar scenarios exist, entities with cycle times equal to
and the productivity is 76.92 m3/h. This example demonstrates how
the weighted average of all the other entities' cycle times are used. For
ABSEMO can be used to plan earthmoving operations. It shows how the
instance, if three T1 trucks, two T2 trucks and two T3 trucks are used,
AB approach of ABSEMO is able to capture a realistic view of the op-
the DES hauling duration becomes (3 × 42 min + 2 × 46 min
eration, without making any unnecessary assumptions that sacrifice the
+ 2 × 63 min) / (3 + 2 + 2) = 49.14 min. This introduces significant
accuracy of results.
inaccuracies to results. Another solution is to break down the DES
model into multiple modules, each having one type of equipment units,
which is time-consuming, error-prone and requires compiling the re- 7. Conclusion
sults of multiple modules.
ABSEMO can model and simulate this example in a very straight- Numerous efforts have been made in planning construction opera-
forward manner. Not only can ABSEMO accommodate all details of the tions through DES. Limitations of DES earthmoving models include
operation including the different types of equipment units, it also keeps primitive methods for calculating activity durations, the inability to
track of the status of every single agent throughout the process. To accommodate equipment units with different specifications performing
demonstrate how ABSMEO can be used for planning such an operation, the same task and ignoring unique scenarios of resource constraints.
an example with a single equipment unit and a single spotter from each The development of a comprehensive AB model of earthmoving op-
type will be simulated. Then a sensitivity analysis will be performed by erations representing bulldozers, loaders, haulers and spotters is de-
varying the number of each type of every agent. To simplify the process monstrated in this paper. Each agent in the proposed model is a smart,
and minimize the results, it is assumed that the contractor uses two adaptive entity which is assigned a set of static and dynamic properties
spotters and one unit of each equipment type, except for bulldozers B1 (attributes and variables) to govern its interactions with the environ-
and B2, loader L1 and truck T2. The contractor can choose one to two ment and with other agents. In addition, each agent has a state chart
agents of both types of bulldozers (B1 and B2) and one to two agents of that acts as a guide for the agent in fulfilling its roles in the operation.
loaders L1 and trucks T2. This makes up 16 combinations (24) which are All interactions are governed by the dynamic exchange of information
individually inputted and run in ABSEMO, as demonstrated by Table 8. on properties and states of agents. This allows for modeling equipment
To briefly analyze the results, it is clear that trucks (haulers) are the units with different specifications performing the same task (e.g. trucks
controlling agent. Adding an additional T2 decreases the duration by of different capacities), accurately calculating activity durations and
approximately 500 h, and increases the productivity by approximately transported earth quantities, as well as accounting for special opera-
16 m3/h. It is also important to notice that adding an additional L2 tional scenarios of resource constraints. A Java-based object-oriented
loader does not enhance productivity. So, earth is being excavated and software application, Agent-Based Simulator for Earthmoving
can be loaded at a much faster rate than it is being transported. To Operations' (ABSEMO), is developed as an implementation of the

13
A. Jabri, T. Zayed
proposed AB model. Contractors will be able to use ABSEMO to plan
earthmoving operations according to the AB approach. ABSEMO is
verified through a real-life case study of the riverbed excavation op-
eration in the SM-3 dam construction project. The available data is
manipulated to fit model inputs and the simulation results are com-
pared with those obtained from a DES model of the same operation. A
percentage difference of 0.42% from the DES results is obtained, ver-
ifying that the model's logic and flow of resources are accurate. A
theoretical case study was also simulated to highlight the strength of
ABMS and demonstrate how ABSEMO can be used for planning earth-
moving operations. The case study showed how ABSEMO was able to
accommodate all details of the operation without the need to make any
assumptions or simplifications. Varying equipment unit combinations
changed the duration and productivity of the operation, highlighting
equipment units that control the process and therefore need to be
carefully studied by the contractor. Creating flexible agents with
adaptive behavior and assigning their properties from an individual AB
prospective allow for capturing a realistic behavior of earthmoving
operations. Hence, the proposed methodology can be further developed
and extended to a more comprehensive level. The strength of ABMS in
building simulation models bottom-up (agent to model) allows for
capturing more complex scenarios in earthmoving operations, and
hence allows for producing more accurate results. Examples of future
development areas include accounting for: Equipment breakdown
probabilities; weather conditions and their effect on transportation
roads and equipment units; fatigue of equipment unit operators and its
effect on the attributes of the equipment unit agents (e.g. longer
durations); and experience of agents in working with other agents (e.g.
truck and loader operators who have experience working together).
References
[1] S. Ahn, S. Lee, R.P. Steel, Effects of workers' social learning: focusing on absence
behavior, J. Constr. Eng. Manag. 139 (8) (2013) 1015–1025, http://dx.doi.org/10.
1061/(ASCE)CO.1943-7862.0000680.
[2] H.S. Alzraiee, Hybrid Simulation for Construction Operations (Ph.D. Dissertation),
Building, Civil and Environmental Engineering Department, Concordia University,
Montreal, QC, Canada, 2013.
[3] H. Alzraiee, O. Moselhi, T. Zayed, Dynamic planning of earthmoving projects using
system dynamics, Gerontechnology 11 (2) (2012) 316, http://dx.doi.org/10.4017/
gt.2012.11.02.191.00.
[5] S. Asgari, R. Awwad, A. Kandil, I. Odeh, Impact of considering need for work and
risk on performance of construction contractors: an agent-based approach, Autom.
Constr. 65 (2016) 9–20, http://dx.doi.org/10.1016/j.autcon.2016.01.004.
[6] S. Biswas, S. Merchawi, Use of discrete event simulation to validate an agent based
scheduling engine, Proceedings of the 32nd Conference on Winter Simulation, SCS,
Orlando, FL, USA, 2000, pp. 1778–1782, , http://dx.doi.org/10.1109/WSC.2000.
899169.
[8] R. Dzeng, Y. Lin, Intelligent agents for supporting construction procurement ne-
gotiation, Expert Syst. Appl. 27 (1) (2004) 107–119, http://dx.doi.org/10.1016/j.
eswa.2003.12.006.
[9] I.H. El-Adaway, A.A. Kandil, Multiagent system for construction dispute resolution
(MAS-COR), J. Constr. Eng. Manag. 136 (3) (2009) 303–315, http://dx.doi.org/10.
1061/(ASCE)CO.1943-7862.0000144.
[10] G. Fortino, A. Garro, W. Russo, A discrete-event simulation framework for the
validation of agent-based and multi-agent systems, Annual Western Orthopaedic
Association Meeting, Camerino, MC, Italy, 2005, pp. 75–84 http://doi.org/
10.1.1.61.572.
[11] N. Gilbert, K. Troitzsch, Simulation for the Social Scientist, McGraw-Hill
International, Maidenhead, Berkshire, 2005 (ISBN 0335216005).
[12] Y.M. Goh, M.J.A. Ali, A hybrid simulation approach for integrating safety behavior
into construction planning: an earthmoving case study, Accid. Anal. Prev. 93 (2016)
310–318, http://dx.doi.org/10.1016/j.aap.2015.09.015.
14
Automation in Construction xxx (xxxx) xxx–xxx
[13] D. Hajjar, S. AbouRizk, AP2-Earth: a simulation based system for the estimating and
planning of earth moving operations, Proceedings of the 29th Conference on Winter
Simulation, IEEE, Atlanta, GA, USA, 1997, pp. 1103–1110, , http://dx.doi.org/10.
1145/268437.268746.
[15] Hydro Quebec, Construction of the Sainte-Marguerite-3 Hydroelectric Development
1994–2002, Bibliotheque Nationale du Quebéc, 2003 (ISBN 2550402979).
[16] J.C. Martinez, Earthmover-simulation tool for earthwork planning, Winter
Simulation Conference Proceedings, ACM, Washington DC, USA, 1998, pp.
1263–1271, , http://dx.doi.org/10.1109/WSC.1998.745988.
[17] M. Marzouk, O. Moselhi, Object-oriented simulation model for earthmoving op-
erations, J. Constr. Eng. Manag. 129 (2) (2003) 173–181, http://dx.doi.org/10.
1061/(ASCE)0733-9364(2003)129:2(173).
[18] M. Marzouk, H. Ali, Modeling safety considerations and space limitations in piling
operations using agent based simulation, Expert Syst. Appl. 40 (12) (2013)
4848–4857, http://dx.doi.org/10.1016/j.eswa.2013.02.021.
[19] J.U. Min, H.C. Bjornsson, Agent-based construction supply chain simulator (CS 2)
for measuring the value of real-time information sharing in construction, J. Constr.
Eng. Manag. 24 (4) (2008) 245–254, http://dx.doi.org/10.1061/(ASCE)0742-597X
(2008)24:4(245).
[20] O. Moselhi, A. Alshibani, Optimization of earthmoving operations in heavy civil
engineering projects, J. Constr. Eng. Manag. 135 (10) (2009) 948–954, http://dx.
doi.org/10.1061/(ASCE)0733-9364(2009)135:10(948).
[21] M.J. North, C.M. Macal, Managing Business Complexity: Discovering Strategic
Solutions with Agent-based Modeling and Simulation, Oxford University Press,
Oxford, UK, 2007.
[22] A.A. Oloufa, Modeling operational activities in object-oriented simulation, J.
Constr. Eng. Manag. 7 (1) (1993) 94–106, http://dx.doi.org/10.1061/(ASCE)0887-
3801(1993)7:1(94) (ISBN 9780195172119).
[23] H. Osman, Agent-based simulation of urban infrastructure asset management ac-
tivities, Autom. Constr. 28 (2012) 45–57, http://dx.doi.org/10.1016/j.autcon.
2012.06.004.
[24] S. Palaniappan, A. Sawhney, M.A. Janssen, K.D. Walsh, Modeling construction
safety as an agent-based emergent phenomenon, The 24th International Symposium
on Automation and Robotics in Construction, Indian Institute of Technology,
Madras, India, 2007.
[25] G. Peer, Powerhouse takes shape far from dam SM-3 project, Heavy Construction
News, 2001, pp. 14–16.
[26] Z. Ren, C.J. Anumba, Learning in multi-agent systems: a case study of construction
claims negotiation, Adv. Eng. Inform. 16 (4) (2002) 265–275, http://dx.doi.org/10.
1016/S1474-0346(03)00015-6.
[27] K. Sanford Bernhardt, S. McNeil, Agent-based modeling: approach for improving
infrastructure management, J. Infrastruct. Syst. 14 (3) (2008) 253–261, http://dx.
doi.org/10.1061/(ASCE)1076-0342(2008)14:3(253).
[28] J. Seo, S. Lee, J. Seo, Simulation-based assessment of workers' muscle fatigue and its
impact on construction operations, J. Constr. Eng. Manag. 142 (11) (2016), http://
dx.doi.org/10.1061/(ASCE)CO.1943-7862.0001182.
[29] J.H. Tah, Towards an agent-based construction supply network modelling and si-
mulation platform, Autom. Constr. 14 (3) (2005) 353–359, http://dx.doi.org/10.
1016/j.autcon.2004.08.003.
[30] O. Tatari, M. Skibniewski, Integrated agent-based construction equipment man-
agement: conceptual design, J. Civ. Eng. Manag. 12 (3) (2006) 231–236, http://dx.
doi.org/10.1080/13923730.2006.9636397.
[31] A. Touran, Integration of simulation with expert systems, J. Constr. Eng. Manag.
116 (3) (1990) 480–493, http://dx.doi.org/10.1061/(ASCE)0733-9364(1990)
116:3(480).
[32] F. Vahdatikhaki, A. Hammad, S.M. Langari, Multi-agent System for Improved Safety
and Productivity of Earthwork Equipment Using Real-time Location Systems,
International Construction Specialty Conference, CSCE, Vancouver, B.C., Canada,
2015, http://dx.doi.org/10.14288/1.0076351.
[33] C. Wartha, M. Peev, A. Borshchev, A. Filippov, Manufacturing supply chain appli-
cations 1: decision support tool-supply chain, Proceedings of the 34th Conference
on Winter Simulation: Exploring New Frontiers, WSC, San Diego, CA, USA, 2002,
pp. 1297–1301, , http://dx.doi.org/10.1109/WSC.2002.1166391.
[34] E. Zankoul, H. Khoury, R. Awwad, Evaluation of agent-based and discrete-event
simulation for modeling construction earthmoving operations, 32nd International
Symposium on Automation and Robotics in Construction and Mining, International
Association for Automation and Robotics in Construction (IAARC), Vilnius,
Lithuania, 2015.
[35] C. Zhang, A. Hammad, H. Bahnassi, Collaborative multi-agent systems for con-
struction equipment based on real-time field data capturing, J. Inf. Technol. Constr.
14 (2009) 204–228 (Special Issues on Next Generation Construction IT: Technology
Foresight, Future Studies, Roadmapping, and Scenario Planning).